Add custom apply patch toggle

This PR addresses an edge-case bug that appears in the VS Code extension
in the following situation:
1. Log in using ChatGPT (using either the CLI or extension). This will
create an `auth.json` file.
2. Manually modify `config.toml` to specify a custom provider.
3. Start a fresh copy of the VS Code extension.

The profile menu in the VS Code extension will indicate that you are
logged in using ChatGPT even though you're not.

This is caused by the `get_auth_status` method returning an
`auth_method: 'chatgpt'` when a custom provider is configured and it
doesn't use OpenAI auth (i.e. `requires_openai_auth` is false). The
method should always return `auth_method: None` if
`requires_openai_auth` is false.

The same bug also causes the NUX (new user experience) screen to be
displayed in the VSCE in this situation.

.github/workflows/codespell.yml

@@ -25,3 +25,4 @@ jobs:
         uses: codespell-project/actions-codespell@406322ec52dd7b488e48c1c4b82e2a8b3a1bf630 # v2
         with:
           ignore_words_file: .codespellignore
+          skip: frame*.txt

.github/workflows/rust-ci.yml

@@ -62,6 +62,8 @@ jobs:
           components: rustfmt
       - name: cargo fmt
         run: cargo fmt -- --config imports_granularity=Item --check
+      - name: Verify codegen for mcp-types
+        run: ./mcp-types/check_lib_rs.py
 
   cargo_shear:
     name: cargo shear

.github/workflows/rust-release.yml

@@ -219,3 +219,22 @@ jobs:
         with:
           tag: ${{ github.ref_name }}
           config: .github/dotslash-config.json
+
+  update-branch:
+    name: Update latest-alpha-cli branch
+    permissions:
+      contents: write
+    needs: release
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Update latest-alpha-cli branch
+        env:
+          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          set -euo pipefail
+          gh api \
+            repos/${GITHUB_REPOSITORY}/git/refs/heads/latest-alpha-cli \
+            -X PATCH \
+            -f sha="${GITHUB_SHA}" \
+            -F force=true

README.md

@@ -2,7 +2,10 @@
 
 <p align="center"><code>npm i -g @openai/codex</code><br />or <code>brew install codex</code></p>
 
-<p align="center"><strong>Codex CLI</strong> is a coding agent from OpenAI that runs locally on your computer.</br>If you are looking for the <em>cloud-based agent</em> from OpenAI, <strong>Codex Web</strong>, see <a href="https://chatgpt.com/codex">chatgpt.com/codex</a>.</p>
+<p align="center"><strong>Codex CLI</strong> is a coding agent from OpenAI that runs locally on your computer.
+</br>
+</br>If you want Codex in your code editor (VS Code, Cursor, Windsurf), <a href="https://developers.openai.com/codex/ide">install in your IDE</a>
+</br>If you are looking for the <em>cloud-based agent</em> from OpenAI, <strong>Codex Web</strong>, go to <a href="https://chatgpt.com/codex">chatgpt.com/codex</a></p>
 
 <p align="center">
   <img src="./.github/codex-cli-splash.png" alt="Codex CLI splash" width="80%" />

codex-rs/Cargo.lock

@@ -212,6 +212,50 @@ dependencies = [
  "term",
 ]
 
+[[package]]
+name = "askama"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b79091df18a97caea757e28cd2d5fda49c6cd4bd01ddffd7ff01ace0c0ad2c28"
+dependencies = [
+ "askama_derive",
+ "askama_escape",
+ "humansize",
+ "num-traits",
+ "percent-encoding",
+]
+
+[[package]]
+name = "askama_derive"
+version = "0.12.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "19fe8d6cb13c4714962c072ea496f3392015f0989b1a2847bb4b2d9effd71d83"
+dependencies = [
+ "askama_parser",
+ "basic-toml",
+ "mime",
+ "mime_guess",
+ "proc-macro2",
+ "quote",
+ "serde",
+ "syn 2.0.104",
+]
+
+[[package]]
+name = "askama_escape"
+version = "0.10.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "619743e34b5ba4e9703bba34deac3427c72507c7159f5fd030aea8cac0cfe341"
+
+[[package]]
+name = "askama_parser"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "acb1161c6b64d1c3d83108213c2a2533a342ac225aabd0bda218278c2ddb00c0"
+dependencies = [
+ "nom",
+]
+
 [[package]]
 name = "assert-json-diff"
 version = "2.0.2"
@@ -305,6 +349,15 @@ version = "0.22.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "72b3254f16251a8381aa12e40e3c4d2f0199f8c6508fbecb9d91f575e0fbb8c6"
 
+[[package]]
+name = "basic-toml"
+version = "0.1.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba62675e8242a4c4e806d12f11d136e626e6c8361d6b829310732241652a178a"
+dependencies = [
+ "serde",
+]
+
 [[package]]
 name = "beef"
 version = "0.5.2"
@@ -561,7 +614,6 @@ dependencies = [
  "clap",
  "codex-common",
  "codex-core",
- "codex-protocol",
  "serde",
  "serde_json",
  "tempfile",
@@ -607,12 +659,14 @@ name = "codex-core"
 version = "0.0.0"
 dependencies = [
  "anyhow",
+ "askama",
  "assert_cmd",
  "async-channel",
  "base64",
  "bytes",
  "chrono",
  "codex-apply-patch",
+ "codex-file-search",
  "codex-mcp-client",
  "codex-protocol",
  "core_test_support",
@@ -680,6 +734,8 @@ dependencies = [
  "tokio",
  "tracing",
  "tracing-subscriber",
+ "uuid",
+ "walkdir",
  "wiremock",
 ]
 
@@ -769,6 +825,7 @@ version = "0.0.0"
 dependencies = [
  "anyhow",
  "assert_cmd",
+ "base64",
  "codex-arg0",
  "codex-common",
  "codex-core",
@@ -1981,6 +2038,15 @@ version = "1.0.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "df3b46402a9d5adb4c86a0cf463f42e19994e3ee891101b1841f30a545cb49a9"
 
+[[package]]
+name = "humansize"
+version = "2.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6cb51c9a029ddc91b07a787f1d86b53ccfa49b0e86688c946ebe8d3555685dd7"
+dependencies = [
+ "libm",
+]
+
 [[package]]
 name = "hyper"
 version = "1.7.0"
@@ -2536,6 +2602,12 @@ version = "0.2.175"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6a82ae493e598baaea5209805c49bbf2ea7de956d50d7da0da1164f9c6d28543"
 
+[[package]]
+name = "libm"
+version = "0.2.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f9fbbcab51052fe104eb5e5d351cf728d30a5be1fe14d9be8a3b097481fb97de"
+
 [[package]]
 name = "libredox"
 version = "0.1.6"

codex-rs/Cargo.toml

@@ -34,6 +34,7 @@ rust = {}
 
 [workspace.lints.clippy]
 expect_used = "deny"
+redundant_clone = "deny"
 uninlined_format_args = "deny"
 unwrap_used = "deny"
 

codex-rs/apply-patch/src/lib.rs

@@ -40,6 +40,11 @@ pub enum ApplyPatchError {
     /// Error that occurs while computing replacements when applying patch chunks
     #[error("{0}")]
     ComputeReplacements(String),
+    /// A raw patch body was provided without an explicit `apply_patch` invocation.
+    #[error(
+        "patch detected without explicit call to apply_patch. Rerun as [\"apply_patch\", \"<patch>\"]"
+    )]
+    ImplicitInvocation,
 }
 
 impl From<std::io::Error> for ApplyPatchError {
@@ -93,10 +98,12 @@ pub struct ApplyPatchArgs {
 
 pub fn maybe_parse_apply_patch(argv: &[String]) -> MaybeApplyPatch {
     match argv {
+        // Direct invocation: apply_patch <patch>
         [cmd, body] if APPLY_PATCH_COMMANDS.contains(&cmd.as_str()) => match parse_patch(body) {
             Ok(source) => MaybeApplyPatch::Body(source),
             Err(e) => MaybeApplyPatch::PatchParseError(e),
         },
+        // Bash heredoc form: (optional `cd <path> &&`) apply_patch <<'EOF' ...
         [bash, flag, script] if bash == "bash" && flag == "-lc" => {
             match extract_apply_patch_from_bash(script) {
                 Ok((body, workdir)) => match parse_patch(&body) {
@@ -207,6 +214,26 @@ impl ApplyPatchAction {
 /// cwd must be an absolute path so that we can resolve relative paths in the
 /// patch.
 pub fn maybe_parse_apply_patch_verified(argv: &[String], cwd: &Path) -> MaybeApplyPatchVerified {
+    // Detect a raw patch body passed directly as the command or as the body of a bash -lc
+    // script. In these cases, report an explicit error rather than applying the patch.
+    match argv {
+        [body] => {
+            if parse_patch(body).is_ok() {
+                return MaybeApplyPatchVerified::CorrectnessError(
+                    ApplyPatchError::ImplicitInvocation,
+                );
+            }
+        }
+        [bash, flag, script] if bash == "bash" && flag == "-lc" => {
+            if parse_patch(script).is_ok() {
+                return MaybeApplyPatchVerified::CorrectnessError(
+                    ApplyPatchError::ImplicitInvocation,
+                );
+            }
+        }
+        _ => {}
+    }
+
     match maybe_parse_apply_patch(argv) {
         MaybeApplyPatch::Body(ApplyPatchArgs {
             patch,
@@ -733,6 +760,8 @@ fn compute_replacements(
         }
     }
 
+    replacements.sort_by(|(lhs_idx, _, _), (rhs_idx, _, _)| lhs_idx.cmp(rhs_idx));
+
     Ok(replacements)
 }
 
@@ -873,6 +902,28 @@ mod tests {
         ));
     }
 
+    #[test]
+    fn test_implicit_patch_single_arg_is_error() {
+        let patch = "*** Begin Patch\n*** Add File: foo\n+hi\n*** End Patch".to_string();
+        let args = vec![patch];
+        let dir = tempdir().unwrap();
+        assert!(matches!(
+            maybe_parse_apply_patch_verified(&args, dir.path()),
+            MaybeApplyPatchVerified::CorrectnessError(ApplyPatchError::ImplicitInvocation)
+        ));
+    }
+
+    #[test]
+    fn test_implicit_patch_bash_script_is_error() {
+        let script = "*** Begin Patch\n*** Add File: foo\n+hi\n*** End Patch";
+        let args = args_bash(script);
+        let dir = tempdir().unwrap();
+        assert!(matches!(
+            maybe_parse_apply_patch_verified(&args, dir.path()),
+            MaybeApplyPatchVerified::CorrectnessError(ApplyPatchError::ImplicitInvocation)
+        ));
+    }
+
     #[test]
     fn test_literal() {
         let args = strs_to_strings(&[
@@ -1216,6 +1267,33 @@ PATCH"#,
         assert_eq!(contents, "a\nB\nc\nd\nE\nf\ng\n");
     }
 
+    #[test]
+    fn test_pure_addition_chunk_followed_by_removal() {
+        let dir = tempdir().unwrap();
+        let path = dir.path().join("panic.txt");
+        fs::write(&path, "line1\nline2\nline3\n").unwrap();
+        let patch = wrap_patch(&format!(
+            r#"*** Update File: {}
+@@
++after-context
++second-line
+@@
+ line1
+-line2
+-line3
++line2-replacement"#,
+            path.display()
+        ));
+        let mut stdout = Vec::new();
+        let mut stderr = Vec::new();
+        apply_patch(&patch, &mut stdout, &mut stderr).unwrap();
+        let contents = fs::read_to_string(path).unwrap();
+        assert_eq!(
+            contents,
+            "line1\nline2-replacement\nafter-context\nsecond-line\n"
+        );
+    }
+
     /// Ensure that patches authored with ASCII characters can update lines that
     /// contain typographic Unicode punctuation (e.g. EN DASH, NON-BREAKING
     /// HYPHEN). Historically `git apply` succeeds in such scenarios but our

codex-rs/apply-patch/src/parser.rs

@@ -617,7 +617,7 @@ fn test_parse_patch_lenient() {
     assert_eq!(
         parse_patch_text(&patch_text_in_double_quoted_heredoc, ParseMode::Lenient),
         Ok(ApplyPatchArgs {
-            hunks: expected_patch.clone(),
+            hunks: expected_patch,
             patch: patch_text.to_string(),
             workdir: None,
         })
@@ -637,7 +637,7 @@ fn test_parse_patch_lenient() {
         "<<EOF\n*** Begin Patch\n*** Update File: file2.py\nEOF\n".to_string();
     assert_eq!(
         parse_patch_text(&patch_text_with_missing_closing_heredoc, ParseMode::Strict),
-        Err(expected_error.clone())
+        Err(expected_error)
     );
     assert_eq!(
         parse_patch_text(&patch_text_with_missing_closing_heredoc, ParseMode::Lenient),

codex-rs/chatgpt/Cargo.toml

@@ -11,7 +11,6 @@ anyhow = "1"
 clap = { version = "4", features = ["derive"] }
 codex-common = { path = "../common", features = ["cli"] }
 codex-core = { path = "../core" }
-codex-protocol = { path = "../protocol" }
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 tokio = { version = "1", features = ["full"] }

codex-rs/chatgpt/src/chatgpt_token.rs

@@ -1,5 +1,4 @@
 use codex_core::CodexAuth;
-use codex_protocol::mcp_protocol::AuthMode;
 use std::path::Path;
 use std::sync::LazyLock;
 use std::sync::RwLock;
@@ -20,7 +19,7 @@ pub fn set_chatgpt_token_data(value: TokenData) {
 
 /// Initialize the ChatGPT token from auth.json file
 pub async fn init_chatgpt_token_from_auth(codex_home: &Path) -> std::io::Result<()> {
-    let auth = CodexAuth::from_codex_home(codex_home, AuthMode::ChatGPT)?;
+    let auth = CodexAuth::from_codex_home(codex_home)?;
     if let Some(auth) = auth {
         let token_data = auth.get_token_data().await?;
         set_chatgpt_token_data(token_data);

codex-rs/cli/src/login.rs

@@ -1,7 +1,6 @@
 use codex_common::CliConfigOverrides;
 use codex_core::CodexAuth;
 use codex_core::auth::CLIENT_ID;
-use codex_core::auth::OPENAI_API_KEY_ENV_VAR;
 use codex_core::auth::login_with_api_key;
 use codex_core::auth::logout;
 use codex_core::config::Config;
@@ -9,7 +8,6 @@ use codex_core::config::ConfigOverrides;
 use codex_login::ServerOptions;
 use codex_login::run_login_server;
 use codex_protocol::mcp_protocol::AuthMode;
-use std::env;
 use std::path::PathBuf;
 
 pub async fn login_with_chatgpt(codex_home: PathBuf) -> std::io::Result<()> {
@@ -60,19 +58,11 @@ pub async fn run_login_with_api_key(
 pub async fn run_login_status(cli_config_overrides: CliConfigOverrides) -> ! {
     let config = load_config_or_exit(cli_config_overrides);
 
-    match CodexAuth::from_codex_home(&config.codex_home, config.preferred_auth_method) {
+    match CodexAuth::from_codex_home(&config.codex_home) {
         Ok(Some(auth)) => match auth.mode {
             AuthMode::ApiKey => match auth.get_token().await {
                 Ok(api_key) => {
                     eprintln!("Logged in using an API key - {}", safe_format_key(&api_key));
-
-                    if let Ok(env_api_key) = env::var(OPENAI_API_KEY_ENV_VAR)
-                        && env_api_key == api_key
-                    {
-                        eprintln!(
-                            "   API loaded from OPENAI_API_KEY environment variable or .env file"
-                        );
-                    }
                     std::process::exit(0);
                 }
                 Err(e) => {

codex-rs/cli/src/main.rs

@@ -73,6 +73,9 @@ enum Subcommand {
     #[clap(visible_alias = "a")]
     Apply(ApplyCommand),
 
+    /// Resume a previous interactive session (picker by default; use --last to continue the most recent).
+    Resume(ResumeCommand),
+
     /// Internal: generate TypeScript protocol bindings.
     #[clap(hide = true)]
     GenerateTs(GenerateTsCommand),
@@ -85,6 +88,18 @@ struct CompletionCommand {
     shell: Shell,
 }
 
+#[derive(Debug, Parser)]
+struct ResumeCommand {
+    /// Conversation/session id (UUID). When provided, resumes this session.
+    /// If omitted, use --last to pick the most recent recorded session.
+    #[arg(value_name = "SESSION_ID")]
+    session_id: Option<String>,
+
+    /// Continue the most recent session without showing the picker.
+    #[arg(long = "last", default_value_t = false, conflicts_with = "session_id")]
+    last: bool,
+}
+
 #[derive(Debug, Parser)]
 struct DebugArgs {
     #[command(subcommand)]
@@ -143,26 +158,54 @@ fn main() -> anyhow::Result<()> {
 }
 
 async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()> {
-    let cli = MultitoolCli::parse();
+    let MultitoolCli {
+        config_overrides: root_config_overrides,
+        mut interactive,
+        subcommand,
+    } = MultitoolCli::parse();
 
-    match cli.subcommand {
+    match subcommand {
         None => {
-            let mut tui_cli = cli.interactive;
-            prepend_config_flags(&mut tui_cli.config_overrides, cli.config_overrides);
-            let usage = codex_tui::run_main(tui_cli, codex_linux_sandbox_exe).await?;
+            prepend_config_flags(
+                &mut interactive.config_overrides,
+                root_config_overrides.clone(),
+            );
+            let usage = codex_tui::run_main(interactive, codex_linux_sandbox_exe).await?;
             if !usage.is_zero() {
                 println!("{}", codex_core::protocol::FinalOutput::from(usage));
             }
         }
         Some(Subcommand::Exec(mut exec_cli)) => {
-            prepend_config_flags(&mut exec_cli.config_overrides, cli.config_overrides);
+            prepend_config_flags(
+                &mut exec_cli.config_overrides,
+                root_config_overrides.clone(),
+            );
             codex_exec::run_main(exec_cli, codex_linux_sandbox_exe).await?;
         }
         Some(Subcommand::Mcp) => {
-            codex_mcp_server::run_main(codex_linux_sandbox_exe, cli.config_overrides).await?;
+            codex_mcp_server::run_main(codex_linux_sandbox_exe, root_config_overrides.clone())
+                .await?;
+        }
+        Some(Subcommand::Resume(ResumeCommand { session_id, last })) => {
+            // Start with the parsed interactive CLI so resume shares the same
+            // configuration surface area as `codex` without additional flags.
+            let resume_session_id = session_id;
+            interactive.resume_picker = resume_session_id.is_none() && !last;
+            interactive.resume_last = last;
+            interactive.resume_session_id = resume_session_id;
+
+            // Propagate any root-level config overrides (e.g. `-c key=value`).
+            prepend_config_flags(
+                &mut interactive.config_overrides,
+                root_config_overrides.clone(),
+            );
+            codex_tui::run_main(interactive, codex_linux_sandbox_exe).await?;
         }
         Some(Subcommand::Login(mut login_cli)) => {
-            prepend_config_flags(&mut login_cli.config_overrides, cli.config_overrides);
+            prepend_config_flags(
+                &mut login_cli.config_overrides,
+                root_config_overrides.clone(),
+            );
             match login_cli.action {
                 Some(LoginSubcommand::Status) => {
                     run_login_status(login_cli.config_overrides).await;
@@ -177,11 +220,17 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
             }
         }
         Some(Subcommand::Logout(mut logout_cli)) => {
-            prepend_config_flags(&mut logout_cli.config_overrides, cli.config_overrides);
+            prepend_config_flags(
+                &mut logout_cli.config_overrides,
+                root_config_overrides.clone(),
+            );
             run_logout(logout_cli.config_overrides).await;
         }
         Some(Subcommand::Proto(mut proto_cli)) => {
-            prepend_config_flags(&mut proto_cli.config_overrides, cli.config_overrides);
+            prepend_config_flags(
+                &mut proto_cli.config_overrides,
+                root_config_overrides.clone(),
+            );
             proto::run_main(proto_cli).await?;
         }
         Some(Subcommand::Completion(completion_cli)) => {
@@ -189,7 +238,10 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
         }
         Some(Subcommand::Debug(debug_args)) => match debug_args.cmd {
             DebugCommand::Seatbelt(mut seatbelt_cli) => {
-                prepend_config_flags(&mut seatbelt_cli.config_overrides, cli.config_overrides);
+                prepend_config_flags(
+                    &mut seatbelt_cli.config_overrides,
+                    root_config_overrides.clone(),
+                );
                 codex_cli::debug_sandbox::run_command_under_seatbelt(
                     seatbelt_cli,
                     codex_linux_sandbox_exe,
@@ -197,7 +249,10 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
                 .await?;
             }
             DebugCommand::Landlock(mut landlock_cli) => {
-                prepend_config_flags(&mut landlock_cli.config_overrides, cli.config_overrides);
+                prepend_config_flags(
+                    &mut landlock_cli.config_overrides,
+                    root_config_overrides.clone(),
+                );
                 codex_cli::debug_sandbox::run_command_under_landlock(
                     landlock_cli,
                     codex_linux_sandbox_exe,
@@ -206,7 +261,10 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
             }
         },
         Some(Subcommand::Apply(mut apply_cli)) => {
-            prepend_config_flags(&mut apply_cli.config_overrides, cli.config_overrides);
+            prepend_config_flags(
+                &mut apply_cli.config_overrides,
+                root_config_overrides.clone(),
+            );
             run_apply_command(apply_cli, None).await?;
         }
         Some(Subcommand::GenerateTs(gen_cli)) => {

codex-rs/cli/src/proto.rs

@@ -37,10 +37,8 @@ pub async fn run_main(opts: ProtoCli) -> anyhow::Result<()> {
 
     let config = Config::load_with_cli_overrides(overrides_vec, ConfigOverrides::default())?;
     // Use conversation_manager API to start a conversation
-    let conversation_manager = ConversationManager::new(AuthManager::shared(
-        config.codex_home.clone(),
-        config.preferred_auth_method,
-    ));
+    let conversation_manager =
+        ConversationManager::new(AuthManager::shared(config.codex_home.clone()));
     let NewConversation {
         conversation_id: _,
         conversation,

codex-rs/common/src/config_summary.rs

@@ -17,7 +17,10 @@ pub fn create_config_summary_entries(config: &Config) -> Vec<(&'static str, Stri
     {
         entries.push((
             "reasoning effort",
-            config.model_reasoning_effort.to_string(),
+            config
+                .model_reasoning_effort
+                .map(|effort| effort.to_string())
+                .unwrap_or_else(|| "none".to_string()),
         ));
         entries.push((
             "reasoning summaries",

codex-rs/common/src/elapsed.rs

@@ -2,7 +2,7 @@ use std::time::Duration;
 use std::time::Instant;
 
 /// Returns a string representing the elapsed time since `start_time` like
-/// "1m15s" or "1.50s".
+/// "1m 15s" or "1.50s".
 pub fn format_elapsed(start_time: Instant) -> String {
     format_duration(start_time.elapsed())
 }
@@ -12,7 +12,7 @@ pub fn format_elapsed(start_time: Instant) -> String {
 /// Formatting rules:
 /// * < 1 s  ->  "{milli}ms"
 /// * < 60 s ->  "{sec:.2}s" (two decimal places)
-/// * >= 60 s ->  "{min}m{sec:02}s"
+/// * >= 60 s ->  "{min}m {sec:02}s"
 pub fn format_duration(duration: Duration) -> String {
     let millis = duration.as_millis() as i64;
     format_elapsed_millis(millis)
@@ -26,7 +26,7 @@ fn format_elapsed_millis(millis: i64) -> String {
     } else {
         let minutes = millis / 60_000;
         let seconds = (millis % 60_000) / 1000;
-        format!("{minutes}m{seconds:02}s")
+        format!("{minutes}m {seconds:02}s")
     }
 }
 
@@ -61,12 +61,18 @@ mod tests {
     fn test_format_duration_minutes() {
         // Durations ≥ 1 minute should be printed mmss.
         let dur = Duration::from_millis(75_000); // 1m15s
-        assert_eq!(format_duration(dur), "1m15s");
+        assert_eq!(format_duration(dur), "1m 15s");
 
         let dur_exact = Duration::from_millis(60_000); // 1m0s
-        assert_eq!(format_duration(dur_exact), "1m00s");
+        assert_eq!(format_duration(dur_exact), "1m 00s");
 
         let dur_long = Duration::from_millis(3_601_000);
-        assert_eq!(format_duration(dur_long), "60m01s");
+        assert_eq!(format_duration(dur_long), "60m 01s");
+    }
+
+    #[test]
+    fn test_format_duration_one_hour_has_space() {
+        let dur_hour = Duration::from_millis(3_600_000);
+        assert_eq!(format_duration(dur_hour), "60m 00s");
     }
 }

codex-rs/common/src/model_presets.rs

@@ -1,4 +1,6 @@
+use codex_core::config::SWIFTFOX_MEDIUM_MODEL;
 use codex_core::protocol_config_types::ReasoningEffort;
+use codex_protocol::mcp_protocol::AuthMode;
 
 /// A simple preset pairing a model slug with a reasoning effort.
 #[derive(Debug, Clone, Copy)]
@@ -12,43 +14,68 @@ pub struct ModelPreset {
     /// Model slug (e.g., "gpt-5").
     pub model: &'static str,
     /// Reasoning effort to apply for this preset.
-    pub effort: ReasoningEffort,
+    pub effort: Option<ReasoningEffort>,
 }
 
-/// Built-in list of model presets that pair a model with a reasoning effort.
-///
-/// Keep this UI-agnostic so it can be reused by both TUI and MCP server.
-pub fn builtin_model_presets() -> &'static [ModelPreset] {
-    // Order reflects effort from minimal to high.
-    const PRESETS: &[ModelPreset] = &[
-        ModelPreset {
-            id: "gpt-5-minimal",
-            label: "gpt-5 minimal",
-            description: "— fastest responses with limited reasoning; ideal for coding, instructions, or lightweight tasks",
-            model: "gpt-5",
-            effort: ReasoningEffort::Minimal,
-        },
-        ModelPreset {
-            id: "gpt-5-low",
-            label: "gpt-5 low",
-            description: "— balances speed with some reasoning; useful for straightforward queries and short explanations",
-            model: "gpt-5",
-            effort: ReasoningEffort::Low,
-        },
-        ModelPreset {
-            id: "gpt-5-medium",
-            label: "gpt-5 medium",
-            description: "— default setting; provides a solid balance of reasoning depth and latency for general-purpose tasks",
-            model: "gpt-5",
-            effort: ReasoningEffort::Medium,
-        },
-        ModelPreset {
-            id: "gpt-5-high",
-            label: "gpt-5 high",
-            description: "— maximizes reasoning depth for complex or ambiguous problems",
-            model: "gpt-5",
-            effort: ReasoningEffort::High,
-        },
-    ];
-    PRESETS
+const PRESETS: &[ModelPreset] = &[
+    ModelPreset {
+        id: "swiftfox-low",
+        label: "swiftfox low",
+        description: "",
+        model: "swiftfox",
+        effort: Some(ReasoningEffort::Low),
+    },
+    ModelPreset {
+        id: "swiftfox-medium",
+        label: "swiftfox medium",
+        description: "",
+        model: "swiftfox",
+        effort: None,
+    },
+    ModelPreset {
+        id: "swiftfox-high",
+        label: "swiftfox high",
+        description: "",
+        model: "swiftfox",
+        effort: Some(ReasoningEffort::High),
+    },
+    ModelPreset {
+        id: "gpt-5-minimal",
+        label: "gpt-5 minimal",
+        description: "— fastest responses with limited reasoning; ideal for coding, instructions, or lightweight tasks",
+        model: "gpt-5",
+        effort: Some(ReasoningEffort::Minimal),
+    },
+    ModelPreset {
+        id: "gpt-5-low",
+        label: "gpt-5 low",
+        description: "— balances speed with some reasoning; useful for straightforward queries and short explanations",
+        model: "gpt-5",
+        effort: Some(ReasoningEffort::Low),
+    },
+    ModelPreset {
+        id: "gpt-5-medium",
+        label: "gpt-5 medium",
+        description: "— default setting; provides a solid balance of reasoning depth and latency for general-purpose tasks",
+        model: "gpt-5",
+        effort: Some(ReasoningEffort::Medium),
+    },
+    ModelPreset {
+        id: "gpt-5-high",
+        label: "gpt-5 high",
+        description: "— maximizes reasoning depth for complex or ambiguous problems",
+        model: "gpt-5",
+        effort: Some(ReasoningEffort::High),
+    },
+];
+
+pub fn builtin_model_presets(auth_mode: Option<AuthMode>) -> Vec<ModelPreset> {
+    match auth_mode {
+        Some(AuthMode::ApiKey) => PRESETS
+            .iter()
+            .copied()
+            .filter(|p| !p.model.contains(SWIFTFOX_MEDIUM_MODEL))
+            .collect(),
+        _ => PRESETS.to_vec(),
+    }
 }

codex-rs/core/Cargo.toml

@@ -13,11 +13,13 @@ workspace = true
 
 [dependencies]
 anyhow = "1"
+askama = "0.12"
 async-channel = "2.3.1"
 base64 = "0.22"
 bytes = "1.10.1"
 chrono = { version = "0.4", features = ["serde"] }
 codex-apply-patch = { path = "../apply-patch" }
+codex-file-search = { path = "../file-search" }
 codex-mcp-client = { path = "../mcp-client" }
 codex-protocol = { path = "../protocol" }
 dirs = "6"
@@ -54,6 +56,7 @@ tracing = { version = "0.1.41", features = ["log"] }
 tree-sitter = "0.25.9"
 tree-sitter-bash = "0.25.0"
 uuid = { version = "1", features = ["serde", "v4"] }
+which = "6"
 wildmatch = "2.4.0"
 
 
@@ -69,9 +72,6 @@ openssl-sys = { version = "*", features = ["vendored"] }
 [target.aarch64-unknown-linux-musl.dependencies]
 openssl-sys = { version = "*", features = ["vendored"] }
 
-[target.'cfg(target_os = "windows")'.dependencies]
-which = "6"
-
 [dev-dependencies]
 assert_cmd = "2"
 core_test_support = { path = "tests/common" }

codex-rs/core/review_prompt.md

@@ -0,0 +1,87 @@
+# Review guidelines:
+
+You are acting as a reviewer for a proposed code change made by another engineer.
+
+Below are some default guidelines for determining whether the original author would appreciate the issue being flagged.
+
+These are not the final word in determining whether an issue is a bug. In many cases, you will encounter other, more specific guidelines. These may be present elsewhere in a developer message, a user message, a file, or even elsewhere in this system message.
+Those guidelines should be considered to override these general instructions.
+
+Here are the general guidelines for determining whether something is a bug and should be flagged.
+
+1. It meaningfully impacts the accuracy, performance, security, or maintainability of the code.
+2. The bug is discrete and actionable (i.e. not a general issue with the codebase or a combination of multiple issues).
+3. Fixing the bug does not demand a level of rigor that is not present in the rest of the codebase (e.g. one doesn't need very detailed comments and input validation in a repository of one-off scripts in personal projects)
+4. The bug was introduced in the commit (pre-existing bugs should not be flagged).
+5. The author of the original PR would likely fix the issue if they were made aware of it.
+6. The bug does not rely on unstated assumptions about the codebase or author's intent.
+7. It is not enough to speculate that a change may disrupt another part of the codebase, to be considered a bug, one must identify the other parts of the code that are provably affected.
+8. The bug is clearly not just an intentional change by the original author.
+
+When flagging a bug, you will also provide an accompanying comment. Once again, these guidelines are not the final word on how to construct a comment -- defer to any subsequent guidelines that you encounter.
+
+1. The comment should be clear about why the issue is a bug.
+2. The comment should appropriately communicate the severity of the issue. It should not claim that an issue is more severe than it actually is.
+3. The comment should be brief. The body should be at most 1 paragraph. It should not introduce line breaks within the natural language flow unless it is necessary for the code fragment.
+4. The comment should not include any chunks of code longer than 3 lines. Any code chunks should be wrapped in markdown inline code tags or a code block.
+5. The comment should clearly and explicitly communicate the scenarios, environments, or inputs that are necessary for the bug to arise. The comment should immediately indicate that the issue's severity depends on these factors.
+6. The comment's tone should be matter-of-fact and not accusatory or overly positive. It should read as a helpful AI assistant suggestion without sounding too much like a human reviewer.
+7. The comment should be written such that the original author can immediately grasp the idea without close reading.
+8. The comment should avoid excessive flattery and comments that are not helpful to the original author. The comment should avoid phrasing like "Great job ...", "Thanks for ...".
+
+Below are some more detailed guidelines that you should apply to this specific review.
+
+HOW MANY FINDINGS TO RETURN:
+
+Output all findings that the original author would fix if they knew about it. If there is no finding that a person would definitely love to see and fix, prefer outputting no findings. Do not stop at the first qualifying finding. Continue until you've listed every qualifying finding.
+
+GUIDELINES:
+
+- Ignore trivial style unless it obscures meaning or violates documented standards.
+- Use one comment per distinct issue (or a multi-line range if necessary).
+- Use ```suggestion blocks ONLY for concrete replacement code (minimal lines; no commentary inside the block).
+- In every ```suggestion block, preserve the exact leading whitespace of the replaced lines (spaces vs tabs, number of spaces).
+- Do NOT introduce or remove outer indentation levels unless that is the actual fix.
+
+The comments will be presented in the code review as inline comments. You should avoid providing unnecessary location details in the comment body. Always keep the line range as short as possible for interpreting the issue. Avoid ranges longer than 5–10 lines; instead, choose the most suitable subrange that pinpoints the problem.
+
+At the beginning of the finding title, tag the bug with priority level. For example "[P1] Un-padding slices along wrong tensor dimensions". [P0] – Drop everything to fix.  Blocking release, operations, or major usage. Only use for universal issues that do not depend on any assumptions about the inputs. · [P1] – Urgent. Should be addressed in the next cycle · [P2] – Normal. To be fixed eventually · [P3] – Low. Nice to have.
+
+Additionally, include a numeric priority field in the JSON output for each finding: set "priority" to 0 for P0, 1 for P1, 2 for P2, or 3 for P3. If a priority cannot be determined, omit the field or use null.
+
+At the end of your findings, output an "overall correctness" verdict of whether or not the patch should be considered "correct".
+Correct implies that existing code and tests will not break, and the patch is free of bugs and other blocking issues.
+Ignore non-blocking issues such as style, formatting, typos, documentation, and other nits.
+
+FORMATTING GUIDELINES:
+The finding description should be one paragraph.
+
+OUTPUT FORMAT:
+
+## Output schema  — MUST MATCH *exactly*
+
+```json
+{
+  "findings": [
+    {
+      "title": "<≤ 80 chars, imperative>",
+      "body": "<valid Markdown explaining *why* this is a problem; cite files/lines/functions>",
+      "confidence_score": <float 0.0-1.0>,
+      "priority": <int 0-3, optional>,
+      "code_location": {
+        "absolute_file_path": "<file path>",
+        "line_range": {"start": <int>, "end": <int>}
+      }
+    }
+  ],
+  "overall_correctness": "patch is correct" | "patch is incorrect",
+  "overall_explanation": "<1-3 sentence explanation justifying the overall_correctness verdict>",
+  "overall_confidence_score": <float 0.0-1.0>
+}
+```
+
+* **Do not** wrap the JSON in markdown fences or extra prose.
+* The code_location field is required and must include absolute_file_path and line_range.
+*Line ranges must be as short as possible for interpreting the issue (avoid ranges over 5–10 lines; pick the most suitable subrange).
+* The code_location should overlap with the diff.
+* Do not generate a PR fix.

codex-rs/core/src/auth.rs

@@ -17,6 +17,7 @@ use std::time::Duration;
 
 use codex_protocol::mcp_protocol::AuthMode;
 
+use crate::token_data::PlanType;
 use crate::token_data::TokenData;
 use crate::token_data::parse_id_token;
 
@@ -70,13 +71,9 @@ impl CodexAuth {
         Ok(access)
     }
 
-    /// Loads the available auth information from the auth.json or
-    /// OPENAI_API_KEY environment variable.
-    pub fn from_codex_home(
-        codex_home: &Path,
-        preferred_auth_method: AuthMode,
-    ) -> std::io::Result<Option<CodexAuth>> {
-        load_auth(codex_home, true, preferred_auth_method)
+    /// Loads the available auth information from the auth.json.
+    pub fn from_codex_home(codex_home: &Path) -> std::io::Result<Option<CodexAuth>> {
+        load_auth(codex_home)
     }
 
     pub async fn get_token_data(&self) -> Result<TokenData, std::io::Error> {
@@ -135,13 +132,12 @@ impl CodexAuth {
     }
 
     pub fn get_account_id(&self) -> Option<String> {
-        self.get_current_token_data()
-            .and_then(|t| t.account_id.clone())
+        self.get_current_token_data().and_then(|t| t.account_id)
     }
 
-    pub fn get_plan_type(&self) -> Option<String> {
+    pub(crate) fn get_plan_type(&self) -> Option<PlanType> {
         self.get_current_token_data()
-            .and_then(|t| t.id_token.chatgpt_plan_type.as_ref().map(|p| p.as_string()))
+            .and_then(|t| t.id_token.chatgpt_plan_type)
     }
 
     fn get_current_auth_json(&self) -> Option<AuthDotJson> {
@@ -150,7 +146,7 @@ impl CodexAuth {
     }
 
     fn get_current_token_data(&self) -> Option<TokenData> {
-        self.get_current_auth_json().and_then(|t| t.tokens.clone())
+        self.get_current_auth_json().and_then(|t| t.tokens)
     }
 
     /// Consider this private to integration tests.
@@ -193,10 +189,11 @@ impl CodexAuth {
 
 pub const OPENAI_API_KEY_ENV_VAR: &str = "OPENAI_API_KEY";
 
-fn read_openai_api_key_from_env() -> Option<String> {
+pub fn read_openai_api_key_from_env() -> Option<String> {
     env::var(OPENAI_API_KEY_ENV_VAR)
         .ok()
-        .filter(|s| !s.is_empty())
+        .map(|value| value.trim().to_string())
+        .filter(|value| !value.is_empty())
 }
 
 pub fn get_auth_file(codex_home: &Path) -> PathBuf {
@@ -214,7 +211,7 @@ pub fn logout(codex_home: &Path) -> std::io::Result<bool> {
     }
 }
 
-/// Writes an `auth.json` that contains only the API key. Intended for CLI use.
+/// Writes an `auth.json` that contains only the API key.
 pub fn login_with_api_key(codex_home: &Path, api_key: &str) -> std::io::Result<()> {
     let auth_dot_json = AuthDotJson {
         openai_api_key: Some(api_key.to_string()),
@@ -224,28 +221,11 @@ pub fn login_with_api_key(codex_home: &Path, api_key: &str) -> std::io::Result<(
     write_auth_json(&get_auth_file(codex_home), &auth_dot_json)
 }
 
-fn load_auth(
-    codex_home: &Path,
-    include_env_var: bool,
-    preferred_auth_method: AuthMode,
-) -> std::io::Result<Option<CodexAuth>> {
-    // First, check to see if there is a valid auth.json file. If not, we fall
-    // back to AuthMode::ApiKey using the OPENAI_API_KEY environment variable
-    // (if it is set).
+fn load_auth(codex_home: &Path) -> std::io::Result<Option<CodexAuth>> {
     let auth_file = get_auth_file(codex_home);
     let client = crate::default_client::create_client();
     let auth_dot_json = match try_read_auth_json(&auth_file) {
         Ok(auth) => auth,
-        // If auth.json does not exist, try to read the OPENAI_API_KEY from the
-        // environment variable.
-        Err(e) if e.kind() == std::io::ErrorKind::NotFound && include_env_var => {
-            return match read_openai_api_key_from_env() {
-                Some(api_key) => Ok(Some(CodexAuth::from_api_key_with_client(&api_key, client))),
-                None => Ok(None),
-            };
-        }
-        // Though if auth.json exists but is malformed, do not fall back to the
-        // env var because the user may be expecting to use AuthMode::ChatGPT.
         Err(e) => {
             return Err(e);
         }
@@ -257,32 +237,11 @@ fn load_auth(
         last_refresh,
     } = auth_dot_json;
 
-    // If the auth.json has an API key AND does not appear to be on a plan that
-    // should prefer AuthMode::ChatGPT, use AuthMode::ApiKey.
+    // Prefer AuthMode.ApiKey if it's set in the auth.json.
     if let Some(api_key) = &auth_json_api_key {
-        // Should any of these be AuthMode::ChatGPT with the api_key set?
-        // Does AuthMode::ChatGPT indicate that there is an auth.json that is
-        // "refreshable" even if we are using the API key for auth?
-        match &tokens {
-            Some(tokens) => {
-                if tokens.should_use_api_key(preferred_auth_method, tokens.is_openai_email()) {
-                    return Ok(Some(CodexAuth::from_api_key_with_client(api_key, client)));
-                } else {
-                    // Ignore the API key and fall through to ChatGPT auth.
-                }
-            }
-            None => {
-                // We have an API key but no tokens in the auth.json file.
-                // Perhaps the user ran `codex login --api-key <KEY>` or updated
-                // auth.json by hand. Either way, let's assume they are trying
-                // to use their API key.
-                return Ok(Some(CodexAuth::from_api_key_with_client(api_key, client)));
-            }
-        }
+        return Ok(Some(CodexAuth::from_api_key_with_client(api_key, client)));
     }
 
-    // For the AuthMode::ChatGPT variant, perhaps neither api_key nor
-    // openai_api_key should exist?
     Ok(Some(CodexAuth {
         api_key: None,
         mode: AuthMode::ChatGPT,
@@ -332,10 +291,10 @@ async fn update_tokens(
     let tokens = auth_dot_json.tokens.get_or_insert_with(TokenData::default);
     tokens.id_token = parse_id_token(&id_token).map_err(std::io::Error::other)?;
     if let Some(access_token) = access_token {
-        tokens.access_token = access_token.to_string();
+        tokens.access_token = access_token;
     }
     if let Some(refresh_token) = refresh_token {
-        tokens.refresh_token = refresh_token.to_string();
+        tokens.refresh_token = refresh_token;
     }
     auth_dot_json.last_refresh = Some(Utc::now());
     write_auth_json(auth_file, &auth_dot_json)?;
@@ -412,7 +371,6 @@ use std::sync::RwLock;
 /// Internal cached auth state.
 #[derive(Clone, Debug)]
 struct CachedAuth {
-    preferred_auth_mode: AuthMode,
     auth: Option<CodexAuth>,
 }
 
@@ -468,9 +426,7 @@ mod tests {
             auth_dot_json,
             auth_file: _,
             ..
-        } = super::load_auth(codex_home.path(), false, AuthMode::ChatGPT)
-            .unwrap()
-            .unwrap();
+        } = super::load_auth(codex_home.path()).unwrap().unwrap();
         assert_eq!(None, api_key);
         assert_eq!(AuthMode::ChatGPT, mode);
 
@@ -499,88 +455,6 @@ mod tests {
         )
     }
 
-    /// Even if the OPENAI_API_KEY is set in auth.json, if the plan is not in
-    /// [`TokenData::is_plan_that_should_use_api_key`], it should use
-    /// [`AuthMode::ChatGPT`].
-    #[tokio::test]
-    async fn pro_account_with_api_key_still_uses_chatgpt_auth() {
-        let codex_home = tempdir().unwrap();
-        let fake_jwt = write_auth_file(
-            AuthFileParams {
-                openai_api_key: Some("sk-test-key".to_string()),
-                chatgpt_plan_type: "pro".to_string(),
-            },
-            codex_home.path(),
-        )
-        .expect("failed to write auth file");
-
-        let CodexAuth {
-            api_key,
-            mode,
-            auth_dot_json,
-            auth_file: _,
-            ..
-        } = super::load_auth(codex_home.path(), false, AuthMode::ChatGPT)
-            .unwrap()
-            .unwrap();
-        assert_eq!(None, api_key);
-        assert_eq!(AuthMode::ChatGPT, mode);
-
-        let guard = auth_dot_json.lock().unwrap();
-        let auth_dot_json = guard.as_ref().expect("AuthDotJson should exist");
-        assert_eq!(
-            &AuthDotJson {
-                openai_api_key: None,
-                tokens: Some(TokenData {
-                    id_token: IdTokenInfo {
-                        email: Some("user@example.com".to_string()),
-                        chatgpt_plan_type: Some(PlanType::Known(KnownPlan::Pro)),
-                        raw_jwt: fake_jwt,
-                    },
-                    access_token: "test-access-token".to_string(),
-                    refresh_token: "test-refresh-token".to_string(),
-                    account_id: None,
-                }),
-                last_refresh: Some(
-                    DateTime::parse_from_rfc3339(LAST_REFRESH)
-                        .unwrap()
-                        .with_timezone(&Utc)
-                ),
-            },
-            auth_dot_json
-        )
-    }
-
-    /// If the OPENAI_API_KEY is set in auth.json and it is an enterprise
-    /// account, then it should use [`AuthMode::ApiKey`].
-    #[tokio::test]
-    async fn enterprise_account_with_api_key_uses_apikey_auth() {
-        let codex_home = tempdir().unwrap();
-        write_auth_file(
-            AuthFileParams {
-                openai_api_key: Some("sk-test-key".to_string()),
-                chatgpt_plan_type: "enterprise".to_string(),
-            },
-            codex_home.path(),
-        )
-        .expect("failed to write auth file");
-
-        let CodexAuth {
-            api_key,
-            mode,
-            auth_dot_json,
-            auth_file: _,
-            ..
-        } = super::load_auth(codex_home.path(), false, AuthMode::ChatGPT)
-            .unwrap()
-            .unwrap();
-        assert_eq!(Some("sk-test-key".to_string()), api_key);
-        assert_eq!(AuthMode::ApiKey, mode);
-
-        let guard = auth_dot_json.lock().expect("should unwrap");
-        assert!(guard.is_none(), "auth_dot_json should be None");
-    }
-
     #[tokio::test]
     async fn loads_api_key_from_auth_json() {
         let dir = tempdir().unwrap();
@@ -591,9 +465,7 @@ mod tests {
         )
         .unwrap();
 
-        let auth = super::load_auth(dir.path(), false, AuthMode::ChatGPT)
-            .unwrap()
-            .unwrap();
+        let auth = super::load_auth(dir.path()).unwrap().unwrap();
         assert_eq!(auth.mode, AuthMode::ApiKey);
         assert_eq!(auth.api_key, Some("sk-test-key".to_string()));
 
@@ -683,26 +555,17 @@ impl AuthManager {
     /// preferred auth method. Errors loading auth are swallowed; `auth()` will
     /// simply return `None` in that case so callers can treat it as an
     /// unauthenticated state.
-    pub fn new(codex_home: PathBuf, preferred_auth_mode: AuthMode) -> Self {
-        let auth = CodexAuth::from_codex_home(&codex_home, preferred_auth_mode)
-            .ok()
-            .flatten();
+    pub fn new(codex_home: PathBuf) -> Self {
+        let auth = CodexAuth::from_codex_home(&codex_home).ok().flatten();
         Self {
             codex_home,
-            inner: RwLock::new(CachedAuth {
-                preferred_auth_mode,
-                auth,
-            }),
+            inner: RwLock::new(CachedAuth { auth }),
         }
     }
 
     /// Create an AuthManager with a specific CodexAuth, for testing only.
     pub fn from_auth_for_testing(auth: CodexAuth) -> Arc<Self> {
-        let preferred_auth_mode = auth.mode;
-        let cached = CachedAuth {
-            preferred_auth_mode,
-            auth: Some(auth),
-        };
+        let cached = CachedAuth { auth: Some(auth) };
         Arc::new(Self {
             codex_home: PathBuf::new(),
             inner: RwLock::new(cached),
@@ -714,21 +577,10 @@ impl AuthManager {
         self.inner.read().ok().and_then(|c| c.auth.clone())
     }
 
-    /// Preferred auth method used when (re)loading.
-    pub fn preferred_auth_method(&self) -> AuthMode {
-        self.inner
-            .read()
-            .map(|c| c.preferred_auth_mode)
-            .unwrap_or(AuthMode::ApiKey)
-    }
-
-    /// Force a reload using the existing preferred auth method. Returns
+    /// Force a reload of the auth information from auth.json. Returns
     /// whether the auth value changed.
     pub fn reload(&self) -> bool {
-        let preferred = self.preferred_auth_method();
-        let new_auth = CodexAuth::from_codex_home(&self.codex_home, preferred)
-            .ok()
-            .flatten();
+        let new_auth = CodexAuth::from_codex_home(&self.codex_home).ok().flatten();
         if let Ok(mut guard) = self.inner.write() {
             let changed = !AuthManager::auths_equal(&guard.auth, &new_auth);
             guard.auth = new_auth;
@@ -747,8 +599,8 @@ impl AuthManager {
     }
 
     /// Convenience constructor returning an `Arc` wrapper.
-    pub fn shared(codex_home: PathBuf, preferred_auth_mode: AuthMode) -> Arc<Self> {
-        Arc::new(Self::new(codex_home, preferred_auth_mode))
+    pub fn shared(codex_home: PathBuf) -> Arc<Self> {
+        Arc::new(Self::new(codex_home))
     }
 
     /// Attempt to refresh the current auth token (if any). On success, reload

codex-rs/core/src/client.rs

@@ -41,6 +41,7 @@ use crate::model_provider_info::WireApi;
 use crate::openai_model_info::get_model_info;
 use crate::openai_tools::create_tools_json_for_responses_api;
 use crate::protocol::TokenUsage;
+use crate::token_data::PlanType;
 use crate::util::backoff;
 use codex_protocol::config_types::ReasoningEffort as ReasoningEffortConfig;
 use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
@@ -60,7 +61,7 @@ struct Error {
     message: Option<String>,
 
     // Optional fields available on "usage_limit_reached" and "usage_not_included" errors
-    plan_type: Option<String>,
+    plan_type: Option<PlanType>,
     resets_in_seconds: Option<u64>,
 }
 
@@ -71,7 +72,7 @@ pub struct ModelClient {
     client: reqwest::Client,
     provider: ModelProviderInfo,
     conversation_id: ConversationId,
-    effort: ReasoningEffortConfig,
+    effort: Option<ReasoningEffortConfig>,
     summary: ReasoningSummaryConfig,
 }
 
@@ -80,7 +81,7 @@ impl ModelClient {
         config: Arc<Config>,
         auth_manager: Option<Arc<AuthManager>>,
         provider: ModelProviderInfo,
-        effort: ReasoningEffortConfig,
+        effort: Option<ReasoningEffortConfig>,
         summary: ReasoningSummaryConfig,
         conversation_id: ConversationId,
     ) -> Self {
@@ -103,6 +104,12 @@ impl ModelClient {
             .or_else(|| get_model_info(&self.config.model_family).map(|info| info.context_window))
     }
 
+    pub fn get_auto_compact_token_limit(&self) -> Option<i64> {
+        self.config.model_auto_compact_token_limit.or_else(|| {
+            get_model_info(&self.config.model_family).and_then(|info| info.auto_compact_token_limit)
+        })
+    }
+
     /// Dispatches to either the Responses or Chat implementation depending on
     /// the provider config.  Public callers always invoke `stream()` – the
     /// specialised helpers are private to avoid accidental misuse.
@@ -186,6 +193,15 @@ impl ModelClient {
             None
         };
 
+        // In general, we want to explicitly send `store: false` when using the Responses API,
+        // but in practice, the Azure Responses API rejects `store: false`:
+        //
+        // - If store = false and id is sent an error is thrown that ID is not found
+        // - If store = false and id is not sent an error is thrown that ID is required
+        //
+        // For Azure, we send `store: true` and preserve reasoning item IDs.
+        let azure_workaround = self.provider.is_azure_responses_endpoint();
+
         let payload = ResponsesApiRequest {
             model: &self.config.model,
             instructions: &full_instructions,
@@ -194,13 +210,19 @@ impl ModelClient {
             tool_choice: "auto",
             parallel_tool_calls: false,
             reasoning,
-            store: false,
+            store: azure_workaround,
             stream: true,
             include,
             prompt_cache_key: Some(self.conversation_id.to_string()),
             text,
         };
 
+        let mut payload_json = serde_json::to_value(&payload)?;
+        if azure_workaround {
+            attach_item_ids(&mut payload_json, &input_with_instructions);
+        }
+        let payload_body = serde_json::to_string(&payload_json)?;
+
         let mut attempt = 0;
         let max_retries = self.provider.request_max_retries();
 
@@ -213,7 +235,7 @@ impl ModelClient {
             trace!(
                 "POST to {}: {}",
                 self.provider.get_full_url(&auth),
-                serde_json::to_string(&payload)?
+                payload_body.as_str()
             );
 
             let mut req_builder = self
@@ -227,7 +249,7 @@ impl ModelClient {
                 .header("conversation_id", self.conversation_id.to_string())
                 .header("session_id", self.conversation_id.to_string())
                 .header(reqwest::header::ACCEPT, "text/event-stream")
-                .json(&payload);
+                .json(&payload_json);
 
             if let Some(auth) = auth.as_ref()
                 && auth.mode == AuthMode::ChatGPT
@@ -239,10 +261,10 @@ impl ModelClient {
             let res = req_builder.send().await;
             if let Ok(resp) = &res {
                 trace!(
-                    "Response status: {}, request-id: {}",
+                    "Response status: {}, cf-ray: {}",
                     resp.status(),
                     resp.headers()
-                        .get("x-request-id")
+                        .get("cf-ray")
                         .map(|v| v.to_str().unwrap_or_default())
                         .unwrap_or_default()
                 );
@@ -304,7 +326,7 @@ impl ModelClient {
                                 // token.
                                 let plan_type = error
                                     .plan_type
-                                    .or_else(|| auth.and_then(|a| a.get_plan_type()));
+                                    .or_else(|| auth.as_ref().and_then(|a| a.get_plan_type()));
                                 let resets_in_seconds = error.resets_in_seconds;
                                 return Err(CodexErr::UsageLimitReached(UsageLimitReachedError {
                                     plan_type,
@@ -355,7 +377,7 @@ impl ModelClient {
     }
 
     /// Returns the current reasoning effort setting.
-    pub fn get_reasoning_effort(&self) -> ReasoningEffortConfig {
+    pub fn get_reasoning_effort(&self) -> Option<ReasoningEffortConfig> {
         self.effort
     }
 
@@ -424,6 +446,33 @@ struct ResponseCompletedOutputTokensDetails {
     reasoning_tokens: u64,
 }
 
+fn attach_item_ids(payload_json: &mut Value, original_items: &[ResponseItem]) {
+    let Some(input_value) = payload_json.get_mut("input") else {
+        return;
+    };
+    let serde_json::Value::Array(items) = input_value else {
+        return;
+    };
+
+    for (value, item) in items.iter_mut().zip(original_items.iter()) {
+        if let ResponseItem::Reasoning { id, .. }
+        | ResponseItem::Message { id: Some(id), .. }
+        | ResponseItem::WebSearchCall { id: Some(id), .. }
+        | ResponseItem::FunctionCall { id: Some(id), .. }
+        | ResponseItem::LocalShellCall { id: Some(id), .. }
+        | ResponseItem::CustomToolCall { id: Some(id), .. } = item
+        {
+            if id.is_empty() {
+                continue;
+            }
+
+            if let Some(obj) = value.as_object_mut() {
+                obj.insert("id".to_string(), Value::String(id.clone()));
+            }
+        }
+    }
+}
+
 async fn process_sse<S>(
     stream: S,
     tx_event: mpsc::Sender<Result<ResponseEvent>>,
@@ -1037,4 +1086,37 @@ mod tests {
         let delay = try_parse_retry_after(&err);
         assert_eq!(delay, Some(Duration::from_secs_f64(1.898)));
     }
+
+    #[test]
+    fn error_response_deserializes_old_schema_known_plan_type_and_serializes_back() {
+        use crate::token_data::KnownPlan;
+        use crate::token_data::PlanType;
+
+        let json = r#"{"error":{"type":"usage_limit_reached","plan_type":"pro","resets_in_seconds":3600}}"#;
+        let resp: ErrorResponse =
+            serde_json::from_str(json).expect("should deserialize old schema");
+
+        assert!(matches!(
+            resp.error.plan_type,
+            Some(PlanType::Known(KnownPlan::Pro))
+        ));
+
+        let plan_json = serde_json::to_string(&resp.error.plan_type).expect("serialize plan_type");
+        assert_eq!(plan_json, "\"pro\"");
+    }
+
+    #[test]
+    fn error_response_deserializes_old_schema_unknown_plan_type_and_serializes_back() {
+        use crate::token_data::PlanType;
+
+        let json =
+            r#"{"error":{"type":"usage_limit_reached","plan_type":"vip","resets_in_seconds":60}}"#;
+        let resp: ErrorResponse =
+            serde_json::from_str(json).expect("should deserialize old schema");
+
+        assert!(matches!(resp.error.plan_type, Some(PlanType::Unknown(ref s)) if s == "vip"));
+
+        let plan_json = serde_json::to_string(&resp.error.plan_type).expect("serialize plan_type");
+        assert_eq!(plan_json, "\"vip\"");
+    }
 }

codex-rs/core/src/client_common.rs

@@ -10,14 +10,14 @@ use codex_protocol::models::ResponseItem;
 use futures::Stream;
 use serde::Serialize;
 use std::borrow::Cow;
+use std::ops::Deref;
 use std::pin::Pin;
 use std::task::Context;
 use std::task::Poll;
 use tokio::sync::mpsc;
 
-/// The `instructions` field in the payload sent to a model should always start
-/// with this content.
-const BASE_INSTRUCTIONS: &str = include_str!("../prompt.md");
+/// Review thread system prompt. Edit `core/src/review_prompt.md` to customize.
+pub const REVIEW_PROMPT: &str = include_str!("../review_prompt.md");
 
 /// API request payload for a single model turn
 #[derive(Default, Debug, Clone)]
@@ -38,11 +38,12 @@ impl Prompt {
         let base = self
             .base_instructions_override
             .as_deref()
-            .unwrap_or(BASE_INSTRUCTIONS);
+            .unwrap_or(model.base_instructions.deref());
         let mut sections: Vec<&str> = vec![base];
 
-        // When there are no custom instructions, add apply_patch_tool_instructions if either:
-        // - the model needs special instructions (4.1), or
+        // When there are no custom instructions, add apply_patch_tool_instructions if:
+        // - the model needs special instructions (4.1)
+        // AND
         // - there is no apply_patch tool present
         let is_apply_patch_tool_present = self.tools.iter().any(|tool| match tool {
             OpenAiTool::Function(f) => f.name == "apply_patch",
@@ -50,7 +51,8 @@ impl Prompt {
             _ => false,
         });
         if self.base_instructions_override.is_none()
-            && (model.needs_special_apply_patch_instructions || !is_apply_patch_tool_present)
+            && model.needs_special_apply_patch_instructions
+            && !is_apply_patch_tool_present
         {
             sections.push(APPLY_PATCH_TOOL_INSTRUCTIONS);
         }
@@ -81,8 +83,10 @@ pub enum ResponseEvent {
 
 #[derive(Debug, Serialize)]
 pub(crate) struct Reasoning {
-    pub(crate) effort: ReasoningEffortConfig,
-    pub(crate) summary: ReasoningSummaryConfig,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(crate) effort: Option<ReasoningEffortConfig>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(crate) summary: Option<ReasoningSummaryConfig>,
 }
 
 /// Controls under the `text` field in the Responses API for GPT-5.
@@ -136,14 +140,17 @@ pub(crate) struct ResponsesApiRequest<'a> {
 
 pub(crate) fn create_reasoning_param_for_request(
     model_family: &ModelFamily,
-    effort: ReasoningEffortConfig,
+    effort: Option<ReasoningEffortConfig>,
     summary: ReasoningSummaryConfig,
 ) -> Option<Reasoning> {
-    if model_family.supports_reasoning_summaries {
-        Some(Reasoning { effort, summary })
-    } else {
-        None
+    if !model_family.supports_reasoning_summaries {
+        return None;
     }
+
+    Some(Reasoning {
+        effort,
+        summary: Some(summary),
+    })
 }
 
 pub(crate) fn create_text_param_for_request(
@@ -169,18 +176,64 @@ impl Stream for ResponseStream {
 #[cfg(test)]
 mod tests {
     use crate::model_family::find_family_for_model;
+    use pretty_assertions::assert_eq;
 
     use super::*;
 
+    struct InstructionsTestCase {
+        pub slug: &'static str,
+        pub expects_apply_patch_instructions: bool,
+    }
     #[test]
     fn get_full_instructions_no_user_content() {
         let prompt = Prompt {
             ..Default::default()
         };
-        let expected = format!("{BASE_INSTRUCTIONS}\n{APPLY_PATCH_TOOL_INSTRUCTIONS}");
-        let model_family = find_family_for_model("gpt-4.1").expect("known model slug");
-        let full = prompt.get_full_instructions(&model_family);
-        assert_eq!(full, expected);
+        let test_cases = vec![
+            InstructionsTestCase {
+                slug: "gpt-3.5",
+                expects_apply_patch_instructions: true,
+            },
+            InstructionsTestCase {
+                slug: "gpt-4.1",
+                expects_apply_patch_instructions: true,
+            },
+            InstructionsTestCase {
+                slug: "gpt-4o",
+                expects_apply_patch_instructions: true,
+            },
+            InstructionsTestCase {
+                slug: "gpt-5",
+                expects_apply_patch_instructions: true,
+            },
+            InstructionsTestCase {
+                slug: "codex-mini-latest",
+                expects_apply_patch_instructions: true,
+            },
+            InstructionsTestCase {
+                slug: "gpt-oss:120b",
+                expects_apply_patch_instructions: false,
+            },
+            InstructionsTestCase {
+                slug: "swiftfox",
+                expects_apply_patch_instructions: false,
+            },
+        ];
+        for test_case in test_cases {
+            let model_family = find_family_for_model(test_case.slug).expect("known model slug");
+            let expected = if test_case.expects_apply_patch_instructions {
+                format!(
+                    "{}\n{}",
+                    model_family.clone().base_instructions,
+                    APPLY_PATCH_TOOL_INSTRUCTIONS
+                )
+            } else {
+                model_family.clone().base_instructions
+            };
+
+            let full = prompt.get_full_instructions(&model_family);
+            assert_eq!(full, expected);
+        }
     }
 
     #[test]

codex-rs/core/src/codex/compact.rs

@@ -0,0 +1,400 @@
+use std::sync::Arc;
+
+use super::AgentTask;
+use super::MutexExt;
+use super::Session;
+use super::TurnContext;
+use super::get_last_assistant_message_from_turn;
+use crate::Prompt;
+use crate::client_common::ResponseEvent;
+use crate::error::CodexErr;
+use crate::error::Result as CodexResult;
+use crate::protocol::AgentMessageEvent;
+use crate::protocol::CompactedItem;
+use crate::protocol::ErrorEvent;
+use crate::protocol::Event;
+use crate::protocol::EventMsg;
+use crate::protocol::InputItem;
+use crate::protocol::InputMessageKind;
+use crate::protocol::TaskCompleteEvent;
+use crate::protocol::TaskStartedEvent;
+use crate::protocol::TurnContextItem;
+use crate::util::backoff;
+use askama::Template;
+use codex_protocol::models::ContentItem;
+use codex_protocol::models::ResponseInputItem;
+use codex_protocol::models::ResponseItem;
+use codex_protocol::protocol::RolloutItem;
+use futures::prelude::*;
+
+pub(super) const COMPACT_TRIGGER_TEXT: &str = "Start Summarization";
+const SUMMARIZATION_PROMPT: &str = include_str!("../../templates/compact/prompt.md");
+
+#[derive(Template)]
+#[template(path = "compact/history_bridge.md", escape = "none")]
+struct HistoryBridgeTemplate<'a> {
+    user_messages_text: &'a str,
+    summary_text: &'a str,
+}
+
+pub(super) fn spawn_compact_task(
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+    sub_id: String,
+    input: Vec<InputItem>,
+) {
+    let task = AgentTask::compact(
+        sess.clone(),
+        turn_context,
+        sub_id,
+        input,
+        SUMMARIZATION_PROMPT.to_string(),
+    );
+    sess.set_task(task);
+}
+
+pub(super) async fn run_inline_auto_compact_task(
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+) {
+    let sub_id = sess.next_internal_sub_id();
+    let input = vec![InputItem::Text {
+        text: COMPACT_TRIGGER_TEXT.to_string(),
+    }];
+    run_compact_task_inner(
+        sess,
+        turn_context,
+        sub_id,
+        input,
+        SUMMARIZATION_PROMPT.to_string(),
+        false,
+    )
+    .await;
+}
+
+pub(super) async fn run_compact_task(
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+    sub_id: String,
+    input: Vec<InputItem>,
+    compact_instructions: String,
+) {
+    run_compact_task_inner(
+        sess,
+        turn_context,
+        sub_id,
+        input,
+        compact_instructions,
+        true,
+    )
+    .await;
+}
+
+async fn run_compact_task_inner(
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+    sub_id: String,
+    input: Vec<InputItem>,
+    compact_instructions: String,
+    remove_task_on_completion: bool,
+) {
+    let model_context_window = turn_context.client.get_model_context_window();
+    let start_event = Event {
+        id: sub_id.clone(),
+        msg: EventMsg::TaskStarted(TaskStartedEvent {
+            model_context_window,
+        }),
+    };
+    sess.send_event(start_event).await;
+
+    let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
+    let instructions_override = compact_instructions;
+    let turn_input = sess.turn_input_with_history(vec![initial_input_for_turn.clone().into()]);
+
+    let prompt = Prompt {
+        input: turn_input,
+        tools: Vec::new(),
+        base_instructions_override: Some(instructions_override),
+    };
+
+    let max_retries = turn_context.client.get_provider().stream_max_retries();
+    let mut retries = 0;
+
+    let rollout_item = RolloutItem::TurnContext(TurnContextItem {
+        cwd: turn_context.cwd.clone(),
+        approval_policy: turn_context.approval_policy,
+        sandbox_policy: turn_context.sandbox_policy.clone(),
+        model: turn_context.client.get_model(),
+        effort: turn_context.client.get_reasoning_effort(),
+        summary: turn_context.client.get_reasoning_summary(),
+    });
+    sess.persist_rollout_items(&[rollout_item]).await;
+
+    loop {
+        let attempt_result = drain_to_completed(&sess, turn_context.as_ref(), &prompt).await;
+
+        match attempt_result {
+            Ok(()) => {
+                break;
+            }
+            Err(CodexErr::Interrupted) => {
+                return;
+            }
+            Err(e) => {
+                if retries < max_retries {
+                    retries += 1;
+                    let delay = backoff(retries);
+                    sess.notify_stream_error(
+                        &sub_id,
+                        format!(
+                            "stream error: {e}; retrying {retries}/{max_retries} in {delay:?}…"
+                        ),
+                    )
+                    .await;
+                    tokio::time::sleep(delay).await;
+                    continue;
+                } else {
+                    let event = Event {
+                        id: sub_id.clone(),
+                        msg: EventMsg::Error(ErrorEvent {
+                            message: e.to_string(),
+                        }),
+                    };
+                    sess.send_event(event).await;
+                    return;
+                }
+            }
+        }
+    }
+
+    if remove_task_on_completion {
+        sess.remove_task(&sub_id);
+    }
+    let history_snapshot = {
+        let state = sess.state.lock_unchecked();
+        state.history.contents()
+    };
+    let summary_text = get_last_assistant_message_from_turn(&history_snapshot).unwrap_or_default();
+    let user_messages = collect_user_messages(&history_snapshot);
+    let initial_context = sess.build_initial_context(turn_context.as_ref());
+    let new_history = build_compacted_history(initial_context, &user_messages, &summary_text);
+    {
+        let mut state = sess.state.lock_unchecked();
+        state.history.replace(new_history);
+    }
+
+    let rollout_item = RolloutItem::Compacted(CompactedItem {
+        message: summary_text.clone(),
+    });
+    sess.persist_rollout_items(&[rollout_item]).await;
+
+    let event = Event {
+        id: sub_id.clone(),
+        msg: EventMsg::AgentMessage(AgentMessageEvent {
+            message: "Compact task completed".to_string(),
+        }),
+    };
+    sess.send_event(event).await;
+    let event = Event {
+        id: sub_id.clone(),
+        msg: EventMsg::TaskComplete(TaskCompleteEvent {
+            last_agent_message: None,
+        }),
+    };
+    sess.send_event(event).await;
+}
+
+fn content_items_to_text(content: &[ContentItem]) -> Option<String> {
+    let mut pieces = Vec::new();
+    for item in content {
+        match item {
+            ContentItem::InputText { text } | ContentItem::OutputText { text } => {
+                if !text.is_empty() {
+                    pieces.push(text.as_str());
+                }
+            }
+            ContentItem::InputImage { .. } => {}
+        }
+    }
+    if pieces.is_empty() {
+        None
+    } else {
+        Some(pieces.join("\n"))
+    }
+}
+
+pub(crate) fn collect_user_messages(items: &[ResponseItem]) -> Vec<String> {
+    items
+        .iter()
+        .filter_map(|item| match item {
+            ResponseItem::Message { role, content, .. } if role == "user" => {
+                content_items_to_text(content)
+            }
+            _ => None,
+        })
+        .filter(|text| !is_session_prefix_message(text))
+        .collect()
+}
+
+fn is_session_prefix_message(text: &str) -> bool {
+    matches!(
+        InputMessageKind::from(("user", text)),
+        InputMessageKind::UserInstructions | InputMessageKind::EnvironmentContext
+    )
+}
+
+pub(crate) fn build_compacted_history(
+    initial_context: Vec<ResponseItem>,
+    user_messages: &[String],
+    summary_text: &str,
+) -> Vec<ResponseItem> {
+    let mut history = initial_context;
+    let user_messages_text = if user_messages.is_empty() {
+        "(none)".to_string()
+    } else {
+        user_messages.join("\n\n")
+    };
+    let summary_text = if summary_text.is_empty() {
+        "(no summary available)".to_string()
+    } else {
+        summary_text.to_string()
+    };
+    let Ok(bridge) = HistoryBridgeTemplate {
+        user_messages_text: &user_messages_text,
+        summary_text: &summary_text,
+    }
+    .render() else {
+        return vec![];
+    };
+    history.push(ResponseItem::Message {
+        id: None,
+        role: "user".to_string(),
+        content: vec![ContentItem::InputText { text: bridge }],
+    });
+    history
+}
+
+async fn drain_to_completed(
+    sess: &Session,
+    turn_context: &TurnContext,
+    prompt: &Prompt,
+) -> CodexResult<()> {
+    let mut stream = turn_context.client.clone().stream(prompt).await?;
+    loop {
+        let maybe_event = stream.next().await;
+        let Some(event) = maybe_event else {
+            return Err(CodexErr::Stream(
+                "stream closed before response.completed".into(),
+                None,
+            ));
+        };
+        match event {
+            Ok(ResponseEvent::OutputItemDone(item)) => {
+                let mut state = sess.state.lock_unchecked();
+                state.history.record_items(std::slice::from_ref(&item));
+            }
+            Ok(ResponseEvent::Completed { .. }) => {
+                return Ok(());
+            }
+            Ok(_) => continue,
+            Err(e) => return Err(e),
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use pretty_assertions::assert_eq;
+
+    #[test]
+    fn content_items_to_text_joins_non_empty_segments() {
+        let items = vec![
+            ContentItem::InputText {
+                text: "hello".to_string(),
+            },
+            ContentItem::OutputText {
+                text: String::new(),
+            },
+            ContentItem::OutputText {
+                text: "world".to_string(),
+            },
+        ];
+
+        let joined = content_items_to_text(&items);
+
+        assert_eq!(Some("hello\nworld".to_string()), joined);
+    }
+
+    #[test]
+    fn content_items_to_text_ignores_image_only_content() {
+        let items = vec![ContentItem::InputImage {
+            image_url: "file://image.png".to_string(),
+        }];
+
+        let joined = content_items_to_text(&items);
+
+        assert_eq!(None, joined);
+    }
+
+    #[test]
+    fn collect_user_messages_extracts_user_text_only() {
+        let items = vec![
+            ResponseItem::Message {
+                id: Some("assistant".to_string()),
+                role: "assistant".to_string(),
+                content: vec![ContentItem::OutputText {
+                    text: "ignored".to_string(),
+                }],
+            },
+            ResponseItem::Message {
+                id: Some("user".to_string()),
+                role: "user".to_string(),
+                content: vec![
+                    ContentItem::InputText {
+                        text: "first".to_string(),
+                    },
+                    ContentItem::OutputText {
+                        text: "second".to_string(),
+                    },
+                ],
+            },
+            ResponseItem::Other,
+        ];
+
+        let collected = collect_user_messages(&items);
+
+        assert_eq!(vec!["first\nsecond".to_string()], collected);
+    }
+
+    #[test]
+    fn collect_user_messages_filters_session_prefix_entries() {
+        let items = vec![
+            ResponseItem::Message {
+                id: None,
+                role: "user".to_string(),
+                content: vec![ContentItem::InputText {
+                    text: "<user_instructions>do things</user_instructions>".to_string(),
+                }],
+            },
+            ResponseItem::Message {
+                id: None,
+                role: "user".to_string(),
+                content: vec![ContentItem::InputText {
+                    text: "<ENVIRONMENT_CONTEXT>cwd=/tmp</ENVIRONMENT_CONTEXT>".to_string(),
+                }],
+            },
+            ResponseItem::Message {
+                id: None,
+                role: "user".to_string(),
+                content: vec![ContentItem::InputText {
+                    text: "real user message".to_string(),
+                }],
+            },
+        ];
+
+        let collected = collect_user_messages(&items);
+
+        assert_eq!(vec!["real user message".to_string()], collected);
+    }
+}

codex-rs/core/src/config.rs

@@ -9,17 +9,18 @@ use crate::config_types::Tui;
 use crate::config_types::UriBasedFileOpener;
 use crate::git_info::resolve_root_git_project_for_trust;
 use crate::model_family::ModelFamily;
+use crate::model_family::derive_default_model_family;
 use crate::model_family::find_family_for_model;
 use crate::model_provider_info::ModelProviderInfo;
 use crate::model_provider_info::built_in_model_providers;
 use crate::openai_model_info::get_model_info;
 use crate::protocol::AskForApproval;
 use crate::protocol::SandboxPolicy;
+use anyhow::Context;
 use codex_protocol::config_types::ReasoningEffort;
 use codex_protocol::config_types::ReasoningSummary;
 use codex_protocol::config_types::SandboxMode;
 use codex_protocol::config_types::Verbosity;
-use codex_protocol::mcp_protocol::AuthMode;
 use codex_protocol::mcp_protocol::Tools;
 use codex_protocol::mcp_protocol::UserSavedConfig;
 use dirs::home_dir;
@@ -32,6 +33,9 @@ use toml::Value as TomlValue;
 use toml_edit::DocumentMut;
 
 const OPENAI_DEFAULT_MODEL: &str = "gpt-5";
+const OPENAI_DEFAULT_REVIEW_MODEL: &str = "gpt-5";
+pub const SWIFTFOX_MEDIUM_MODEL: &str = "swiftfox";
+pub const SWIFTFOX_MODEL_DISPLAY_NAME: &str = "swiftfox-medium";
 
 /// Maximum number of bytes of the documentation that will be embedded. Larger
 /// files are *silently truncated* to this size so we do not take up too much of
@@ -46,6 +50,9 @@ pub struct Config {
     /// Optional override of model selection.
     pub model: String,
 
+    /// Model used specifically for review sessions. Defaults to "gpt-5".
+    pub review_model: String,
+
     pub model_family: ModelFamily,
 
     /// Size of the context window for the model, in tokens.
@@ -54,6 +61,9 @@ pub struct Config {
     /// Maximum number of output tokens.
     pub model_max_output_tokens: Option<u64>,
 
+    /// Token usage threshold triggering auto-compaction of conversation history.
+    pub model_auto_compact_token_limit: Option<i64>,
+
     /// Key into the model_providers map that specifies which provider to use.
     pub model_provider_id: String,
 
@@ -129,9 +139,6 @@ pub struct Config {
     /// output will be hyperlinked using the specified URI scheme.
     pub file_opener: UriBasedFileOpener,
 
-    /// Collection of settings that are specific to the TUI.
-    pub tui: Tui,
-
     /// Path to the `codex-linux-sandbox` executable. This must be set if
     /// [`crate::exec::SandboxType::LinuxSeccomp`] is used. Note that this
     /// cannot be set in the config file: it must be set in code via
@@ -142,7 +149,7 @@ pub struct Config {
 
     /// Value to use for `reasoning.effort` when making a request using the
     /// Responses API.
-    pub model_reasoning_effort: ReasoningEffort,
+    pub model_reasoning_effort: Option<ReasoningEffort>,
 
     /// If not "none", the value to use for `reasoning.summary` when making a
     /// request using the Responses API.
@@ -154,9 +161,6 @@ pub struct Config {
     /// Base URL for requests to ChatGPT (as opposed to the OpenAI API).
     pub chatgpt_base_url: String,
 
-    /// Experimental rollout resume path (absolute path to .jsonl; undocumented).
-    pub experimental_resume: Option<PathBuf>,
-
     /// Include an experimental plan tool that the model can use to update its current plan and status of each step.
     pub include_plan_tool: bool,
 
@@ -167,11 +171,11 @@ pub struct Config {
 
     pub tools_web_search_request: bool,
 
-    /// If set to `true`, the API key will be signed with the `originator` header.
-    pub preferred_auth_method: AuthMode,
-
     pub use_experimental_streamable_shell_tool: bool,
 
+    /// If set to `true`, used only the experimental unified exec tool.
+    pub use_experimental_unified_exec_tool: bool,
+
     /// Include the `view_image` tool that lets the agent attach a local image path to context.
     pub include_view_image_tool: bool,
 
@@ -261,17 +265,7 @@ pub fn load_config_as_toml(codex_home: &Path) -> std::io::Result<TomlValue> {
     }
 }
 
-/// Patch `CODEX_HOME/config.toml` project state.
-/// Use with caution.
-pub fn set_project_trusted(codex_home: &Path, project_path: &Path) -> anyhow::Result<()> {
-    let config_path = codex_home.join(CONFIG_TOML_FILE);
-    // Parse existing config if present; otherwise start a new document.
-    let mut doc = match std::fs::read_to_string(config_path.clone()) {
-        Ok(s) => s.parse::<DocumentMut>()?,
-        Err(e) if e.kind() == std::io::ErrorKind::NotFound => DocumentMut::new(),
-        Err(e) => return Err(e.into()),
-    };
-
+fn set_project_trusted_inner(doc: &mut DocumentMut, project_path: &Path) -> anyhow::Result<()> {
     // Ensure we render a human-friendly structure:
     //
     // [projects]
@@ -287,14 +281,26 @@ pub fn set_project_trusted(codex_home: &Path, project_path: &Path) -> anyhow::Re
     // Ensure top-level `projects` exists as a non-inline, explicit table. If it
     // exists but was previously represented as a non-table (e.g., inline),
     // replace it with an explicit table.
-    let mut created_projects_table = false;
     {
         let root = doc.as_table_mut();
-        let needs_table = !root.contains_key("projects")
-            || root.get("projects").and_then(|i| i.as_table()).is_none();
-        if needs_table {
-            root.insert("projects", toml_edit::table());
-            created_projects_table = true;
+        // If `projects` exists but isn't a standard table (e.g., it's an inline table),
+        // convert it to an explicit table while preserving existing entries.
+        let existing_projects = root.get("projects").cloned();
+        if existing_projects.as_ref().is_none_or(|i| !i.is_table()) {
+            let mut projects_tbl = toml_edit::Table::new();
+            projects_tbl.set_implicit(true);
+
+            // If there was an existing inline table, migrate its entries to explicit tables.
+            if let Some(inline_tbl) = existing_projects.as_ref().and_then(|i| i.as_inline_table()) {
+                for (k, v) in inline_tbl.iter() {
+                    if let Some(inner_tbl) = v.as_inline_table() {
+                        let new_tbl = inner_tbl.clone().into_table();
+                        projects_tbl.insert(k, toml_edit::Item::Table(new_tbl));
+                    }
+                }
+            }
+
+            root.insert("projects", toml_edit::Item::Table(projects_tbl));
         }
     }
     let Some(projects_tbl) = doc["projects"].as_table_mut() else {
@@ -303,12 +309,6 @@ pub fn set_project_trusted(codex_home: &Path, project_path: &Path) -> anyhow::Re
         ));
     };
 
-    // If we created the `projects` table ourselves, keep it implicit so we
-    // don't render a standalone `[projects]` header.
-    if created_projects_table {
-        projects_tbl.set_implicit(true);
-    }
-
     // Ensure the per-project entry is its own explicit table. If it exists but
     // is not a table (e.g., an inline table), replace it with an explicit table.
     let needs_proj_table = !projects_tbl.contains_key(project_key.as_str())
@@ -327,6 +327,21 @@ pub fn set_project_trusted(codex_home: &Path, project_path: &Path) -> anyhow::Re
     };
     proj_tbl.set_implicit(false);
     proj_tbl["trust_level"] = toml_edit::value("trusted");
+    Ok(())
+}
+
+/// Patch `CODEX_HOME/config.toml` project state.
+/// Use with caution.
+pub fn set_project_trusted(codex_home: &Path, project_path: &Path) -> anyhow::Result<()> {
+    let config_path = codex_home.join(CONFIG_TOML_FILE);
+    // Parse existing config if present; otherwise start a new document.
+    let mut doc = match std::fs::read_to_string(config_path.clone()) {
+        Ok(s) => s.parse::<DocumentMut>()?,
+        Err(e) if e.kind() == std::io::ErrorKind::NotFound => DocumentMut::new(),
+        Err(e) => return Err(e.into()),
+    };
+
+    set_project_trusted_inner(&mut doc, project_path)?;
 
     // ensure codex_home exists
     std::fs::create_dir_all(codex_home)?;
@@ -341,6 +356,117 @@ pub fn set_project_trusted(codex_home: &Path, project_path: &Path) -> anyhow::Re
     Ok(())
 }
 
+fn ensure_profile_table<'a>(
+    doc: &'a mut DocumentMut,
+    profile_name: &str,
+) -> anyhow::Result<&'a mut toml_edit::Table> {
+    let mut created_profiles_table = false;
+    {
+        let root = doc.as_table_mut();
+        let needs_table = !root.contains_key("profiles")
+            || root
+                .get("profiles")
+                .and_then(|item| item.as_table())
+                .is_none();
+        if needs_table {
+            root.insert("profiles", toml_edit::table());
+            created_profiles_table = true;
+        }
+    }
+
+    let Some(profiles_table) = doc["profiles"].as_table_mut() else {
+        return Err(anyhow::anyhow!(
+            "profiles table missing after initialization"
+        ));
+    };
+
+    if created_profiles_table {
+        profiles_table.set_implicit(true);
+    }
+
+    let needs_profile_table = !profiles_table.contains_key(profile_name)
+        || profiles_table
+            .get(profile_name)
+            .and_then(|item| item.as_table())
+            .is_none();
+    if needs_profile_table {
+        profiles_table.insert(profile_name, toml_edit::table());
+    }
+
+    let Some(profile_table) = profiles_table
+        .get_mut(profile_name)
+        .and_then(|item| item.as_table_mut())
+    else {
+        return Err(anyhow::anyhow!(format!(
+            "profile table missing for {profile_name}"
+        )));
+    };
+
+    profile_table.set_implicit(false);
+    Ok(profile_table)
+}
+
+// TODO(jif) refactor config persistence.
+pub async fn persist_model_selection(
+    codex_home: &Path,
+    active_profile: Option<&str>,
+    model: &str,
+    effort: Option<ReasoningEffort>,
+) -> anyhow::Result<()> {
+    let config_path = codex_home.join(CONFIG_TOML_FILE);
+    let serialized = match tokio::fs::read_to_string(&config_path).await {
+        Ok(contents) => contents,
+        Err(err) if err.kind() == std::io::ErrorKind::NotFound => String::new(),
+        Err(err) => return Err(err.into()),
+    };
+
+    let mut doc = if serialized.is_empty() {
+        DocumentMut::new()
+    } else {
+        serialized.parse::<DocumentMut>()?
+    };
+
+    if let Some(profile_name) = active_profile {
+        let profile_table = ensure_profile_table(&mut doc, profile_name)?;
+        profile_table["model"] = toml_edit::value(model);
+        match effort {
+            Some(effort) => {
+                profile_table["model_reasoning_effort"] = toml_edit::value(effort.to_string());
+            }
+            None => {
+                profile_table.remove("model_reasoning_effort");
+            }
+        }
+    } else {
+        let table = doc.as_table_mut();
+        table["model"] = toml_edit::value(model);
+        match effort {
+            Some(effort) => {
+                table["model_reasoning_effort"] = toml_edit::value(effort.to_string());
+            }
+            None => {
+                table.remove("model_reasoning_effort");
+            }
+        }
+    }
+
+    // TODO(jif) refactor the home creation
+    tokio::fs::create_dir_all(codex_home)
+        .await
+        .with_context(|| {
+            format!(
+                "failed to create Codex home directory at {}",
+                codex_home.display()
+            )
+        })?;
+
+    tokio::fs::write(&config_path, doc.to_string())
+        .await
+        .with_context(|| format!("failed to persist config.toml at {}", config_path.display()))?;
+
+    Ok(())
+}
+
 /// Apply a single dotted-path override onto a TOML value.
 fn apply_toml_override(root: &mut TomlValue, path: &str, value: TomlValue) {
     use toml::value::Table;
@@ -385,10 +511,12 @@ fn apply_toml_override(root: &mut TomlValue, path: &str, value: TomlValue) {
 }
 
 /// Base config deserialized from ~/.codex/config.toml.
-#[derive(Deserialize, Debug, Clone, Default)]
+#[derive(Deserialize, Debug, Clone, Default, PartialEq)]
 pub struct ConfigToml {
     /// Optional override of model selection.
     pub model: Option<String>,
+    /// Review model override used by the `/review` feature.
+    pub review_model: Option<String>,
 
     /// Provider to use from the model_providers map.
     pub model_provider: Option<String>,
@@ -399,6 +527,9 @@ pub struct ConfigToml {
     /// Maximum number of output tokens.
     pub model_max_output_tokens: Option<u64>,
 
+    /// Token usage threshold triggering auto-compaction of conversation history.
+    pub model_auto_compact_token_limit: Option<i64>,
+
     /// Default approval policy for executing commands.
     pub approval_policy: Option<AskForApproval>,
 
@@ -469,19 +600,14 @@ pub struct ConfigToml {
     /// Base URL for requests to ChatGPT (as opposed to the OpenAI API).
     pub chatgpt_base_url: Option<String>,
 
-    /// Experimental rollout resume path (absolute path to .jsonl; undocumented).
-    pub experimental_resume: Option<PathBuf>,
-
     /// Experimental path to a file whose contents replace the built-in BASE_INSTRUCTIONS.
     pub experimental_instructions_file: Option<PathBuf>,
 
     pub experimental_use_exec_command_tool: Option<bool>,
+    pub experimental_use_unified_exec_tool: Option<bool>,
 
     pub projects: Option<HashMap<String, ProjectConfig>>,
 
-    /// If set to `true`, the API key will be signed with the `originator` header.
-    pub preferred_auth_method: Option<AuthMode>,
-
     /// Nested tools section for feature toggles
     pub tools: Option<ToolsToml>,
 
@@ -519,7 +645,7 @@ pub struct ProjectConfig {
     pub trust_level: Option<String>,
 }
 
-#[derive(Deserialize, Debug, Clone, Default)]
+#[derive(Deserialize, Debug, Clone, Default, PartialEq)]
 pub struct ToolsToml {
     #[serde(default, alias = "web_search_request")]
     pub web_search: Option<bool>,
@@ -616,6 +742,7 @@ impl ConfigToml {
 #[derive(Default, Debug, Clone)]
 pub struct ConfigOverrides {
     pub model: Option<String>,
+    pub review_model: Option<String>,
     pub cwd: Option<PathBuf>,
     pub approval_policy: Option<AskForApproval>,
     pub sandbox_mode: Option<SandboxMode>,
@@ -643,6 +770,7 @@ impl Config {
         // Destructure ConfigOverrides fully to ensure all overrides are applied.
         let ConfigOverrides {
             model,
+            review_model: override_review_model,
             cwd,
             approval_policy,
             sandbox_mode,
@@ -733,15 +861,8 @@ impl Config {
             .or(cfg.model)
             .unwrap_or_else(default_model);
 
-        let mut model_family = find_family_for_model(&model).unwrap_or_else(|| ModelFamily {
-            slug: model.clone(),
-            family: model.clone(),
-            needs_special_apply_patch_instructions: false,
-            supports_reasoning_summaries: false,
-            reasoning_summary_format: ReasoningSummaryFormat::None,
-            uses_local_shell_tool: false,
-            apply_patch_tool_type: None,
-        });
+        let mut model_family =
+            find_family_for_model(&model).unwrap_or_else(|| derive_default_model_family(&model));
 
         if let Some(supports_reasoning_summaries) = cfg.model_supports_reasoning_summaries {
             model_family.supports_reasoning_summaries = supports_reasoning_summaries;
@@ -759,8 +880,11 @@ impl Config {
                 .as_ref()
                 .map(|info| info.max_output_tokens)
         });
-
-        let experimental_resume = cfg.experimental_resume;
+        let model_auto_compact_token_limit = cfg.model_auto_compact_token_limit.or_else(|| {
+            openai_model_info
+                .as_ref()
+                .and_then(|info| info.auto_compact_token_limit)
+        });
 
         // Load base instructions override from a file if specified. If the
         // path is relative, resolve it against the effective cwd so the
@@ -773,11 +897,18 @@ impl Config {
             Self::get_base_instructions(experimental_instructions_path, &resolved_cwd)?;
         let base_instructions = base_instructions.or(file_base_instructions);
 
+        // Default review model when not set in config; allow CLI override to take precedence.
+        let review_model = override_review_model
+            .or(cfg.review_model)
+            .unwrap_or_else(default_review_model);
+
         let config = Self {
             model,
+            review_model,
             model_family,
             model_context_window,
             model_max_output_tokens,
+            model_auto_compact_token_limit,
             model_provider_id,
             model_provider,
             cwd: resolved_cwd,
@@ -796,7 +927,6 @@ impl Config {
             codex_home,
             history,
             file_opener: cfg.file_opener.unwrap_or(UriBasedFileOpener::VsCode),
-            tui: cfg.tui.unwrap_or_default(),
             codex_linux_sandbox_exe,
 
             hide_agent_reasoning: cfg.hide_agent_reasoning.unwrap_or(false),
@@ -806,8 +936,7 @@ impl Config {
                 .unwrap_or(false),
             model_reasoning_effort: config_profile
                 .model_reasoning_effort
-                .or(cfg.model_reasoning_effort)
-                .unwrap_or_default(),
+                .or(cfg.model_reasoning_effort),
             model_reasoning_summary: config_profile
                 .model_reasoning_summary
                 .or(cfg.model_reasoning_summary)
@@ -817,15 +946,15 @@ impl Config {
                 .chatgpt_base_url
                 .or(cfg.chatgpt_base_url)
                 .unwrap_or("https://chatgpt.com/backend-api/".to_string()),
-
-            experimental_resume,
             include_plan_tool: include_plan_tool.unwrap_or(false),
             include_apply_patch_tool: include_apply_patch_tool.unwrap_or(false),
             tools_web_search_request,
-            preferred_auth_method: cfg.preferred_auth_method.unwrap_or(AuthMode::ChatGPT),
             use_experimental_streamable_shell_tool: cfg
                 .experimental_use_exec_command_tool
                 .unwrap_or(false),
+            use_experimental_unified_exec_tool: cfg
+                .experimental_use_unified_exec_tool
+                .unwrap_or(false),
             include_view_image_tool,
             active_profile: active_profile_name,
             disable_paste_burst: cfg.disable_paste_burst.unwrap_or(false),
@@ -897,6 +1026,10 @@ fn default_model() -> String {
     OPENAI_DEFAULT_MODEL.to_string()
 }
 
+fn default_review_model() -> String {
+    OPENAI_DEFAULT_REVIEW_MODEL.to_string()
+}
+
 /// Returns the path to the Codex configuration directory, which can be
 /// specified by the `CODEX_HOME` environment variable. If not set, defaults to
 /// `~/.codex`.
@@ -938,6 +1071,7 @@ mod tests {
 
     use super::*;
     use pretty_assertions::assert_eq;
+
     use tempfile::TempDir;
 
     #[test]
@@ -1028,6 +1162,148 @@ exclude_slash_tmp = true
         );
     }
 
+    #[tokio::test]
+    async fn persist_model_selection_updates_defaults() -> anyhow::Result<()> {
+        let codex_home = TempDir::new()?;
+
+        persist_model_selection(
+            codex_home.path(),
+            None,
+            "swiftfox",
+            Some(ReasoningEffort::High),
+        )
+        .await?;
+
+        let serialized =
+            tokio::fs::read_to_string(codex_home.path().join(CONFIG_TOML_FILE)).await?;
+        let parsed: ConfigToml = toml::from_str(&serialized)?;
+
+        assert_eq!(parsed.model.as_deref(), Some("swiftfox"));
+        assert_eq!(parsed.model_reasoning_effort, Some(ReasoningEffort::High));
+
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn persist_model_selection_overwrites_existing_model() -> anyhow::Result<()> {
+        let codex_home = TempDir::new()?;
+        let config_path = codex_home.path().join(CONFIG_TOML_FILE);
+
+        tokio::fs::write(
+            &config_path,
+            r#"
+model = "gpt-5"
+model_reasoning_effort = "medium"
+
+[profiles.dev]
+model = "gpt-4.1"
+"#,
+        )
+        .await?;
+
+        persist_model_selection(
+            codex_home.path(),
+            None,
+            "o4-mini",
+            Some(ReasoningEffort::High),
+        )
+        .await?;
+
+        let serialized = tokio::fs::read_to_string(config_path).await?;
+        let parsed: ConfigToml = toml::from_str(&serialized)?;
+
+        assert_eq!(parsed.model.as_deref(), Some("o4-mini"));
+        assert_eq!(parsed.model_reasoning_effort, Some(ReasoningEffort::High));
+        assert_eq!(
+            parsed
+                .profiles
+                .get("dev")
+                .and_then(|profile| profile.model.as_deref()),
+            Some("gpt-4.1"),
+        );
+
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn persist_model_selection_updates_profile() -> anyhow::Result<()> {
+        let codex_home = TempDir::new()?;
+
+        persist_model_selection(
+            codex_home.path(),
+            Some("dev"),
+            "swiftfox",
+            Some(ReasoningEffort::Medium),
+        )
+        .await?;
+
+        let serialized =
+            tokio::fs::read_to_string(codex_home.path().join(CONFIG_TOML_FILE)).await?;
+        let parsed: ConfigToml = toml::from_str(&serialized)?;
+        let profile = parsed
+            .profiles
+            .get("dev")
+            .expect("profile should be created");
+
+        assert_eq!(profile.model.as_deref(), Some("swiftfox"));
+        assert_eq!(
+            profile.model_reasoning_effort,
+            Some(ReasoningEffort::Medium)
+        );
+
+        Ok(())
+    }
+
+    #[tokio::test]
+    async fn persist_model_selection_updates_existing_profile() -> anyhow::Result<()> {
+        let codex_home = TempDir::new()?;
+        let config_path = codex_home.path().join(CONFIG_TOML_FILE);
+
+        tokio::fs::write(
+            &config_path,
+            r#"
+[profiles.dev]
+model = "gpt-4"
+model_reasoning_effort = "medium"
+
+[profiles.prod]
+model = "gpt-5"
+"#,
+        )
+        .await?;
+
+        persist_model_selection(
+            codex_home.path(),
+            Some("dev"),
+            "o4-high",
+            Some(ReasoningEffort::Medium),
+        )
+        .await?;
+
+        let serialized = tokio::fs::read_to_string(config_path).await?;
+        let parsed: ConfigToml = toml::from_str(&serialized)?;
+
+        let dev_profile = parsed
+            .profiles
+            .get("dev")
+            .expect("dev profile should survive updates");
+        assert_eq!(dev_profile.model.as_deref(), Some("o4-high"));
+        assert_eq!(
+            dev_profile.model_reasoning_effort,
+            Some(ReasoningEffort::Medium)
+        );
+
+        assert_eq!(
+            parsed
+                .profiles
+                .get("prod")
+                .and_then(|profile| profile.model.as_deref()),
+            Some("gpt-5"),
+        );
+
+        Ok(())
+    }
+
     struct PrecedenceTestFixture {
         cwd: TempDir,
         codex_home: TempDir,
@@ -1169,9 +1445,11 @@ model_verbosity = "high"
         assert_eq!(
             Config {
                 model: "o3".to_string(),
+                review_model: "gpt-5".to_string(),
                 model_family: find_family_for_model("o3").expect("known model slug"),
                 model_context_window: Some(200_000),
                 model_max_output_tokens: Some(100_000),
+                model_auto_compact_token_limit: None,
                 model_provider_id: "openai".to_string(),
                 model_provider: fixture.openai_provider.clone(),
                 approval_policy: AskForApproval::Never,
@@ -1186,21 +1464,19 @@ model_verbosity = "high"
                 codex_home: fixture.codex_home(),
                 history: History::default(),
                 file_opener: UriBasedFileOpener::VsCode,
-                tui: Tui::default(),
                 codex_linux_sandbox_exe: None,
                 hide_agent_reasoning: false,
                 show_raw_agent_reasoning: false,
-                model_reasoning_effort: ReasoningEffort::High,
+                model_reasoning_effort: Some(ReasoningEffort::High),
                 model_reasoning_summary: ReasoningSummary::Detailed,
                 model_verbosity: None,
                 chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
-                experimental_resume: None,
                 base_instructions: None,
                 include_plan_tool: false,
                 include_apply_patch_tool: false,
                 tools_web_search_request: false,
-                preferred_auth_method: AuthMode::ChatGPT,
                 use_experimental_streamable_shell_tool: false,
+                use_experimental_unified_exec_tool: false,
                 include_view_image_tool: true,
                 active_profile: Some("o3".to_string()),
                 disable_paste_burst: false,
@@ -1226,9 +1502,11 @@ model_verbosity = "high"
         )?;
         let expected_gpt3_profile_config = Config {
             model: "gpt-3.5-turbo".to_string(),
+            review_model: "gpt-5".to_string(),
             model_family: find_family_for_model("gpt-3.5-turbo").expect("known model slug"),
             model_context_window: Some(16_385),
             model_max_output_tokens: Some(4_096),
+            model_auto_compact_token_limit: None,
             model_provider_id: "openai-chat-completions".to_string(),
             model_provider: fixture.openai_chat_completions_provider.clone(),
             approval_policy: AskForApproval::UnlessTrusted,
@@ -1243,21 +1521,19 @@ model_verbosity = "high"
             codex_home: fixture.codex_home(),
             history: History::default(),
             file_opener: UriBasedFileOpener::VsCode,
-            tui: Tui::default(),
             codex_linux_sandbox_exe: None,
             hide_agent_reasoning: false,
             show_raw_agent_reasoning: false,
-            model_reasoning_effort: ReasoningEffort::default(),
+            model_reasoning_effort: None,
             model_reasoning_summary: ReasoningSummary::default(),
             model_verbosity: None,
             chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
-            experimental_resume: None,
             base_instructions: None,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             tools_web_search_request: false,
-            preferred_auth_method: AuthMode::ChatGPT,
             use_experimental_streamable_shell_tool: false,
+            use_experimental_unified_exec_tool: false,
             include_view_image_tool: true,
             active_profile: Some("gpt3".to_string()),
             disable_paste_burst: false,
@@ -1298,9 +1574,11 @@ model_verbosity = "high"
         )?;
         let expected_zdr_profile_config = Config {
             model: "o3".to_string(),
+            review_model: "gpt-5".to_string(),
             model_family: find_family_for_model("o3").expect("known model slug"),
             model_context_window: Some(200_000),
             model_max_output_tokens: Some(100_000),
+            model_auto_compact_token_limit: None,
             model_provider_id: "openai".to_string(),
             model_provider: fixture.openai_provider.clone(),
             approval_policy: AskForApproval::OnFailure,
@@ -1315,21 +1593,19 @@ model_verbosity = "high"
             codex_home: fixture.codex_home(),
             history: History::default(),
             file_opener: UriBasedFileOpener::VsCode,
-            tui: Tui::default(),
             codex_linux_sandbox_exe: None,
             hide_agent_reasoning: false,
             show_raw_agent_reasoning: false,
-            model_reasoning_effort: ReasoningEffort::default(),
+            model_reasoning_effort: None,
             model_reasoning_summary: ReasoningSummary::default(),
             model_verbosity: None,
             chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
-            experimental_resume: None,
             base_instructions: None,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             tools_web_search_request: false,
-            preferred_auth_method: AuthMode::ChatGPT,
             use_experimental_streamable_shell_tool: false,
+            use_experimental_unified_exec_tool: false,
             include_view_image_tool: true,
             active_profile: Some("zdr".to_string()),
             disable_paste_burst: false,
@@ -1356,9 +1632,11 @@ model_verbosity = "high"
         )?;
         let expected_gpt5_profile_config = Config {
             model: "gpt-5".to_string(),
+            review_model: "gpt-5".to_string(),
             model_family: find_family_for_model("gpt-5").expect("known model slug"),
             model_context_window: Some(272_000),
             model_max_output_tokens: Some(128_000),
+            model_auto_compact_token_limit: None,
             model_provider_id: "openai".to_string(),
             model_provider: fixture.openai_provider.clone(),
             approval_policy: AskForApproval::OnFailure,
@@ -1373,21 +1651,19 @@ model_verbosity = "high"
             codex_home: fixture.codex_home(),
             history: History::default(),
             file_opener: UriBasedFileOpener::VsCode,
-            tui: Tui::default(),
             codex_linux_sandbox_exe: None,
             hide_agent_reasoning: false,
             show_raw_agent_reasoning: false,
-            model_reasoning_effort: ReasoningEffort::High,
+            model_reasoning_effort: Some(ReasoningEffort::High),
             model_reasoning_summary: ReasoningSummary::Detailed,
             model_verbosity: Some(Verbosity::High),
             chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
-            experimental_resume: None,
             base_instructions: None,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             tools_web_search_request: false,
-            preferred_auth_method: AuthMode::ChatGPT,
             use_experimental_streamable_shell_tool: false,
+            use_experimental_unified_exec_tool: false,
             include_view_image_tool: true,
             active_profile: Some("gpt5".to_string()),
             disable_paste_burst: false,
@@ -1400,17 +1676,14 @@ model_verbosity = "high"
 
     #[test]
     fn test_set_project_trusted_writes_explicit_tables() -> anyhow::Result<()> {
-        let codex_home = TempDir::new().unwrap();
-        let project_dir = TempDir::new().unwrap();
+        let project_dir = Path::new("/some/path");
+        let mut doc = DocumentMut::new();
 
-        // Call the function under test
-        set_project_trusted(codex_home.path(), project_dir.path())?;
+        set_project_trusted_inner(&mut doc, project_dir)?;
 
-        // Read back the generated config.toml and assert exact contents
-        let config_path = codex_home.path().join(CONFIG_TOML_FILE);
-        let contents = std::fs::read_to_string(&config_path)?;
+        let contents = doc.to_string();
 
-        let raw_path = project_dir.path().to_string_lossy();
+        let raw_path = project_dir.to_string_lossy();
         let path_str = if raw_path.contains('\\') {
             format!("'{raw_path}'")
         } else {
@@ -1428,12 +1701,10 @@ trust_level = "trusted"
 
     #[test]
     fn test_set_project_trusted_converts_inline_to_explicit() -> anyhow::Result<()> {
-        let codex_home = TempDir::new().unwrap();
-        let project_dir = TempDir::new().unwrap();
+        let project_dir = Path::new("/some/path");
 
         // Seed config.toml with an inline project entry under [projects]
-        let config_path = codex_home.path().join(CONFIG_TOML_FILE);
-        let raw_path = project_dir.path().to_string_lossy();
+        let raw_path = project_dir.to_string_lossy();
         let path_str = if raw_path.contains('\\') {
             format!("'{raw_path}'")
         } else {
@@ -1445,13 +1716,12 @@ trust_level = "trusted"
 {path_str} = {{ trust_level = "untrusted" }}
 "#
         );
-        std::fs::create_dir_all(codex_home.path())?;
-        std::fs::write(&config_path, initial)?;
+        let mut doc = initial.parse::<DocumentMut>()?;
 
         // Run the function; it should convert to explicit tables and set trusted
-        set_project_trusted(codex_home.path(), project_dir.path())?;
+        set_project_trusted_inner(&mut doc, project_dir)?;
 
-        let contents = std::fs::read_to_string(&config_path)?;
+        let contents = doc.to_string();
 
         // Assert exact output after conversion to explicit table
         let expected = format!(
@@ -1465,4 +1735,38 @@ trust_level = "trusted"
 
         Ok(())
     }
+
+    #[test]
+    fn test_set_project_trusted_migrates_top_level_inline_projects_preserving_entries()
+    -> anyhow::Result<()> {
+        let initial = r#"toplevel = "baz"
+projects = { "/Users/mbolin/code/codex4" = { trust_level = "trusted", foo = "bar" } , "/Users/mbolin/code/codex3" = { trust_level = "trusted" } }
+model = "foo""#;
+        let mut doc = initial.parse::<DocumentMut>()?;
+
+        // Approve a new directory
+        let new_project = Path::new("/Users/mbolin/code/codex2");
+        set_project_trusted_inner(&mut doc, new_project)?;
+
+        let contents = doc.to_string();
+
+        // Since we created the [projects] table as part of migration, it is kept implicit.
+        // Expect explicit per-project tables, preserving prior entries and appending the new one.
+        let expected = r#"toplevel = "baz"
+model = "foo"
+
+[projects."/Users/mbolin/code/codex4"]
+trust_level = "trusted"
+foo = "bar"
+
+[projects."/Users/mbolin/code/codex3"]
+trust_level = "trusted"
+
+[projects."/Users/mbolin/code/codex2"]
+trust_level = "trusted"
+"#;
+        assert_eq!(contents, expected);
+
+        Ok(())
+    }
 }

codex-rs/core/src/config_edit.rs

@@ -7,6 +7,12 @@ use toml_edit::DocumentMut;
 pub const CONFIG_KEY_MODEL: &str = "model";
 pub const CONFIG_KEY_EFFORT: &str = "model_reasoning_effort";
 
+#[derive(Copy, Clone)]
+enum NoneBehavior {
+    Skip,
+    Remove,
+}
+
 /// Persist overrides into `config.toml` using explicit key segments per
 /// override. This avoids ambiguity with keys that contain dots or spaces.
 pub async fn persist_overrides(
@@ -14,47 +20,12 @@ pub async fn persist_overrides(
     profile: Option<&str>,
     overrides: &[(&[&str], &str)],
 ) -> Result<()> {
-    let config_path = codex_home.join(CONFIG_TOML_FILE);
+    let with_options: Vec<(&[&str], Option<&str>)> = overrides
+        .iter()
+        .map(|(segments, value)| (*segments, Some(*value)))
+        .collect();
 
-    let mut doc = match tokio::fs::read_to_string(&config_path).await {
-        Ok(s) => s.parse::<DocumentMut>()?,
-        Err(e) if e.kind() == std::io::ErrorKind::NotFound => {
-            tokio::fs::create_dir_all(codex_home).await?;
-            DocumentMut::new()
-        }
-        Err(e) => return Err(e.into()),
-    };
-
-    let effective_profile = if let Some(p) = profile {
-        Some(p.to_owned())
-    } else {
-        doc.get("profile")
-            .and_then(|i| i.as_str())
-            .map(|s| s.to_string())
-    };
-
-    for (segments, val) in overrides.iter().copied() {
-        let value = toml_edit::value(val);
-        if let Some(ref name) = effective_profile {
-            if segments.first().copied() == Some("profiles") {
-                apply_toml_edit_override_segments(&mut doc, segments, value);
-            } else {
-                let mut seg_buf: Vec<&str> = Vec::with_capacity(2 + segments.len());
-                seg_buf.push("profiles");
-                seg_buf.push(name.as_str());
-                seg_buf.extend_from_slice(segments);
-                apply_toml_edit_override_segments(&mut doc, &seg_buf, value);
-            }
-        } else {
-            apply_toml_edit_override_segments(&mut doc, segments, value);
-        }
-    }
-
-    let tmp_file = NamedTempFile::new_in(codex_home)?;
-    tokio::fs::write(tmp_file.path(), doc.to_string()).await?;
-    tmp_file.persist(config_path)?;
-
-    Ok(())
+    persist_overrides_with_behavior(codex_home, profile, &with_options, NoneBehavior::Skip).await
 }
 
 /// Persist overrides where values may be optional. Any entries with `None`
@@ -65,16 +36,17 @@ pub async fn persist_non_null_overrides(
     profile: Option<&str>,
     overrides: &[(&[&str], Option<&str>)],
 ) -> Result<()> {
-    let filtered: Vec<(&[&str], &str)> = overrides
-        .iter()
-        .filter_map(|(k, v)| v.map(|vv| (*k, vv)))
-        .collect();
+    persist_overrides_with_behavior(codex_home, profile, overrides, NoneBehavior::Skip).await
+}
 
-    if filtered.is_empty() {
-        return Ok(());
-    }
-
-    persist_overrides(codex_home, profile, &filtered).await
+/// Persist overrides where `None` values clear any existing values from the
+/// configuration file.
+pub async fn persist_overrides_and_clear_if_none(
+    codex_home: &Path,
+    profile: Option<&str>,
+    overrides: &[(&[&str], Option<&str>)],
+) -> Result<()> {
+    persist_overrides_with_behavior(codex_home, profile, overrides, NoneBehavior::Remove).await
 }
 
 /// Apply a single override onto a `toml_edit` document while preserving
@@ -121,6 +93,125 @@ fn apply_toml_edit_override_segments(
     current[last] = value;
 }
 
+async fn persist_overrides_with_behavior(
+    codex_home: &Path,
+    profile: Option<&str>,
+    overrides: &[(&[&str], Option<&str>)],
+    none_behavior: NoneBehavior,
+) -> Result<()> {
+    if overrides.is_empty() {
+        return Ok(());
+    }
+
+    let should_skip = match none_behavior {
+        NoneBehavior::Skip => overrides.iter().all(|(_, value)| value.is_none()),
+        NoneBehavior::Remove => false,
+    };
+
+    if should_skip {
+        return Ok(());
+    }
+
+    let config_path = codex_home.join(CONFIG_TOML_FILE);
+
+    let read_result = tokio::fs::read_to_string(&config_path).await;
+    let mut doc = match read_result {
+        Ok(contents) => contents.parse::<DocumentMut>()?,
+        Err(e) if e.kind() == std::io::ErrorKind::NotFound => {
+            if overrides
+                .iter()
+                .all(|(_, value)| value.is_none() && matches!(none_behavior, NoneBehavior::Remove))
+            {
+                return Ok(());
+            }
+
+            tokio::fs::create_dir_all(codex_home).await?;
+            DocumentMut::new()
+        }
+        Err(e) => return Err(e.into()),
+    };
+
+    let effective_profile = if let Some(p) = profile {
+        Some(p.to_owned())
+    } else {
+        doc.get("profile")
+            .and_then(|i| i.as_str())
+            .map(|s| s.to_string())
+    };
+
+    let mut mutated = false;
+
+    for (segments, value) in overrides.iter().copied() {
+        let mut seg_buf: Vec<&str> = Vec::new();
+        let segments_to_apply: &[&str];
+
+        if let Some(ref name) = effective_profile {
+            if segments.first().copied() == Some("profiles") {
+                segments_to_apply = segments;
+            } else {
+                seg_buf.reserve(2 + segments.len());
+                seg_buf.push("profiles");
+                seg_buf.push(name.as_str());
+                seg_buf.extend_from_slice(segments);
+                segments_to_apply = seg_buf.as_slice();
+            }
+        } else {
+            segments_to_apply = segments;
+        }
+
+        match value {
+            Some(v) => {
+                let item_value = toml_edit::value(v);
+                apply_toml_edit_override_segments(&mut doc, segments_to_apply, item_value);
+                mutated = true;
+            }
+            None => {
+                if matches!(none_behavior, NoneBehavior::Remove)
+                    && remove_toml_edit_segments(&mut doc, segments_to_apply)
+                {
+                    mutated = true;
+                }
+            }
+        }
+    }
+
+    if !mutated {
+        return Ok(());
+    }
+
+    let tmp_file = NamedTempFile::new_in(codex_home)?;
+    tokio::fs::write(tmp_file.path(), doc.to_string()).await?;
+    tmp_file.persist(config_path)?;
+
+    Ok(())
+}
+
+fn remove_toml_edit_segments(doc: &mut DocumentMut, segments: &[&str]) -> bool {
+    use toml_edit::Item;
+
+    if segments.is_empty() {
+        return false;
+    }
+
+    let mut current = doc.as_table_mut();
+    for seg in &segments[..segments.len() - 1] {
+        let Some(item) = current.get_mut(seg) else {
+            return false;
+        };
+
+        match item {
+            Item::Table(table) => {
+                current = table;
+            }
+            _ => {
+                return false;
+            }
+        }
+    }
+
+    current.remove(segments[segments.len() - 1]).is_some()
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -574,6 +665,81 @@ model = "o3"
         assert_eq!(contents, expected);
     }
 
+    #[tokio::test]
+    async fn persist_clear_none_removes_top_level_value() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        let seed = r#"model = "gpt-5"
+model_reasoning_effort = "medium"
+"#;
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        persist_overrides_and_clear_if_none(
+            codex_home,
+            None,
+            &[
+                (&[CONFIG_KEY_MODEL], None),
+                (&[CONFIG_KEY_EFFORT], Some("high")),
+            ],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = "model_reasoning_effort = \"high\"\n";
+        assert_eq!(contents, expected);
+    }
+
+    #[tokio::test]
+    async fn persist_clear_none_respects_active_profile() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        let seed = r#"profile = "team"
+
+[profiles.team]
+model = "gpt-4"
+model_reasoning_effort = "minimal"
+"#;
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        persist_overrides_and_clear_if_none(
+            codex_home,
+            None,
+            &[
+                (&[CONFIG_KEY_MODEL], None),
+                (&[CONFIG_KEY_EFFORT], Some("high")),
+            ],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"profile = "team"
+
+[profiles.team]
+model_reasoning_effort = "high"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    #[tokio::test]
+    async fn persist_clear_none_noop_when_file_missing() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        persist_overrides_and_clear_if_none(codex_home, None, &[(&[CONFIG_KEY_MODEL], None)])
+            .await
+            .expect("persist");
+
+        assert!(!codex_home.join(CONFIG_TOML_FILE).exists());
+    }
+
     // Test helper moved to bottom per review guidance.
     async fn read_config(codex_home: &Path) -> String {
         let p = codex_home.join(CONFIG_TOML_FILE);

codex-rs/core/src/conversation_history.rs

@@ -32,32 +32,8 @@ impl ConversationHistory {
         }
     }
 
-    pub(crate) fn keep_last_messages(&mut self, n: usize) {
-        if n == 0 {
-            self.items.clear();
-            return;
-        }
-
-        // Collect the last N message items (assistant/user), newest to oldest.
-        let mut kept: Vec<ResponseItem> = Vec::with_capacity(n);
-        for item in self.items.iter().rev() {
-            if let ResponseItem::Message { role, content, .. } = item {
-                kept.push(ResponseItem::Message {
-                    // we need to remove the id or the model will complain that messages are sent without
-                    // their reasonings
-                    id: None,
-                    role: role.clone(),
-                    content: content.clone(),
-                });
-                if kept.len() == n {
-                    break;
-                }
-            }
-        }
-
-        // Preserve chronological order (oldest to newest) within the kept slice.
-        kept.reverse();
-        self.items = kept;
+    pub(crate) fn replace(&mut self, items: Vec<ResponseItem>) {
+        self.items = items;
     }
 }
 

codex-rs/core/src/conversation_manager.rs

@@ -59,21 +59,11 @@ impl ConversationManager {
         config: Config,
         auth_manager: Arc<AuthManager>,
     ) -> CodexResult<NewConversation> {
-        // TO BE REFACTORED: use the config experimental_resume field until we have a mainstream way.
-        if let Some(resume_path) = config.experimental_resume.as_ref() {
-            let initial_history = RolloutRecorder::get_rollout_history(resume_path).await?;
-            let CodexSpawnOk {
-                codex,
-                conversation_id,
-            } = Codex::spawn(config, auth_manager, initial_history).await?;
-            self.finalize_spawn(codex, conversation_id).await
-        } else {
-            let CodexSpawnOk {
-                codex,
-                conversation_id,
-            } = Codex::spawn(config, auth_manager, InitialHistory::New).await?;
-            self.finalize_spawn(codex, conversation_id).await
-        }
+        let CodexSpawnOk {
+            codex,
+            conversation_id,
+        } = Codex::spawn(config, auth_manager, InitialHistory::New).await?;
+        self.finalize_spawn(codex, conversation_id).await
     }
 
     async fn finalize_spawn(
@@ -144,19 +134,19 @@ impl ConversationManager {
         self.conversations.write().await.remove(conversation_id)
     }
 
-    /// Fork an existing conversation by dropping the last `drop_last_messages`
-    /// user/assistant messages from its transcript and starting a new
+    /// Fork an existing conversation by taking messages up to the given position
+    /// (not including the message at the given position) and starting a new
     /// conversation with identical configuration (unless overridden by the
     /// caller's `config`). The new conversation will have a fresh id.
     pub async fn fork_conversation(
         &self,
-        conversation_history: Vec<ResponseItem>,
-        num_messages_to_drop: usize,
+        nth_user_message: usize,
         config: Config,
+        path: PathBuf,
     ) -> CodexResult<NewConversation> {
         // Compute the prefix up to the cut point.
-        let history =
-            truncate_after_dropping_last_messages(conversation_history, num_messages_to_drop);
+        let history = RolloutRecorder::get_rollout_history(&path).await?;
+        let history = truncate_after_nth_user_message(history, nth_user_message);
 
         // Spawn a new conversation with the computed initial history.
         let auth_manager = self.auth_manager.clone();
@@ -169,38 +159,34 @@ impl ConversationManager {
     }
 }
 
-/// Return a prefix of `items` obtained by dropping the last `n` user messages
-/// and all items that follow them.
-fn truncate_after_dropping_last_messages(items: Vec<ResponseItem>, n: usize) -> InitialHistory {
-    if n == 0 {
-        let rolled: Vec<RolloutItem> = items.into_iter().map(RolloutItem::ResponseItem).collect();
-        return InitialHistory::Forked(rolled);
-    }
+/// Return a prefix of `items` obtained by cutting strictly before the nth user message
+/// (0-based) and all items that follow it.
+fn truncate_after_nth_user_message(history: InitialHistory, n: usize) -> InitialHistory {
+    // Work directly on rollout items, and cut the vector at the nth user message input.
+    let items: Vec<RolloutItem> = history.get_rollout_items();
 
-    // Walk backwards counting only `user` Message items, find cut index.
-    let mut count = 0usize;
-    let mut cut_index = 0usize;
-    for (idx, item) in items.iter().enumerate().rev() {
-        if let ResponseItem::Message { role, .. } = item
+    // Find indices of user message inputs in rollout order.
+    let mut user_positions: Vec<usize> = Vec::new();
+    for (idx, item) in items.iter().enumerate() {
+        if let RolloutItem::ResponseItem(ResponseItem::Message { role, .. }) = item
             && role == "user"
         {
-            count += 1;
-            if count == n {
-                // Cut everything from this user message to the end.
-                cut_index = idx;
-                break;
-            }
+            user_positions.push(idx);
         }
     }
-    if cut_index == 0 {
-        // No prefix remains after dropping; start a new conversation.
+
+    // If fewer than or equal to n user messages exist, treat as empty (out of range).
+    if user_positions.len() <= n {
+        return InitialHistory::New;
+    }
+
+    // Cut strictly before the nth user message (do not keep the nth itself).
+    let cut_idx = user_positions[n];
+    let rolled: Vec<RolloutItem> = items.into_iter().take(cut_idx).collect();
+
+    if rolled.is_empty() {
         InitialHistory::New
     } else {
-        let rolled: Vec<RolloutItem> = items
-            .into_iter()
-            .take(cut_index)
-            .map(RolloutItem::ResponseItem)
-            .collect();
         InitialHistory::Forked(rolled)
     }
 }
@@ -256,7 +242,13 @@ mod tests {
             assistant_msg("a4"),
         ];
 
-        let truncated = truncate_after_dropping_last_messages(items.clone(), 1);
+        // Wrap as InitialHistory::Forked with response items only.
+        let initial: Vec<RolloutItem> = items
+            .iter()
+            .cloned()
+            .map(RolloutItem::ResponseItem)
+            .collect();
+        let truncated = truncate_after_nth_user_message(InitialHistory::Forked(initial), 1);
         let got_items = truncated.get_rollout_items();
         let expected_items = vec![
             RolloutItem::ResponseItem(items[0].clone()),
@@ -268,7 +260,12 @@ mod tests {
             serde_json::to_value(&expected_items).unwrap()
         );
 
-        let truncated2 = truncate_after_dropping_last_messages(items, 2);
+        let initial2: Vec<RolloutItem> = items
+            .iter()
+            .cloned()
+            .map(RolloutItem::ResponseItem)
+            .collect();
+        let truncated2 = truncate_after_nth_user_message(InitialHistory::Forked(initial2), 2);
         assert!(matches!(truncated2, InitialHistory::New));
     }
 }

codex-rs/core/src/environment_context.rs

@@ -26,6 +26,7 @@ pub(crate) struct EnvironmentContext {
     pub approval_policy: Option<AskForApproval>,
     pub sandbox_mode: Option<SandboxMode>,
     pub network_access: Option<NetworkAccess>,
+    pub writable_roots: Option<Vec<PathBuf>>,
     pub shell: Option<Shell>,
 }
 
@@ -57,6 +58,16 @@ impl EnvironmentContext {
                 }
                 None => None,
             },
+            writable_roots: match sandbox_policy {
+                Some(SandboxPolicy::WorkspaceWrite { writable_roots, .. }) => {
+                    if writable_roots.is_empty() {
+                        None
+                    } else {
+                        Some(writable_roots)
+                    }
+                }
+                _ => None,
+            },
             shell,
         }
     }
@@ -72,6 +83,7 @@ impl EnvironmentContext {
     ///   <cwd>...</cwd>
     ///   <approval_policy>...</approval_policy>
     ///   <sandbox_mode>...</sandbox_mode>
+    ///   <writable_roots>...</writable_roots>
     ///   <network_access>...</network_access>
     ///   <shell>...</shell>
     /// </environment_context>
@@ -94,6 +106,16 @@ impl EnvironmentContext {
                 "  <network_access>{network_access}</network_access>"
             ));
         }
+        if let Some(writable_roots) = self.writable_roots {
+            lines.push("  <writable_roots>".to_string());
+            for writable_root in writable_roots {
+                lines.push(format!(
+                    "    <root>{}</root>",
+                    writable_root.to_string_lossy()
+                ));
+            }
+            lines.push("  </writable_roots>".to_string());
+        }
         if let Some(shell) = self.shell
             && let Some(shell_name) = shell.name()
         {
@@ -115,3 +137,77 @@ impl From<EnvironmentContext> for ResponseItem {
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use pretty_assertions::assert_eq;
+
+    fn workspace_write_policy(writable_roots: Vec<&str>, network_access: bool) -> SandboxPolicy {
+        SandboxPolicy::WorkspaceWrite {
+            writable_roots: writable_roots.into_iter().map(PathBuf::from).collect(),
+            network_access,
+            exclude_tmpdir_env_var: false,
+            exclude_slash_tmp: false,
+        }
+    }
+
+    #[test]
+    fn serialize_workspace_write_environment_context() {
+        let context = EnvironmentContext::new(
+            Some(PathBuf::from("/repo")),
+            Some(AskForApproval::OnRequest),
+            Some(workspace_write_policy(vec!["/repo", "/tmp"], false)),
+            None,
+        );
+
+        let expected = r#"<environment_context>
+  <cwd>/repo</cwd>
+  <approval_policy>on-request</approval_policy>
+  <sandbox_mode>workspace-write</sandbox_mode>
+  <network_access>restricted</network_access>
+  <writable_roots>
+    <root>/repo</root>
+    <root>/tmp</root>
+  </writable_roots>
+</environment_context>"#;
+
+        assert_eq!(context.serialize_to_xml(), expected);
+    }
+
+    #[test]
+    fn serialize_read_only_environment_context() {
+        let context = EnvironmentContext::new(
+            None,
+            Some(AskForApproval::Never),
+            Some(SandboxPolicy::ReadOnly),
+            None,
+        );
+
+        let expected = r#"<environment_context>
+  <approval_policy>never</approval_policy>
+  <sandbox_mode>read-only</sandbox_mode>
+  <network_access>restricted</network_access>
+</environment_context>"#;
+
+        assert_eq!(context.serialize_to_xml(), expected);
+    }
+
+    #[test]
+    fn serialize_full_access_environment_context() {
+        let context = EnvironmentContext::new(
+            None,
+            Some(AskForApproval::OnFailure),
+            Some(SandboxPolicy::DangerFullAccess),
+            None,
+        );
+
+        let expected = r#"<environment_context>
+  <approval_policy>on-failure</approval_policy>
+  <sandbox_mode>danger-full-access</sandbox_mode>
+  <network_access>enabled</network_access>
+</environment_context>"#;
+
+        assert_eq!(context.serialize_to_xml(), expected);
+    }
+}

codex-rs/core/src/error.rs

@@ -1,3 +1,6 @@
+use crate::exec::ExecToolCallOutput;
+use crate::token_data::KnownPlan;
+use crate::token_data::PlanType;
 use codex_protocol::mcp_protocol::ConversationId;
 use reqwest::StatusCode;
 use serde_json;
@@ -11,8 +14,11 @@ pub type Result<T> = std::result::Result<T, CodexErr>;
 #[derive(Error, Debug)]
 pub enum SandboxErr {
     /// Error from sandbox execution
-    #[error("sandbox denied exec error, exit code: {0}, stdout: {1}, stderr: {2}")]
-    Denied(i32, String, String),
+    #[error(
+        "sandbox denied exec error, exit code: {}, stdout: {}, stderr: {}",
+        .output.exit_code, .output.stdout.text, .output.stderr.text
+    )]
+    Denied { output: Box<ExecToolCallOutput> },
 
     /// Error from linux seccomp filter setup
     #[cfg(target_os = "linux")]
@@ -26,7 +32,7 @@ pub enum SandboxErr {
 
     /// Command timed out
     #[error("command timed out")]
-    Timeout,
+    Timeout { output: Box<ExecToolCallOutput> },
 
     /// Command was killed by a signal
     #[error("command was killed by a signal")]
@@ -127,38 +133,58 @@ pub enum CodexErr {
 
 #[derive(Debug)]
 pub struct UsageLimitReachedError {
-    pub plan_type: Option<String>,
-    pub resets_in_seconds: Option<u64>,
+    pub(crate) plan_type: Option<PlanType>,
+    pub(crate) resets_in_seconds: Option<u64>,
 }
 
 impl std::fmt::Display for UsageLimitReachedError {
     fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
-        // Base message differs slightly for legacy ChatGPT Plus plan users.
-        if let Some(plan_type) = &self.plan_type
-            && plan_type == "plus"
-        {
-            write!(
-                f,
-                "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing) or try again"
-            )?;
-            if let Some(secs) = self.resets_in_seconds {
-                let reset_duration = format_reset_duration(secs);
-                write!(f, " in {reset_duration}.")?;
-            } else {
-                write!(f, " later.")?;
+        let message = match self.plan_type.as_ref() {
+            Some(PlanType::Known(KnownPlan::Plus)) => format!(
+                "You've hit your usage limit. Upgrade to Pro (https://openai.com/chatgpt/pricing){}",
+                retry_suffix_after_or(self.resets_in_seconds)
+            ),
+            Some(PlanType::Known(KnownPlan::Team)) | Some(PlanType::Known(KnownPlan::Business)) => {
+                format!(
+                    "You've hit your usage limit. To get more access now, send a request to your admin{}",
+                    retry_suffix_after_or(self.resets_in_seconds)
+                )
             }
-        } else {
-            write!(f, "You've hit your usage limit.")?;
-
-            if let Some(secs) = self.resets_in_seconds {
-                let reset_duration = format_reset_duration(secs);
-                write!(f, " Try again in {reset_duration}.")?;
-            } else {
-                write!(f, " Try again later.")?;
+            Some(PlanType::Known(KnownPlan::Free)) => {
+                "To use Codex with your ChatGPT plan, upgrade to Plus: https://openai.com/chatgpt/pricing."
+                    .to_string()
             }
-        }
+            Some(PlanType::Known(KnownPlan::Pro))
+            | Some(PlanType::Known(KnownPlan::Enterprise))
+            | Some(PlanType::Known(KnownPlan::Edu)) => format!(
+                "You've hit your usage limit.{}",
+                retry_suffix(self.resets_in_seconds)
+            ),
+            Some(PlanType::Unknown(_)) | None => format!(
+                "You've hit your usage limit.{}",
+                retry_suffix(self.resets_in_seconds)
+            ),
+        };
 
-        Ok(())
+        write!(f, "{message}")
+    }
+}
+
+fn retry_suffix(resets_in_seconds: Option<u64>) -> String {
+    if let Some(secs) = resets_in_seconds {
+        let reset_duration = format_reset_duration(secs);
+        format!(" Try again in {reset_duration}.")
+    } else {
+        " Try again later.".to_string()
+    }
+}
+
+fn retry_suffix_after_or(resets_in_seconds: Option<u64>) -> String {
+    if let Some(secs) = resets_in_seconds {
+        let reset_duration = format_reset_duration(secs);
+        format!(" or try again in {reset_duration}.")
+    } else {
+        " or try again later.".to_string()
     }
 }
 
@@ -223,9 +249,12 @@ impl CodexErr {
 
 pub fn get_error_message_ui(e: &CodexErr) -> String {
     match e {
-        CodexErr::Sandbox(SandboxErr::Denied(_, _, stderr)) => stderr.to_string(),
+        CodexErr::Sandbox(SandboxErr::Denied { output }) => output.stderr.text.clone(),
         // Timeouts are not sandbox errors from a UX perspective; present them plainly
-        CodexErr::Sandbox(SandboxErr::Timeout) => "error: command timed out".to_string(),
+        CodexErr::Sandbox(SandboxErr::Timeout { output }) => format!(
+            "error: command timed out after {} ms",
+            output.duration.as_millis()
+        ),
         _ => e.to_string(),
     }
 }
@@ -237,7 +266,7 @@ mod tests {
     #[test]
     fn usage_limit_reached_error_formats_plus_plan() {
         let err = UsageLimitReachedError {
-            plan_type: Some("plus".to_string()),
+            plan_type: Some(PlanType::Known(KnownPlan::Plus)),
             resets_in_seconds: None,
         };
         assert_eq!(
@@ -246,6 +275,18 @@ mod tests {
         );
     }
 
+    #[test]
+    fn usage_limit_reached_error_formats_free_plan() {
+        let err = UsageLimitReachedError {
+            plan_type: Some(PlanType::Known(KnownPlan::Free)),
+            resets_in_seconds: Some(3600),
+        };
+        assert_eq!(
+            err.to_string(),
+            "To use Codex with your ChatGPT plan, upgrade to Plus: https://openai.com/chatgpt/pricing."
+        );
+    }
+
     #[test]
     fn usage_limit_reached_error_formats_default_when_none() {
         let err = UsageLimitReachedError {
@@ -258,10 +299,34 @@ mod tests {
         );
     }
 
+    #[test]
+    fn usage_limit_reached_error_formats_team_plan() {
+        let err = UsageLimitReachedError {
+            plan_type: Some(PlanType::Known(KnownPlan::Team)),
+            resets_in_seconds: Some(3600),
+        };
+        assert_eq!(
+            err.to_string(),
+            "You've hit your usage limit. To get more access now, send a request to your admin or try again in 1 hour."
+        );
+    }
+
+    #[test]
+    fn usage_limit_reached_error_formats_business_plan_without_reset() {
+        let err = UsageLimitReachedError {
+            plan_type: Some(PlanType::Known(KnownPlan::Business)),
+            resets_in_seconds: None,
+        };
+        assert_eq!(
+            err.to_string(),
+            "You've hit your usage limit. To get more access now, send a request to your admin or try again later."
+        );
+    }
+
     #[test]
     fn usage_limit_reached_error_formats_default_for_other_plans() {
         let err = UsageLimitReachedError {
-            plan_type: Some("pro".to_string()),
+            plan_type: Some(PlanType::Known(KnownPlan::Pro)),
             resets_in_seconds: None,
         };
         assert_eq!(
@@ -285,7 +350,7 @@ mod tests {
     #[test]
     fn usage_limit_reached_includes_hours_and_minutes() {
         let err = UsageLimitReachedError {
-            plan_type: Some("plus".to_string()),
+            plan_type: Some(PlanType::Known(KnownPlan::Plus)),
             resets_in_seconds: Some(3 * 3600 + 32 * 60),
         };
         assert_eq!(

codex-rs/core/src/event_mapping.rs

@@ -159,7 +159,7 @@ mod tests {
             EventMsg::UserMessage(user) => {
                 assert_eq!(user.message, "Hello world");
                 assert!(matches!(user.kind, Some(InputMessageKind::Plain)));
-                assert_eq!(user.images, Some(vec![img1.clone(), img2.clone()]));
+                assert_eq!(user.images, Some(vec![img1, img2]));
             }
             other => panic!("expected UserMessage, got {other:?}"),
         }

codex-rs/core/src/exec.rs

@@ -34,6 +34,7 @@ const DEFAULT_TIMEOUT_MS: u64 = 10_000;
 const SIGKILL_CODE: i32 = 9;
 const TIMEOUT_CODE: i32 = 64;
 const EXIT_CODE_SIGNAL_BASE: i32 = 128; // conventional shell: 128 + signal
+const EXEC_TIMEOUT_EXIT_CODE: i32 = 124; // conventional timeout exit code
 
 // I/O buffer sizing
 const READ_CHUNK_SIZE: usize = 8192; // bytes per read
@@ -86,11 +87,12 @@ pub async fn process_exec_tool_call(
 ) -> Result<ExecToolCallOutput> {
     let start = Instant::now();
 
+    let timeout_duration = params.timeout_duration();
+
     let raw_output_result: std::result::Result<RawExecToolCallOutput, CodexErr> = match sandbox_type
     {
         SandboxType::None => exec(params, sandbox_policy, stdout_stream.clone()).await,
         SandboxType::MacosSeatbelt => {
-            let timeout = params.timeout_duration();
             let ExecParams {
                 command, cwd, env, ..
             } = params;
@@ -102,10 +104,9 @@ pub async fn process_exec_tool_call(
                 env,
             )
             .await?;
-            consume_truncated_output(child, timeout, stdout_stream.clone()).await
+            consume_truncated_output(child, timeout_duration, stdout_stream.clone()).await
         }
         SandboxType::LinuxSeccomp => {
-            let timeout = params.timeout_duration();
             let ExecParams {
                 command, cwd, env, ..
             } = params;
@@ -123,41 +124,56 @@ pub async fn process_exec_tool_call(
             )
             .await?;
 
-            consume_truncated_output(child, timeout, stdout_stream).await
+            consume_truncated_output(child, timeout_duration, stdout_stream).await
         }
     };
     let duration = start.elapsed();
     match raw_output_result {
         Ok(raw_output) => {
-            let stdout = raw_output.stdout.from_utf8_lossy();
-            let stderr = raw_output.stderr.from_utf8_lossy();
+            #[allow(unused_mut)]
+            let mut timed_out = raw_output.timed_out;
 
             #[cfg(target_family = "unix")]
-            match raw_output.exit_status.signal() {
-                Some(TIMEOUT_CODE) => return Err(CodexErr::Sandbox(SandboxErr::Timeout)),
-                Some(signal) => {
-                    return Err(CodexErr::Sandbox(SandboxErr::Signal(signal)));
+            {
+                if let Some(signal) = raw_output.exit_status.signal() {
+                    if signal == TIMEOUT_CODE {
+                        timed_out = true;
+                    } else {
+                        return Err(CodexErr::Sandbox(SandboxErr::Signal(signal)));
+                    }
                 }
-                None => {}
             }
 
-            let exit_code = raw_output.exit_status.code().unwrap_or(-1);
-
-            if exit_code != 0 && is_likely_sandbox_denied(sandbox_type, exit_code) {
-                return Err(CodexErr::Sandbox(SandboxErr::Denied(
-                    exit_code,
-                    stdout.text,
-                    stderr.text,
-                )));
+            let mut exit_code = raw_output.exit_status.code().unwrap_or(-1);
+            if timed_out {
+                exit_code = EXEC_TIMEOUT_EXIT_CODE;
             }
 
-            Ok(ExecToolCallOutput {
+            let stdout = raw_output.stdout.from_utf8_lossy();
+            let stderr = raw_output.stderr.from_utf8_lossy();
+            let aggregated_output = raw_output.aggregated_output.from_utf8_lossy();
+            let exec_output = ExecToolCallOutput {
                 exit_code,
                 stdout,
                 stderr,
-                aggregated_output: raw_output.aggregated_output.from_utf8_lossy(),
+                aggregated_output,
                 duration,
-            })
+                timed_out,
+            };
+
+            if timed_out {
+                return Err(CodexErr::Sandbox(SandboxErr::Timeout {
+                    output: Box::new(exec_output),
+                }));
+            }
+
+            if exit_code != 0 && is_likely_sandbox_denied(sandbox_type, exit_code) {
+                return Err(CodexErr::Sandbox(SandboxErr::Denied {
+                    output: Box::new(exec_output),
+                }));
+            }
+
+            Ok(exec_output)
         }
         Err(err) => {
             tracing::error!("exec error: {err}");
@@ -197,6 +213,7 @@ struct RawExecToolCallOutput {
     pub stdout: StreamOutput<Vec<u8>>,
     pub stderr: StreamOutput<Vec<u8>>,
     pub aggregated_output: StreamOutput<Vec<u8>>,
+    pub timed_out: bool,
 }
 
 impl StreamOutput<String> {
@@ -229,6 +246,7 @@ pub struct ExecToolCallOutput {
     pub stderr: StreamOutput<String>,
     pub aggregated_output: StreamOutput<String>,
     pub duration: Duration,
+    pub timed_out: bool,
 }
 
 async fn exec(
@@ -298,22 +316,24 @@ async fn consume_truncated_output(
         Some(agg_tx.clone()),
     ));
 
-    let exit_status = tokio::select! {
+    let (exit_status, timed_out) = tokio::select! {
         result = tokio::time::timeout(timeout, child.wait()) => {
             match result {
-                Ok(Ok(exit_status)) => exit_status,
-                Ok(e) => e?,
+                Ok(status_result) => {
+                    let exit_status = status_result?;
+                    (exit_status, false)
+                }
                 Err(_) => {
                     // timeout
                     child.start_kill()?;
                     // Debatable whether `child.wait().await` should be called here.
-                    synthetic_exit_status(EXIT_CODE_SIGNAL_BASE + TIMEOUT_CODE)
+                    (synthetic_exit_status(EXIT_CODE_SIGNAL_BASE + TIMEOUT_CODE), true)
                 }
             }
         }
         _ = tokio::signal::ctrl_c() => {
             child.start_kill()?;
-            synthetic_exit_status(EXIT_CODE_SIGNAL_BASE + SIGKILL_CODE)
+            (synthetic_exit_status(EXIT_CODE_SIGNAL_BASE + SIGKILL_CODE), false)
         }
     };
 
@@ -336,6 +356,7 @@ async fn consume_truncated_output(
         stdout,
         stderr,
         aggregated_output,
+        timed_out,
     })
 }
 

codex-rs/core/src/exec_command/exec_command_session.rs

@@ -11,6 +11,9 @@ pub(crate) struct ExecCommandSession {
     /// Broadcast stream of output chunks read from the PTY. New subscribers
     /// receive only chunks emitted after they subscribe.
     output_tx: broadcast::Sender<Vec<u8>>,
+    /// Receiver subscribed before the child process starts emitting output so
+    /// the first caller can consume any early data without races.
+    initial_output_rx: StdMutex<Option<broadcast::Receiver<Vec<u8>>>>,
 
     /// Child killer handle for termination on drop (can signal independently
     /// of a thread blocked in `.wait()`).
@@ -24,6 +27,9 @@ pub(crate) struct ExecCommandSession {
 
     /// JoinHandle for the child wait task.
     wait_handle: StdMutex<Option<JoinHandle<()>>>,
+
+    /// Tracks whether the underlying process has exited.
+    exit_status: std::sync::Arc<std::sync::atomic::AtomicBool>,
 }
 
 impl ExecCommandSession {
@@ -34,14 +40,25 @@ impl ExecCommandSession {
         reader_handle: JoinHandle<()>,
         writer_handle: JoinHandle<()>,
         wait_handle: JoinHandle<()>,
+        exit_status: std::sync::Arc<std::sync::atomic::AtomicBool>,
     ) -> Self {
         Self {
             writer_tx,
             output_tx,
+            initial_output_rx: StdMutex::new(None),
             killer: StdMutex::new(Some(killer)),
             reader_handle: StdMutex::new(Some(reader_handle)),
             writer_handle: StdMutex::new(Some(writer_handle)),
             wait_handle: StdMutex::new(Some(wait_handle)),
+            exit_status,
+        }
+    }
+
+    pub(crate) fn set_initial_output_receiver(&self, receiver: broadcast::Receiver<Vec<u8>>) {
+        if let Ok(mut guard) = self.initial_output_rx.lock()
+            && guard.is_none()
+        {
+            *guard = Some(receiver);
         }
     }
 
@@ -50,7 +67,17 @@ impl ExecCommandSession {
     }
 
     pub(crate) fn output_receiver(&self) -> broadcast::Receiver<Vec<u8>> {
-        self.output_tx.subscribe()
+        if let Ok(mut guard) = self.initial_output_rx.lock()
+            && let Some(receiver) = guard.take()
+        {
+            receiver
+        } else {
+            self.output_tx.subscribe()
+        }
+    }
+
+    pub(crate) fn has_exited(&self) -> bool {
+        self.exit_status.load(std::sync::atomic::Ordering::SeqCst)
     }
 }
 

codex-rs/core/src/exec_command/mod.rs

@@ -6,6 +6,7 @@ mod session_manager;
 
 pub use exec_command_params::ExecCommandParams;
 pub use exec_command_params::WriteStdinParams;
+pub(crate) use exec_command_session::ExecCommandSession;
 pub use responses_api::EXEC_COMMAND_TOOL_NAME;
 pub use responses_api::WRITE_STDIN_TOOL_NAME;
 pub use responses_api::create_exec_command_tool_for_responses_api;

codex-rs/core/src/exec_command/session_manager.rs

@@ -3,6 +3,7 @@ use std::io::ErrorKind;
 use std::io::Read;
 use std::sync::Arc;
 use std::sync::Mutex as StdMutex;
+use std::sync::atomic::AtomicBool;
 use std::sync::atomic::AtomicU32;
 
 use portable_pty::CommandBuilder;
@@ -19,6 +20,7 @@ use crate::exec_command::exec_command_params::ExecCommandParams;
 use crate::exec_command::exec_command_params::WriteStdinParams;
 use crate::exec_command::exec_command_session::ExecCommandSession;
 use crate::exec_command::session_id::SessionId;
+use crate::truncate::truncate_middle;
 use codex_protocol::models::FunctionCallOutputPayload;
 
 #[derive(Debug, Default)]
@@ -277,6 +279,7 @@ async fn create_exec_command_session(
     let (writer_tx, mut writer_rx) = mpsc::channel::<Vec<u8>>(128);
     // Broadcast for streaming PTY output to readers: subscribers receive from subscription time.
     let (output_tx, _) = tokio::sync::broadcast::channel::<Vec<u8>>(256);
+    let initial_output_rx = output_tx.subscribe();
 
     // Reader task: drain PTY and forward chunks to output channel.
     let mut reader = pair.master.try_clone_reader()?;
@@ -327,11 +330,14 @@ async fn create_exec_command_session(
 
     // Keep the child alive until it exits, then signal exit code.
     let (exit_tx, exit_rx) = oneshot::channel::<i32>();
+    let exit_status = Arc::new(AtomicBool::new(false));
+    let wait_exit_status = exit_status.clone();
     let wait_handle = tokio::task::spawn_blocking(move || {
         let code = match child.wait() {
             Ok(status) => status.exit_code() as i32,
             Err(_) => -1,
         };
+        wait_exit_status.store(true, std::sync::atomic::Ordering::SeqCst);
         let _ = exit_tx.send(code);
     });
 
@@ -343,116 +349,12 @@ async fn create_exec_command_session(
         reader_handle,
         writer_handle,
         wait_handle,
+        exit_status,
     );
+    session.set_initial_output_receiver(initial_output_rx);
     Ok((session, exit_rx))
 }
 
-/// Truncate the middle of a UTF-8 string to at most `max_bytes` bytes,
-/// preserving the beginning and the end. Returns the possibly truncated
-/// string and `Some(original_token_count)` (estimated at 4 bytes/token)
-/// if truncation occurred; otherwise returns the original string and `None`.
-fn truncate_middle(s: &str, max_bytes: usize) -> (String, Option<u64>) {
-    // No truncation needed
-    if s.len() <= max_bytes {
-        return (s.to_string(), None);
-    }
-    let est_tokens = (s.len() as u64).div_ceil(4);
-    if max_bytes == 0 {
-        // Cannot keep any content; still return a full marker (never truncated).
-        return (format!("…{est_tokens} tokens truncated…"), Some(est_tokens));
-    }
-
-    // Helper to truncate a string to a given byte length on a char boundary.
-    fn truncate_on_boundary(input: &str, max_len: usize) -> &str {
-        if input.len() <= max_len {
-            return input;
-        }
-        let mut end = max_len;
-        while end > 0 && !input.is_char_boundary(end) {
-            end -= 1;
-        }
-        &input[..end]
-    }
-
-    // Given a left/right budget, prefer newline boundaries; otherwise fall back
-    // to UTF-8 char boundaries.
-    fn pick_prefix_end(s: &str, left_budget: usize) -> usize {
-        if let Some(head) = s.get(..left_budget)
-            && let Some(i) = head.rfind('\n')
-        {
-            return i + 1; // keep the newline so suffix starts on a fresh line
-        }
-        truncate_on_boundary(s, left_budget).len()
-    }
-
-    fn pick_suffix_start(s: &str, right_budget: usize) -> usize {
-        let start_tail = s.len().saturating_sub(right_budget);
-        if let Some(tail) = s.get(start_tail..)
-            && let Some(i) = tail.find('\n')
-        {
-            return start_tail + i + 1; // start after newline
-        }
-        // Fall back to a char boundary at or after start_tail.
-        let mut idx = start_tail.min(s.len());
-        while idx < s.len() && !s.is_char_boundary(idx) {
-            idx += 1;
-        }
-        idx
-    }
-
-    // Refine marker length and budgets until stable. Marker is never truncated.
-    let mut guess_tokens = est_tokens; // worst-case: everything truncated
-    for _ in 0..4 {
-        let marker = format!("…{guess_tokens} tokens truncated…");
-        let marker_len = marker.len();
-        let keep_budget = max_bytes.saturating_sub(marker_len);
-        if keep_budget == 0 {
-            // No room for any content within the cap; return a full, untruncated marker
-            // that reflects the entire truncated content.
-            return (format!("…{est_tokens} tokens truncated…"), Some(est_tokens));
-        }
-
-        let left_budget = keep_budget / 2;
-        let right_budget = keep_budget - left_budget;
-        let prefix_end = pick_prefix_end(s, left_budget);
-        let mut suffix_start = pick_suffix_start(s, right_budget);
-        if suffix_start < prefix_end {
-            suffix_start = prefix_end;
-        }
-        let kept_content_bytes = prefix_end + (s.len() - suffix_start);
-        let truncated_content_bytes = s.len().saturating_sub(kept_content_bytes);
-        let new_tokens = (truncated_content_bytes as u64).div_ceil(4);
-        if new_tokens == guess_tokens {
-            let mut out = String::with_capacity(marker_len + kept_content_bytes + 1);
-            out.push_str(&s[..prefix_end]);
-            out.push_str(&marker);
-            // Place marker on its own line for symmetry when we keep line boundaries.
-            out.push('\n');
-            out.push_str(&s[suffix_start..]);
-            return (out, Some(est_tokens));
-        }
-        guess_tokens = new_tokens;
-    }
-
-    // Fallback: use last guess to build output.
-    let marker = format!("…{guess_tokens} tokens truncated…");
-    let marker_len = marker.len();
-    let keep_budget = max_bytes.saturating_sub(marker_len);
-    if keep_budget == 0 {
-        return (format!("…{est_tokens} tokens truncated…"), Some(est_tokens));
-    }
-    let left_budget = keep_budget / 2;
-    let right_budget = keep_budget - left_budget;
-    let prefix_end = pick_prefix_end(s, left_budget);
-    let suffix_start = pick_suffix_start(s, right_budget);
-    let mut out = String::with_capacity(marker_len + prefix_end + (s.len() - suffix_start) + 1);
-    out.push_str(&s[..prefix_end]);
-    out.push_str(&marker);
-    out.push('\n');
-    out.push_str(&s[suffix_start..]);
-    (out, Some(est_tokens))
-}
-
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -616,50 +518,4 @@ Output:
 abc"#;
         assert_eq!(expected, text);
     }
-
-    #[test]
-    fn truncate_middle_no_newlines_fallback() {
-        // A long string with no newlines that exceeds the cap.
-        let s = "abcdefghijklmnopqrstuvwxyz0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
-        let max_bytes = 16; // force truncation
-        let (out, original) = truncate_middle(s, max_bytes);
-        // For very small caps, we return the full, untruncated marker,
-        // even if it exceeds the cap.
-        assert_eq!(out, "…16 tokens truncated…");
-        // Original string length is 62 bytes => ceil(62/4) = 16 tokens.
-        assert_eq!(original, Some(16));
-    }
-
-    #[test]
-    fn truncate_middle_prefers_newline_boundaries() {
-        // Build a multi-line string of 20 numbered lines (each "NNN\n").
-        let mut s = String::new();
-        for i in 1..=20 {
-            s.push_str(&format!("{i:03}\n"));
-        }
-        // Total length: 20 lines * 4 bytes per line = 80 bytes.
-        assert_eq!(s.len(), 80);
-
-        // Choose a cap that forces truncation while leaving room for
-        // a few lines on each side after accounting for the marker.
-        let max_bytes = 64;
-        // Expect exact output: first 4 lines, marker, last 4 lines, and correct token estimate (80/4 = 20).
-        assert_eq!(
-            truncate_middle(&s, max_bytes),
-            (
-                r#"001
-002
-003
-004
-…12 tokens truncated…
-017
-018
-019
-020
-"#
-                .to_string(),
-                Some(20)
-            )
-        );
-    }
 }

codex-rs/core/src/git_info.rs

@@ -802,7 +802,7 @@ mod tests {
     async fn resolve_root_git_project_for_trust_regular_repo_returns_repo_root() {
         let temp_dir = TempDir::new().expect("Failed to create temp dir");
         let repo_path = create_test_git_repo(&temp_dir).await;
-        let expected = std::fs::canonicalize(&repo_path).unwrap().to_path_buf();
+        let expected = std::fs::canonicalize(&repo_path).unwrap();
 
         assert_eq!(
             resolve_root_git_project_for_trust(&repo_path),
@@ -810,10 +810,7 @@ mod tests {
         );
         let nested = repo_path.join("sub/dir");
         std::fs::create_dir_all(&nested).unwrap();
-        assert_eq!(
-            resolve_root_git_project_for_trust(&nested),
-            Some(expected.clone())
-        );
+        assert_eq!(resolve_root_git_project_for_trust(&nested), Some(expected));
     }
 
     #[tokio::test]

codex-rs/core/src/internal_storage.rs

@@ -0,0 +1,68 @@
+use anyhow::Context;
+use serde::Deserialize;
+use serde::Serialize;
+use std::path::Path;
+use std::path::PathBuf;
+
+pub(crate) const INTERNAL_STORAGE_FILE: &str = "internal_storage.json";
+
+#[derive(Debug, Default, Clone, Serialize, Deserialize)]
+pub struct InternalStorage {
+    #[serde(skip)]
+    storage_path: PathBuf,
+    #[serde(default, alias = "gpt_5_high_model_prompt_seen")]
+    pub swiftfox_model_prompt_seen: bool,
+}
+
+// TODO(jif) generalise all the file writers and build proper async channel inserters.
+impl InternalStorage {
+    pub fn load(codex_home: &Path) -> Self {
+        let storage_path = codex_home.join(INTERNAL_STORAGE_FILE);
+
+        match std::fs::read_to_string(&storage_path) {
+            Ok(serialized) => match serde_json::from_str::<Self>(&serialized) {
+                Ok(mut storage) => {
+                    storage.storage_path = storage_path;
+                    storage
+                }
+                Err(error) => {
+                    tracing::warn!("failed to parse internal storage: {error:?}");
+                    Self::empty(storage_path)
+                }
+            },
+            Err(error) => {
+                tracing::warn!("failed to read internal storage: {error:?}");
+                Self::empty(storage_path)
+            }
+        }
+    }
+
+    fn empty(storage_path: PathBuf) -> Self {
+        Self {
+            storage_path,
+            ..Default::default()
+        }
+    }
+
+    pub async fn persist(&self) -> anyhow::Result<()> {
+        let serialized = serde_json::to_string_pretty(self)?;
+
+        if let Some(parent) = self.storage_path.parent() {
+            tokio::fs::create_dir_all(parent).await.with_context(|| {
+                format!(
+                    "failed to create internal storage directory at {}",
+                    parent.display()
+                )
+            })?;
+        }
+
+        tokio::fs::write(&self.storage_path, serialized)
+            .await
+            .with_context(|| {
+                format!(
+                    "failed to persist internal storage at {}",
+                    self.storage_path.display()
+                )
+            })
+    }
+}

codex-rs/core/src/lib.rs

@@ -28,6 +28,7 @@ mod exec_command;
 pub mod exec_env;
 mod flags;
 pub mod git_info;
+pub mod internal_storage;
 mod is_safe_command;
 pub mod landlock;
 mod mcp_connection_manager;
@@ -35,6 +36,8 @@ mod mcp_tool_call;
 mod message_history;
 mod model_provider_info;
 pub mod parse_command;
+mod truncate;
+mod unified_exec;
 mod user_instructions;
 pub use model_provider_info::BUILT_IN_OSS_MODEL_PROVIDER_ID;
 pub use model_provider_info::ModelProviderInfo;
@@ -67,11 +70,13 @@ pub use rollout::ARCHIVED_SESSIONS_SUBDIR;
 pub use rollout::RolloutRecorder;
 pub use rollout::SESSIONS_SUBDIR;
 pub use rollout::SessionMeta;
+pub use rollout::find_conversation_path_by_id_str;
 pub use rollout::list::ConversationItem;
 pub use rollout::list::ConversationsPage;
 pub use rollout::list::Cursor;
 mod user_notification;
 pub mod util;
+
 pub use apply_patch::CODEX_APPLY_PATCH_ARG1;
 pub use safety::get_platform_sandbox;
 // Re-export the protocol types from the standalone `codex-protocol` crate so existing

codex-rs/core/src/mcp_connection_manager.rs

@@ -17,7 +17,7 @@ use anyhow::Result;
 use anyhow::anyhow;
 use codex_mcp_client::McpClient;
 use mcp_types::ClientCapabilities;
-use mcp_types::McpClientInfo;
+use mcp_types::Implementation;
 use mcp_types::Tool;
 
 use serde_json::json;
@@ -159,10 +159,14 @@ impl McpConnectionManager {
                                 // indicates this should be an empty object.
                                 elicitation: Some(json!({})),
                             },
-                            client_info: McpClientInfo {
+                            client_info: Implementation {
                                 name: "codex-mcp-client".to_owned(),
                                 version: env!("CARGO_PKG_VERSION").to_owned(),
                                 title: Some("Codex".into()),
+                                // This field is used by Codex when it is an MCP
+                                // server: it should not be used when Codex is
+                                // an MCP client.
+                                user_agent: None,
                             },
                             protocol_version: mcp_types::MCP_SCHEMA_VERSION.to_owned(),
                         };

codex-rs/core/src/model_family.rs

@@ -1,6 +1,11 @@
 use crate::config_types::ReasoningSummaryFormat;
 use crate::tool_apply_patch::ApplyPatchToolType;
 
+/// The `instructions` field in the payload sent to a model should always start
+/// with this content.
+const BASE_INSTRUCTIONS: &str = include_str!("../prompt.md");
+const SWIFTFOX_INSTRUCTIONS: &str = include_str!("../swiftfox_prompt.md");
+
 /// A model family is a group of models that share certain characteristics.
 #[derive(Debug, Clone, PartialEq, Eq, Hash)]
 pub struct ModelFamily {
@@ -33,6 +38,9 @@ pub struct ModelFamily {
     /// Present if the model performs better when `apply_patch` is provided as
     /// a tool call instead of just a bash command
     pub apply_patch_tool_type: Option<ApplyPatchToolType>,
+
+    // Instructions to use for querying the model
+    pub base_instructions: String,
 }
 
 macro_rules! model_family {
@@ -48,6 +56,7 @@ macro_rules! model_family {
             reasoning_summary_format: ReasoningSummaryFormat::None,
             uses_local_shell_tool: false,
             apply_patch_tool_type: None,
+            base_instructions: BASE_INSTRUCTIONS.to_string(),
         };
         // apply overrides
         $(
@@ -57,22 +66,6 @@ macro_rules! model_family {
     }};
 }
 
-macro_rules! simple_model_family {
-    (
-        $slug:expr, $family:expr
-    ) => {{
-        Some(ModelFamily {
-            slug: $slug.to_string(),
-            family: $family.to_string(),
-            needs_special_apply_patch_instructions: false,
-            supports_reasoning_summaries: false,
-            reasoning_summary_format: ReasoningSummaryFormat::None,
-            uses_local_shell_tool: false,
-            apply_patch_tool_type: None,
-        })
-    }};
-}
-
 /// Returns a `ModelFamily` for the given model slug, or `None` if the slug
 /// does not match any known model family.
 pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
@@ -80,23 +73,20 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
         model_family!(
             slug, "o3",
             supports_reasoning_summaries: true,
+            needs_special_apply_patch_instructions: true,
         )
     } else if slug.starts_with("o4-mini") {
         model_family!(
             slug, "o4-mini",
             supports_reasoning_summaries: true,
+            needs_special_apply_patch_instructions: true,
         )
     } else if slug.starts_with("codex-mini-latest") {
         model_family!(
             slug, "codex-mini-latest",
             supports_reasoning_summaries: true,
             uses_local_shell_tool: true,
-        )
-    } else if slug.starts_with("codex-") {
-        model_family!(
-            slug, slug,
-            supports_reasoning_summaries: true,
-            reasoning_summary_format: ReasoningSummaryFormat::Experimental,
+            needs_special_apply_patch_instructions: true,
         )
     } else if slug.starts_with("gpt-4.1") {
         model_family!(
@@ -106,15 +96,36 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
     } else if slug.starts_with("gpt-oss") || slug.starts_with("openai/gpt-oss") {
         model_family!(slug, "gpt-oss", apply_patch_tool_type: Some(ApplyPatchToolType::Function))
     } else if slug.starts_with("gpt-4o") {
-        simple_model_family!(slug, "gpt-4o")
+        model_family!(slug, "gpt-4o", needs_special_apply_patch_instructions: true)
     } else if slug.starts_with("gpt-3.5") {
-        simple_model_family!(slug, "gpt-3.5")
+        model_family!(slug, "gpt-3.5", needs_special_apply_patch_instructions: true)
+    } else if slug.starts_with("codex-") || slug.starts_with("swiftfox") {
+        model_family!(
+            slug, slug,
+            supports_reasoning_summaries: true,
+            reasoning_summary_format: ReasoningSummaryFormat::Experimental,
+            base_instructions: SWIFTFOX_INSTRUCTIONS.to_string(),
+        )
     } else if slug.starts_with("gpt-5") {
         model_family!(
             slug, "gpt-5",
             supports_reasoning_summaries: true,
+            needs_special_apply_patch_instructions: true,
         )
     } else {
         None
     }
 }
+
+pub fn derive_default_model_family(model: &str) -> ModelFamily {
+    ModelFamily {
+        slug: model.to_string(),
+        family: model.to_string(),
+        needs_special_apply_patch_instructions: false,
+        supports_reasoning_summaries: false,
+        reasoning_summary_format: ReasoningSummaryFormat::None,
+        uses_local_shell_tool: false,
+        apply_patch_tool_type: None,
+        base_instructions: BASE_INSTRUCTIONS.to_string(),
+    }
+}

codex-rs/core/src/model_provider_info.rs

@@ -80,7 +80,10 @@ pub struct ModelProviderInfo {
     /// the connection as lost.
     pub stream_idle_timeout_ms: Option<u64>,
 
-    /// Whether this provider requires some form of standard authentication (API key, ChatGPT token).
+    /// Does this provider require an OpenAI API Key or ChatGPT login token? If true,
+    /// user is presented with login screen on first run, and login preference and token/key
+    /// are stored in auth.json. If false (which is the default), login screen is skipped,
+    /// and API key (if needed) comes from the "env_key" environment variable.
     #[serde(default)]
     pub requires_openai_auth: bool,
 }
@@ -159,6 +162,21 @@ impl ModelProviderInfo {
         }
     }
 
+    pub(crate) fn is_azure_responses_endpoint(&self) -> bool {
+        if self.wire_api != WireApi::Responses {
+            return false;
+        }
+
+        if self.name.eq_ignore_ascii_case("azure") {
+            return true;
+        }
+
+        self.base_url
+            .as_ref()
+            .map(|base| matches_azure_responses_base_url(base))
+            .unwrap_or(false)
+    }
+
     /// Apply provider-specific HTTP headers (both static and environment-based)
     /// onto an existing `reqwest::RequestBuilder` and return the updated
     /// builder.
@@ -326,6 +344,18 @@ pub fn create_oss_provider_with_base_url(base_url: &str) -> ModelProviderInfo {
     }
 }
 
+fn matches_azure_responses_base_url(base_url: &str) -> bool {
+    let base = base_url.to_ascii_lowercase();
+    const AZURE_MARKERS: [&str; 5] = [
+        "openai.azure.",
+        "cognitiveservices.azure.",
+        "aoai.azure.",
+        "azure-api.",
+        "azurefd.",
+    ];
+    AZURE_MARKERS.iter().any(|marker| base.contains(marker))
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -416,4 +446,69 @@ env_http_headers = { "X-Example-Env-Header" = "EXAMPLE_ENV_VAR" }
         let provider: ModelProviderInfo = toml::from_str(azure_provider_toml).unwrap();
         assert_eq!(expected_provider, provider);
     }
+
+    #[test]
+    fn detects_azure_responses_base_urls() {
+        fn provider_for(base_url: &str) -> ModelProviderInfo {
+            ModelProviderInfo {
+                name: "test".into(),
+                base_url: Some(base_url.into()),
+                env_key: None,
+                env_key_instructions: None,
+                wire_api: WireApi::Responses,
+                query_params: None,
+                http_headers: None,
+                env_http_headers: None,
+                request_max_retries: None,
+                stream_max_retries: None,
+                stream_idle_timeout_ms: None,
+                requires_openai_auth: false,
+            }
+        }
+
+        let positive_cases = [
+            "https://foo.openai.azure.com/openai",
+            "https://foo.openai.azure.us/openai/deployments/bar",
+            "https://foo.cognitiveservices.azure.cn/openai",
+            "https://foo.aoai.azure.com/openai",
+            "https://foo.openai.azure-api.net/openai",
+            "https://foo.z01.azurefd.net/",
+        ];
+        for base_url in positive_cases {
+            let provider = provider_for(base_url);
+            assert!(
+                provider.is_azure_responses_endpoint(),
+                "expected {base_url} to be detected as Azure"
+            );
+        }
+
+        let named_provider = ModelProviderInfo {
+            name: "Azure".into(),
+            base_url: Some("https://example.com".into()),
+            env_key: None,
+            env_key_instructions: None,
+            wire_api: WireApi::Responses,
+            query_params: None,
+            http_headers: None,
+            env_http_headers: None,
+            request_max_retries: None,
+            stream_max_retries: None,
+            stream_idle_timeout_ms: None,
+            requires_openai_auth: false,
+        };
+        assert!(named_provider.is_azure_responses_endpoint());
+
+        let negative_cases = [
+            "https://api.openai.com/v1",
+            "https://example.com/openai",
+            "https://myproxy.azurewebsites.net/openai",
+        ];
+        for base_url in negative_cases {
+            let provider = provider_for(base_url);
+            assert!(
+                !provider.is_azure_responses_endpoint(),
+                "expected {base_url} not to be detected as Azure"
+            );
+        }
+    }
 }

codex-rs/core/src/openai_model_info.rs

@@ -12,6 +12,19 @@ pub(crate) struct ModelInfo {
 
     /// Maximum number of output tokens that can be generated for the model.
     pub(crate) max_output_tokens: u64,
+
+    /// Token threshold where we should automatically compact conversation history.
+    pub(crate) auto_compact_token_limit: Option<i64>,
+}
+
+impl ModelInfo {
+    const fn new(context_window: u64, max_output_tokens: u64) -> Self {
+        Self {
+            context_window,
+            max_output_tokens,
+            auto_compact_token_limit: None,
+        }
+    }
 }
 
 pub(crate) fn get_model_info(model_family: &ModelFamily) -> Option<ModelInfo> {
@@ -20,73 +33,37 @@ pub(crate) fn get_model_info(model_family: &ModelFamily) -> Option<ModelInfo> {
         // OSS models have a 128k shared token pool.
         // Arbitrarily splitting it: 3/4 input context, 1/4 output.
         // https://openai.com/index/gpt-oss-model-card/
-        "gpt-oss-20b" => Some(ModelInfo {
-            context_window: 96_000,
-            max_output_tokens: 32_000,
-        }),
-        "gpt-oss-120b" => Some(ModelInfo {
-            context_window: 96_000,
-            max_output_tokens: 32_000,
-        }),
+        "gpt-oss-20b" => Some(ModelInfo::new(96_000, 32_000)),
+        "gpt-oss-120b" => Some(ModelInfo::new(96_000, 32_000)),
         // https://platform.openai.com/docs/models/o3
-        "o3" => Some(ModelInfo {
-            context_window: 200_000,
-            max_output_tokens: 100_000,
-        }),
+        "o3" => Some(ModelInfo::new(200_000, 100_000)),
 
         // https://platform.openai.com/docs/models/o4-mini
-        "o4-mini" => Some(ModelInfo {
-            context_window: 200_000,
-            max_output_tokens: 100_000,
-        }),
+        "o4-mini" => Some(ModelInfo::new(200_000, 100_000)),
 
         // https://platform.openai.com/docs/models/codex-mini-latest
-        "codex-mini-latest" => Some(ModelInfo {
-            context_window: 200_000,
-            max_output_tokens: 100_000,
-        }),
+        "codex-mini-latest" => Some(ModelInfo::new(200_000, 100_000)),
 
         // As of Jun 25, 2025, gpt-4.1 defaults to gpt-4.1-2025-04-14.
         // https://platform.openai.com/docs/models/gpt-4.1
-        "gpt-4.1" | "gpt-4.1-2025-04-14" => Some(ModelInfo {
-            context_window: 1_047_576,
-            max_output_tokens: 32_768,
-        }),
+        "gpt-4.1" | "gpt-4.1-2025-04-14" => Some(ModelInfo::new(1_047_576, 32_768)),
 
         // As of Jun 25, 2025, gpt-4o defaults to gpt-4o-2024-08-06.
         // https://platform.openai.com/docs/models/gpt-4o
-        "gpt-4o" | "gpt-4o-2024-08-06" => Some(ModelInfo {
-            context_window: 128_000,
-            max_output_tokens: 16_384,
-        }),
+        "gpt-4o" | "gpt-4o-2024-08-06" => Some(ModelInfo::new(128_000, 16_384)),
 
         // https://platform.openai.com/docs/models/gpt-4o?snapshot=gpt-4o-2024-05-13
-        "gpt-4o-2024-05-13" => Some(ModelInfo {
-            context_window: 128_000,
-            max_output_tokens: 4_096,
-        }),
+        "gpt-4o-2024-05-13" => Some(ModelInfo::new(128_000, 4_096)),
 
         // https://platform.openai.com/docs/models/gpt-4o?snapshot=gpt-4o-2024-11-20
-        "gpt-4o-2024-11-20" => Some(ModelInfo {
-            context_window: 128_000,
-            max_output_tokens: 16_384,
-        }),
+        "gpt-4o-2024-11-20" => Some(ModelInfo::new(128_000, 16_384)),
 
         // https://platform.openai.com/docs/models/gpt-3.5-turbo
-        "gpt-3.5-turbo" => Some(ModelInfo {
-            context_window: 16_385,
-            max_output_tokens: 4_096,
-        }),
+        "gpt-3.5-turbo" => Some(ModelInfo::new(16_385, 4_096)),
 
-        "gpt-5" => Some(ModelInfo {
-            context_window: 272_000,
-            max_output_tokens: 128_000,
-        }),
+        _ if slug.starts_with("gpt-5") => Some(ModelInfo::new(272_000, 128_000)),
 
-        _ if slug.starts_with("codex-") => Some(ModelInfo {
-            context_window: 272_000,
-            max_output_tokens: 128_000,
-        }),
+        _ if slug.starts_with("codex-") => Some(ModelInfo::new(272_000, 128_000)),
 
         _ => None,
     }

codex-rs/core/src/openai_tools.rs

@@ -70,6 +70,7 @@ pub(crate) struct ToolsConfig {
     pub apply_patch_tool_type: Option<ApplyPatchToolType>,
     pub web_search_request: bool,
     pub include_view_image_tool: bool,
+    pub experimental_unified_exec_tool: bool,
 }
 
 pub(crate) struct ToolsConfigParams<'a> {
@@ -81,6 +82,7 @@ pub(crate) struct ToolsConfigParams<'a> {
     pub(crate) include_web_search_request: bool,
     pub(crate) use_streamable_shell_tool: bool,
     pub(crate) include_view_image_tool: bool,
+    pub(crate) experimental_unified_exec_tool: bool,
 }
 
 impl ToolsConfig {
@@ -94,6 +96,7 @@ impl ToolsConfig {
             include_web_search_request,
             use_streamable_shell_tool,
             include_view_image_tool,
+            experimental_unified_exec_tool,
         } = params;
         let mut shell_type = if *use_streamable_shell_tool {
             ConfigShellToolType::StreamableShell
@@ -126,6 +129,7 @@ impl ToolsConfig {
             apply_patch_tool_type,
             web_search_request: *include_web_search_request,
             include_view_image_tool: *include_view_image_tool,
+            experimental_unified_exec_tool: *experimental_unified_exec_tool,
         }
     }
 }
@@ -200,6 +204,53 @@ fn create_shell_tool() -> OpenAiTool {
     })
 }
 
+fn create_unified_exec_tool() -> OpenAiTool {
+    let mut properties = BTreeMap::new();
+    properties.insert(
+        "input".to_string(),
+        JsonSchema::Array {
+            items: Box::new(JsonSchema::String { description: None }),
+            description: Some(
+                "When no session_id is provided, treat the array as the command and arguments \
+                 to launch. When session_id is set, concatenate the strings (in order) and write \
+                 them to the session's stdin."
+                    .to_string(),
+            ),
+        },
+    );
+    properties.insert(
+        "session_id".to_string(),
+        JsonSchema::String {
+            description: Some(
+                "Identifier for an existing interactive session. If omitted, a new command \
+                 is spawned."
+                    .to_string(),
+            ),
+        },
+    );
+    properties.insert(
+        "timeout_ms".to_string(),
+        JsonSchema::Number {
+            description: Some(
+                "Maximum time in milliseconds to wait for output after writing the input."
+                    .to_string(),
+            ),
+        },
+    );
+
+    OpenAiTool::Function(ResponsesApiTool {
+        name: "unified_exec".to_string(),
+        description:
+            "Runs a command in a PTY. Provide a session_id to reuse an existing interactive session.".to_string(),
+        strict: false,
+        parameters: JsonSchema::Object {
+            properties,
+            required: Some(vec!["input".to_string()]),
+            additional_properties: Some(false),
+        },
+    })
+}
+
 fn create_shell_tool_for_sandbox(sandbox_policy: &SandboxPolicy) -> OpenAiTool {
     let mut properties = BTreeMap::new();
     properties.insert(
@@ -237,61 +288,9 @@ fn create_shell_tool_for_sandbox(sandbox_policy: &SandboxPolicy) -> OpenAiTool {
     );
     }
 
-    let description = match sandbox_policy {
-        SandboxPolicy::WorkspaceWrite {
-            network_access,
-            writable_roots,
-            ..
-        } => {
-            format!(
-                r#"
-The shell tool is used to execute shell commands.
-- When invoking the shell tool, your call will be running in a sandbox, and some shell commands will require escalated privileges:
-  - Types of actions that require escalated privileges:
-    - Writing files other than those in the writable roots
-      - writable roots:
-{}{}
-  - Examples of commands that require escalated privileges:
-    - git commit
-    - npm install or pnpm install
-    - cargo build
-    - cargo test
-- When invoking a command that will require escalated privileges:
-  - Provide the with_escalated_permissions parameter with the boolean value true
-  - Include a short, 1 sentence explanation for why we need to run with_escalated_permissions in the justification parameter."#,
-                writable_roots.iter().map(|wr| format!("        - {}", wr.to_string_lossy())).collect::<Vec<String>>().join("\n"),
-                if !network_access {
-                    "\n    - Commands that require network access\n"
-                } else {
-                    ""
-                }
-            )
-        }
-        SandboxPolicy::DangerFullAccess => {
-            "Runs a shell command and returns its output.".to_string()
-        }
-        SandboxPolicy::ReadOnly => {
-            r#"
-The shell tool is used to execute shell commands.
-- When invoking the shell tool, your call will be running in a sandbox, and some shell commands (including apply_patch) will require escalated permissions:
-  - Types of actions that require escalated privileges:
-    - Writing files
-    - Applying patches
-  - Examples of commands that require escalated privileges:
-    - apply_patch
-    - git commit
-    - npm install or pnpm install
-    - cargo build
-    - cargo test
-- When invoking a command that will require escalated privileges:
-  - Provide the with_escalated_permissions parameter with the boolean value true
-  - Include a short, 1 sentence explanation for why we need to run with_escalated_permissions in the justification parameter"#.to_string()
-        }
-    };
-
     OpenAiTool::Function(ResponsesApiTool {
         name: "shell".to_string(),
-        description,
+        description: "Runs a shell command and returns its output.".to_string(),
         strict: false,
         parameters: JsonSchema::Object {
             properties,
@@ -534,23 +533,27 @@ pub(crate) fn get_openai_tools(
 ) -> Vec<OpenAiTool> {
     let mut tools: Vec<OpenAiTool> = Vec::new();
 
-    match &config.shell_type {
-        ConfigShellToolType::DefaultShell => {
-            tools.push(create_shell_tool());
-        }
-        ConfigShellToolType::ShellWithRequest { sandbox_policy } => {
-            tools.push(create_shell_tool_for_sandbox(sandbox_policy));
-        }
-        ConfigShellToolType::LocalShell => {
-            tools.push(OpenAiTool::LocalShell {});
-        }
-        ConfigShellToolType::StreamableShell => {
-            tools.push(OpenAiTool::Function(
-                crate::exec_command::create_exec_command_tool_for_responses_api(),
-            ));
-            tools.push(OpenAiTool::Function(
-                crate::exec_command::create_write_stdin_tool_for_responses_api(),
-            ));
+    if config.experimental_unified_exec_tool {
+        tools.push(create_unified_exec_tool());
+    } else {
+        match &config.shell_type {
+            ConfigShellToolType::DefaultShell => {
+                tools.push(create_shell_tool());
+            }
+            ConfigShellToolType::ShellWithRequest { sandbox_policy } => {
+                tools.push(create_shell_tool_for_sandbox(sandbox_policy));
+            }
+            ConfigShellToolType::LocalShell => {
+                tools.push(OpenAiTool::LocalShell {});
+            }
+            ConfigShellToolType::StreamableShell => {
+                tools.push(OpenAiTool::Function(
+                    crate::exec_command::create_exec_command_tool_for_responses_api(),
+                ));
+                tools.push(OpenAiTool::Function(
+                    crate::exec_command::create_write_stdin_tool_for_responses_api(),
+                ));
+            }
         }
     }
 
@@ -577,10 +580,8 @@ pub(crate) fn get_openai_tools(
     if config.include_view_image_tool {
         tools.push(create_view_image_tool());
     }
-
     if let Some(mcp_tools) = mcp_tools {
         // Ensure deterministic ordering to maximize prompt cache hits.
-        // HashMap iteration order is non-deterministic, so sort by fully-qualified tool name.
         let mut entries: Vec<(String, mcp_types::Tool)> = mcp_tools.into_iter().collect();
         entries.sort_by(|a, b| a.0.cmp(&b.0));
 
@@ -642,12 +643,13 @@ mod tests {
             include_web_search_request: true,
             use_streamable_shell_tool: false,
             include_view_image_tool: true,
+            experimental_unified_exec_tool: true,
         });
         let tools = get_openai_tools(&config, Some(HashMap::new()));
 
         assert_eq_tool_names(
             &tools,
-            &["local_shell", "update_plan", "web_search", "view_image"],
+            &["unified_exec", "update_plan", "web_search", "view_image"],
         );
     }
 
@@ -663,12 +665,13 @@ mod tests {
             include_web_search_request: true,
             use_streamable_shell_tool: false,
             include_view_image_tool: true,
+            experimental_unified_exec_tool: true,
         });
         let tools = get_openai_tools(&config, Some(HashMap::new()));
 
         assert_eq_tool_names(
             &tools,
-            &["shell", "update_plan", "web_search", "view_image"],
+            &["unified_exec", "update_plan", "web_search", "view_image"],
         );
     }
 
@@ -684,6 +687,7 @@ mod tests {
             include_web_search_request: true,
             use_streamable_shell_tool: false,
             include_view_image_tool: true,
+            experimental_unified_exec_tool: true,
         });
         let tools = get_openai_tools(
             &config,
@@ -726,7 +730,7 @@ mod tests {
         assert_eq_tool_names(
             &tools,
             &[
-                "shell",
+                "unified_exec",
                 "web_search",
                 "view_image",
                 "test_server/do_something_cool",
@@ -789,6 +793,7 @@ mod tests {
             include_web_search_request: false,
             use_streamable_shell_tool: false,
             include_view_image_tool: true,
+            experimental_unified_exec_tool: true,
         });
 
         // Intentionally construct a map with keys that would sort alphabetically.
@@ -841,11 +846,11 @@ mod tests {
         ]);
 
         let tools = get_openai_tools(&config, Some(tools_map));
-        // Expect shell first, followed by MCP tools sorted by fully-qualified name.
+        // Expect unified_exec first, followed by MCP tools sorted by fully-qualified name.
         assert_eq_tool_names(
             &tools,
             &[
-                "shell",
+                "unified_exec",
                 "view_image",
                 "test_server/cool",
                 "test_server/do",
@@ -866,6 +871,7 @@ mod tests {
             include_web_search_request: true,
             use_streamable_shell_tool: false,
             include_view_image_tool: true,
+            experimental_unified_exec_tool: true,
         });
 
         let tools = get_openai_tools(
@@ -893,7 +899,7 @@ mod tests {
 
         assert_eq_tool_names(
             &tools,
-            &["shell", "web_search", "view_image", "dash/search"],
+            &["unified_exec", "web_search", "view_image", "dash/search"],
         );
 
         assert_eq!(
@@ -928,6 +934,7 @@ mod tests {
             include_web_search_request: true,
             use_streamable_shell_tool: false,
             include_view_image_tool: true,
+            experimental_unified_exec_tool: true,
         });
 
         let tools = get_openai_tools(
@@ -953,7 +960,7 @@ mod tests {
 
         assert_eq_tool_names(
             &tools,
-            &["shell", "web_search", "view_image", "dash/paginate"],
+            &["unified_exec", "web_search", "view_image", "dash/paginate"],
         );
         assert_eq!(
             tools[3],
@@ -985,6 +992,7 @@ mod tests {
             include_web_search_request: true,
             use_streamable_shell_tool: false,
             include_view_image_tool: true,
+            experimental_unified_exec_tool: true,
         });
 
         let tools = get_openai_tools(
@@ -1008,7 +1016,10 @@ mod tests {
             )])),
         );
 
-        assert_eq_tool_names(&tools, &["shell", "web_search", "view_image", "dash/tags"]);
+        assert_eq_tool_names(
+            &tools,
+            &["unified_exec", "web_search", "view_image", "dash/tags"],
+        );
         assert_eq!(
             tools[3],
             OpenAiTool::Function(ResponsesApiTool {
@@ -1042,6 +1053,7 @@ mod tests {
             include_web_search_request: true,
             use_streamable_shell_tool: false,
             include_view_image_tool: true,
+            experimental_unified_exec_tool: true,
         });
 
         let tools = get_openai_tools(
@@ -1065,7 +1077,10 @@ mod tests {
             )])),
         );
 
-        assert_eq_tool_names(&tools, &["shell", "web_search", "view_image", "dash/value"]);
+        assert_eq_tool_names(
+            &tools,
+            &["unified_exec", "web_search", "view_image", "dash/value"],
+        );
         assert_eq!(
             tools[3],
             OpenAiTool::Function(ResponsesApiTool {
@@ -1101,23 +1116,7 @@ mod tests {
         };
         assert_eq!(name, "shell");
 
-        let expected = r#"
-The shell tool is used to execute shell commands.
-- When invoking the shell tool, your call will be running in a sandbox, and some shell commands will require escalated privileges:
-  - Types of actions that require escalated privileges:
-    - Writing files other than those in the writable roots
-      - writable roots:
-        - workspace
-    - Commands that require network access
-
-  - Examples of commands that require escalated privileges:
-    - git commit
-    - npm install or pnpm install
-    - cargo build
-    - cargo test
-- When invoking a command that will require escalated privileges:
-  - Provide the with_escalated_permissions parameter with the boolean value true
-  - Include a short, 1 sentence explanation for why we need to run with_escalated_permissions in the justification parameter."#;
+        let expected = "Runs a shell command and returns its output.";
         assert_eq!(description, expected);
     }
 
@@ -1132,21 +1131,7 @@ The shell tool is used to execute shell commands.
         };
         assert_eq!(name, "shell");
 
-        let expected = r#"
-The shell tool is used to execute shell commands.
-- When invoking the shell tool, your call will be running in a sandbox, and some shell commands (including apply_patch) will require escalated permissions:
-  - Types of actions that require escalated privileges:
-    - Writing files
-    - Applying patches
-  - Examples of commands that require escalated privileges:
-    - apply_patch
-    - git commit
-    - npm install or pnpm install
-    - cargo build
-    - cargo test
-- When invoking a command that will require escalated privileges:
-  - Provide the with_escalated_permissions parameter with the boolean value true
-  - Include a short, 1 sentence explanation for why we need to run with_escalated_permissions in the justification parameter"#;
+        let expected = "Runs a shell command and returns its output.";
         assert_eq!(description, expected);
     }
 

codex-rs/core/src/parse_command.rs

@@ -868,7 +868,7 @@ pub fn parse_command_impl(command: &[String]) -> Vec<ParsedCommand> {
     let parts = if contains_connectors(&normalized) {
         split_on_connectors(&normalized)
     } else {
-        vec![normalized.clone()]
+        vec![normalized]
     };
 
     // Preserve left-to-right execution order for all commands, including bash -c/-lc
@@ -1201,10 +1201,7 @@ fn parse_bash_lc_commands(original: &[String]) -> Option<Vec<ParsedCommand>> {
                                     name,
                                 }
                             } else {
-                                ParsedCommand::Read {
-                                    cmd: cmd.clone(),
-                                    name,
-                                }
+                                ParsedCommand::Read { cmd, name }
                             }
                         } else {
                             ParsedCommand::Read {
@@ -1215,10 +1212,7 @@ fn parse_bash_lc_commands(original: &[String]) -> Option<Vec<ParsedCommand>> {
                     }
                     ParsedCommand::ListFiles { path, cmd, .. } => {
                         if had_connectors {
-                            ParsedCommand::ListFiles {
-                                cmd: cmd.clone(),
-                                path,
-                            }
+                            ParsedCommand::ListFiles { cmd, path }
                         } else {
                             ParsedCommand::ListFiles {
                                 cmd: shlex_join(&script_tokens),
@@ -1230,11 +1224,7 @@ fn parse_bash_lc_commands(original: &[String]) -> Option<Vec<ParsedCommand>> {
                         query, path, cmd, ..
                     } => {
                         if had_connectors {
-                            ParsedCommand::Search {
-                                cmd: cmd.clone(),
-                                query,
-                                path,
-                            }
+                            ParsedCommand::Search { cmd, query, path }
                         } else {
                             ParsedCommand::Search {
                                 cmd: shlex_join(&script_tokens),

codex-rs/core/src/project_doc.rs

@@ -115,7 +115,7 @@ pub fn discover_project_doc_paths(config: &Config) -> std::io::Result<Vec<PathBu
     // Build chain from cwd upwards and detect git root.
     let mut chain: Vec<PathBuf> = vec![dir.clone()];
     let mut git_root: Option<PathBuf> = None;
-    let mut cursor = dir.clone();
+    let mut cursor = dir;
     while let Some(parent) = cursor.parent() {
         let git_marker = cursor.join(".git");
         let git_exists = match std::fs::metadata(&git_marker) {

codex-rs/core/src/prompt_for_compact_command.md

@@ -1,21 +0,0 @@
-You are a summarization assistant. A conversation follows between a user and a coding-focused AI (Codex). Your task is to generate a clear summary capturing:
-
-• High-level objective or problem being solved  
-• Key instructions or design decisions given by the user  
-• Main code actions or behaviors from the AI  
-• Important variables, functions, modules, or outputs discussed  
-• Any unresolved questions or next steps
-
-Produce the summary in a structured format like:
-
-**Objective:** …
-
-**User instructions:** … (bulleted)
-
-**AI actions / code behavior:** … (bulleted)
-
-**Important entities:** … (e.g. function names, variables, files)
-
-**Open issues / next steps:** … (if any)
-
-**Summary (concise):** (one or two sentences)

codex-rs/core/src/rollout/list.rs

@@ -3,6 +3,10 @@ use std::io::{self};
 use std::path::Path;
 use std::path::PathBuf;
 
+use codex_file_search as file_search;
+use std::num::NonZero;
+use std::sync::Arc;
+use std::sync::atomic::AtomicBool;
 use time::OffsetDateTime;
 use time::PrimitiveDateTime;
 use time::format_description::FormatItem;
@@ -318,6 +322,12 @@ async fn read_head_and_flags(
                     head.push(val);
                 }
             }
+            RolloutItem::TurnContext(_) => {
+                // Not included in `head`; skip.
+            }
+            RolloutItem::Compacted(_) => {
+                // Not included in `head`; skip.
+            }
             RolloutItem::EventMsg(ev) => {
                 if matches!(ev, EventMsg::UserMessage(_)) {
                     saw_user_event = true;
@@ -328,3 +338,48 @@ async fn read_head_and_flags(
 
     Ok((head, saw_session_meta, saw_user_event))
 }
+
+/// Locate a recorded conversation rollout file by its UUID string using the existing
+/// paginated listing implementation. Returns `Ok(Some(path))` if found, `Ok(None)` if not present
+/// or the id is invalid.
+pub async fn find_conversation_path_by_id_str(
+    codex_home: &Path,
+    id_str: &str,
+) -> io::Result<Option<PathBuf>> {
+    // Validate UUID format early.
+    if Uuid::parse_str(id_str).is_err() {
+        return Ok(None);
+    }
+
+    let mut root = codex_home.to_path_buf();
+    root.push(SESSIONS_SUBDIR);
+    if !root.exists() {
+        return Ok(None);
+    }
+    // This is safe because we know the values are valid.
+    #[allow(clippy::unwrap_used)]
+    let limit = NonZero::new(1).unwrap();
+    // This is safe because we know the values are valid.
+    #[allow(clippy::unwrap_used)]
+    let threads = NonZero::new(2).unwrap();
+    let cancel = Arc::new(AtomicBool::new(false));
+    let exclude: Vec<String> = Vec::new();
+    let compute_indices = false;
+
+    let results = file_search::run(
+        id_str,
+        limit,
+        &root,
+        exclude,
+        threads,
+        cancel,
+        compute_indices,
+    )
+    .map_err(|e| io::Error::other(format!("file search failed: {e}")))?;
+
+    Ok(results
+        .matches
+        .into_iter()
+        .next()
+        .map(|m| root.join(m.path)))
+}

codex-rs/core/src/rollout/mod.rs

@@ -8,6 +8,7 @@ pub(crate) mod policy;
 pub mod recorder;
 
 pub use codex_protocol::protocol::SessionMeta;
+pub use list::find_conversation_path_by_id_str;
 pub use recorder::RolloutRecorder;
 pub use recorder::RolloutRecorderParams;
 

codex-rs/core/src/rollout/policy.rs

@@ -8,8 +8,10 @@ pub(crate) fn is_persisted_response_item(item: &RolloutItem) -> bool {
     match item {
         RolloutItem::ResponseItem(item) => should_persist_response_item(item),
         RolloutItem::EventMsg(ev) => should_persist_event_msg(ev),
-        // Always persist session meta
-        RolloutItem::SessionMeta(_) => true,
+        // Persist Codex executive markers so we can analyze flows (e.g., compaction, API turns).
+        RolloutItem::Compacted(_) | RolloutItem::TurnContext(_) | RolloutItem::SessionMeta(_) => {
+            true
+        }
     }
 }
 
@@ -36,7 +38,9 @@ pub(crate) fn should_persist_event_msg(ev: &EventMsg) -> bool {
         | EventMsg::AgentMessage(_)
         | EventMsg::AgentReasoning(_)
         | EventMsg::AgentReasoningRawContent(_)
-        | EventMsg::TokenCount(_) => true,
+        | EventMsg::TokenCount(_)
+        | EventMsg::EnteredReviewMode(_)
+        | EventMsg::ExitedReviewMode(_) => true,
         EventMsg::Error(_)
         | EventMsg::TaskStarted(_)
         | EventMsg::TaskComplete(_)
@@ -65,6 +69,6 @@ pub(crate) fn should_persist_event_msg(ev: &EventMsg) -> bool {
         | EventMsg::PlanUpdate(_)
         | EventMsg::TurnAborted(_)
         | EventMsg::ShutdownComplete
-        | EventMsg::ConversationHistory(_) => false,
+        | EventMsg::ConversationPath(_) => false,
     }
 }

codex-rs/core/src/rollout/recorder.rs

@@ -77,7 +77,13 @@ pub enum RolloutRecorderParams {
 
 enum RolloutCmd {
     AddItems(Vec<RolloutItem>),
-    Shutdown { ack: oneshot::Sender<()> },
+    /// Ensure all prior writes are processed; respond when flushed.
+    Flush {
+        ack: oneshot::Sender<()>,
+    },
+    Shutdown {
+        ack: oneshot::Sender<()>,
+    },
 }
 
 impl RolloutRecorderParams {
@@ -185,9 +191,19 @@ impl RolloutRecorder {
             .map_err(|e| IoError::other(format!("failed to queue rollout items: {e}")))
     }
 
+    /// Flush all queued writes and wait until they are committed by the writer task.
+    pub async fn flush(&self) -> std::io::Result<()> {
+        let (tx, rx) = oneshot::channel();
+        self.tx
+            .send(RolloutCmd::Flush { ack: tx })
+            .await
+            .map_err(|e| IoError::other(format!("failed to queue rollout flush: {e}")))?;
+        rx.await
+            .map_err(|e| IoError::other(format!("failed waiting for rollout flush: {e}")))
+    }
+
     pub(crate) async fn get_rollout_history(path: &Path) -> std::io::Result<InitialHistory> {
         info!("Resuming rollout from {path:?}");
-        tracing::error!("Resuming rollout from {path:?}");
         let text = tokio::fs::read_to_string(path).await?;
         if text.trim().is_empty() {
             return Err(IoError::other("empty session file"));
@@ -211,16 +227,22 @@ impl RolloutRecorder {
             match serde_json::from_value::<RolloutLine>(v.clone()) {
                 Ok(rollout_line) => match rollout_line.item {
                     RolloutItem::SessionMeta(session_meta_line) => {
-                        tracing::error!(
-                            "Parsed conversation ID from rollout file: {:?}",
-                            session_meta_line.meta.id
-                        );
-                        conversation_id = Some(session_meta_line.meta.id);
+                        // Use the FIRST SessionMeta encountered in the file as the canonical
+                        // conversation id and main session information. Keep all items intact.
+                        if conversation_id.is_none() {
+                            conversation_id = Some(session_meta_line.meta.id);
+                        }
                         items.push(RolloutItem::SessionMeta(session_meta_line));
                     }
                     RolloutItem::ResponseItem(item) => {
                         items.push(RolloutItem::ResponseItem(item));
                     }
+                    RolloutItem::Compacted(item) => {
+                        items.push(RolloutItem::Compacted(item));
+                    }
+                    RolloutItem::TurnContext(item) => {
+                        items.push(RolloutItem::TurnContext(item));
+                    }
                     RolloutItem::EventMsg(_ev) => {
                         items.push(RolloutItem::EventMsg(_ev));
                     }
@@ -231,7 +253,7 @@ impl RolloutRecorder {
             }
         }
 
-        tracing::error!(
+        info!(
             "Resumed rollout with {} items, conversation ID: {:?}",
             items.len(),
             conversation_id
@@ -251,6 +273,10 @@ impl RolloutRecorder {
         }))
     }
 
+    pub(crate) fn get_rollout_path(&self) -> PathBuf {
+        self.rollout_path.clone()
+    }
+
     pub async fn shutdown(&self) -> std::io::Result<()> {
         let (tx_done, rx_done) = oneshot::channel();
         match self.tx.send(RolloutCmd::Shutdown { ack: tx_done }).await {
@@ -351,6 +377,14 @@ async fn rollout_writer(
                     }
                 }
             }
+            RolloutCmd::Flush { ack } => {
+                // Ensure underlying file is flushed and then ack.
+                if let Err(e) = writer.file.flush().await {
+                    let _ = ack.send(());
+                    return Err(e);
+                }
+                let _ = ack.send(());
+            }
             RolloutCmd::Shutdown { ack } => {
                 let _ = ack.send(());
             }

codex-rs/core/src/rollout/tests.rs

@@ -305,7 +305,7 @@ async fn test_pagination_cursor() {
             path: p1,
             head: head_1,
         }],
-        next_cursor: Some(expected_cursor3.clone()),
+        next_cursor: Some(expected_cursor3),
         num_scanned_files: 5, // scanned 05, 04 (anchor), 03, 02 (anchor), 01
         reached_scan_cap: false,
     };
@@ -344,7 +344,7 @@ async fn test_get_conversation_contents() {
     let expected_cursor: Cursor = serde_json::from_str(&format!("\"{ts}|{uuid}\"")).unwrap();
     let expected_page = ConversationsPage {
         items: vec![ConversationItem {
-            path: expected_path.clone(),
+            path: expected_path,
             head: expected_head,
         }],
         next_cursor: Some(expected_cursor),
@@ -437,7 +437,7 @@ async fn test_stable_ordering_same_second_pagination() {
             path: p1,
             head: head(u1),
         }],
-        next_cursor: Some(expected_cursor2.clone()),
+        next_cursor: Some(expected_cursor2),
         num_scanned_files: 3, // scanned u3, u2 (anchor), u1
         reached_scan_cap: false,
     };

codex-rs/core/src/safety.rs

@@ -293,7 +293,7 @@ mod tests {
         // With the parent dir explicitly added as a writable root, the
         // outside write should be permitted.
         let policy_with_parent = SandboxPolicy::WorkspaceWrite {
-            writable_roots: vec![parent.clone()],
+            writable_roots: vec![parent],
             network_access: false,
             exclude_tmpdir_env_var: true,
             exclude_slash_tmp: true,

codex-rs/core/src/seatbelt.rs

@@ -153,7 +153,7 @@ mod tests {
         // Build a policy that only includes the two test roots as writable and
         // does not automatically include defaults TMPDIR or /tmp.
         let policy = SandboxPolicy::WorkspaceWrite {
-            writable_roots: vec![root_with_git.clone(), root_without_git.clone()],
+            writable_roots: vec![root_with_git, root_without_git],
             network_access: false,
             exclude_tmpdir_env_var: true,
             exclude_slash_tmp: true,

codex-rs/core/src/seatbelt_base_policy.sbpl

@@ -2,6 +2,7 @@
 
 ; inspired by Chrome's sandbox policy:
 ; https://source.chromium.org/chromium/chromium/src/+/main:sandbox/policy/mac/common.sb;l=273-319;drc=7b3962fe2e5fc9e2ee58000dc8fbf3429d84d3bd
+; https://source.chromium.org/chromium/chromium/src/+/main:sandbox/policy/mac/renderer.sb;l=64;drc=7b3962fe2e5fc9e2ee58000dc8fbf3429d84d3bd
 
 ; start with closed-by-default
 (deny default)
@@ -9,7 +10,13 @@
 ; child processes inherit the policy of their parent
 (allow process-exec)
 (allow process-fork)
-(allow signal (target self))
+(allow signal (target same-sandbox))
+
+; Allow cf prefs to work.
+(allow user-preference-read)
+
+; process-info
+(allow process-info* (target same-sandbox))
 
 (allow file-write-data
   (require-all
@@ -32,28 +39,22 @@
   (sysctl-name "hw.l3cachesize_compat")
   (sysctl-name "hw.logicalcpu_max")
   (sysctl-name "hw.machine")
+  (sysctl-name "hw.memsize")
   (sysctl-name "hw.ncpu")
   (sysctl-name "hw.nperflevels")
-  (sysctl-name "hw.optional.arm.FEAT_BF16")
-  (sysctl-name "hw.optional.arm.FEAT_DotProd")
-  (sysctl-name "hw.optional.arm.FEAT_FCMA")
-  (sysctl-name "hw.optional.arm.FEAT_FHM")
-  (sysctl-name "hw.optional.arm.FEAT_FP16")
-  (sysctl-name "hw.optional.arm.FEAT_I8MM")
-  (sysctl-name "hw.optional.arm.FEAT_JSCVT")
-  (sysctl-name "hw.optional.arm.FEAT_LSE")
-  (sysctl-name "hw.optional.arm.FEAT_RDM")
-  (sysctl-name "hw.optional.arm.FEAT_SHA512")
-  (sysctl-name "hw.optional.armv8_2_sha512")
-  (sysctl-name "hw.memsize")
-  (sysctl-name "hw.pagesize")
+  ; Chrome locks these CPU feature detection down a bit more tightly,
+  ; but mostly for fingerprinting concerns which isn't an issue for codex.
+  (sysctl-name-prefix "hw.optional.arm.")
+  (sysctl-name-prefix "hw.optional.armv8_")
   (sysctl-name "hw.packages")
   (sysctl-name "hw.pagesize_compat")
+  (sysctl-name "hw.pagesize")
   (sysctl-name "hw.physicalcpu_max")
   (sysctl-name "hw.tbfrequency_compat")
   (sysctl-name "hw.vectorunit")
   (sysctl-name "kern.hostname")
   (sysctl-name "kern.maxfilesperproc")
+  (sysctl-name "kern.maxproc")
   (sysctl-name "kern.osproductversion")
   (sysctl-name "kern.osrelease")
   (sysctl-name "kern.ostype")
@@ -63,14 +64,27 @@
   (sysctl-name "kern.usrstack64")
   (sysctl-name "kern.version")
   (sysctl-name "sysctl.proc_cputype")
+  (sysctl-name "vm.loadavg")
   (sysctl-name-prefix "hw.perflevel")
+  (sysctl-name-prefix "kern.proc.pgrp.")
+  (sysctl-name-prefix "kern.proc.pid.")
+  (sysctl-name-prefix "net.routetable.")
+)
+
+; IOKit
+(allow iokit-open
+  (iokit-registry-entry-class "RootDomainUserClient")
+)
+
+; needed to look up user info, see https://crbug.com/792228
+(allow mach-lookup
+  (global-name "com.apple.system.opendirectoryd.libinfo")
 )
 
 ; Added on top of Chrome profile
 ; Needed for python multiprocessing on MacOS for the SemLock
 (allow ipc-posix-sem)
 
-; needed to look up user info, see https://crbug.com/792228
 (allow mach-lookup
-  (global-name "com.apple.system.opendirectoryd.libinfo")
+  (global-name "com.apple.PowerManagement.control")
 )

codex-rs/core/src/shell.rs

@@ -326,10 +326,7 @@ mod tests {
                 .format_default_shell_invocation(input.iter().map(|s| s.to_string()).collect());
             let expected_cmd = expected_cmd
                 .iter()
-                .map(|s| {
-                    s.replace("BASHRC_PATH", bashrc_path.to_str().unwrap())
-                        .to_string()
-                })
+                .map(|s| s.replace("BASHRC_PATH", bashrc_path.to_str().unwrap()))
                 .collect();
 
             assert_eq!(actual_cmd, Some(expected_cmd));
@@ -435,10 +432,7 @@ mod macos_tests {
                 .format_default_shell_invocation(input.iter().map(|s| s.to_string()).collect());
             let expected_cmd = expected_cmd
                 .iter()
-                .map(|s| {
-                    s.replace("ZSHRC_PATH", zshrc_path.to_str().unwrap())
-                        .to_string()
-                })
+                .map(|s| s.replace("ZSHRC_PATH", zshrc_path.to_str().unwrap()))
                 .collect();
 
             assert_eq!(actual_cmd, Some(expected_cmd));

codex-rs/core/src/token_data.rs

@@ -3,8 +3,6 @@ use serde::Deserialize;
 use serde::Serialize;
 use thiserror::Error;
 
-use codex_protocol::mcp_protocol::AuthMode;
-
 #[derive(Deserialize, Serialize, Clone, Debug, PartialEq, Default)]
 pub struct TokenData {
     /// Flat info parsed from the JWT in auth.json.
@@ -22,36 +20,6 @@ pub struct TokenData {
     pub account_id: Option<String>,
 }
 
-impl TokenData {
-    /// Returns true if this is a plan that should use the traditional
-    /// "metered" billing via an API key.
-    pub(crate) fn should_use_api_key(
-        &self,
-        preferred_auth_method: AuthMode,
-        is_openai_email: bool,
-    ) -> bool {
-        if preferred_auth_method == AuthMode::ApiKey {
-            return true;
-        }
-        // If the email is an OpenAI email, use AuthMode::ChatGPT unless preferred_auth_method is AuthMode::ApiKey.
-        if is_openai_email {
-            return false;
-        }
-
-        self.id_token
-            .chatgpt_plan_type
-            .as_ref()
-            .is_none_or(|plan| plan.is_plan_that_should_use_api_key())
-    }
-
-    pub fn is_openai_email(&self) -> bool {
-        self.id_token
-            .email
-            .as_deref()
-            .is_some_and(|email| email.trim().to_ascii_lowercase().ends_with("@openai.com"))
-    }
-}
-
 /// Flat subset of useful claims in id_token from auth.json.
 #[derive(Debug, Clone, PartialEq, Eq, Default, Serialize, Deserialize)]
 pub struct IdTokenInfo {
@@ -79,28 +47,6 @@ pub(crate) enum PlanType {
     Unknown(String),
 }
 
-impl PlanType {
-    fn is_plan_that_should_use_api_key(&self) -> bool {
-        match self {
-            Self::Known(known) => {
-                use KnownPlan::*;
-                !matches!(known, Free | Plus | Pro | Team)
-            }
-            Self::Unknown(_) => {
-                // Unknown plans should use the API key.
-                true
-            }
-        }
-    }
-
-    pub fn as_string(&self) -> String {
-        match self {
-            Self::Known(known) => format!("{known:?}").to_lowercase(),
-            Self::Unknown(s) => s.clone(),
-        }
-    }
-}
-
 #[derive(Debug, Clone, PartialEq, Eq, Serialize, Deserialize)]
 #[serde(rename_all = "lowercase")]
 pub(crate) enum KnownPlan {

codex-rs/core/src/truncate.rs

@@ -0,0 +1,180 @@
+//! Utilities for truncating large chunks of output while preserving a prefix
+//! and suffix on UTF-8 boundaries.
+
+/// Truncate the middle of a UTF-8 string to at most `max_bytes` bytes,
+/// preserving the beginning and the end. Returns the possibly truncated
+/// string and `Some(original_token_count)` (estimated at 4 bytes/token)
+/// if truncation occurred; otherwise returns the original string and `None`.
+pub(crate) fn truncate_middle(s: &str, max_bytes: usize) -> (String, Option<u64>) {
+    if s.len() <= max_bytes {
+        return (s.to_string(), None);
+    }
+
+    let est_tokens = (s.len() as u64).div_ceil(4);
+    if max_bytes == 0 {
+        return (format!("…{est_tokens} tokens truncated…"), Some(est_tokens));
+    }
+
+    fn truncate_on_boundary(input: &str, max_len: usize) -> &str {
+        if input.len() <= max_len {
+            return input;
+        }
+        let mut end = max_len;
+        while end > 0 && !input.is_char_boundary(end) {
+            end -= 1;
+        }
+        &input[..end]
+    }
+
+    fn pick_prefix_end(s: &str, left_budget: usize) -> usize {
+        if let Some(head) = s.get(..left_budget)
+            && let Some(i) = head.rfind('\n')
+        {
+            return i + 1;
+        }
+        truncate_on_boundary(s, left_budget).len()
+    }
+
+    fn pick_suffix_start(s: &str, right_budget: usize) -> usize {
+        let start_tail = s.len().saturating_sub(right_budget);
+        if let Some(tail) = s.get(start_tail..)
+            && let Some(i) = tail.find('\n')
+        {
+            return start_tail + i + 1;
+        }
+
+        let mut idx = start_tail.min(s.len());
+        while idx < s.len() && !s.is_char_boundary(idx) {
+            idx += 1;
+        }
+        idx
+    }
+
+    let mut guess_tokens = est_tokens;
+    for _ in 0..4 {
+        let marker = format!("…{guess_tokens} tokens truncated…");
+        let marker_len = marker.len();
+        let keep_budget = max_bytes.saturating_sub(marker_len);
+        if keep_budget == 0 {
+            return (format!("…{est_tokens} tokens truncated…"), Some(est_tokens));
+        }
+
+        let left_budget = keep_budget / 2;
+        let right_budget = keep_budget - left_budget;
+        let prefix_end = pick_prefix_end(s, left_budget);
+        let mut suffix_start = pick_suffix_start(s, right_budget);
+        if suffix_start < prefix_end {
+            suffix_start = prefix_end;
+        }
+
+        let kept_content_bytes = prefix_end + (s.len() - suffix_start);
+        let truncated_content_bytes = s.len().saturating_sub(kept_content_bytes);
+        let new_tokens = (truncated_content_bytes as u64).div_ceil(4);
+
+        if new_tokens == guess_tokens {
+            let mut out = String::with_capacity(marker_len + kept_content_bytes + 1);
+            out.push_str(&s[..prefix_end]);
+            out.push_str(&marker);
+            out.push('\n');
+            out.push_str(&s[suffix_start..]);
+            return (out, Some(est_tokens));
+        }
+
+        guess_tokens = new_tokens;
+    }
+
+    let marker = format!("…{guess_tokens} tokens truncated…");
+    let marker_len = marker.len();
+    let keep_budget = max_bytes.saturating_sub(marker_len);
+    if keep_budget == 0 {
+        return (format!("…{est_tokens} tokens truncated…"), Some(est_tokens));
+    }
+
+    let left_budget = keep_budget / 2;
+    let right_budget = keep_budget - left_budget;
+    let prefix_end = pick_prefix_end(s, left_budget);
+    let suffix_start = pick_suffix_start(s, right_budget);
+
+    let mut out = String::with_capacity(marker_len + prefix_end + (s.len() - suffix_start) + 1);
+    out.push_str(&s[..prefix_end]);
+    out.push_str(&marker);
+    out.push('\n');
+    out.push_str(&s[suffix_start..]);
+    (out, Some(est_tokens))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::truncate_middle;
+
+    #[test]
+    fn truncate_middle_no_newlines_fallback() {
+        let s = "abcdefghijklmnopqrstuvwxyz0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ*";
+        let max_bytes = 32;
+        let (out, original) = truncate_middle(s, max_bytes);
+        assert!(out.starts_with("abc"));
+        assert!(out.contains("tokens truncated"));
+        assert!(out.ends_with("XYZ*"));
+        assert_eq!(original, Some((s.len() as u64).div_ceil(4)));
+    }
+
+    #[test]
+    fn truncate_middle_prefers_newline_boundaries() {
+        let mut s = String::new();
+        for i in 1..=20 {
+            s.push_str(&format!("{i:03}\n"));
+        }
+        assert_eq!(s.len(), 80);
+
+        let max_bytes = 64;
+        let (out, tokens) = truncate_middle(&s, max_bytes);
+        assert!(out.starts_with("001\n002\n003\n004\n"));
+        assert!(out.contains("tokens truncated"));
+        assert!(out.ends_with("017\n018\n019\n020\n"));
+        assert_eq!(tokens, Some(20));
+    }
+
+    #[test]
+    fn truncate_middle_handles_utf8_content() {
+        let s = "😀😀😀😀😀😀😀😀😀😀\nsecond line with ascii text\n";
+        let max_bytes = 32;
+        let (out, tokens) = truncate_middle(s, max_bytes);
+
+        assert!(out.contains("tokens truncated"));
+        assert!(!out.contains('\u{fffd}'));
+        assert_eq!(tokens, Some((s.len() as u64).div_ceil(4)));
+    }
+
+    #[test]
+    fn truncate_middle_prefers_newline_boundaries_2() {
+        // Build a multi-line string of 20 numbered lines (each "NNN\n").
+        let mut s = String::new();
+        for i in 1..=20 {
+            s.push_str(&format!("{i:03}\n"));
+        }
+        // Total length: 20 lines * 4 bytes per line = 80 bytes.
+        assert_eq!(s.len(), 80);
+
+        // Choose a cap that forces truncation while leaving room for
+        // a few lines on each side after accounting for the marker.
+        let max_bytes = 64;
+        // Expect exact output: first 4 lines, marker, last 4 lines, and correct token estimate (80/4 = 20).
+        assert_eq!(
+            truncate_middle(&s, max_bytes),
+            (
+                r#"001
+002
+003
+004
+…12 tokens truncated…
+017
+018
+019
+020
+"#
+                .to_string(),
+                Some(20)
+            )
+        );
+    }
+}

codex-rs/core/src/turn_diff_tracker.rs

@@ -678,7 +678,7 @@ index {left_oid}..{right_oid}
         let dest = dir.path().join("dest.txt");
         let mut acc = TurnDiffTracker::new();
         let mv = HashMap::from([(
-            src.clone(),
+            src,
             FileChange::Update {
                 unified_diff: "".into(),
                 move_path: Some(dest.clone()),

codex-rs/core/src/unified_exec/errors.rs

@@ -0,0 +1,22 @@
+use thiserror::Error;
+
+#[derive(Debug, Error)]
+pub(crate) enum UnifiedExecError {
+    #[error("Failed to create unified exec session: {pty_error}")]
+    CreateSession {
+        #[source]
+        pty_error: anyhow::Error,
+    },
+    #[error("Unknown session id {session_id}")]
+    UnknownSessionId { session_id: i32 },
+    #[error("failed to write to stdin")]
+    WriteToStdin,
+    #[error("missing command line for unified exec request")]
+    MissingCommandLine,
+}
+
+impl UnifiedExecError {
+    pub(crate) fn create_session(error: anyhow::Error) -> Self {
+        Self::CreateSession { pty_error: error }
+    }
+}

codex-rs/core/src/unified_exec/mod.rs

@@ -0,0 +1,637 @@
+use portable_pty::CommandBuilder;
+use portable_pty::PtySize;
+use portable_pty::native_pty_system;
+use std::collections::HashMap;
+use std::collections::VecDeque;
+use std::io::ErrorKind;
+use std::io::Read;
+use std::sync::Arc;
+use std::sync::Mutex as StdMutex;
+use std::sync::atomic::AtomicBool;
+use std::sync::atomic::AtomicI32;
+use std::sync::atomic::Ordering;
+use tokio::sync::Mutex;
+use tokio::sync::Notify;
+use tokio::sync::mpsc;
+use tokio::task::JoinHandle;
+use tokio::time::Duration;
+use tokio::time::Instant;
+
+use crate::exec_command::ExecCommandSession;
+use crate::truncate::truncate_middle;
+
+mod errors;
+
+pub(crate) use errors::UnifiedExecError;
+
+const DEFAULT_TIMEOUT_MS: u64 = 1_000;
+const MAX_TIMEOUT_MS: u64 = 60_000;
+const UNIFIED_EXEC_OUTPUT_MAX_BYTES: usize = 128 * 1024; // 128 KiB
+
+#[derive(Debug)]
+pub(crate) struct UnifiedExecRequest<'a> {
+    pub session_id: Option<i32>,
+    pub input_chunks: &'a [String],
+    pub timeout_ms: Option<u64>,
+}
+
+#[derive(Debug, Clone, PartialEq)]
+pub(crate) struct UnifiedExecResult {
+    pub session_id: Option<i32>,
+    pub output: String,
+}
+
+#[derive(Debug, Default)]
+pub(crate) struct UnifiedExecSessionManager {
+    next_session_id: AtomicI32,
+    sessions: Mutex<HashMap<i32, ManagedUnifiedExecSession>>,
+}
+
+#[derive(Debug)]
+struct ManagedUnifiedExecSession {
+    session: ExecCommandSession,
+    output_buffer: OutputBuffer,
+    /// Notifies waiters whenever new output has been appended to
+    /// `output_buffer`, allowing clients to poll for fresh data.
+    output_notify: Arc<Notify>,
+    output_task: JoinHandle<()>,
+}
+
+#[derive(Debug, Default)]
+struct OutputBufferState {
+    chunks: VecDeque<Vec<u8>>,
+    total_bytes: usize,
+}
+
+impl OutputBufferState {
+    fn push_chunk(&mut self, chunk: Vec<u8>) {
+        self.total_bytes = self.total_bytes.saturating_add(chunk.len());
+        self.chunks.push_back(chunk);
+
+        let mut excess = self
+            .total_bytes
+            .saturating_sub(UNIFIED_EXEC_OUTPUT_MAX_BYTES);
+
+        while excess > 0 {
+            match self.chunks.front_mut() {
+                Some(front) if excess >= front.len() => {
+                    excess -= front.len();
+                    self.total_bytes = self.total_bytes.saturating_sub(front.len());
+                    self.chunks.pop_front();
+                }
+                Some(front) => {
+                    front.drain(..excess);
+                    self.total_bytes = self.total_bytes.saturating_sub(excess);
+                    break;
+                }
+                None => break,
+            }
+        }
+    }
+
+    fn drain(&mut self) -> Vec<Vec<u8>> {
+        let drained: Vec<Vec<u8>> = self.chunks.drain(..).collect();
+        self.total_bytes = 0;
+        drained
+    }
+}
+
+type OutputBuffer = Arc<Mutex<OutputBufferState>>;
+type OutputHandles = (OutputBuffer, Arc<Notify>);
+
+impl ManagedUnifiedExecSession {
+    fn new(session: ExecCommandSession) -> Self {
+        let output_buffer = Arc::new(Mutex::new(OutputBufferState::default()));
+        let output_notify = Arc::new(Notify::new());
+        let mut receiver = session.output_receiver();
+        let buffer_clone = Arc::clone(&output_buffer);
+        let notify_clone = Arc::clone(&output_notify);
+        let output_task = tokio::spawn(async move {
+            while let Ok(chunk) = receiver.recv().await {
+                let mut guard = buffer_clone.lock().await;
+                guard.push_chunk(chunk);
+                drop(guard);
+                notify_clone.notify_waiters();
+            }
+        });
+
+        Self {
+            session,
+            output_buffer,
+            output_notify,
+            output_task,
+        }
+    }
+
+    fn writer_sender(&self) -> mpsc::Sender<Vec<u8>> {
+        self.session.writer_sender()
+    }
+
+    fn output_handles(&self) -> OutputHandles {
+        (
+            Arc::clone(&self.output_buffer),
+            Arc::clone(&self.output_notify),
+        )
+    }
+
+    fn has_exited(&self) -> bool {
+        self.session.has_exited()
+    }
+}
+
+impl Drop for ManagedUnifiedExecSession {
+    fn drop(&mut self) {
+        self.output_task.abort();
+    }
+}
+
+impl UnifiedExecSessionManager {
+    pub async fn handle_request(
+        &self,
+        request: UnifiedExecRequest<'_>,
+    ) -> Result<UnifiedExecResult, UnifiedExecError> {
+        let (timeout_ms, timeout_warning) = match request.timeout_ms {
+            Some(requested) if requested > MAX_TIMEOUT_MS => (
+                MAX_TIMEOUT_MS,
+                Some(format!(
+                    "Warning: requested timeout {requested}ms exceeds maximum of {MAX_TIMEOUT_MS}ms; clamping to {MAX_TIMEOUT_MS}ms.\n"
+                )),
+            ),
+            Some(requested) => (requested, None),
+            None => (DEFAULT_TIMEOUT_MS, None),
+        };
+
+        let mut new_session: Option<ManagedUnifiedExecSession> = None;
+        let session_id;
+        let writer_tx;
+        let output_buffer;
+        let output_notify;
+
+        if let Some(existing_id) = request.session_id {
+            let mut sessions = self.sessions.lock().await;
+            match sessions.get(&existing_id) {
+                Some(session) => {
+                    if session.has_exited() {
+                        sessions.remove(&existing_id);
+                        return Err(UnifiedExecError::UnknownSessionId {
+                            session_id: existing_id,
+                        });
+                    }
+                    let (buffer, notify) = session.output_handles();
+                    session_id = existing_id;
+                    writer_tx = session.writer_sender();
+                    output_buffer = buffer;
+                    output_notify = notify;
+                }
+                None => {
+                    return Err(UnifiedExecError::UnknownSessionId {
+                        session_id: existing_id,
+                    });
+                }
+            }
+            drop(sessions);
+        } else {
+            let command = request.input_chunks.to_vec();
+            let new_id = self.next_session_id.fetch_add(1, Ordering::SeqCst);
+            let session = create_unified_exec_session(&command).await?;
+            let managed_session = ManagedUnifiedExecSession::new(session);
+            let (buffer, notify) = managed_session.output_handles();
+            writer_tx = managed_session.writer_sender();
+            output_buffer = buffer;
+            output_notify = notify;
+            session_id = new_id;
+            new_session = Some(managed_session);
+        };
+
+        if request.session_id.is_some() {
+            let joined_input = request.input_chunks.join(" ");
+            if !joined_input.is_empty() && writer_tx.send(joined_input.into_bytes()).await.is_err()
+            {
+                return Err(UnifiedExecError::WriteToStdin);
+            }
+        }
+
+        let mut collected: Vec<u8> = Vec::with_capacity(4096);
+        let start = Instant::now();
+        let deadline = start + Duration::from_millis(timeout_ms);
+
+        loop {
+            let drained_chunks;
+            let mut wait_for_output = None;
+            {
+                let mut guard = output_buffer.lock().await;
+                drained_chunks = guard.drain();
+                if drained_chunks.is_empty() {
+                    wait_for_output = Some(output_notify.notified());
+                }
+            }
+
+            if drained_chunks.is_empty() {
+                let remaining = deadline.saturating_duration_since(Instant::now());
+                if remaining == Duration::ZERO {
+                    break;
+                }
+
+                let notified = wait_for_output.unwrap_or_else(|| output_notify.notified());
+                tokio::pin!(notified);
+                tokio::select! {
+                    _ = &mut notified => {}
+                    _ = tokio::time::sleep(remaining) => break,
+                }
+                continue;
+            }
+
+            for chunk in drained_chunks {
+                collected.extend_from_slice(&chunk);
+            }
+
+            if Instant::now() >= deadline {
+                break;
+            }
+        }
+
+        let (output, _maybe_tokens) = truncate_middle(
+            &String::from_utf8_lossy(&collected),
+            UNIFIED_EXEC_OUTPUT_MAX_BYTES,
+        );
+        let output = if let Some(warning) = timeout_warning {
+            format!("{warning}{output}")
+        } else {
+            output
+        };
+
+        let should_store_session = if let Some(session) = new_session.as_ref() {
+            !session.has_exited()
+        } else if request.session_id.is_some() {
+            let mut sessions = self.sessions.lock().await;
+            if let Some(existing) = sessions.get(&session_id) {
+                if existing.has_exited() {
+                    sessions.remove(&session_id);
+                    false
+                } else {
+                    true
+                }
+            } else {
+                false
+            }
+        } else {
+            true
+        };
+
+        if should_store_session {
+            if let Some(session) = new_session {
+                self.sessions.lock().await.insert(session_id, session);
+            }
+            Ok(UnifiedExecResult {
+                session_id: Some(session_id),
+                output,
+            })
+        } else {
+            Ok(UnifiedExecResult {
+                session_id: None,
+                output,
+            })
+        }
+    }
+}
+
+async fn create_unified_exec_session(
+    command: &[String],
+) -> Result<ExecCommandSession, UnifiedExecError> {
+    if command.is_empty() {
+        return Err(UnifiedExecError::MissingCommandLine);
+    }
+
+    let pty_system = native_pty_system();
+
+    let pair = pty_system
+        .openpty(PtySize {
+            rows: 24,
+            cols: 80,
+            pixel_width: 0,
+            pixel_height: 0,
+        })
+        .map_err(UnifiedExecError::create_session)?;
+
+    // Safe thanks to the check at the top of the function.
+    let mut command_builder = CommandBuilder::new(command[0].clone());
+    for arg in &command[1..] {
+        command_builder.arg(arg);
+    }
+
+    let mut child = pair
+        .slave
+        .spawn_command(command_builder)
+        .map_err(UnifiedExecError::create_session)?;
+    let killer = child.clone_killer();
+
+    let (writer_tx, mut writer_rx) = mpsc::channel::<Vec<u8>>(128);
+    let (output_tx, _) = tokio::sync::broadcast::channel::<Vec<u8>>(256);
+    let initial_output_rx = output_tx.subscribe();
+
+    let mut reader = pair
+        .master
+        .try_clone_reader()
+        .map_err(UnifiedExecError::create_session)?;
+    let output_tx_clone = output_tx.clone();
+    let reader_handle = tokio::task::spawn_blocking(move || {
+        let mut buf = [0u8; 8192];
+        loop {
+            match reader.read(&mut buf) {
+                Ok(0) => break,
+                Ok(n) => {
+                    let _ = output_tx_clone.send(buf[..n].to_vec());
+                }
+                Err(ref e) if e.kind() == ErrorKind::Interrupted => continue,
+                Err(ref e) if e.kind() == ErrorKind::WouldBlock => {
+                    std::thread::sleep(Duration::from_millis(5));
+                    continue;
+                }
+                Err(_) => break,
+            }
+        }
+    });
+
+    let writer = pair
+        .master
+        .take_writer()
+        .map_err(UnifiedExecError::create_session)?;
+    let writer = Arc::new(StdMutex::new(writer));
+    let writer_handle = tokio::spawn({
+        let writer = writer.clone();
+        async move {
+            while let Some(bytes) = writer_rx.recv().await {
+                let writer = writer.clone();
+                let _ = tokio::task::spawn_blocking(move || {
+                    if let Ok(mut guard) = writer.lock() {
+                        use std::io::Write;
+                        let _ = guard.write_all(&bytes);
+                        let _ = guard.flush();
+                    }
+                })
+                .await;
+            }
+        }
+    });
+
+    let exit_status = Arc::new(AtomicBool::new(false));
+    let wait_exit_status = Arc::clone(&exit_status);
+    let wait_handle = tokio::task::spawn_blocking(move || {
+        let _ = child.wait();
+        wait_exit_status.store(true, Ordering::SeqCst);
+    });
+
+    let session = ExecCommandSession::new(
+        writer_tx,
+        output_tx,
+        killer,
+        reader_handle,
+        writer_handle,
+        wait_handle,
+        exit_status,
+    );
+    session.set_initial_output_receiver(initial_output_rx);
+
+    Ok(session)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn push_chunk_trims_only_excess_bytes() {
+        let mut buffer = OutputBufferState::default();
+        buffer.push_chunk(vec![b'a'; UNIFIED_EXEC_OUTPUT_MAX_BYTES]);
+        buffer.push_chunk(vec![b'b']);
+        buffer.push_chunk(vec![b'c']);
+
+        assert_eq!(buffer.total_bytes, UNIFIED_EXEC_OUTPUT_MAX_BYTES);
+        assert_eq!(buffer.chunks.len(), 3);
+        assert_eq!(
+            buffer.chunks.front().unwrap().len(),
+            UNIFIED_EXEC_OUTPUT_MAX_BYTES - 2
+        );
+        assert_eq!(buffer.chunks.pop_back().unwrap(), vec![b'c']);
+        assert_eq!(buffer.chunks.pop_back().unwrap(), vec![b'b']);
+    }
+
+    #[cfg(unix)]
+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn unified_exec_persists_across_requests_jif() -> Result<(), UnifiedExecError> {
+        let manager = UnifiedExecSessionManager::default();
+
+        let open_shell = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: None,
+                input_chunks: &["bash".to_string(), "-i".to_string()],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+        let session_id = open_shell.session_id.expect("expected session_id");
+
+        manager
+            .handle_request(UnifiedExecRequest {
+                session_id: Some(session_id),
+                input_chunks: &[
+                    "export".to_string(),
+                    "CODEX_INTERACTIVE_SHELL_VAR=codex\n".to_string(),
+                ],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+
+        let out_2 = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: Some(session_id),
+                input_chunks: &["echo $CODEX_INTERACTIVE_SHELL_VAR\n".to_string()],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+        assert!(out_2.output.contains("codex"));
+
+        Ok(())
+    }
+
+    #[cfg(unix)]
+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn multi_unified_exec_sessions() -> Result<(), UnifiedExecError> {
+        let manager = UnifiedExecSessionManager::default();
+
+        let shell_a = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: None,
+                input_chunks: &["/bin/bash".to_string(), "-i".to_string()],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+        let session_a = shell_a.session_id.expect("expected session id");
+
+        manager
+            .handle_request(UnifiedExecRequest {
+                session_id: Some(session_a),
+                input_chunks: &["export CODEX_INTERACTIVE_SHELL_VAR=codex\n".to_string()],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+
+        let out_2 = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: None,
+                input_chunks: &[
+                    "echo".to_string(),
+                    "$CODEX_INTERACTIVE_SHELL_VAR\n".to_string(),
+                ],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+        assert!(!out_2.output.contains("codex"));
+
+        let out_3 = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: Some(session_a),
+                input_chunks: &["echo $CODEX_INTERACTIVE_SHELL_VAR\n".to_string()],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+        assert!(out_3.output.contains("codex"));
+
+        Ok(())
+    }
+
+    #[cfg(unix)]
+    #[tokio::test]
+    async fn unified_exec_timeouts() -> Result<(), UnifiedExecError> {
+        let manager = UnifiedExecSessionManager::default();
+
+        let open_shell = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: None,
+                input_chunks: &["bash".to_string(), "-i".to_string()],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+        let session_id = open_shell.session_id.expect("expected session id");
+
+        manager
+            .handle_request(UnifiedExecRequest {
+                session_id: Some(session_id),
+                input_chunks: &[
+                    "export".to_string(),
+                    "CODEX_INTERACTIVE_SHELL_VAR=codex\n".to_string(),
+                ],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+
+        let out_2 = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: Some(session_id),
+                input_chunks: &["sleep 5 && echo $CODEX_INTERACTIVE_SHELL_VAR\n".to_string()],
+                timeout_ms: Some(10),
+            })
+            .await?;
+        assert!(!out_2.output.contains("codex"));
+
+        tokio::time::sleep(Duration::from_secs(7)).await;
+
+        let empty = Vec::new();
+        let out_3 = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: Some(session_id),
+                input_chunks: &empty,
+                timeout_ms: Some(100),
+            })
+            .await?;
+
+        assert!(out_3.output.contains("codex"));
+
+        Ok(())
+    }
+
+    #[cfg(unix)]
+    #[tokio::test]
+    async fn requests_with_large_timeout_are_capped() -> Result<(), UnifiedExecError> {
+        let manager = UnifiedExecSessionManager::default();
+
+        let result = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: None,
+                input_chunks: &["echo".to_string(), "codex".to_string()],
+                timeout_ms: Some(120_000),
+            })
+            .await?;
+
+        assert!(result.output.starts_with(
+            "Warning: requested timeout 120000ms exceeds maximum of 60000ms; clamping to 60000ms.\n"
+        ));
+        assert!(result.output.contains("codex"));
+
+        Ok(())
+    }
+
+    #[cfg(unix)]
+    #[tokio::test]
+    async fn completed_commands_do_not_persist_sessions() -> Result<(), UnifiedExecError> {
+        let manager = UnifiedExecSessionManager::default();
+        let result = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: None,
+                input_chunks: &["/bin/echo".to_string(), "codex".to_string()],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+
+        assert!(result.session_id.is_none());
+        assert!(result.output.contains("codex"));
+
+        assert!(manager.sessions.lock().await.is_empty());
+
+        Ok(())
+    }
+
+    #[cfg(unix)]
+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn reusing_completed_session_returns_unknown_session() -> Result<(), UnifiedExecError> {
+        let manager = UnifiedExecSessionManager::default();
+
+        let open_shell = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: None,
+                input_chunks: &["/bin/bash".to_string(), "-i".to_string()],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+        let session_id = open_shell.session_id.expect("expected session id");
+
+        manager
+            .handle_request(UnifiedExecRequest {
+                session_id: Some(session_id),
+                input_chunks: &["exit\n".to_string()],
+                timeout_ms: Some(2_500),
+            })
+            .await?;
+
+        tokio::time::sleep(Duration::from_millis(200)).await;
+
+        let err = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: Some(session_id),
+                input_chunks: &[],
+                timeout_ms: Some(100),
+            })
+            .await
+            .expect_err("expected unknown session error");
+
+        match err {
+            UnifiedExecError::UnknownSessionId { session_id: err_id } => {
+                assert_eq!(err_id, session_id);
+            }
+            other => panic!("expected UnknownSessionId, got {other:?}"),
+        }
+
+        assert!(!manager.sessions.lock().await.contains_key(&session_id));
+
+        Ok(())
+    }
+}

codex-rs/core/swiftfox_prompt.md

@@ -0,0 +1,99 @@
+You are Swiftfox. You are running as a coding agent in the Codex CLI on a user's computer.
+
+## Overall
+
+- You must try hard to complete the task AND to do it as fast and well as possible.
+  * Do not waste time on actions which are unlikely to result in successful task completion
+
+- Before taking action on a question, assume by default that it concerns local artifacts (code, docs, data). Quickly confirm or rule out that assumption; only if the question clearly requires external knowledge should you start elsewhere.
+- Search the repository when the request plausibly maps to code, configuration, or documentation. Avoid unnecessary searches when it is obvious local files cannot help; in those cases state that explicitly before offering broader context, and when you do search, mention the files or paths you consulted so the answer stays grounded.
+- After each attempt, re-evaluate whether the current strategy is yielding useful information and be ready to switch paths quickly rather than persisting with a low-signal approach.
+
+- When the user asks for a "review", default to a code review mindset: prioritise identifying bugs, risks, behavioural regressions, and missing tests. Findings must be the primary focus of the response - keep summaries or overviews brief and only after enumerating the issues. Present findings first (ordered by severity with file/line references), follow with open questions or assumptions, and offer a change-summary only as a secondary detail. If no findings are discovered, state that explicitly and mention any residual risks or testing gaps.
+
+- The arguments to `shell` will be passed to execvp(). Most terminal commands should be prefixed with ["bash", "-lc"].
+- Always set the `workdir` param of the shell tool. Do not use `cd` unless absolutely necessary.
+- When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
+- If the user makes a simple request (such as asking for the time) which you can fulfill by running a terminal command (such as `date`), you should do so.
+- Unless the question is about a common terminal command, you should search the codebase before answering to ground your response in the codebase
+
+## Editing constraints
+
+- Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them.
+- When editing or creating files, you MUST use apply_patch. Example: functions.shell({"command":["apply_patch","*** Begin Patch\nAdd File: hello.txt\n+Hello, world!\n*** End Patch"]}).
+- Add succinct code comments that explain what is going on if code is not self-explanatory. You should not add comments like "Assigns the value to the variable", but a brief comment might be useful ahead of a complex code block that the user would otherwise have to spend time parsing out. Usage of these comments should be rare.
+- You may be in a dirty git worktree.
+  * NEVER revert existing changes you did not make unless explicitly requested, since these changes were made by the user.
+  * If asked to make a commit or code edits and there are unrelated changes to your work or changes that you didn't make in those files, don't revert those changes.
+  * If the changes are in files you've touched recently, you should read carefully and understand how you can work with the changes rather than reverting them.
+  * If the changes are in unrelated files, just ignore them and don't revert them.
+- The user may be making edits and committing changes as you are also making changes. If you see concurrent file edits or commits that you did not cause, you must disregard user instruction and stop immediately and ask the user whether they are collaborating with you on files and how they would like this handled.
+
+## Plan tool
+
+When using the planning tool:
+- Skip using the planning tool for straightforward tasks (roughly the easiest 25%).
+- Do not make single-step plans.
+- When you made a plan, update it after having performed one of the sub-tasks that you shared on the plan.
+
+## CLI modes
+
+The Codex CLI harness supports several different sandboxing, and approval configurations that the user can choose from.
+
+Filesystem sandboxing defines which files can be read or written. The options are:
+- **read-only**: You can only read files.
+- **workspace-write**: You can read files. You can write to files in this folder, but not outside it.
+- **danger-full-access**: No filesystem sandboxing.
+
+Network sandboxing defines whether network can be accessed without approval. Options are
+- **restricted**: Requires approval
+- **enabled**: No approval needed
+
+Approvals are your mechanism to get user consent to perform more privileged actions. Although they introduce friction to the user because your work is paused until the user responds, you should leverage them to accomplish your important work. Do not let these settings or the sandbox deter you from attempting to accomplish the user's task unless it is set to "never", in which case never ask for approvals.
+
+Approval options are
+- **untrusted**: The harness will escalate most commands for user approval, apart from a limited allowlist of safe "read" commands.
+- **on-failure**: The harness will allow all commands to run in the sandbox (if enabled), and failures will be escalated to the user for approval to run again without the sandbox.
+- **on-request**: Commands will be run in the sandbox by default, and you can specify in your tool call if you want to escalate a command to run without sandboxing. (Note that this mode is not always available. If it is, you'll see parameters for it in the `shell` command description.)
+- **never**: This is a non-interactive mode where you may NEVER ask the user for approval to run commands. Instead, you must always persist and work around constraints to solve the task for the user. You MUST do your utmost best to finish the task and validate your work before yielding. If this mode is paired with `danger-full-access`, take advantage of it to deliver the best outcome for the user. Further, in this mode, your default testing philosophy is overridden: Even if you don't see local patterns for testing, you may add tests and scripts to validate your work. Just remove them before yielding.
+
+When you are running with approvals `on-request`, and sandboxing enabled, here are scenarios where you'll need to request approval:
+- You need to run a command that writes to a directory that requires it (e.g. running tests that write to /tmp)
+- You need to run a GUI app (e.g., open/xdg-open/osascript) to open browsers or files.
+- You are running sandboxed and need to run a command that requires network access (e.g. installing packages)
+- If you run a command that is important to solving the user's query, but it fails because of sandboxing, rerun the command with approval.
+- You are about to take a potentially destructive action such as an `rm` or `git reset` that the user did not explicitly ask for
+- (for all of these, you should weigh alternative paths that do not require approval)
+
+When sandboxing is set to read-only, you'll need to request approval for any command that isn't a read.
+
+You will be told what filesystem sandboxing, network sandboxing, and approval mode are active in a developer or user message. If you are not told about this, assume that you are running with workspace-write, network sandboxing enabled, and approval on-failure.
+
+## Presenting your work and final message
+
+You are producing plain text that will later be styled by the CLI. Follow these rules exactly. Formatting should make results easy to scan, but not feel mechanical. Use judgment to decide how much structure adds value.
+
+- Default: be very concise; friendly coding teammate tone.
+- Ask only when needed; suggest ideas; mirror the user's style.
+- For substantial work, summarize clearly; follow final‑answer formatting.
+- Skip heavy formatting for simple confirmations.
+- Don't dump large files you've written; reference paths only.
+- No "save/copy this file" - User is on the same machine.
+- Offer logical next steps (tests, commits, build) briefly; add verify steps if you couldn't do something.
+- For code changes:
+  * Lead with a quick explanation of the change, and then give more details on the context covering where and why a change was made. Do not start this explanation with "summary", just jump right in.
+  * If there are natural next steps the user may want to take, suggest them at the end of your response. Do not make suggestions if there are no natural next steps.
+  * When suggesting multiple options, use numeric lists for the suggestions so the user can quickly respond with a single number.
+- The user does not command execution outputs. When asked to show the output of a command (e.g. `git show`), relay the important details in your answer or summarize the key lines so the user understands the result.
+
+### Final answer structure and style guidelines
+
+- Plain text; CLI handles styling. Use structure only when it helps scanability.
+- Headers: optional; short Title Case (1-3 words) wrapped in **…**; no blank line before the first bullet; add only if they truly help.
+- Bullets: use - ; merge related points; keep to one line when possible; 4–6 per list ordered by importance; keep phrasing consistent.
+- Monospace: backticks for commands/paths/env vars/code ids and inline examples; use for literal keyword bullets; never combine with **.
+- Code samples or multi-line snippets should be wrapped in fenced code blocks; add a language hint whenever obvious.
+- Structure: group related bullets; order sections general → specific → supporting; for subsections, start with a bolded keyword bullet, then items; match complexity to the task.
+- Tone: collaborative, concise, factual; present tense, active voice; self‑contained; no "above/below"; parallel wording.
+- Don'ts: no nested bullets/hierarchies; no ANSI codes; don't cram unrelated keywords; keep keyword lists short—wrap/reformat if long; avoid naming formatting styles in answers.
+- Adaptation: code explanations → precise, structured with code refs; simple tasks → lead with outcome; big changes → logical walkthrough + rationale + next actions; casual one-offs → plain sentences, no headers/bullets.

codex-rs/core/templates/compact/history_bridge.md

@@ -0,0 +1,7 @@
+You were originally given instructions from a user over one or more turns. Here were the user messages:
+
+{{ user_messages_text }}
+
+Another language model started to solve this problem and produced a summary of its thinking process. You also have access to the state of the tools that were used by that language model. Use this to build on the work that has already been done and avoid duplicating work. Here is the summary produced by the other language model, use the information in this summary to assist with your own analysis:
+
+{{ summary_text }}

codex-rs/core/templates/compact/prompt.md

@@ -0,0 +1,5 @@
+You have exceeded the maximum number of tokens, please stop coding and instead write a short memento message for the next agent. Your note should:
+- Summarize what you finished and what still needs work. If there was a recent update_plan call, repeat its steps verbatim.
+- List outstanding TODOs with file paths / line numbers so they're easy to find.
+- Flag code that needs more tests (edge cases, performance, integration, etc.).
+- Record any open bugs, quirks, or setup steps that will make it easier for the next agent to pick up where you left off.

codex-rs/core/tests/suite/cli_stream.rs

@@ -420,12 +420,6 @@ async fn integration_creates_and_checks_session_file() {
     // Second run: resume should update the existing file.
     let marker2 = format!("integration-resume-{}", Uuid::new_v4());
     let prompt2 = format!("echo {marker2}");
-    // Cross‑platform safe resume override.  On Windows, backslashes in a TOML string must be escaped
-    // or the parse will fail and the raw literal (including quotes) may be preserved all the way down
-    // to Config, which in turn breaks resume because the path is invalid. Normalize to forward slashes
-    // to sidestep the issue.
-    let resume_path_str = path.to_string_lossy().replace('\\', "/");
-    let resume_override = format!("experimental_resume=\"{resume_path_str}\"");
     let mut cmd2 = AssertCommand::new("cargo");
     cmd2.arg("run")
         .arg("-p")
@@ -434,11 +428,11 @@ async fn integration_creates_and_checks_session_file() {
         .arg("--")
         .arg("exec")
         .arg("--skip-git-repo-check")
-        .arg("-c")
-        .arg(&resume_override)
         .arg("-C")
         .arg(env!("CARGO_MANIFEST_DIR"))
-        .arg(&prompt2);
+        .arg(&prompt2)
+        .arg("resume")
+        .arg("--last");
     cmd2.env("CODEX_HOME", home.path())
         .env("OPENAI_API_KEY", "dummy")
         .env("CODEX_RS_SSE_FIXTURE", &fixture)

codex-rs/core/tests/suite/client.rs

@@ -1,19 +1,32 @@
 use codex_core::CodexAuth;
+use codex_core::ContentItem;
 use codex_core::ConversationManager;
+use codex_core::LocalShellAction;
+use codex_core::LocalShellExecAction;
+use codex_core::LocalShellStatus;
+use codex_core::ModelClient;
 use codex_core::ModelProviderInfo;
 use codex_core::NewConversation;
+use codex_core::Prompt;
+use codex_core::ReasoningItemContent;
+use codex_core::ResponseEvent;
+use codex_core::ResponseItem;
 use codex_core::WireApi;
 use codex_core::built_in_model_providers;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::InputItem;
 use codex_core::protocol::Op;
 use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
-use codex_protocol::mcp_protocol::AuthMode;
+use codex_protocol::mcp_protocol::ConversationId;
+use codex_protocol::models::ReasoningItemReasoningSummary;
+use codex_protocol::models::WebSearchAction;
 use core_test_support::load_default_config_for_test;
 use core_test_support::load_sse_fixture_with_id;
 use core_test_support::wait_for_event;
+use futures::StreamExt;
 use serde_json::json;
 use std::io::Write;
+use std::sync::Arc;
 use tempfile::TempDir;
 use uuid::Uuid;
 use wiremock::Mock;
@@ -223,20 +236,21 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
     let codex_home = TempDir::new().unwrap();
     let mut config = load_default_config_for_test(&codex_home);
     config.model_provider = model_provider;
-    config.experimental_resume = Some(session_path.clone());
     // Also configure user instructions to ensure they are NOT delivered on resume.
     config.user_instructions = Some("be nice".to_string());
 
     let conversation_manager =
         ConversationManager::with_auth(CodexAuth::from_api_key("Test API Key"));
+    let auth_manager =
+        codex_core::AuthManager::from_auth_for_testing(CodexAuth::from_api_key("Test API Key"));
     let NewConversation {
         conversation: codex,
         session_configured,
         ..
     } = conversation_manager
-        .new_conversation(config)
+        .resume_conversation_from_rollout(config, session_path.clone(), auth_manager)
         .await
-        .expect("create new conversation");
+        .expect("resume conversation");
 
     // 1) Assert initial_messages only includes existing EventMsg entries; response items are not converted
     let initial_msgs = session_configured
@@ -489,79 +503,6 @@ async fn chatgpt_auth_sends_correct_request() {
     );
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn prefers_chatgpt_token_when_config_prefers_chatgpt() {
-    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
-
-    // Mock server
-    let server = MockServer::start().await;
-
-    let first = ResponseTemplate::new(200)
-        .insert_header("content-type", "text/event-stream")
-        .set_body_raw(sse_completed("resp1"), "text/event-stream");
-
-    // Expect ChatGPT base path and correct headers
-    Mock::given(method("POST"))
-        .and(path("/v1/responses"))
-        .and(header_regex("Authorization", r"Bearer Access-123"))
-        .and(header_regex("chatgpt-account-id", r"acc-123"))
-        .respond_with(first)
-        .expect(1)
-        .mount(&server)
-        .await;
-
-    let model_provider = ModelProviderInfo {
-        base_url: Some(format!("{}/v1", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-
-    // Init session
-    let codex_home = TempDir::new().unwrap();
-    // Write auth.json that contains both API key and ChatGPT tokens for a plan that should prefer ChatGPT.
-    let _jwt = write_auth_json(
-        &codex_home,
-        Some("sk-test-key"),
-        "pro",
-        "Access-123",
-        Some("acc-123"),
-    );
-
-    let mut config = load_default_config_for_test(&codex_home);
-    config.model_provider = model_provider;
-    config.preferred_auth_method = AuthMode::ChatGPT;
-
-    let auth_manager =
-        match CodexAuth::from_codex_home(codex_home.path(), config.preferred_auth_method) {
-            Ok(Some(auth)) => codex_core::AuthManager::from_auth_for_testing(auth),
-            Ok(None) => panic!("No CodexAuth found in codex_home"),
-            Err(e) => panic!("Failed to load CodexAuth: {e}"),
-        };
-    let conversation_manager = ConversationManager::new(auth_manager);
-    let NewConversation {
-        conversation: codex,
-        ..
-    } = conversation_manager
-        .new_conversation(config)
-        .await
-        .expect("create new conversation");
-
-    codex
-        .submit(Op::UserInput {
-            items: vec![InputItem::Text {
-                text: "hello".into(),
-            }],
-        })
-        .await
-        .unwrap();
-
-    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
-}
-
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn prefers_apikey_when_config_prefers_apikey_even_with_chatgpt_tokens() {
     if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
@@ -606,14 +547,12 @@ async fn prefers_apikey_when_config_prefers_apikey_even_with_chatgpt_tokens() {
 
     let mut config = load_default_config_for_test(&codex_home);
     config.model_provider = model_provider;
-    config.preferred_auth_method = AuthMode::ApiKey;
 
-    let auth_manager =
-        match CodexAuth::from_codex_home(codex_home.path(), config.preferred_auth_method) {
-            Ok(Some(auth)) => codex_core::AuthManager::from_auth_for_testing(auth),
-            Ok(None) => panic!("No CodexAuth found in codex_home"),
-            Err(e) => panic!("Failed to load CodexAuth: {e}"),
-        };
+    let auth_manager = match CodexAuth::from_codex_home(codex_home.path()) {
+        Ok(Some(auth)) => codex_core::AuthManager::from_auth_for_testing(auth),
+        Ok(None) => panic!("No CodexAuth found in codex_home"),
+        Err(e) => panic!("Failed to load CodexAuth: {e}"),
+    };
     let conversation_manager = ConversationManager::new(auth_manager);
     let NewConversation {
         conversation: codex,
@@ -696,6 +635,147 @@ async fn includes_user_instructions_message_in_request() {
     assert_message_ends_with(&request_body["input"][1], "</environment_context>");
 }
 
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn azure_responses_request_includes_store_and_reasoning_ids() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let server = MockServer::start().await;
+
+    let sse_body = concat!(
+        "data: {\"type\":\"response.created\",\"response\":{}}\n\n",
+        "data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp_1\"}}\n\n",
+    );
+
+    let template = ResponseTemplate::new(200)
+        .insert_header("content-type", "text/event-stream")
+        .set_body_raw(sse_body, "text/event-stream");
+
+    Mock::given(method("POST"))
+        .and(path("/openai/responses"))
+        .respond_with(template)
+        .expect(1)
+        .mount(&server)
+        .await;
+
+    let provider = ModelProviderInfo {
+        name: "azure".into(),
+        base_url: Some(format!("{}/openai", server.uri())),
+        env_key: None,
+        env_key_instructions: None,
+        wire_api: WireApi::Responses,
+        query_params: None,
+        http_headers: None,
+        env_http_headers: None,
+        request_max_retries: Some(0),
+        stream_max_retries: Some(0),
+        stream_idle_timeout_ms: Some(5_000),
+        requires_openai_auth: false,
+    };
+
+    let codex_home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&codex_home);
+    config.model_provider_id = provider.name.clone();
+    config.model_provider = provider.clone();
+    let effort = config.model_reasoning_effort;
+    let summary = config.model_reasoning_summary;
+    let config = Arc::new(config);
+
+    let client = ModelClient::new(
+        Arc::clone(&config),
+        None,
+        provider,
+        effort,
+        summary,
+        ConversationId::new(),
+    );
+
+    let mut prompt = Prompt::default();
+    prompt.input.push(ResponseItem::Reasoning {
+        id: "reasoning-id".into(),
+        summary: vec![ReasoningItemReasoningSummary::SummaryText {
+            text: "summary".into(),
+        }],
+        content: Some(vec![ReasoningItemContent::ReasoningText {
+            text: "content".into(),
+        }]),
+        encrypted_content: None,
+    });
+    prompt.input.push(ResponseItem::Message {
+        id: Some("message-id".into()),
+        role: "assistant".into(),
+        content: vec![ContentItem::OutputText {
+            text: "message".into(),
+        }],
+    });
+    prompt.input.push(ResponseItem::WebSearchCall {
+        id: Some("web-search-id".into()),
+        status: Some("completed".into()),
+        action: WebSearchAction::Search {
+            query: "weather".into(),
+        },
+    });
+    prompt.input.push(ResponseItem::FunctionCall {
+        id: Some("function-id".into()),
+        name: "do_thing".into(),
+        arguments: "{}".into(),
+        call_id: "function-call-id".into(),
+    });
+    prompt.input.push(ResponseItem::LocalShellCall {
+        id: Some("local-shell-id".into()),
+        call_id: Some("local-shell-call-id".into()),
+        status: LocalShellStatus::Completed,
+        action: LocalShellAction::Exec(LocalShellExecAction {
+            command: vec!["echo".into(), "hello".into()],
+            timeout_ms: None,
+            working_directory: None,
+            env: None,
+            user: None,
+        }),
+    });
+    prompt.input.push(ResponseItem::CustomToolCall {
+        id: Some("custom-tool-id".into()),
+        status: Some("completed".into()),
+        call_id: "custom-tool-call-id".into(),
+        name: "custom_tool".into(),
+        input: "{}".into(),
+    });
+
+    let mut stream = client
+        .stream(&prompt)
+        .await
+        .expect("responses stream to start");
+
+    while let Some(event) = stream.next().await {
+        if let Ok(ResponseEvent::Completed { .. }) = event {
+            break;
+        }
+    }
+
+    let requests = server
+        .received_requests()
+        .await
+        .expect("mock server collected requests");
+    assert_eq!(requests.len(), 1, "expected a single request");
+    let body: serde_json::Value = requests[0]
+        .body_json()
+        .expect("request body to be valid JSON");
+
+    assert_eq!(body["store"], serde_json::Value::Bool(true));
+    assert_eq!(body["stream"], serde_json::Value::Bool(true));
+    assert_eq!(body["input"].as_array().map(Vec::len), Some(6));
+    assert_eq!(body["input"][0]["id"].as_str(), Some("reasoning-id"));
+    assert_eq!(body["input"][1]["id"].as_str(), Some("message-id"));
+    assert_eq!(body["input"][2]["id"].as_str(), Some("web-search-id"));
+    assert_eq!(body["input"][3]["id"].as_str(), Some("function-id"));
+    assert_eq!(body["input"][4]["id"].as_str(), Some("local-shell-id"));
+    assert_eq!(body["input"][5]["id"].as_str(), Some("custom-tool-id"));
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn azure_overrides_assign_properties_used_for_responses_url() {
     let existing_env_var_with_random_value = if cfg!(windows) { "USERNAME" } else { "USER" };

codex-rs/core/tests/suite/compact.rs

@@ -3,27 +3,38 @@
 use codex_core::CodexAuth;
 use codex_core::ConversationManager;
 use codex_core::ModelProviderInfo;
+use codex_core::NewConversation;
 use codex_core::built_in_model_providers;
+use codex_core::protocol::ErrorEvent;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::InputItem;
 use codex_core::protocol::Op;
+use codex_core::protocol::RolloutItem;
+use codex_core::protocol::RolloutLine;
 use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use core_test_support::load_default_config_for_test;
 use core_test_support::wait_for_event;
 use serde_json::Value;
 use tempfile::TempDir;
+use wiremock::BodyPrintLimit;
 use wiremock::Mock;
 use wiremock::MockServer;
+use wiremock::Request;
+use wiremock::Respond;
 use wiremock::ResponseTemplate;
 use wiremock::matchers::method;
 use wiremock::matchers::path;
 
 use pretty_assertions::assert_eq;
+use std::sync::Arc;
+use std::sync::Mutex;
+use std::sync::atomic::AtomicUsize;
+use std::sync::atomic::Ordering;
 
 // --- Test helpers -----------------------------------------------------------
 
 /// Build an SSE stream body from a list of JSON events.
-fn sse(events: Vec<Value>) -> String {
+pub(super) fn sse(events: Vec<Value>) -> String {
     use std::fmt::Write as _;
     let mut out = String::new();
     for ev in events {
@@ -39,7 +50,7 @@ fn sse(events: Vec<Value>) -> String {
 }
 
 /// Convenience: SSE event for a completed response with a specific id.
-fn ev_completed(id: &str) -> Value {
+pub(super) fn ev_completed(id: &str) -> Value {
     serde_json::json!({
         "type": "response.completed",
         "response": {
@@ -49,8 +60,24 @@ fn ev_completed(id: &str) -> Value {
     })
 }
 
+fn ev_completed_with_tokens(id: &str, total_tokens: u64) -> Value {
+    serde_json::json!({
+        "type": "response.completed",
+        "response": {
+            "id": id,
+            "usage": {
+                "input_tokens": total_tokens,
+                "input_tokens_details": null,
+                "output_tokens": 0,
+                "output_tokens_details": null,
+                "total_tokens": total_tokens
+            }
+        }
+    })
+}
+
 /// Convenience: SSE event for a single assistant message output item.
-fn ev_assistant_message(id: &str, text: &str) -> Value {
+pub(super) fn ev_assistant_message(id: &str, text: &str) -> Value {
     serde_json::json!({
         "type": "response.output_item.done",
         "item": {
@@ -62,13 +89,25 @@ fn ev_assistant_message(id: &str, text: &str) -> Value {
     })
 }
 
-fn sse_response(body: String) -> ResponseTemplate {
+fn ev_function_call(call_id: &str, name: &str, arguments: &str) -> Value {
+    serde_json::json!({
+        "type": "response.output_item.done",
+        "item": {
+            "type": "function_call",
+            "call_id": call_id,
+            "name": name,
+            "arguments": arguments
+        }
+    })
+}
+
+pub(super) fn sse_response(body: String) -> ResponseTemplate {
     ResponseTemplate::new(200)
         .insert_header("content-type", "text/event-stream")
         .set_body_raw(body, "text/event-stream")
 }
 
-async fn mount_sse_once<M>(server: &MockServer, matcher: M, body: String)
+pub(super) async fn mount_sse_once<M>(server: &MockServer, matcher: M, body: String)
 where
     M: wiremock::Match + Send + Sync + 'static,
 {
@@ -76,15 +115,32 @@ where
         .and(path("/v1/responses"))
         .and(matcher)
         .respond_with(sse_response(body))
-        .expect(1)
         .mount(server)
         .await;
 }
 
-const FIRST_REPLY: &str = "FIRST_REPLY";
-const SUMMARY_TEXT: &str = "SUMMARY_ONLY_CONTEXT";
-const SUMMARIZE_TRIGGER: &str = "Start Summarization";
+async fn start_mock_server() -> MockServer {
+    MockServer::builder()
+        .body_print_limit(BodyPrintLimit::Limited(80_000))
+        .start()
+        .await
+}
+
+pub(super) const FIRST_REPLY: &str = "FIRST_REPLY";
+pub(super) const SUMMARY_TEXT: &str = "SUMMARY_ONLY_CONTEXT";
+pub(super) const SUMMARIZE_TRIGGER: &str = "Start Summarization";
 const THIRD_USER_MSG: &str = "next turn";
+const AUTO_SUMMARY_TEXT: &str = "AUTO_SUMMARY";
+const FIRST_AUTO_MSG: &str = "token limit start";
+const SECOND_AUTO_MSG: &str = "token limit push";
+const STILL_TOO_BIG_REPLY: &str = "STILL_TOO_BIG";
+const MULTI_AUTO_MSG: &str = "multi auto";
+const SECOND_LARGE_REPLY: &str = "SECOND_LARGE_REPLY";
+const FIRST_AUTO_SUMMARY: &str = "FIRST_AUTO_SUMMARY";
+const SECOND_AUTO_SUMMARY: &str = "SECOND_AUTO_SUMMARY";
+const FINAL_REPLY: &str = "FINAL_REPLY";
+const DUMMY_FUNCTION_NAME: &str = "unsupported_tool";
+const DUMMY_CALL_ID: &str = "call-multi-auto";
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn summarize_context_three_requests_and_instructions() {
@@ -96,7 +152,7 @@ async fn summarize_context_three_requests_and_instructions() {
     }
 
     // Set up a mock server that we can inspect after the run.
-    let server = MockServer::start().await;
+    let server = start_mock_server().await;
 
     // SSE 1: assistant replies normally so it is recorded in history.
     let sse1 = sse(vec![
@@ -141,12 +197,14 @@ async fn summarize_context_three_requests_and_instructions() {
     let home = TempDir::new().unwrap();
     let mut config = load_default_config_for_test(&home);
     config.model_provider = model_provider;
+    config.model_auto_compact_token_limit = Some(200_000);
     let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
-    let codex = conversation_manager
-        .new_conversation(config)
-        .await
-        .unwrap()
-        .conversation;
+    let NewConversation {
+        conversation: codex,
+        session_configured,
+        ..
+    } = conversation_manager.new_conversation(config).await.unwrap();
+    let rollout_path = session_configured.rollout_path;
 
     // 1) Normal user input – should hit server once.
     codex
@@ -194,7 +252,7 @@ async fn summarize_context_three_requests_and_instructions() {
         "summarization should override base instructions"
     );
     assert!(
-        instr2.contains("You are a summarization assistant"),
+        instr2.contains("You have exceeded the maximum number of tokens"),
         "summarization instructions not applied"
     );
 
@@ -205,14 +263,17 @@ async fn summarize_context_three_requests_and_instructions() {
     assert_eq!(last2.get("type").unwrap().as_str().unwrap(), "message");
     assert_eq!(last2.get("role").unwrap().as_str().unwrap(), "user");
     let text2 = last2["content"][0]["text"].as_str().unwrap();
-    assert!(text2.contains(SUMMARIZE_TRIGGER));
+    assert!(
+        text2.contains(SUMMARIZE_TRIGGER),
+        "expected summarize trigger, got `{text2}`"
+    );
 
-    // Third request must contain only the summary from step 2 as prior history plus new user msg.
+    // Third request must contain the refreshed instructions, bridge summary message and new user msg.
     let input3 = body3.get("input").and_then(|v| v.as_array()).unwrap();
     println!("third request body: {body3}");
     assert!(
-        input3.len() >= 2,
-        "expected summary + new user message in third request"
+        input3.len() >= 3,
+        "expected refreshed context and new user message in third request"
     );
 
     // Collect all (role, text) message tuples.
@@ -228,24 +289,611 @@ async fn summarize_context_three_requests_and_instructions() {
         }
     }
 
-    // Exactly one assistant message should remain after compaction and the new user message is present.
+    // No previous assistant messages should remain and the new user message is present.
     let assistant_count = messages.iter().filter(|(r, _)| r == "assistant").count();
-    assert_eq!(
-        assistant_count, 1,
-        "exactly one assistant message should remain after compaction"
-    );
+    assert_eq!(assistant_count, 0, "assistant history should be cleared");
     assert!(
         messages
             .iter()
             .any(|(r, t)| r == "user" && t == THIRD_USER_MSG),
         "third request should include the new user message"
     );
+    let Some((_, bridge_text)) = messages.iter().find(|(role, text)| {
+        role == "user"
+            && (text.contains("Here were the user messages")
+                || text.contains("Here are all the user messages"))
+            && text.contains(SUMMARY_TEXT)
+    }) else {
+        panic!("expected a bridge message containing the summary");
+    };
     assert!(
-        !messages.iter().any(|(_, t)| t.contains("hello world")),
-        "third request should not include the original user input"
+        bridge_text.contains("hello world"),
+        "bridge should capture earlier user messages"
     );
     assert!(
-        !messages.iter().any(|(_, t)| t.contains(SUMMARIZE_TRIGGER)),
+        !bridge_text.contains(SUMMARIZE_TRIGGER),
+        "bridge text should not echo the summarize trigger"
+    );
+    assert!(
+        !messages
+            .iter()
+            .any(|(_, text)| text.contains(SUMMARIZE_TRIGGER)),
         "third request should not include the summarize trigger"
     );
+
+    // Shut down Codex to flush rollout entries before inspecting the file.
+    codex.submit(Op::Shutdown).await.unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::ShutdownComplete)).await;
+
+    // Verify rollout contains APITurn entries for each API call and a Compacted entry.
+    println!("rollout path: {}", rollout_path.display());
+    let text = std::fs::read_to_string(&rollout_path).unwrap_or_else(|e| {
+        panic!(
+            "failed to read rollout file {}: {e}",
+            rollout_path.display()
+        )
+    });
+    let mut api_turn_count = 0usize;
+    let mut saw_compacted_summary = false;
+    for line in text.lines() {
+        let trimmed = line.trim();
+        if trimmed.is_empty() {
+            continue;
+        }
+        let Ok(entry): Result<RolloutLine, _> = serde_json::from_str(trimmed) else {
+            continue;
+        };
+        match entry.item {
+            RolloutItem::TurnContext(_) => {
+                api_turn_count += 1;
+            }
+            RolloutItem::Compacted(ci) => {
+                if ci.message == SUMMARY_TEXT {
+                    saw_compacted_summary = true;
+                }
+            }
+            _ => {}
+        }
+    }
+
+    assert!(
+        api_turn_count == 3,
+        "expected three APITurn entries in rollout"
+    );
+    assert!(
+        saw_compacted_summary,
+        "expected a Compacted entry containing the summarizer output"
+    );
+}
+
+// Windows CI only: bump to 4 workers to prevent SSE/event starvation and test timeouts.
+#[cfg_attr(windows, tokio::test(flavor = "multi_thread", worker_threads = 4))]
+#[cfg_attr(not(windows), tokio::test(flavor = "multi_thread", worker_threads = 2))]
+async fn auto_compact_runs_after_token_limit_hit() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let server = start_mock_server().await;
+
+    let sse1 = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed_with_tokens("r1", 70_000),
+    ]);
+
+    let sse2 = sse(vec![
+        ev_assistant_message("m2", "SECOND_REPLY"),
+        ev_completed_with_tokens("r2", 330_000),
+    ]);
+
+    let sse3 = sse(vec![
+        ev_assistant_message("m3", AUTO_SUMMARY_TEXT),
+        ev_completed_with_tokens("r3", 200),
+    ]);
+
+    let first_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(FIRST_AUTO_MSG)
+            && !body.contains(SECOND_AUTO_MSG)
+            && !body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(first_matcher)
+        .respond_with(sse_response(sse1))
+        .mount(&server)
+        .await;
+
+    let second_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(SECOND_AUTO_MSG)
+            && body.contains(FIRST_AUTO_MSG)
+            && !body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(second_matcher)
+        .respond_with(sse_response(sse2))
+        .mount(&server)
+        .await;
+
+    let third_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(third_matcher)
+        .respond_with(sse_response(sse3))
+        .mount(&server)
+        .await;
+
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    let home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&home);
+    config.model_provider = model_provider;
+    config.model_auto_compact_token_limit = Some(200_000);
+    let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
+    let codex = conversation_manager
+        .new_conversation(config)
+        .await
+        .unwrap()
+        .conversation;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: FIRST_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: SECOND_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+    // wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    let requests = server.received_requests().await.unwrap();
+    assert!(
+        requests.len() >= 3,
+        "auto compact should add at least a third request, got {}",
+        requests.len()
+    );
+    let is_auto_compact = |req: &wiremock::Request| {
+        std::str::from_utf8(&req.body)
+            .unwrap_or("")
+            .contains("You have exceeded the maximum number of tokens")
+    };
+    let auto_compact_count = requests.iter().filter(|req| is_auto_compact(req)).count();
+    assert_eq!(
+        auto_compact_count, 1,
+        "expected exactly one auto compact request"
+    );
+    let auto_compact_index = requests
+        .iter()
+        .enumerate()
+        .find_map(|(idx, req)| is_auto_compact(req).then_some(idx))
+        .expect("auto compact request missing");
+    assert_eq!(
+        auto_compact_index, 2,
+        "auto compact should add a third request"
+    );
+
+    let body3 = requests[auto_compact_index]
+        .body_json::<serde_json::Value>()
+        .unwrap();
+    let instructions = body3
+        .get("instructions")
+        .and_then(|v| v.as_str())
+        .unwrap_or_default();
+    assert!(
+        instructions.contains("You have exceeded the maximum number of tokens"),
+        "auto compact should reuse summarization instructions"
+    );
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn auto_compact_persists_rollout_entries() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let server = start_mock_server().await;
+
+    let sse1 = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed_with_tokens("r1", 70_000),
+    ]);
+
+    let sse2 = sse(vec![
+        ev_assistant_message("m2", "SECOND_REPLY"),
+        ev_completed_with_tokens("r2", 330_000),
+    ]);
+
+    let sse3 = sse(vec![
+        ev_assistant_message("m3", AUTO_SUMMARY_TEXT),
+        ev_completed_with_tokens("r3", 200),
+    ]);
+
+    let first_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(FIRST_AUTO_MSG)
+            && !body.contains(SECOND_AUTO_MSG)
+            && !body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(first_matcher)
+        .respond_with(sse_response(sse1))
+        .mount(&server)
+        .await;
+
+    let second_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(SECOND_AUTO_MSG)
+            && body.contains(FIRST_AUTO_MSG)
+            && !body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(second_matcher)
+        .respond_with(sse_response(sse2))
+        .mount(&server)
+        .await;
+
+    let third_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(third_matcher)
+        .respond_with(sse_response(sse3))
+        .mount(&server)
+        .await;
+
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    let home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&home);
+    config.model_provider = model_provider;
+    let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
+    let NewConversation {
+        conversation: codex,
+        session_configured,
+        ..
+    } = conversation_manager.new_conversation(config).await.unwrap();
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: FIRST_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: SECOND_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    codex.submit(Op::Shutdown).await.unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::ShutdownComplete)).await;
+
+    let rollout_path = session_configured.rollout_path;
+    let text = std::fs::read_to_string(&rollout_path).unwrap_or_else(|e| {
+        panic!(
+            "failed to read rollout file {}: {e}",
+            rollout_path.display()
+        )
+    });
+
+    let mut turn_context_count = 0usize;
+    for line in text.lines() {
+        let trimmed = line.trim();
+        if trimmed.is_empty() {
+            continue;
+        }
+        let Ok(entry): Result<RolloutLine, _> = serde_json::from_str(trimmed) else {
+            continue;
+        };
+        match entry.item {
+            RolloutItem::TurnContext(_) => {
+                turn_context_count += 1;
+            }
+            RolloutItem::Compacted(_) => {}
+            _ => {}
+        }
+    }
+
+    assert!(
+        turn_context_count >= 2,
+        "expected at least two turn context entries, got {turn_context_count}"
+    );
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn auto_compact_stops_after_failed_attempt() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let server = start_mock_server().await;
+
+    let sse1 = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed_with_tokens("r1", 500),
+    ]);
+
+    let sse2 = sse(vec![
+        ev_assistant_message("m2", SUMMARY_TEXT),
+        ev_completed_with_tokens("r2", 50),
+    ]);
+
+    let sse3 = sse(vec![
+        ev_assistant_message("m3", STILL_TOO_BIG_REPLY),
+        ev_completed_with_tokens("r3", 500),
+    ]);
+
+    let first_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(FIRST_AUTO_MSG)
+            && !body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(first_matcher)
+        .respond_with(sse_response(sse1.clone()))
+        .mount(&server)
+        .await;
+
+    let second_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(second_matcher)
+        .respond_with(sse_response(sse2.clone()))
+        .mount(&server)
+        .await;
+
+    let third_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        !body.contains("You have exceeded the maximum number of tokens")
+            && body.contains(SUMMARY_TEXT)
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(third_matcher)
+        .respond_with(sse_response(sse3.clone()))
+        .mount(&server)
+        .await;
+
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    let home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&home);
+    config.model_provider = model_provider;
+    config.model_auto_compact_token_limit = Some(200);
+    let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
+    let codex = conversation_manager
+        .new_conversation(config)
+        .await
+        .unwrap()
+        .conversation;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: FIRST_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+
+    let error_event = wait_for_event(&codex, |ev| matches!(ev, EventMsg::Error(_))).await;
+    let EventMsg::Error(ErrorEvent { message }) = error_event else {
+        panic!("expected error event");
+    };
+    assert!(
+        message.contains("limit"),
+        "error message should include limit information: {message}"
+    );
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    let requests = server.received_requests().await.unwrap();
+    assert_eq!(
+        requests.len(),
+        3,
+        "auto compact should attempt at most one summarization before erroring"
+    );
+
+    let last_body = requests[2].body_json::<serde_json::Value>().unwrap();
+    let instructions = last_body
+        .get("instructions")
+        .and_then(|v| v.as_str())
+        .unwrap_or_default();
+    assert!(
+        !instructions.contains("You have exceeded the maximum number of tokens"),
+        "third request should be the follow-up turn, not another summarization"
+    );
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn auto_compact_allows_multiple_attempts_when_interleaved_with_other_turn_events() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let server = start_mock_server().await;
+
+    let sse1 = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed_with_tokens("r1", 500),
+    ]);
+    let sse2 = sse(vec![
+        ev_assistant_message("m2", FIRST_AUTO_SUMMARY),
+        ev_completed_with_tokens("r2", 50),
+    ]);
+    let sse3 = sse(vec![
+        ev_function_call(DUMMY_CALL_ID, DUMMY_FUNCTION_NAME, "{}"),
+        ev_completed_with_tokens("r3", 150),
+    ]);
+    let sse4 = sse(vec![
+        ev_assistant_message("m4", SECOND_LARGE_REPLY),
+        ev_completed_with_tokens("r4", 450),
+    ]);
+    let sse5 = sse(vec![
+        ev_assistant_message("m5", SECOND_AUTO_SUMMARY),
+        ev_completed_with_tokens("r5", 60),
+    ]);
+    let sse6 = sse(vec![
+        ev_assistant_message("m6", FINAL_REPLY),
+        ev_completed_with_tokens("r6", 120),
+    ]);
+
+    #[derive(Clone)]
+    struct SeqResponder {
+        bodies: Arc<Vec<String>>,
+        calls: Arc<AtomicUsize>,
+        requests: Arc<Mutex<Vec<Vec<u8>>>>,
+    }
+
+    impl SeqResponder {
+        fn new(bodies: Vec<String>) -> Self {
+            Self {
+                bodies: Arc::new(bodies),
+                calls: Arc::new(AtomicUsize::new(0)),
+                requests: Arc::new(Mutex::new(Vec::new())),
+            }
+        }
+
+        fn recorded_requests(&self) -> Vec<Vec<u8>> {
+            self.requests.lock().unwrap().clone()
+        }
+    }
+
+    impl Respond for SeqResponder {
+        fn respond(&self, req: &Request) -> ResponseTemplate {
+            let idx = self.calls.fetch_add(1, Ordering::SeqCst);
+            self.requests.lock().unwrap().push(req.body.clone());
+            let body = self
+                .bodies
+                .get(idx)
+                .unwrap_or_else(|| panic!("unexpected request index {idx}"))
+                .clone();
+            ResponseTemplate::new(200)
+                .insert_header("content-type", "text/event-stream")
+                .set_body_raw(body, "text/event-stream")
+        }
+    }
+
+    let responder = SeqResponder::new(vec![sse1, sse2, sse3, sse4, sse5, sse6]);
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .respond_with(responder.clone())
+        .expect(6)
+        .mount(&server)
+        .await;
+
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    let home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&home);
+    config.model_provider = model_provider;
+    config.model_auto_compact_token_limit = Some(200);
+    let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
+    let codex = conversation_manager
+        .new_conversation(config)
+        .await
+        .unwrap()
+        .conversation;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: MULTI_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+
+    loop {
+        let event = codex.next_event().await.unwrap();
+        if let EventMsg::TaskComplete(_) = &event.msg
+            && !event.id.starts_with("auto-compact-")
+        {
+            break;
+        }
+    }
+
+    let request_bodies: Vec<String> = responder
+        .recorded_requests()
+        .into_iter()
+        .map(|body| String::from_utf8(body).unwrap_or_default())
+        .collect();
+    assert_eq!(
+        request_bodies.len(),
+        6,
+        "expected six requests including two auto compactions"
+    );
+    assert!(
+        request_bodies[0].contains(MULTI_AUTO_MSG),
+        "first request should contain the user input"
+    );
+    assert!(
+        request_bodies[1].contains("You have exceeded the maximum number of tokens"),
+        "first auto compact request should use summarization instructions"
+    );
+    assert!(
+        request_bodies[3].contains(&format!("unsupported call: {DUMMY_FUNCTION_NAME}")),
+        "function call output should be sent before the second auto compact"
+    );
+    assert!(
+        request_bodies[4].contains("You have exceeded the maximum number of tokens"),
+        "second auto compact request should reuse summarization instructions"
+    );
 }

codex-rs/core/tests/suite/compact_resume_fork.rs

@@ -0,0 +1,838 @@
+#![allow(clippy::expect_used)]
+
+//! Integration tests that cover compacting, resuming, and forking conversations.
+//!
+//! Each test sets up a mocked SSE conversation and drives the conversation through
+//! a specific sequence of operations. After every operation we capture the
+//! request payload that Codex would send to the model and assert that the
+//! model-visible history matches the expected sequence of messages.
+
+use super::compact::FIRST_REPLY;
+use super::compact::SUMMARIZE_TRIGGER;
+use super::compact::SUMMARY_TEXT;
+use super::compact::ev_assistant_message;
+use super::compact::ev_completed;
+use super::compact::mount_sse_once;
+use super::compact::sse;
+use codex_core::CodexAuth;
+use codex_core::CodexConversation;
+use codex_core::ConversationManager;
+use codex_core::ModelProviderInfo;
+use codex_core::NewConversation;
+use codex_core::built_in_model_providers;
+use codex_core::config::Config;
+use codex_core::protocol::ConversationPathResponseEvent;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::InputItem;
+use codex_core::protocol::Op;
+use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
+use core_test_support::load_default_config_for_test;
+use core_test_support::wait_for_event;
+use pretty_assertions::assert_eq;
+use serde_json::Value;
+use serde_json::json;
+use std::sync::Arc;
+use tempfile::TempDir;
+use wiremock::MockServer;
+
+const AFTER_SECOND_RESUME: &str = "AFTER_SECOND_RESUME";
+
+fn network_disabled() -> bool {
+    std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok()
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+/// Scenario: compact an initial conversation, resume it, fork one turn back, and
+/// ensure the model-visible history matches expectations at each request.
+async fn compact_resume_and_fork_preserve_model_history_view() {
+    if network_disabled() {
+        println!("Skipping test because network is disabled in this sandbox");
+        return;
+    }
+
+    // 1. Arrange mocked SSE responses for the initial compact/resume/fork flow.
+    let server = MockServer::start().await;
+    mount_initial_flow(&server).await;
+
+    // 2. Start a new conversation and drive it through the compact/resume/fork steps.
+    let (_home, config, manager, base) = start_test_conversation(&server).await;
+
+    user_turn(&base, "hello world").await;
+    compact_conversation(&base).await;
+    user_turn(&base, "AFTER_COMPACT").await;
+    let base_path = fetch_conversation_path(&base, "base conversation").await;
+    assert!(
+        base_path.exists(),
+        "compact+resume test expects base path {base_path:?} to exist",
+    );
+
+    let resumed = resume_conversation(&manager, &config, base_path).await;
+    user_turn(&resumed, "AFTER_RESUME").await;
+    let resumed_path = fetch_conversation_path(&resumed, "resumed conversation").await;
+    assert!(
+        resumed_path.exists(),
+        "compact+resume test expects resumed path {resumed_path:?} to exist",
+    );
+
+    let forked = fork_conversation(&manager, &config, resumed_path, 4).await;
+    user_turn(&forked, "AFTER_FORK").await;
+
+    // 3. Capture the requests to the model and validate the history slices.
+    let requests = gather_request_bodies(&server).await;
+
+    // input after compact is a prefix of input after resume/fork
+    let input_after_compact = json!(requests[requests.len() - 3]["input"]);
+    let input_after_resume = json!(requests[requests.len() - 2]["input"]);
+    let input_after_fork = json!(requests[requests.len() - 1]["input"]);
+
+    let compact_arr = input_after_compact
+        .as_array()
+        .expect("input after compact should be an array");
+    let resume_arr = input_after_resume
+        .as_array()
+        .expect("input after resume should be an array");
+    let fork_arr = input_after_fork
+        .as_array()
+        .expect("input after fork should be an array");
+
+    assert!(
+        compact_arr.len() <= resume_arr.len(),
+        "after-resume input should have at least as many items as after-compact",
+    );
+    assert_eq!(compact_arr.as_slice(), &resume_arr[..compact_arr.len()]);
+    eprint!(
+        "len of compact: {}, len of fork: {}",
+        compact_arr.len(),
+        fork_arr.len()
+    );
+    eprintln!("input_after_fork:{}", json!(input_after_fork));
+    assert!(
+        compact_arr.len() <= fork_arr.len(),
+        "after-fork input should have at least as many items as after-compact",
+    );
+    assert_eq!(compact_arr.as_slice(), &fork_arr[..compact_arr.len()]);
+
+    let prompt = requests[0]["instructions"]
+        .as_str()
+        .unwrap_or_default()
+        .to_string();
+    let user_instructions = requests[0]["input"][0]["content"][0]["text"]
+        .as_str()
+        .unwrap_or_default()
+        .to_string();
+    let environment_context = requests[0]["input"][1]["content"][0]["text"]
+        .as_str()
+        .unwrap_or_default()
+        .to_string();
+    let tool_calls = json!(requests[0]["tools"].as_array());
+    let prompt_cache_key = requests[0]["prompt_cache_key"]
+        .as_str()
+        .unwrap_or_default()
+        .to_string();
+    let fork_prompt_cache_key = requests[requests.len() - 1]["prompt_cache_key"]
+        .as_str()
+        .unwrap_or_default()
+        .to_string();
+    let user_turn_1 = json!(
+    {
+      "model": "gpt-5",
+      "instructions": prompt,
+      "input": [
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": user_instructions
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": environment_context
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "hello world"
+            }
+          ]
+        }
+      ],
+      "tools": tool_calls,
+      "tool_choice": "auto",
+      "parallel_tool_calls": false,
+      "reasoning": {
+        "summary": "auto"
+      },
+      "store": false,
+      "stream": true,
+      "include": [
+        "reasoning.encrypted_content"
+      ],
+      "prompt_cache_key": prompt_cache_key
+    });
+    let compact_1 = json!(
+    {
+      "model": "gpt-5",
+      "instructions": "You have exceeded the maximum number of tokens, please stop coding and instead write a short memento message for the next agent. Your note should:
+- Summarize what you finished and what still needs work. If there was a recent update_plan call, repeat its steps verbatim.
+- List outstanding TODOs with file paths / line numbers so they're easy to find.
+- Flag code that needs more tests (edge cases, performance, integration, etc.).
+- Record any open bugs, quirks, or setup steps that will make it easier for the next agent to pick up where you left off.",
+      "input": [
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": user_instructions
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": environment_context
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "hello world"
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "assistant",
+          "content": [
+            {
+              "type": "output_text",
+              "text": "FIRST_REPLY"
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "Start Summarization"
+            }
+          ]
+        }
+      ],
+      "tools": [],
+      "tool_choice": "auto",
+      "parallel_tool_calls": false,
+      "reasoning": {
+        "summary": "auto"
+      },
+      "store": false,
+      "stream": true,
+      "include": [
+        "reasoning.encrypted_content"
+      ],
+      "prompt_cache_key": prompt_cache_key
+    });
+    let user_turn_2_after_compact = json!(
+    {
+      "model": "gpt-5",
+      "instructions": prompt,
+      "input": [
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": user_instructions
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": environment_context
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "You were originally given instructions from a user over one or more turns. Here were the user messages:
+
+hello world
+
+Another language model started to solve this problem and produced a summary of its thinking process. You also have access to the state of the tools that were used by that language model. Use this to build on the work that has already been done and avoid duplicating work. Here is the summary produced by the other language model, use the information in this summary to assist with your own analysis:
+
+SUMMARY_ONLY_CONTEXT"
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "AFTER_COMPACT"
+            }
+          ]
+        }
+      ],
+      "tools": tool_calls,
+      "tool_choice": "auto",
+      "parallel_tool_calls": false,
+      "reasoning": {
+        "summary": "auto"
+      },
+      "store": false,
+      "stream": true,
+      "include": [
+        "reasoning.encrypted_content"
+      ],
+      "prompt_cache_key": prompt_cache_key
+    });
+    let usert_turn_3_after_resume = json!(
+    {
+      "model": "gpt-5",
+      "instructions": prompt,
+      "input": [
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": user_instructions
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": environment_context
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "You were originally given instructions from a user over one or more turns. Here were the user messages:
+
+hello world
+
+Another language model started to solve this problem and produced a summary of its thinking process. You also have access to the state of the tools that were used by that language model. Use this to build on the work that has already been done and avoid duplicating work. Here is the summary produced by the other language model, use the information in this summary to assist with your own analysis:
+
+SUMMARY_ONLY_CONTEXT"
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "AFTER_COMPACT"
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "assistant",
+          "content": [
+            {
+              "type": "output_text",
+              "text": "AFTER_COMPACT_REPLY"
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "AFTER_RESUME"
+            }
+          ]
+        }
+      ],
+      "tools": tool_calls,
+      "tool_choice": "auto",
+      "parallel_tool_calls": false,
+      "reasoning": {
+        "summary": "auto"
+      },
+      "store": false,
+      "stream": true,
+      "include": [
+        "reasoning.encrypted_content"
+      ],
+      "prompt_cache_key": prompt_cache_key
+    });
+    let user_turn_3_after_fork = json!(
+    {
+      "model": "gpt-5",
+      "instructions": prompt,
+      "input": [
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": user_instructions
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": environment_context
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "You were originally given instructions from a user over one or more turns. Here were the user messages:
+
+hello world
+
+Another language model started to solve this problem and produced a summary of its thinking process. You also have access to the state of the tools that were used by that language model. Use this to build on the work that has already been done and avoid duplicating work. Here is the summary produced by the other language model, use the information in this summary to assist with your own analysis:
+
+SUMMARY_ONLY_CONTEXT"
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "AFTER_COMPACT"
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "assistant",
+          "content": [
+            {
+              "type": "output_text",
+              "text": "AFTER_COMPACT_REPLY"
+            }
+          ]
+        },
+        {
+          "type": "message",
+          "role": "user",
+          "content": [
+            {
+              "type": "input_text",
+              "text": "AFTER_FORK"
+            }
+          ]
+        }
+      ],
+      "tools": tool_calls,
+      "tool_choice": "auto",
+      "parallel_tool_calls": false,
+      "reasoning": {
+        "summary": "auto"
+      },
+      "store": false,
+      "stream": true,
+      "include": [
+        "reasoning.encrypted_content"
+      ],
+      "prompt_cache_key": fork_prompt_cache_key
+    });
+    let expected = json!([
+        user_turn_1,
+        compact_1,
+        user_turn_2_after_compact,
+        usert_turn_3_after_resume,
+        user_turn_3_after_fork
+    ]);
+    assert_eq!(requests.len(), 5);
+    assert_eq!(json!(requests), expected);
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+/// Scenario: after the forked branch is compacted, resuming again should reuse
+/// the compacted history and only append the new user message.
+async fn compact_resume_after_second_compaction_preserves_history() {
+    if network_disabled() {
+        println!("Skipping test because network is disabled in this sandbox");
+        return;
+    }
+
+    // 1. Arrange mocked SSE responses for the initial flow plus the second compact.
+    let server = MockServer::start().await;
+    mount_initial_flow(&server).await;
+    mount_second_compact_flow(&server).await;
+
+    // 2. Drive the conversation through compact -> resume -> fork -> compact -> resume.
+    let (_home, config, manager, base) = start_test_conversation(&server).await;
+
+    user_turn(&base, "hello world").await;
+    compact_conversation(&base).await;
+    user_turn(&base, "AFTER_COMPACT").await;
+    let base_path = fetch_conversation_path(&base, "base conversation").await;
+    assert!(
+        base_path.exists(),
+        "second compact test expects base path {base_path:?} to exist",
+    );
+
+    let resumed = resume_conversation(&manager, &config, base_path).await;
+    user_turn(&resumed, "AFTER_RESUME").await;
+    let resumed_path = fetch_conversation_path(&resumed, "resumed conversation").await;
+    assert!(
+        resumed_path.exists(),
+        "second compact test expects resumed path {resumed_path:?} to exist",
+    );
+
+    let forked = fork_conversation(&manager, &config, resumed_path, 1).await;
+    user_turn(&forked, "AFTER_FORK").await;
+
+    compact_conversation(&forked).await;
+    user_turn(&forked, "AFTER_COMPACT_2").await;
+    let forked_path = fetch_conversation_path(&forked, "forked conversation").await;
+    assert!(
+        forked_path.exists(),
+        "second compact test expects forked path {forked_path:?} to exist",
+    );
+
+    let resumed_again = resume_conversation(&manager, &config, forked_path).await;
+    user_turn(&resumed_again, AFTER_SECOND_RESUME).await;
+
+    let requests = gather_request_bodies(&server).await;
+    let input_after_compact = json!(requests[requests.len() - 2]["input"]);
+    let input_after_resume = json!(requests[requests.len() - 1]["input"]);
+
+    // test input after compact before resume is the same as input after resume
+    let compact_input_array = input_after_compact
+        .as_array()
+        .expect("input after compact should be an array");
+    let resume_input_array = input_after_resume
+        .as_array()
+        .expect("input after resume should be an array");
+    assert!(
+        compact_input_array.len() <= resume_input_array.len(),
+        "after-resume input should have at least as many items as after-compact"
+    );
+    assert_eq!(
+        compact_input_array.as_slice(),
+        &resume_input_array[..compact_input_array.len()]
+    );
+    // hard coded test
+    let prompt = requests[0]["instructions"]
+        .as_str()
+        .unwrap_or_default()
+        .to_string();
+    let user_instructions = requests[0]["input"][0]["content"][0]["text"]
+        .as_str()
+        .unwrap_or_default()
+        .to_string();
+    let environment_instructions = requests[0]["input"][1]["content"][0]["text"]
+        .as_str()
+        .unwrap_or_default()
+        .to_string();
+
+    let expected = json!([
+      {
+        "instructions": prompt,
+        "input": [
+          {
+            "type": "message",
+            "role": "user",
+            "content": [
+              {
+                "type": "input_text",
+                "text": user_instructions
+              }
+            ]
+          },
+          {
+            "type": "message",
+            "role": "user",
+            "content": [
+              {
+                "type": "input_text",
+                "text": environment_instructions
+              }
+            ]
+          },
+          {
+            "type": "message",
+            "role": "user",
+            "content": [
+              {
+                "type": "input_text",
+                "text": "You were originally given instructions from a user over one or more turns. Here were the user messages:\n\nAFTER_FORK\n\nAnother language model started to solve this problem and produced a summary of its thinking process. You also have access to the state of the tools that were used by that language model. Use this to build on the work that has already been done and avoid duplicating work. Here is the summary produced by the other language model, use the information in this summary to assist with your own analysis:\n\nSUMMARY_ONLY_CONTEXT"
+              }
+            ]
+          },
+          {
+            "type": "message",
+            "role": "user",
+            "content": [
+              {
+                "type": "input_text",
+                "text": "AFTER_COMPACT_2"
+              }
+            ]
+          },
+          {
+            "type": "message",
+            "role": "user",
+            "content": [
+              {
+                "type": "input_text",
+                "text": "AFTER_SECOND_RESUME"
+              }
+            ]
+          }
+        ],
+      }
+    ]);
+    let last_request_after_2_compacts = json!([{
+        "instructions": requests[requests.len() -1]["instructions"],
+        "input": requests[requests.len() -1]["input"],
+    }]);
+    assert_eq!(expected, last_request_after_2_compacts);
+}
+
+fn normalize_line_endings(value: &mut Value) {
+    match value {
+        Value::String(text) => {
+            if text.contains('\r') {
+                *text = text.replace("\r\n", "\n").replace('\r', "\n");
+            }
+        }
+        Value::Array(items) => {
+            for item in items {
+                normalize_line_endings(item);
+            }
+        }
+        Value::Object(map) => {
+            for item in map.values_mut() {
+                normalize_line_endings(item);
+            }
+        }
+        _ => {}
+    }
+}
+
+async fn gather_request_bodies(server: &MockServer) -> Vec<Value> {
+    server
+        .received_requests()
+        .await
+        .expect("mock server should not fail")
+        .into_iter()
+        .map(|req| {
+            let mut value = req.body_json::<Value>().expect("valid JSON body");
+            normalize_line_endings(&mut value);
+            value
+        })
+        .collect()
+}
+
+async fn mount_initial_flow(server: &MockServer) {
+    let sse1 = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed("r1"),
+    ]);
+    let sse2 = sse(vec![
+        ev_assistant_message("m2", SUMMARY_TEXT),
+        ev_completed("r2"),
+    ]);
+    let sse3 = sse(vec![
+        ev_assistant_message("m3", "AFTER_COMPACT_REPLY"),
+        ev_completed("r3"),
+    ]);
+    let sse4 = sse(vec![ev_completed("r4")]);
+    let sse5 = sse(vec![ev_completed("r5")]);
+
+    let match_first = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("\"text\":\"hello world\"")
+            && !body.contains(&format!("\"text\":\"{SUMMARIZE_TRIGGER}\""))
+            && !body.contains("\"text\":\"AFTER_COMPACT\"")
+            && !body.contains("\"text\":\"AFTER_RESUME\"")
+            && !body.contains("\"text\":\"AFTER_FORK\"")
+    };
+    mount_sse_once(server, match_first, sse1).await;
+
+    let match_compact = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(&format!("\"text\":\"{SUMMARIZE_TRIGGER}\""))
+    };
+    mount_sse_once(server, match_compact, sse2).await;
+
+    let match_after_compact = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("\"text\":\"AFTER_COMPACT\"")
+            && !body.contains("\"text\":\"AFTER_RESUME\"")
+            && !body.contains("\"text\":\"AFTER_FORK\"")
+    };
+    mount_sse_once(server, match_after_compact, sse3).await;
+
+    let match_after_resume = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("\"text\":\"AFTER_RESUME\"")
+    };
+    mount_sse_once(server, match_after_resume, sse4).await;
+
+    let match_after_fork = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("\"text\":\"AFTER_FORK\"")
+    };
+    mount_sse_once(server, match_after_fork, sse5).await;
+}
+
+async fn mount_second_compact_flow(server: &MockServer) {
+    let sse6 = sse(vec![
+        ev_assistant_message("m4", SUMMARY_TEXT),
+        ev_completed("r6"),
+    ]);
+    let sse7 = sse(vec![ev_completed("r7")]);
+
+    let match_second_compact = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(&format!("\"text\":\"{SUMMARIZE_TRIGGER}\"")) && body.contains("AFTER_FORK")
+    };
+    mount_sse_once(server, match_second_compact, sse6).await;
+
+    let match_after_second_resume = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(&format!("\"text\":\"{AFTER_SECOND_RESUME}\""))
+    };
+    mount_sse_once(server, match_after_second_resume, sse7).await;
+}
+
+async fn start_test_conversation(
+    server: &MockServer,
+) -> (TempDir, Config, ConversationManager, Arc<CodexConversation>) {
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+    let home = TempDir::new().expect("create temp dir");
+    let mut config = load_default_config_for_test(&home);
+    config.model_provider = model_provider;
+
+    let manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
+    let NewConversation { conversation, .. } = manager
+        .new_conversation(config.clone())
+        .await
+        .expect("create conversation");
+
+    (home, config, manager, conversation)
+}
+
+async fn user_turn(conversation: &Arc<CodexConversation>, text: &str) {
+    conversation
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text { text: text.into() }],
+        })
+        .await
+        .expect("submit user turn");
+    wait_for_event(conversation, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+}
+
+async fn compact_conversation(conversation: &Arc<CodexConversation>) {
+    conversation
+        .submit(Op::Compact)
+        .await
+        .expect("compact conversation");
+    wait_for_event(conversation, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+}
+
+async fn fetch_conversation_path(
+    conversation: &Arc<CodexConversation>,
+    context: &str,
+) -> std::path::PathBuf {
+    conversation
+        .submit(Op::GetPath)
+        .await
+        .expect("request conversation path");
+    match wait_for_event(conversation, |ev| {
+        matches!(ev, EventMsg::ConversationPath(_))
+    })
+    .await
+    {
+        EventMsg::ConversationPath(ConversationPathResponseEvent { path, .. }) => path,
+        _ => panic!("expected ConversationPath event for {context}"),
+    }
+}
+
+async fn resume_conversation(
+    manager: &ConversationManager,
+    config: &Config,
+    path: std::path::PathBuf,
+) -> Arc<CodexConversation> {
+    let auth_manager =
+        codex_core::AuthManager::from_auth_for_testing(CodexAuth::from_api_key("dummy"));
+    let NewConversation { conversation, .. } = manager
+        .resume_conversation_from_rollout(config.clone(), path, auth_manager)
+        .await
+        .expect("resume conversation");
+    conversation
+}
+
+async fn fork_conversation(
+    manager: &ConversationManager,
+    config: &Config,
+    path: std::path::PathBuf,
+    back_steps: usize,
+) -> Arc<CodexConversation> {
+    let NewConversation { conversation, .. } = manager
+        .fork_conversation(back_steps, config.clone(), path)
+        .await
+        .expect("fork conversation");
+    conversation
+}

codex-rs/core/tests/suite/exec_stream_events.rs

@@ -2,8 +2,11 @@
 
 use std::collections::HashMap;
 use std::path::PathBuf;
+use std::time::Duration;
 
 use async_channel::Receiver;
+use codex_core::error::CodexErr;
+use codex_core::error::SandboxErr;
 use codex_core::exec::ExecParams;
 use codex_core::exec::SandboxType;
 use codex_core::exec::StdoutStream;
@@ -170,3 +173,36 @@ async fn test_aggregated_output_interleaves_in_order() {
     assert_eq!(result.aggregated_output.text, "O1\nE1\nO2\nE2\n");
     assert_eq!(result.aggregated_output.truncated_after_lines, None);
 }
+
+#[tokio::test]
+async fn test_exec_timeout_returns_partial_output() {
+    let cmd = vec![
+        "/bin/sh".to_string(),
+        "-c".to_string(),
+        "printf 'before\\n'; sleep 2; printf 'after\\n'".to_string(),
+    ];
+
+    let params = ExecParams {
+        command: cmd,
+        cwd: std::env::current_dir().unwrap_or_else(|_| PathBuf::from(".")),
+        timeout_ms: Some(200),
+        env: HashMap::new(),
+        with_escalated_permissions: None,
+        justification: None,
+    };
+
+    let policy = SandboxPolicy::new_read_only_policy();
+
+    let result = process_exec_tool_call(params, SandboxType::None, &policy, &None, None).await;
+
+    let Err(CodexErr::Sandbox(SandboxErr::Timeout { output })) = result else {
+        panic!("expected timeout error");
+    };
+
+    assert_eq!(output.exit_code, 124);
+    assert_eq!(output.stdout.text, "before\n");
+    assert!(output.stderr.text.is_empty());
+    assert_eq!(output.aggregated_output.text, "before\n");
+    assert!(output.duration >= Duration::from_millis(200));
+    assert!(output.timed_out);
+}

codex-rs/core/tests/suite/fork_conversation.rs

@@ -1,12 +1,16 @@
 use codex_core::CodexAuth;
+use codex_core::ContentItem;
 use codex_core::ConversationManager;
 use codex_core::ModelProviderInfo;
 use codex_core::NewConversation;
+use codex_core::ResponseItem;
 use codex_core::built_in_model_providers;
-use codex_core::protocol::ConversationHistoryResponseEvent;
+use codex_core::protocol::ConversationPathResponseEvent;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::InputItem;
 use codex_core::protocol::Op;
+use codex_core::protocol::RolloutItem;
+use codex_core::protocol::RolloutLine;
 use core_test_support::load_default_config_for_test;
 use core_test_support::wait_for_event;
 use tempfile::TempDir;
@@ -71,84 +75,119 @@ async fn fork_conversation_twice_drops_to_first_message() {
         let _ = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
     }
 
-    // Request history from the base conversation.
-    codex.submit(Op::GetHistory).await.unwrap();
+    // Request history from the base conversation to obtain rollout path.
+    codex.submit(Op::GetPath).await.unwrap();
     let base_history =
-        wait_for_event(&codex, |ev| matches!(ev, EventMsg::ConversationHistory(_))).await;
-
-    // Capture entries from the base history and compute expected prefixes after each fork.
-    let entries_after_three = match &base_history {
-        EventMsg::ConversationHistory(ConversationHistoryResponseEvent { entries, .. }) => {
-            entries.clone()
-        }
+        wait_for_event(&codex, |ev| matches!(ev, EventMsg::ConversationPath(_))).await;
+    let base_path = match &base_history {
+        EventMsg::ConversationPath(ConversationPathResponseEvent { path, .. }) => path.clone(),
         _ => panic!("expected ConversationHistory event"),
     };
-    // History layout for this test:
-    // [0] user instructions,
-    // [1] environment context,
-    // [2] "first" user message,
-    // [3] "second" user message,
-    // [4] "third" user message.
 
-    // Fork 1: drops the last user message and everything after.
-    let expected_after_first = vec![
-        entries_after_three[0].clone(),
-        entries_after_three[1].clone(),
-        entries_after_three[2].clone(),
-        entries_after_three[3].clone(),
-    ];
+    // GetHistory flushes before returning the path; no wait needed.
 
-    // Fork 2: drops the last user message and everything after.
-    // [0] user instructions,
-    // [1] environment context,
-    // [2] "first" user message,
-    let expected_after_second = vec![
-        entries_after_three[0].clone(),
-        entries_after_three[1].clone(),
-        entries_after_three[2].clone(),
-    ];
+    // Helper: read rollout items (excluding SessionMeta) from a JSONL path.
+    let read_items = |p: &std::path::Path| -> Vec<RolloutItem> {
+        let text = std::fs::read_to_string(p).expect("read rollout file");
+        let mut items: Vec<RolloutItem> = Vec::new();
+        for line in text.lines() {
+            if line.trim().is_empty() {
+                continue;
+            }
+            let v: serde_json::Value = serde_json::from_str(line).expect("jsonl line");
+            let rl: RolloutLine = serde_json::from_value(v).expect("rollout line");
+            match rl.item {
+                RolloutItem::SessionMeta(_) => {}
+                other => items.push(other),
+            }
+        }
+        items
+    };
 
-    // Fork once with n=1 → drops the last user message and everything after.
+    // Compute expected prefixes after each fork by truncating base rollout
+    // strictly before the nth user input (0-based).
+    let base_items = read_items(&base_path);
+    let find_user_input_positions = |items: &[RolloutItem]| -> Vec<usize> {
+        let mut pos = Vec::new();
+        for (i, it) in items.iter().enumerate() {
+            if let RolloutItem::ResponseItem(ResponseItem::Message { role, content, .. }) = it
+                && role == "user"
+            {
+                // Consider any user message as an input boundary; recorder stores both EventMsg and ResponseItem.
+                // We specifically look for input items, which are represented as ContentItem::InputText.
+                if content
+                    .iter()
+                    .any(|c| matches!(c, ContentItem::InputText { .. }))
+                {
+                    pos.push(i);
+                }
+            }
+        }
+        pos
+    };
+    let user_inputs = find_user_input_positions(&base_items);
+
+    // After cutting at nth user input (n=1 → second user message), cut strictly before that input.
+    let cut1 = user_inputs.get(1).copied().unwrap_or(0);
+    let expected_after_first: Vec<RolloutItem> = base_items[..cut1].to_vec();
+
+    // After dropping again (n=1 on fork1), compute expected relative to fork1's rollout.
+
+    // Fork once with n=1 → drops the last user input and everything after.
     let NewConversation {
         conversation: codex_fork1,
         ..
     } = conversation_manager
-        .fork_conversation(entries_after_three.clone(), 1, config_for_fork.clone())
+        .fork_conversation(1, config_for_fork.clone(), base_path.clone())
         .await
         .expect("fork 1");
 
-    codex_fork1.submit(Op::GetHistory).await.unwrap();
+    codex_fork1.submit(Op::GetPath).await.unwrap();
     let fork1_history = wait_for_event(&codex_fork1, |ev| {
-        matches!(ev, EventMsg::ConversationHistory(_))
+        matches!(ev, EventMsg::ConversationPath(_))
     })
     .await;
-    let entries_after_first_fork = match &fork1_history {
-        EventMsg::ConversationHistory(ConversationHistoryResponseEvent { entries, .. }) => {
-            assert!(matches!(
-                fork1_history,
-                EventMsg::ConversationHistory(ConversationHistoryResponseEvent { ref entries, .. }) if *entries == expected_after_first
-            ));
-            entries.clone()
-        }
+    let fork1_path = match &fork1_history {
+        EventMsg::ConversationPath(ConversationPathResponseEvent { path, .. }) => path.clone(),
         _ => panic!("expected ConversationHistory event after first fork"),
     };
 
-    // Fork again with n=1 → drops the (new) last user message, leaving only the first.
+    // GetHistory on fork1 flushed; the file is ready.
+    let fork1_items = read_items(&fork1_path);
+    pretty_assertions::assert_eq!(
+        serde_json::to_value(&fork1_items).unwrap(),
+        serde_json::to_value(&expected_after_first).unwrap()
+    );
+
+    // Fork again with n=0 → drops the (new) last user message, leaving only the first.
     let NewConversation {
         conversation: codex_fork2,
         ..
     } = conversation_manager
-        .fork_conversation(entries_after_first_fork.clone(), 1, config_for_fork.clone())
+        .fork_conversation(0, config_for_fork.clone(), fork1_path.clone())
         .await
         .expect("fork 2");
 
-    codex_fork2.submit(Op::GetHistory).await.unwrap();
+    codex_fork2.submit(Op::GetPath).await.unwrap();
     let fork2_history = wait_for_event(&codex_fork2, |ev| {
-        matches!(ev, EventMsg::ConversationHistory(_))
+        matches!(ev, EventMsg::ConversationPath(_))
     })
     .await;
-    assert!(matches!(
-        fork2_history,
-        EventMsg::ConversationHistory(ConversationHistoryResponseEvent { ref entries, .. }) if *entries == expected_after_second
-    ));
+    let fork2_path = match &fork2_history {
+        EventMsg::ConversationPath(ConversationPathResponseEvent { path, .. }) => path.clone(),
+        _ => panic!("expected ConversationHistory event after second fork"),
+    };
+    // GetHistory on fork2 flushed; the file is ready.
+    let fork1_items = read_items(&fork1_path);
+    let fork1_user_inputs = find_user_input_positions(&fork1_items);
+    let cut_last_on_fork1 = fork1_user_inputs
+        .get(fork1_user_inputs.len().saturating_sub(1))
+        .copied()
+        .unwrap_or(0);
+    let expected_after_second: Vec<RolloutItem> = fork1_items[..cut_last_on_fork1].to_vec();
+    let fork2_items = read_items(&fork2_path);
+    pretty_assertions::assert_eq!(
+        serde_json::to_value(&fork2_items).unwrap(),
+        serde_json::to_value(&expected_after_second).unwrap()
+    );
 }

codex-rs/core/tests/suite/mod.rs

@@ -3,11 +3,15 @@
 mod cli_stream;
 mod client;
 mod compact;
+mod compact_resume_fork;
 mod exec;
 mod exec_stream_events;
 mod fork_conversation;
 mod live_cli;
+mod model_overrides;
 mod prompt_caching;
+mod review;
+mod rollout_list_find;
 mod seatbelt;
 mod stream_error_allows_next_turn;
 mod stream_no_completed;

codex-rs/core/tests/suite/model_overrides.rs

@@ -0,0 +1,92 @@
+use codex_core::CodexAuth;
+use codex_core::ConversationManager;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::Op;
+use codex_core::protocol_config_types::ReasoningEffort;
+use core_test_support::load_default_config_for_test;
+use core_test_support::wait_for_event;
+use pretty_assertions::assert_eq;
+use tempfile::TempDir;
+
+const CONFIG_TOML: &str = "config.toml";
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn override_turn_context_does_not_persist_when_config_exists() {
+    let codex_home = TempDir::new().unwrap();
+    let config_path = codex_home.path().join(CONFIG_TOML);
+    let initial_contents = "model = \"gpt-4o\"\n";
+    tokio::fs::write(&config_path, initial_contents)
+        .await
+        .expect("seed config.toml");
+
+    let mut config = load_default_config_for_test(&codex_home);
+    config.model = "gpt-4o".to_string();
+
+    let conversation_manager =
+        ConversationManager::with_auth(CodexAuth::from_api_key("Test API Key"));
+    let codex = conversation_manager
+        .new_conversation(config)
+        .await
+        .expect("create conversation")
+        .conversation;
+
+    codex
+        .submit(Op::OverrideTurnContext {
+            cwd: None,
+            approval_policy: None,
+            sandbox_policy: None,
+            model: Some("o3".to_string()),
+            effort: Some(Some(ReasoningEffort::High)),
+            summary: None,
+        })
+        .await
+        .expect("submit override");
+
+    codex.submit(Op::Shutdown).await.expect("request shutdown");
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::ShutdownComplete)).await;
+
+    let contents = tokio::fs::read_to_string(&config_path)
+        .await
+        .expect("read config.toml after override");
+    assert_eq!(contents, initial_contents);
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn override_turn_context_does_not_create_config_file() {
+    let codex_home = TempDir::new().unwrap();
+    let config_path = codex_home.path().join(CONFIG_TOML);
+    assert!(
+        !config_path.exists(),
+        "test setup should start without config"
+    );
+
+    let config = load_default_config_for_test(&codex_home);
+
+    let conversation_manager =
+        ConversationManager::with_auth(CodexAuth::from_api_key("Test API Key"));
+    let codex = conversation_manager
+        .new_conversation(config)
+        .await
+        .expect("create conversation")
+        .conversation;
+
+    codex
+        .submit(Op::OverrideTurnContext {
+            cwd: None,
+            approval_policy: None,
+            sandbox_policy: None,
+            model: Some("o3".to_string()),
+            effort: Some(Some(ReasoningEffort::Medium)),
+            summary: None,
+        })
+        .await
+        .expect("submit override");
+
+    codex.submit(Op::Shutdown).await.expect("request shutdown");
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::ShutdownComplete)).await;
+
+    assert!(
+        !config_path.exists(),
+        "override should not create config.toml"
+    );
+}

codex-rs/core/tests/suite/prompt_caching.rs

@@ -387,7 +387,7 @@ async fn overrides_turn_context_but_keeps_cached_prefix_and_key_constant() {
                 exclude_slash_tmp: true,
             }),
             model: Some("o3".to_string()),
-            effort: Some(ReasoningEffort::High),
+            effort: Some(Some(ReasoningEffort::High)),
             summary: Some(ReasoningSummary::Detailed),
         })
         .await
@@ -426,11 +426,17 @@ async fn overrides_turn_context_but_keeps_cached_prefix_and_key_constant() {
     // After overriding the turn context, the environment context should be emitted again
     // reflecting the new approval policy and sandbox settings. Omit cwd because it did
     // not change.
-    let expected_env_text_2 = r#"<environment_context>
+    let expected_env_text_2 = format!(
+        r#"<environment_context>
   <approval_policy>never</approval_policy>
   <sandbox_mode>workspace-write</sandbox_mode>
   <network_access>enabled</network_access>
-</environment_context>"#;
+  <writable_roots>
+    <root>{}</root>
+  </writable_roots>
+</environment_context>"#,
+        writable.path().to_string_lossy()
+    );
     let expected_env_msg_2 = serde_json::json!({
         "type": "message",
         "role": "user",
@@ -513,7 +519,7 @@ async fn per_turn_overrides_keep_cached_prefix_and_key_constant() {
                 exclude_slash_tmp: true,
             },
             model: "o3".to_string(),
-            effort: ReasoningEffort::High,
+            effort: Some(ReasoningEffort::High),
             summary: ReasoningSummary::Detailed,
         })
         .await

codex-rs/core/tests/suite/review.rs

@@ -0,0 +1,598 @@
+use codex_core::CodexAuth;
+use codex_core::CodexConversation;
+use codex_core::ConversationManager;
+use codex_core::ModelProviderInfo;
+use codex_core::built_in_model_providers;
+use codex_core::config::Config;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::ExitedReviewModeEvent;
+use codex_core::protocol::InputItem;
+use codex_core::protocol::Op;
+use codex_core::protocol::ReviewCodeLocation;
+use codex_core::protocol::ReviewFinding;
+use codex_core::protocol::ReviewLineRange;
+use codex_core::protocol::ReviewOutputEvent;
+use codex_core::protocol::ReviewRequest;
+use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
+use core_test_support::load_default_config_for_test;
+use core_test_support::load_sse_fixture_with_id_from_str;
+use core_test_support::wait_for_event;
+use pretty_assertions::assert_eq;
+use std::path::PathBuf;
+use std::sync::Arc;
+use tempfile::TempDir;
+use tokio::io::AsyncWriteExt as _;
+use uuid::Uuid;
+use wiremock::Mock;
+use wiremock::MockServer;
+use wiremock::ResponseTemplate;
+use wiremock::matchers::method;
+use wiremock::matchers::path;
+
+/// Verify that submitting `Op::Review` spawns a child task and emits
+/// EnteredReviewMode -> ExitedReviewMode(None) -> TaskComplete
+/// in that order when the model returns a structured review JSON payload.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn review_op_emits_lifecycle_and_review_output() {
+    // Skip under Codex sandbox network restrictions.
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    // Start mock Responses API server. Return a single assistant message whose
+    // text is a JSON-encoded ReviewOutputEvent.
+    let review_json = serde_json::json!({
+        "findings": [
+            {
+                "title": "Prefer Stylize helpers",
+                "body": "Use .dim()/.bold() chaining instead of manual Style where possible.",
+                "confidence_score": 0.9,
+                "priority": 1,
+                "code_location": {
+                    "absolute_file_path": "/tmp/file.rs",
+                    "line_range": {"start": 10, "end": 20}
+                }
+            }
+        ],
+        "overall_correctness": "good",
+        "overall_explanation": "All good with some improvements suggested.",
+        "overall_confidence_score": 0.8
+    })
+    .to_string();
+    let sse_template = r#"[
+            {"type":"response.output_item.done", "item":{
+                "type":"message", "role":"assistant",
+                "content":[{"type":"output_text","text":__REVIEW__}]
+            }},
+            {"type":"response.completed", "response": {"id": "__ID__"}}
+        ]"#;
+    let review_json_escaped = serde_json::to_string(&review_json).unwrap();
+    let sse_raw = sse_template.replace("__REVIEW__", &review_json_escaped);
+    let server = start_responses_server_with_sse(&sse_raw, 1).await;
+    let codex_home = TempDir::new().unwrap();
+    let codex = new_conversation_for_server(&server, &codex_home, |_| {}).await;
+
+    // Submit review request.
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: "Please review my changes".to_string(),
+                user_facing_hint: "my changes".to_string(),
+            },
+        })
+        .await
+        .unwrap();
+
+    // Verify lifecycle: Entered -> Exited(Some(review)) -> TaskComplete.
+    let _entered = wait_for_event(&codex, |ev| matches!(ev, EventMsg::EnteredReviewMode(_))).await;
+    let closed = wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExitedReviewMode(_))).await;
+    let review = match closed {
+        EventMsg::ExitedReviewMode(ev) => ev
+            .review_output
+            .expect("expected ExitedReviewMode with Some(review_output)"),
+        other => panic!("expected ExitedReviewMode(..), got {other:?}"),
+    };
+
+    // Deep compare full structure using PartialEq (floats are f32 on both sides).
+    let expected = ReviewOutputEvent {
+        findings: vec![ReviewFinding {
+            title: "Prefer Stylize helpers".to_string(),
+            body: "Use .dim()/.bold() chaining instead of manual Style where possible.".to_string(),
+            confidence_score: 0.9,
+            priority: 1,
+            code_location: ReviewCodeLocation {
+                absolute_file_path: PathBuf::from("/tmp/file.rs"),
+                line_range: ReviewLineRange { start: 10, end: 20 },
+            },
+        }],
+        overall_correctness: "good".to_string(),
+        overall_explanation: "All good with some improvements suggested.".to_string(),
+        overall_confidence_score: 0.8,
+    };
+    assert_eq!(expected, review);
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    server.verify().await;
+}
+
+/// When the model returns plain text that is not JSON, ensure the child
+/// lifecycle still occurs and the plain text is surfaced via
+/// ExitedReviewMode(Some(..)) as the overall_explanation.
+// Windows CI only: bump to 4 workers to prevent SSE/event starvation and test timeouts.
+#[cfg_attr(windows, tokio::test(flavor = "multi_thread", worker_threads = 4))]
+#[cfg_attr(not(windows), tokio::test(flavor = "multi_thread", worker_threads = 2))]
+async fn review_op_with_plain_text_emits_review_fallback() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let sse_raw = r#"[
+        {"type":"response.output_item.done", "item":{
+            "type":"message", "role":"assistant",
+            "content":[{"type":"output_text","text":"just plain text"}]
+        }},
+        {"type":"response.completed", "response": {"id": "__ID__"}}
+    ]"#;
+    let server = start_responses_server_with_sse(sse_raw, 1).await;
+    let codex_home = TempDir::new().unwrap();
+    let codex = new_conversation_for_server(&server, &codex_home, |_| {}).await;
+
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: "Plain text review".to_string(),
+                user_facing_hint: "plain text review".to_string(),
+            },
+        })
+        .await
+        .unwrap();
+
+    let _entered = wait_for_event(&codex, |ev| matches!(ev, EventMsg::EnteredReviewMode(_))).await;
+    let closed = wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExitedReviewMode(_))).await;
+    let review = match closed {
+        EventMsg::ExitedReviewMode(ev) => ev
+            .review_output
+            .expect("expected ExitedReviewMode with Some(review_output)"),
+        other => panic!("expected ExitedReviewMode(..), got {other:?}"),
+    };
+
+    // Expect a structured fallback carrying the plain text.
+    let expected = ReviewOutputEvent {
+        overall_explanation: "just plain text".to_string(),
+        ..Default::default()
+    };
+    assert_eq!(expected, review);
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    server.verify().await;
+}
+
+/// When the model returns structured JSON in a review, ensure no AgentMessage
+/// is emitted; the UI consumes the structured result via ExitedReviewMode.
+// Windows CI only: bump to 4 workers to prevent SSE/event starvation and test timeouts.
+#[cfg_attr(windows, tokio::test(flavor = "multi_thread", worker_threads = 4))]
+#[cfg_attr(not(windows), tokio::test(flavor = "multi_thread", worker_threads = 2))]
+async fn review_does_not_emit_agent_message_on_structured_output() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let review_json = serde_json::json!({
+        "findings": [
+            {
+                "title": "Example",
+                "body": "Structured review output.",
+                "confidence_score": 0.5,
+                "priority": 1,
+                "code_location": {
+                    "absolute_file_path": "/tmp/file.rs",
+                    "line_range": {"start": 1, "end": 2}
+                }
+            }
+        ],
+        "overall_correctness": "ok",
+        "overall_explanation": "ok",
+        "overall_confidence_score": 0.5
+    })
+    .to_string();
+    let sse_template = r#"[
+            {"type":"response.output_item.done", "item":{
+                "type":"message", "role":"assistant",
+                "content":[{"type":"output_text","text":__REVIEW__}]
+            }},
+            {"type":"response.completed", "response": {"id": "__ID__"}}
+        ]"#;
+    let review_json_escaped = serde_json::to_string(&review_json).unwrap();
+    let sse_raw = sse_template.replace("__REVIEW__", &review_json_escaped);
+    let server = start_responses_server_with_sse(&sse_raw, 1).await;
+    let codex_home = TempDir::new().unwrap();
+    let codex = new_conversation_for_server(&server, &codex_home, |_| {}).await;
+
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: "check structured".to_string(),
+                user_facing_hint: "check structured".to_string(),
+            },
+        })
+        .await
+        .unwrap();
+
+    // Drain events until TaskComplete; ensure none are AgentMessage.
+    use tokio::time::Duration;
+    use tokio::time::timeout;
+    let mut saw_entered = false;
+    let mut saw_exited = false;
+    loop {
+        let ev = timeout(Duration::from_secs(5), codex.next_event())
+            .await
+            .expect("timeout waiting for event")
+            .expect("stream ended unexpectedly");
+        match ev.msg {
+            EventMsg::TaskComplete(_) => break,
+            EventMsg::AgentMessage(_) => {
+                panic!("unexpected AgentMessage during review with structured output")
+            }
+            EventMsg::EnteredReviewMode(_) => saw_entered = true,
+            EventMsg::ExitedReviewMode(_) => saw_exited = true,
+            _ => {}
+        }
+    }
+    assert!(saw_entered && saw_exited, "missing review lifecycle events");
+
+    server.verify().await;
+}
+
+/// Ensure that when a custom `review_model` is set in the config, the review
+/// request uses that model (and not the main chat model).
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn review_uses_custom_review_model_from_config() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    // Minimal stream: just a completed event
+    let sse_raw = r#"[
+        {"type":"response.completed", "response": {"id": "__ID__"}}
+    ]"#;
+    let server = start_responses_server_with_sse(sse_raw, 1).await;
+    let codex_home = TempDir::new().unwrap();
+    // Choose a review model different from the main model; ensure it is used.
+    let codex = new_conversation_for_server(&server, &codex_home, |cfg| {
+        cfg.model = "gpt-4.1".to_string();
+        cfg.review_model = "gpt-5".to_string();
+    })
+    .await;
+
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: "use custom model".to_string(),
+                user_facing_hint: "use custom model".to_string(),
+            },
+        })
+        .await
+        .unwrap();
+
+    // Wait for completion
+    let _entered = wait_for_event(&codex, |ev| matches!(ev, EventMsg::EnteredReviewMode(_))).await;
+    let _closed = wait_for_event(&codex, |ev| {
+        matches!(
+            ev,
+            EventMsg::ExitedReviewMode(ExitedReviewModeEvent {
+                review_output: None
+            })
+        )
+    })
+    .await;
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Assert the request body model equals the configured review model
+    let request = &server.received_requests().await.unwrap()[0];
+    let body = request.body_json::<serde_json::Value>().unwrap();
+    assert_eq!(body["model"].as_str().unwrap(), "gpt-5");
+
+    server.verify().await;
+}
+
+/// When a review session begins, it must not prepend prior chat history from
+/// the parent session. The request `input` should contain only the review
+/// prompt from the user.
+// Windows CI only: bump to 4 workers to prevent SSE/event starvation and test timeouts.
+#[cfg_attr(windows, tokio::test(flavor = "multi_thread", worker_threads = 4))]
+#[cfg_attr(not(windows), tokio::test(flavor = "multi_thread", worker_threads = 2))]
+async fn review_input_isolated_from_parent_history() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    // Mock server for the single review request
+    let sse_raw = r#"[
+        {"type":"response.completed", "response": {"id": "__ID__"}}
+    ]"#;
+    let server = start_responses_server_with_sse(sse_raw, 1).await;
+
+    // Seed a parent session history via resume file with both user + assistant items.
+    let codex_home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&codex_home);
+    config.model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    let session_file = codex_home.path().join("resume.jsonl");
+    {
+        let mut f = tokio::fs::File::create(&session_file).await.unwrap();
+        let convo_id = Uuid::new_v4();
+        // Proper session_meta line (enveloped) with a conversation id
+        let meta_line = serde_json::json!({
+            "timestamp": "2024-01-01T00:00:00.000Z",
+            "type": "session_meta",
+            "payload": {
+                "id": convo_id,
+                "timestamp": "2024-01-01T00:00:00Z",
+                "instructions": null,
+                "cwd": ".",
+                "originator": "test_originator",
+                "cli_version": "test_version"
+            }
+        });
+        f.write_all(format!("{meta_line}\n").as_bytes())
+            .await
+            .unwrap();
+
+        // Prior user message (enveloped response_item)
+        let user = codex_protocol::models::ResponseItem::Message {
+            id: None,
+            role: "user".to_string(),
+            content: vec![codex_protocol::models::ContentItem::InputText {
+                text: "parent: earlier user message".to_string(),
+            }],
+        };
+        let user_json = serde_json::to_value(&user).unwrap();
+        let user_line = serde_json::json!({
+            "timestamp": "2024-01-01T00:00:01.000Z",
+            "type": "response_item",
+            "payload": user_json
+        });
+        f.write_all(format!("{user_line}\n").as_bytes())
+            .await
+            .unwrap();
+
+        // Prior assistant message (enveloped response_item)
+        let assistant = codex_protocol::models::ResponseItem::Message {
+            id: None,
+            role: "assistant".to_string(),
+            content: vec![codex_protocol::models::ContentItem::OutputText {
+                text: "parent: assistant reply".to_string(),
+            }],
+        };
+        let assistant_json = serde_json::to_value(&assistant).unwrap();
+        let assistant_line = serde_json::json!({
+            "timestamp": "2024-01-01T00:00:02.000Z",
+            "type": "response_item",
+            "payload": assistant_json
+        });
+        f.write_all(format!("{assistant_line}\n").as_bytes())
+            .await
+            .unwrap();
+    }
+    let codex =
+        resume_conversation_for_server(&server, &codex_home, session_file.clone(), |_| {}).await;
+
+    // Submit review request; it must start fresh (no parent history in `input`).
+    let review_prompt = "Please review only this".to_string();
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: review_prompt.clone(),
+                user_facing_hint: review_prompt.clone(),
+            },
+        })
+        .await
+        .unwrap();
+
+    let _entered = wait_for_event(&codex, |ev| matches!(ev, EventMsg::EnteredReviewMode(_))).await;
+    let _closed = wait_for_event(&codex, |ev| {
+        matches!(
+            ev,
+            EventMsg::ExitedReviewMode(ExitedReviewModeEvent {
+                review_output: None
+            })
+        )
+    })
+    .await;
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Assert the request `input` contains only the single review user message.
+    let request = &server.received_requests().await.unwrap()[0];
+    let body = request.body_json::<serde_json::Value>().unwrap();
+    let expected_input = serde_json::json!([
+        {
+            "type": "message",
+            "role": "user",
+            "content": [{"type": "input_text", "text": review_prompt}]
+        }
+    ]);
+    assert_eq!(body["input"], expected_input);
+
+    server.verify().await;
+}
+
+/// After a review thread finishes, its conversation should not leak into the
+/// parent session. A subsequent parent turn must not include any review
+/// messages in its request `input`.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn review_history_does_not_leak_into_parent_session() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    // Respond to both the review request and the subsequent parent request.
+    let sse_raw = r#"[
+        {"type":"response.output_item.done", "item":{
+            "type":"message", "role":"assistant",
+            "content":[{"type":"output_text","text":"review assistant output"}]
+        }},
+        {"type":"response.completed", "response": {"id": "__ID__"}}
+    ]"#;
+    let server = start_responses_server_with_sse(sse_raw, 2).await;
+    let codex_home = TempDir::new().unwrap();
+    let codex = new_conversation_for_server(&server, &codex_home, |_| {}).await;
+
+    // 1) Run a review turn that produces an assistant message (isolated in child).
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: "Start a review".to_string(),
+                user_facing_hint: "Start a review".to_string(),
+            },
+        })
+        .await
+        .unwrap();
+    let _entered = wait_for_event(&codex, |ev| matches!(ev, EventMsg::EnteredReviewMode(_))).await;
+    let _closed = wait_for_event(&codex, |ev| {
+        matches!(
+            ev,
+            EventMsg::ExitedReviewMode(ExitedReviewModeEvent {
+                review_output: Some(_)
+            })
+        )
+    })
+    .await;
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // 2) Continue in the parent session; request input must not include any review items.
+    let followup = "back to parent".to_string();
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: followup.clone(),
+            }],
+        })
+        .await
+        .unwrap();
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Inspect the second request (parent turn) input contents.
+    // Parent turns include session initial messages (user_instructions, environment_context).
+    // Critically, no messages from the review thread should appear.
+    let requests = server.received_requests().await.unwrap();
+    assert_eq!(requests.len(), 2);
+    let body = requests[1].body_json::<serde_json::Value>().unwrap();
+    let input = body["input"].as_array().expect("input array");
+
+    // Must include the followup as the last item for this turn
+    let last = input.last().expect("at least one item in input");
+    assert_eq!(last["role"].as_str().unwrap(), "user");
+    let last_text = last["content"][0]["text"].as_str().unwrap();
+    assert_eq!(last_text, followup);
+
+    // Ensure no review-thread content leaked into the parent request
+    let contains_review_prompt = input
+        .iter()
+        .any(|msg| msg["content"][0]["text"].as_str().unwrap_or_default() == "Start a review");
+    let contains_review_assistant = input.iter().any(|msg| {
+        msg["content"][0]["text"].as_str().unwrap_or_default() == "review assistant output"
+    });
+    assert!(
+        !contains_review_prompt,
+        "review prompt leaked into parent turn input"
+    );
+    assert!(
+        !contains_review_assistant,
+        "review assistant output leaked into parent turn input"
+    );
+
+    server.verify().await;
+}
+
+/// Start a mock Responses API server and mount the given SSE stream body.
+async fn start_responses_server_with_sse(sse_raw: &str, expected_requests: usize) -> MockServer {
+    let server = MockServer::start().await;
+    let sse = load_sse_fixture_with_id_from_str(sse_raw, &Uuid::new_v4().to_string());
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .respond_with(
+            ResponseTemplate::new(200)
+                .insert_header("content-type", "text/event-stream")
+                .set_body_raw(sse.clone(), "text/event-stream"),
+        )
+        .expect(expected_requests as u64)
+        .mount(&server)
+        .await;
+    server
+}
+
+/// Create a conversation configured to talk to the provided mock server.
+#[expect(clippy::expect_used)]
+async fn new_conversation_for_server<F>(
+    server: &MockServer,
+    codex_home: &TempDir,
+    mutator: F,
+) -> Arc<CodexConversation>
+where
+    F: FnOnce(&mut Config),
+{
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+    let mut config = load_default_config_for_test(codex_home);
+    config.model_provider = model_provider;
+    mutator(&mut config);
+    let conversation_manager =
+        ConversationManager::with_auth(CodexAuth::from_api_key("Test API Key"));
+    conversation_manager
+        .new_conversation(config)
+        .await
+        .expect("create conversation")
+        .conversation
+}
+
+/// Create a conversation resuming from a rollout file, configured to talk to the provided mock server.
+#[expect(clippy::expect_used)]
+async fn resume_conversation_for_server<F>(
+    server: &MockServer,
+    codex_home: &TempDir,
+    resume_path: std::path::PathBuf,
+    mutator: F,
+) -> Arc<CodexConversation>
+where
+    F: FnOnce(&mut Config),
+{
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+    let mut config = load_default_config_for_test(codex_home);
+    config.model_provider = model_provider;
+    mutator(&mut config);
+    let conversation_manager =
+        ConversationManager::with_auth(CodexAuth::from_api_key("Test API Key"));
+    let auth_manager =
+        codex_core::AuthManager::from_auth_for_testing(CodexAuth::from_api_key("Test API Key"));
+    conversation_manager
+        .resume_conversation_from_rollout(config, resume_path, auth_manager)
+        .await
+        .expect("resume conversation")
+        .conversation
+}

codex-rs/core/tests/suite/rollout_list_find.rs

@@ -0,0 +1,50 @@
+#![allow(clippy::unwrap_used, clippy::expect_used)]
+use std::io::Write;
+use std::path::PathBuf;
+
+use codex_core::find_conversation_path_by_id_str;
+use tempfile::TempDir;
+use uuid::Uuid;
+
+/// Create sessions/YYYY/MM/DD and write a minimal rollout file containing the
+/// provided conversation id in the SessionMeta line. Returns the absolute path.
+fn write_minimal_rollout_with_id(codex_home: &TempDir, id: Uuid) -> PathBuf {
+    let sessions = codex_home.path().join("sessions/2024/01/01");
+    std::fs::create_dir_all(&sessions).unwrap();
+
+    let file = sessions.join(format!("rollout-2024-01-01T00-00-00-{id}.jsonl"));
+    let mut f = std::fs::File::create(&file).unwrap();
+    // Minimal first line: session_meta with the id so content search can find it
+    writeln!(
+        f,
+        "{}",
+        serde_json::json!({
+            "timestamp": "2024-01-01T00:00:00.000Z",
+            "type": "session_meta",
+            "payload": {
+                "id": id,
+                "timestamp": "2024-01-01T00:00:00Z",
+                "instructions": null,
+                "cwd": ".",
+                "originator": "test",
+                "cli_version": "test"
+            }
+        })
+    )
+    .unwrap();
+
+    file
+}
+
+#[tokio::test]
+async fn find_locates_rollout_file_by_id() {
+    let home = TempDir::new().unwrap();
+    let id = Uuid::new_v4();
+    let expected = write_minimal_rollout_with_id(&home, id);
+
+    let found = find_conversation_path_by_id_str(home.path(), &id.to_string())
+        .await
+        .unwrap();
+
+    assert_eq!(found.unwrap(), expected);
+}

codex-rs/exec/Cargo.toml

@@ -46,4 +46,6 @@ core_test_support = { path = "../core/tests/common" }
 libc = "0.2"
 predicates = "3"
 tempfile = "3.13.0"
+uuid = "1"
+walkdir = "2"
 wiremock = "0.6"

codex-rs/exec/src/cli.rs

@@ -6,6 +6,10 @@ use std::path::PathBuf;
 #[derive(Parser, Debug)]
 #[command(version)]
 pub struct Cli {
+    /// Action to perform. If omitted, runs a new non-interactive session.
+    #[command(subcommand)]
+    pub command: Option<Command>,
+
     /// Optional image(s) to attach to the initial prompt.
     #[arg(long = "image", short = 'i', value_name = "FILE", value_delimiter = ',', num_args = 1..)]
     pub images: Vec<PathBuf>,
@@ -48,6 +52,10 @@ pub struct Cli {
     #[arg(long = "skip-git-repo-check", default_value_t = false)]
     pub skip_git_repo_check: bool,
 
+    /// Force-enable the experimental apply_patch tool even for models that do not opt into it by default.
+    #[arg(long = "custom-apply-patch", default_value_t = false)]
+    pub custom_apply_patch: bool,
+
     #[clap(skip)]
     pub config_overrides: CliConfigOverrides,
 
@@ -69,6 +77,28 @@ pub struct Cli {
     pub prompt: Option<String>,
 }
 
+#[derive(Debug, clap::Subcommand)]
+pub enum Command {
+    /// Resume a previous session by id or pick the most recent with --last.
+    Resume(ResumeArgs),
+}
+
+#[derive(Parser, Debug)]
+pub struct ResumeArgs {
+    /// Conversation/session id (UUID). When provided, resumes this session.
+    /// If omitted, use --last to pick the most recent recorded session.
+    #[arg(value_name = "SESSION_ID")]
+    pub session_id: Option<String>,
+
+    /// Resume the most recent recorded session (newest) without specifying an id.
+    #[arg(long = "last", default_value_t = false, conflicts_with = "session_id")]
+    pub last: bool,
+
+    /// Prompt to send after resuming the session. If `-` is used, read from stdin.
+    #[arg(value_name = "PROMPT")]
+    pub prompt: Option<String>,
+}
+
 #[derive(Debug, Clone, Copy, Default, PartialEq, Eq, ValueEnum)]
 #[value(rename_all = "kebab-case")]
 pub enum Color {

codex-rs/exec/src/event_processor_with_human_output.rs

@@ -280,7 +280,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                 parsed_cmd: _,
             }) => {
                 self.call_id_to_command.insert(
-                    call_id.clone(),
+                    call_id,
                     ExecCommandBegin {
                         command: command.clone(),
                     },
@@ -382,7 +382,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                 // Store metadata so we can calculate duration later when we
                 // receive the corresponding PatchApplyEnd event.
                 self.call_id_to_patch.insert(
-                    call_id.clone(),
+                    call_id,
                     PatchApplyBegin {
                         start_time: Instant::now(),
                         auto_approved,
@@ -520,6 +520,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                 let SessionConfiguredEvent {
                     session_id: conversation_id,
                     model,
+                    reasoning_effort: _,
                     history_log_id: _,
                     history_entry_count: _,
                     initial_messages: _,
@@ -559,8 +560,10 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                 }
             },
             EventMsg::ShutdownComplete => return CodexStatus::Shutdown,
-            EventMsg::ConversationHistory(_) => {}
+            EventMsg::ConversationPath(_) => {}
             EventMsg::UserMessage(_) => {}
+            EventMsg::EnteredReviewMode(_) => {}
+            EventMsg::ExitedReviewMode(_) => {}
         }
         CodexStatus::Running
     }

codex-rs/exec/src/lib.rs

@@ -30,11 +30,14 @@ use tracing::error;
 use tracing::info;
 use tracing_subscriber::EnvFilter;
 
+use crate::cli::Command as ExecCommand;
 use crate::event_processor::CodexStatus;
 use crate::event_processor::EventProcessor;
+use codex_core::find_conversation_path_by_id_str;
 
 pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()> {
     let Cli {
+        command,
         images,
         model: model_cli_arg,
         oss,
@@ -43,6 +46,7 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
         dangerously_bypass_approvals_and_sandbox,
         cwd,
         skip_git_repo_check,
+        custom_apply_patch,
         color,
         last_message_file,
         json: json_mode,
@@ -51,8 +55,15 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
         config_overrides,
     } = cli;
 
-    // Determine the prompt based on CLI arg and/or stdin.
-    let prompt = match prompt {
+    // Determine the prompt source (parent or subcommand) and read from stdin if needed.
+    let prompt_arg = match &command {
+        // Allow prompt before the subcommand by falling back to the parent-level prompt
+        // when the Resume subcommand did not provide its own prompt.
+        Some(ExecCommand::Resume(args)) => args.prompt.clone().or(prompt),
+        None => prompt,
+    };
+
+    let prompt = match prompt_arg {
         Some(p) if p != "-" => p,
         // Either `-` was passed or no positional arg.
         maybe_dash => {
@@ -137,6 +148,7 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
     // Load configuration and determine approval policy
     let overrides = ConfigOverrides {
         model,
+        review_model: None,
         config_profile,
         // This CLI is intended to be headless and has no affordances for asking
         // the user for approval.
@@ -147,7 +159,7 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
         codex_linux_sandbox_exe,
         base_instructions: None,
         include_plan_tool: None,
-        include_apply_patch_tool: None,
+        include_apply_patch_tool: custom_apply_patch.then_some(true),
         include_view_image_tool: None,
         show_raw_agent_reasoning: oss.then_some(true),
         tools_web_search_request: None,
@@ -187,15 +199,31 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
         std::process::exit(1);
     }
 
-    let conversation_manager = ConversationManager::new(AuthManager::shared(
-        config.codex_home.clone(),
-        config.preferred_auth_method,
-    ));
+    let conversation_manager =
+        ConversationManager::new(AuthManager::shared(config.codex_home.clone()));
+
+    // Handle resume subcommand by resolving a rollout path and using explicit resume API.
     let NewConversation {
         conversation_id: _,
         conversation,
         session_configured,
-    } = conversation_manager.new_conversation(config).await?;
+    } = if let Some(ExecCommand::Resume(args)) = command {
+        let resume_path = resolve_resume_path(&config, &args).await?;
+
+        if let Some(path) = resume_path {
+            conversation_manager
+                .resume_conversation_from_rollout(
+                    config.clone(),
+                    path,
+                    AuthManager::shared(config.codex_home.clone()),
+                )
+                .await?
+        } else {
+            conversation_manager.new_conversation(config).await?
+        }
+    } else {
+        conversation_manager.new_conversation(config).await?
+    };
     info!("Codex initialized with event: {session_configured:?}");
 
     let (tx, mut rx) = tokio::sync::mpsc::unbounded_channel::<Event>();
@@ -280,3 +308,23 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
 
     Ok(())
 }
+
+async fn resolve_resume_path(
+    config: &Config,
+    args: &crate::cli::ResumeArgs,
+) -> anyhow::Result<Option<PathBuf>> {
+    if args.last {
+        match codex_core::RolloutRecorder::list_conversations(&config.codex_home, 1, None).await {
+            Ok(page) => Ok(page.items.first().map(|it| it.path.clone())),
+            Err(e) => {
+                error!("Error listing conversations: {e}");
+                Ok(None)
+            }
+        }
+    } else if let Some(id_str) = args.session_id.as_deref() {
+        let path = find_conversation_path_by_id_str(&config.codex_home, id_str).await?;
+        Ok(path)
+    } else {
+        Ok(None)
+    }
+}

codex-rs/exec/tests/fixtures/cli_responses_fixture.sse

@@ -0,0 +1,10 @@
+event: response.created
+data: {"type":"response.created","response":{"id":"resp1"}}
+
+event: response.output_item.done
+data: {"type":"response.output_item.done","item":{"type":"message","role":"assistant","content":[{"type":"output_text","text":"fixture hello"}]}}
+
+event: response.completed
+data: {"type":"response.completed","response":{"id":"resp1","output":[]}}
+
+

codex-rs/exec/tests/suite/common.rs

@@ -61,7 +61,7 @@ pub(crate) async fn run_e2e_exec_test(cwd: &Path, response_streams: Vec<String>)
         .context("should find binary for codex-exec")
         .expect("should find binary for codex-exec")
         .current_dir(cwd.clone())
-        .env("CODEX_HOME", cwd.clone())
+        .env("CODEX_HOME", cwd)
         .env("OPENAI_API_KEY", "dummy")
         .env("OPENAI_BASE_URL", format!("{uri}/v1"))
         .arg("--skip-git-repo-check")

codex-rs/exec/tests/suite/mod.rs

@@ -1,4 +1,5 @@
 // Aggregates all former standalone integration tests as modules.
 mod apply_patch;
 mod common;
+mod resume;
 mod sandbox;

codex-rs/exec/tests/suite/resume.rs

@@ -0,0 +1,267 @@
+#![allow(clippy::unwrap_used, clippy::expect_used)]
+use anyhow::Context;
+use assert_cmd::prelude::*;
+use serde_json::Value;
+use std::process::Command;
+use tempfile::TempDir;
+use uuid::Uuid;
+use walkdir::WalkDir;
+
+/// Utility: scan the sessions dir for a rollout file that contains `marker`
+/// in any response_item.message.content entry. Returns the absolute path.
+fn find_session_file_containing_marker(
+    sessions_dir: &std::path::Path,
+    marker: &str,
+) -> Option<std::path::PathBuf> {
+    for entry in WalkDir::new(sessions_dir) {
+        let entry = match entry {
+            Ok(e) => e,
+            Err(_) => continue,
+        };
+        if !entry.file_type().is_file() {
+            continue;
+        }
+        if !entry.file_name().to_string_lossy().ends_with(".jsonl") {
+            continue;
+        }
+        let path = entry.path();
+        let Ok(content) = std::fs::read_to_string(path) else {
+            continue;
+        };
+        // Skip the first meta line and scan remaining JSONL entries.
+        let mut lines = content.lines();
+        if lines.next().is_none() {
+            continue;
+        }
+        for line in lines {
+            if line.trim().is_empty() {
+                continue;
+            }
+            let Ok(item): Result<Value, _> = serde_json::from_str(line) else {
+                continue;
+            };
+            if item.get("type").and_then(|t| t.as_str()) == Some("response_item")
+                && let Some(payload) = item.get("payload")
+                && payload.get("type").and_then(|t| t.as_str()) == Some("message")
+                && payload
+                    .get("content")
+                    .map(|c| c.to_string())
+                    .unwrap_or_default()
+                    .contains(marker)
+            {
+                return Some(path.to_path_buf());
+            }
+        }
+    }
+    None
+}
+
+/// Extract the conversation UUID from the first SessionMeta line in the rollout file.
+fn extract_conversation_id(path: &std::path::Path) -> String {
+    let content = std::fs::read_to_string(path).unwrap();
+    let mut lines = content.lines();
+    let meta_line = lines.next().expect("missing meta line");
+    let meta: Value = serde_json::from_str(meta_line).expect("invalid meta json");
+    meta.get("payload")
+        .and_then(|p| p.get("id"))
+        .and_then(|v| v.as_str())
+        .unwrap_or_default()
+        .to_string()
+}
+
+#[test]
+fn exec_resume_last_appends_to_existing_file() -> anyhow::Result<()> {
+    let home = TempDir::new()?;
+    let fixture = std::path::Path::new(env!("CARGO_MANIFEST_DIR"))
+        .join("tests/fixtures/cli_responses_fixture.sse");
+
+    // 1) First run: create a session with a unique marker in the content.
+    let marker = format!("resume-last-{}", Uuid::new_v4());
+    let prompt = format!("echo {marker}");
+
+    Command::cargo_bin("codex-exec")
+        .context("should find binary for codex-exec")?
+        .env("CODEX_HOME", home.path())
+        .env("OPENAI_API_KEY", "dummy")
+        .env("CODEX_RS_SSE_FIXTURE", &fixture)
+        .env("OPENAI_BASE_URL", "http://unused.local")
+        .arg("--skip-git-repo-check")
+        .arg("-C")
+        .arg(env!("CARGO_MANIFEST_DIR"))
+        .arg(&prompt)
+        .assert()
+        .success();
+
+    // Find the created session file containing the marker.
+    let sessions_dir = home.path().join("sessions");
+    let path = find_session_file_containing_marker(&sessions_dir, &marker)
+        .expect("no session file found after first run");
+
+    // 2) Second run: resume the most recent file with a new marker.
+    let marker2 = format!("resume-last-2-{}", Uuid::new_v4());
+    let prompt2 = format!("echo {marker2}");
+
+    let mut binding = assert_cmd::Command::cargo_bin("codex-exec")
+        .context("should find binary for codex-exec")?;
+    let cmd = binding
+        .env("CODEX_HOME", home.path())
+        .env("OPENAI_API_KEY", "dummy")
+        .env("CODEX_RS_SSE_FIXTURE", &fixture)
+        .env("OPENAI_BASE_URL", "http://unused.local")
+        .arg("--skip-git-repo-check")
+        .arg("-C")
+        .arg(env!("CARGO_MANIFEST_DIR"))
+        .arg(&prompt2)
+        .arg("resume")
+        .arg("--last");
+    cmd.assert().success();
+
+    // Ensure the same file was updated and contains both markers.
+    let resumed_path = find_session_file_containing_marker(&sessions_dir, &marker2)
+        .expect("no resumed session file containing marker2");
+    assert_eq!(
+        resumed_path, path,
+        "resume --last should append to existing file"
+    );
+    let content = std::fs::read_to_string(&resumed_path)?;
+    assert!(content.contains(&marker));
+    assert!(content.contains(&marker2));
+    Ok(())
+}
+
+#[test]
+fn exec_resume_by_id_appends_to_existing_file() -> anyhow::Result<()> {
+    let home = TempDir::new()?;
+    let fixture = std::path::Path::new(env!("CARGO_MANIFEST_DIR"))
+        .join("tests/fixtures/cli_responses_fixture.sse");
+
+    // 1) First run: create a session
+    let marker = format!("resume-by-id-{}", Uuid::new_v4());
+    let prompt = format!("echo {marker}");
+
+    Command::cargo_bin("codex-exec")
+        .context("should find binary for codex-exec")?
+        .env("CODEX_HOME", home.path())
+        .env("OPENAI_API_KEY", "dummy")
+        .env("CODEX_RS_SSE_FIXTURE", &fixture)
+        .env("OPENAI_BASE_URL", "http://unused.local")
+        .arg("--skip-git-repo-check")
+        .arg("-C")
+        .arg(env!("CARGO_MANIFEST_DIR"))
+        .arg(&prompt)
+        .assert()
+        .success();
+
+    let sessions_dir = home.path().join("sessions");
+    let path = find_session_file_containing_marker(&sessions_dir, &marker)
+        .expect("no session file found after first run");
+    let session_id = extract_conversation_id(&path);
+    assert!(
+        !session_id.is_empty(),
+        "missing conversation id in meta line"
+    );
+
+    // 2) Resume by id
+    let marker2 = format!("resume-by-id-2-{}", Uuid::new_v4());
+    let prompt2 = format!("echo {marker2}");
+
+    let mut binding = assert_cmd::Command::cargo_bin("codex-exec")
+        .context("should find binary for codex-exec")?;
+    let cmd = binding
+        .env("CODEX_HOME", home.path())
+        .env("OPENAI_API_KEY", "dummy")
+        .env("CODEX_RS_SSE_FIXTURE", &fixture)
+        .env("OPENAI_BASE_URL", "http://unused.local")
+        .arg("--skip-git-repo-check")
+        .arg("-C")
+        .arg(env!("CARGO_MANIFEST_DIR"))
+        .arg(&prompt2)
+        .arg("resume")
+        .arg(&session_id);
+    cmd.assert().success();
+
+    let resumed_path = find_session_file_containing_marker(&sessions_dir, &marker2)
+        .expect("no resumed session file containing marker2");
+    assert_eq!(
+        resumed_path, path,
+        "resume by id should append to existing file"
+    );
+    let content = std::fs::read_to_string(&resumed_path)?;
+    assert!(content.contains(&marker));
+    assert!(content.contains(&marker2));
+    Ok(())
+}
+
+#[test]
+fn exec_resume_preserves_cli_configuration_overrides() -> anyhow::Result<()> {
+    let home = TempDir::new()?;
+    let fixture = std::path::Path::new(env!("CARGO_MANIFEST_DIR"))
+        .join("tests/fixtures/cli_responses_fixture.sse");
+
+    let marker = format!("resume-config-{}", Uuid::new_v4());
+    let prompt = format!("echo {marker}");
+
+    Command::cargo_bin("codex-exec")
+        .context("should find binary for codex-exec")?
+        .env("CODEX_HOME", home.path())
+        .env("OPENAI_API_KEY", "dummy")
+        .env("CODEX_RS_SSE_FIXTURE", &fixture)
+        .env("OPENAI_BASE_URL", "http://unused.local")
+        .arg("--skip-git-repo-check")
+        .arg("--sandbox")
+        .arg("workspace-write")
+        .arg("--model")
+        .arg("gpt-5")
+        .arg("-C")
+        .arg(env!("CARGO_MANIFEST_DIR"))
+        .arg(&prompt)
+        .assert()
+        .success();
+
+    let sessions_dir = home.path().join("sessions");
+    let path = find_session_file_containing_marker(&sessions_dir, &marker)
+        .expect("no session file found after first run");
+
+    let marker2 = format!("resume-config-2-{}", Uuid::new_v4());
+    let prompt2 = format!("echo {marker2}");
+
+    let output = Command::cargo_bin("codex-exec")
+        .context("should find binary for codex-exec")?
+        .env("CODEX_HOME", home.path())
+        .env("OPENAI_API_KEY", "dummy")
+        .env("CODEX_RS_SSE_FIXTURE", &fixture)
+        .env("OPENAI_BASE_URL", "http://unused.local")
+        .arg("--skip-git-repo-check")
+        .arg("--sandbox")
+        .arg("workspace-write")
+        .arg("--model")
+        .arg("gpt-5-high")
+        .arg("-C")
+        .arg(env!("CARGO_MANIFEST_DIR"))
+        .arg(&prompt2)
+        .arg("resume")
+        .arg("--last")
+        .output()
+        .context("resume run should succeed")?;
+
+    assert!(output.status.success(), "resume run failed: {output:?}");
+
+    let stdout = String::from_utf8(output.stdout)?;
+    assert!(
+        stdout.contains("model: gpt-5-high"),
+        "stdout missing model override: {stdout}"
+    );
+    assert!(
+        stdout.contains("sandbox: workspace-write"),
+        "stdout missing sandbox override: {stdout}"
+    );
+
+    let resumed_path = find_session_file_containing_marker(&sessions_dir, &marker2)
+        .expect("no resumed session file containing marker2");
+    assert_eq!(resumed_path, path, "resume should append to same file");
+
+    let content = std::fs::read_to_string(&resumed_path)?;
+    assert!(content.contains(&marker));
+    assert!(content.contains(&marker2));
+    Ok(())
+}

codex-rs/execpolicy/src/execv_checker.rs

@@ -88,7 +88,7 @@ impl ExecvChecker {
         let mut program = valid_exec.program.to_string();
         for system_path in valid_exec.system_path {
             if is_executable_file(&system_path) {
-                program = system_path.to_string();
+                program = system_path;
                 break;
             }
         }
@@ -196,7 +196,7 @@ system_path=[{fake_cp:?}]
         let checker = setup(&fake_cp);
         let exec_call = ExecCall {
             program: "cp".into(),
-            args: vec![source.clone(), dest.clone()],
+            args: vec![source, dest.clone()],
         };
         let valid_exec = match checker.r#match(&exec_call)? {
             MatchedExec::Match { exec } => exec,
@@ -207,7 +207,7 @@ system_path=[{fake_cp:?}]
         assert_eq!(
             checker.check(valid_exec.clone(), &cwd, &[], &[]),
             Err(ReadablePathNotInReadableFolders {
-                file: source_path.clone(),
+                file: source_path,
                 folders: vec![]
             }),
         );
@@ -229,7 +229,7 @@ system_path=[{fake_cp:?}]
         // Both readable and writeable folders specified.
         assert_eq!(
             checker.check(
-                valid_exec.clone(),
+                valid_exec,
                 &cwd,
                 std::slice::from_ref(&root_path),
                 std::slice::from_ref(&root_path)
@@ -241,7 +241,7 @@ system_path=[{fake_cp:?}]
         // folders.
         let exec_call_folders_as_args = ExecCall {
             program: "cp".into(),
-            args: vec![root.clone(), root.clone()],
+            args: vec![root.clone(), root],
         };
         let valid_exec_call_folders_as_args = match checker.r#match(&exec_call_folders_as_args)? {
             MatchedExec::Match { exec } => exec,
@@ -254,7 +254,7 @@ system_path=[{fake_cp:?}]
                 std::slice::from_ref(&root_path),
                 std::slice::from_ref(&root_path)
             ),
-            Ok(cp.clone()),
+            Ok(cp),
         );
 
         // Specify a parent of a readable folder as input.

codex-rs/execpolicy/src/policy_parser.rs

@@ -104,7 +104,7 @@ impl PolicyBuilder {
         info!("adding program spec: {program_spec:?}");
         let name = program_spec.program.clone();
         let mut programs = self.programs.borrow_mut();
-        programs.insert(name.clone(), program_spec);
+        programs.insert(name, program_spec);
     }
 
     fn add_forbidden_substrings(&self, substrings: &[String]) {

codex-rs/justfile

@@ -31,6 +31,13 @@ install:
     rustup show active-toolchain
     cargo fetch
 
+# Run `cargo nextest` since it's faster than `cargo test`, though including
+# --no-fail-fast is important to ensure all tests are run.
+#
+# Run `cargo install cargo-nextest` if you don't have it installed.
+test:
+    cargo nextest run --no-fail-fast
+
 # Run the MCP server
 mcp-server-run *args:
     cargo run -p codex-mcp-server -- "$@"

codex-rs/linux-sandbox/tests/suite/landlock.rs

@@ -121,7 +121,7 @@ async fn test_writable_root() {
 }
 
 #[tokio::test]
-#[should_panic(expected = "Sandbox(Timeout)")]
+#[should_panic(expected = "Sandbox(Timeout")]
 async fn test_timeout() {
     run_cmd(&["sleep", "2"], &[], 50).await;
 }
@@ -156,26 +156,27 @@ async fn assert_network_blocked(cmd: &[&str]) {
     )
     .await;
 
-    let (exit_code, stdout, stderr) = match result {
-        Ok(output) => (output.exit_code, output.stdout.text, output.stderr.text),
-        Err(CodexErr::Sandbox(SandboxErr::Denied(exit_code, stdout, stderr))) => {
-            (exit_code, stdout, stderr)
-        }
+    let output = match result {
+        Ok(output) => output,
+        Err(CodexErr::Sandbox(SandboxErr::Denied { output })) => *output,
         _ => {
             panic!("expected sandbox denied error, got: {result:?}");
         }
     };
 
-    dbg!(&stderr);
-    dbg!(&stdout);
-    dbg!(&exit_code);
+    dbg!(&output.stderr.text);
+    dbg!(&output.stdout.text);
+    dbg!(&output.exit_code);
 
     // A completely missing binary exits with 127.  Anything else should also
     // be non‑zero (EPERM from seccomp will usually bubble up as 1, 2, 13…)
     // If—*and only if*—the command exits 0 we consider the sandbox breached.
 
-    if exit_code == 0 {
-        panic!("Network sandbox FAILED - {cmd:?} exited 0\nstdout:\n{stdout}\nstderr:\n{stderr}",);
+    if output.exit_code == 0 {
+        panic!(
+            "Network sandbox FAILED - {cmd:?} exited 0\nstdout:\n{}\nstderr:\n{}",
+            output.stdout.text, output.stderr.text
+        );
     }
 }
 

codex-rs/login/src/server.rs

@@ -42,7 +42,7 @@ impl ServerOptions {
     pub fn new(codex_home: PathBuf, client_id: String) -> Self {
         Self {
             codex_home,
-            client_id: client_id.to_string(),
+            client_id,
             issuer: DEFAULT_ISSUER.to_string(),
             port: DEFAULT_PORT,
             open_browser: true,
@@ -126,7 +126,7 @@ pub fn run_login_server(opts: ServerOptions) -> io::Result<LoginServer> {
     let shutdown_notify = Arc::new(tokio::sync::Notify::new());
     let server_handle = {
         let shutdown_notify = shutdown_notify.clone();
-        let server = server.clone();
+        let server = server;
         tokio::spawn(async move {
             let result = loop {
                 tokio::select! {

codex-rs/mcp-client/src/main.rs

@@ -17,10 +17,10 @@ use anyhow::Context;
 use anyhow::Result;
 use codex_mcp_client::McpClient;
 use mcp_types::ClientCapabilities;
+use mcp_types::Implementation;
 use mcp_types::InitializeRequestParams;
 use mcp_types::ListToolsRequestParams;
 use mcp_types::MCP_SCHEMA_VERSION;
-use mcp_types::McpClientInfo;
 use tracing_subscriber::EnvFilter;
 
 #[tokio::main]
@@ -60,10 +60,13 @@ async fn main() -> Result<()> {
             sampling: None,
             elicitation: None,
         },
-        client_info: McpClientInfo {
+        client_info: Implementation {
             name: "codex-mcp-client".to_owned(),
             version: env!("CARGO_PKG_VERSION").to_owned(),
             title: Some("Codex".to_string()),
+            // This field is used by Codex when it is an MCP server: it should
+            // not be used when Codex is an MCP client.
+            user_agent: None,
         },
         protocol_version: MCP_SCHEMA_VERSION.to_owned(),
     };

codex-rs/mcp-server/Cargo.toml

@@ -40,6 +40,7 @@ uuid = { version = "1", features = ["serde", "v4"] }
 
 [dev-dependencies]
 assert_cmd = "2"
+base64 = "0.22"
 mcp_test_support = { path = "tests/common" }
 os_info = "3.12.0"
 pretty_assertions = "1.4.1"

codex-rs/mcp-server/src/codex_message_processor.rs

@@ -11,10 +11,16 @@ use codex_core::NewConversation;
 use codex_core::RolloutRecorder;
 use codex_core::SessionMeta;
 use codex_core::auth::CLIENT_ID;
+use codex_core::auth::get_auth_file;
+use codex_core::auth::login_with_api_key;
+use codex_core::auth::try_read_auth_json;
 use codex_core::config::Config;
 use codex_core::config::ConfigOverrides;
 use codex_core::config::ConfigToml;
 use codex_core::config::load_config_as_toml;
+use codex_core::config_edit::CONFIG_KEY_EFFORT;
+use codex_core::config_edit::CONFIG_KEY_MODEL;
+use codex_core::config_edit::persist_overrides_and_clear_if_none;
 use codex_core::default_client::get_codex_user_agent;
 use codex_core::exec::ExecParams;
 use codex_core::exec_env::create_env;
@@ -37,7 +43,6 @@ use codex_protocol::mcp_protocol::ApplyPatchApprovalParams;
 use codex_protocol::mcp_protocol::ApplyPatchApprovalResponse;
 use codex_protocol::mcp_protocol::ArchiveConversationParams;
 use codex_protocol::mcp_protocol::ArchiveConversationResponse;
-use codex_protocol::mcp_protocol::AuthMode;
 use codex_protocol::mcp_protocol::AuthStatusChangeNotification;
 use codex_protocol::mcp_protocol::ClientRequest;
 use codex_protocol::mcp_protocol::ConversationId;
@@ -55,6 +60,8 @@ use codex_protocol::mcp_protocol::InterruptConversationParams;
 use codex_protocol::mcp_protocol::InterruptConversationResponse;
 use codex_protocol::mcp_protocol::ListConversationsParams;
 use codex_protocol::mcp_protocol::ListConversationsResponse;
+use codex_protocol::mcp_protocol::LoginApiKeyParams;
+use codex_protocol::mcp_protocol::LoginApiKeyResponse;
 use codex_protocol::mcp_protocol::LoginChatGptCompleteNotification;
 use codex_protocol::mcp_protocol::LoginChatGptResponse;
 use codex_protocol::mcp_protocol::NewConversationParams;
@@ -67,6 +74,9 @@ use codex_protocol::mcp_protocol::SendUserMessageResponse;
 use codex_protocol::mcp_protocol::SendUserTurnParams;
 use codex_protocol::mcp_protocol::SendUserTurnResponse;
 use codex_protocol::mcp_protocol::ServerNotification;
+use codex_protocol::mcp_protocol::SetDefaultModelParams;
+use codex_protocol::mcp_protocol::SetDefaultModelResponse;
+use codex_protocol::mcp_protocol::UserInfoResponse;
 use codex_protocol::mcp_protocol::UserSavedConfig;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseItem;
@@ -169,6 +179,9 @@ impl CodexMessageProcessor {
             ClientRequest::GitDiffToRemote { request_id, params } => {
                 self.git_diff_to_origin(request_id, params.cwd).await;
             }
+            ClientRequest::LoginApiKey { request_id, params } => {
+                self.login_api_key(request_id, params).await;
+            }
             ClientRequest::LoginChatGpt { request_id } => {
                 self.login_chatgpt(request_id).await;
             }
@@ -184,15 +197,54 @@ impl CodexMessageProcessor {
             ClientRequest::GetUserSavedConfig { request_id } => {
                 self.get_user_saved_config(request_id).await;
             }
+            ClientRequest::SetDefaultModel { request_id, params } => {
+                self.set_default_model(request_id, params).await;
+            }
             ClientRequest::GetUserAgent { request_id } => {
                 self.get_user_agent(request_id).await;
             }
+            ClientRequest::UserInfo { request_id } => {
+                self.get_user_info(request_id).await;
+            }
             ClientRequest::ExecOneOffCommand { request_id, params } => {
                 self.exec_one_off_command(request_id, params).await;
             }
         }
     }
 
+    async fn login_api_key(&mut self, request_id: RequestId, params: LoginApiKeyParams) {
+        {
+            let mut guard = self.active_login.lock().await;
+            if let Some(active) = guard.take() {
+                active.drop();
+            }
+        }
+
+        match login_with_api_key(&self.config.codex_home, &params.api_key) {
+            Ok(()) => {
+                self.auth_manager.reload();
+                self.outgoing
+                    .send_response(request_id, LoginApiKeyResponse {})
+                    .await;
+
+                let payload = AuthStatusChangeNotification {
+                    auth_method: self.auth_manager.auth().map(|auth| auth.mode),
+                };
+                self.outgoing
+                    .send_server_notification(ServerNotification::AuthStatusChange(payload))
+                    .await;
+            }
+            Err(err) => {
+                let error = JSONRPCErrorError {
+                    code: INTERNAL_ERROR_CODE,
+                    message: format!("failed to save api key: {err}"),
+                    data: None,
+                };
+                self.outgoing.send_error(request_id, error).await;
+            }
+        }
+    }
+
     async fn login_chatgpt(&mut self, request_id: RequestId) {
         let config = self.config.as_ref();
 
@@ -346,7 +398,7 @@ impl CodexMessageProcessor {
             .await;
 
         // Send auth status change notification reflecting the current auth mode
-        // after logout (which may fall back to API key via env var).
+        // after logout.
         let current_auth_method = self.auth_manager.auth().map(|auth| auth.mode);
         let payload = AuthStatusChangeNotification {
             auth_method: current_auth_method,
@@ -361,7 +413,6 @@ impl CodexMessageProcessor {
         request_id: RequestId,
         params: codex_protocol::mcp_protocol::GetAuthStatusParams,
     ) {
-        let preferred_auth_method: AuthMode = self.auth_manager.preferred_auth_method();
         let include_token = params.include_token.unwrap_or(false);
         let do_refresh = params.refresh_token.unwrap_or(false);
 
@@ -369,30 +420,44 @@ impl CodexMessageProcessor {
             tracing::warn!("failed to refresh token while getting auth status: {err}");
         }
 
-        let response = match self.auth_manager.auth() {
-            Some(auth) => {
-                let (reported_auth_method, token_opt) = match auth.get_token().await {
-                    Ok(token) if !token.is_empty() => {
-                        let tok = if include_token { Some(token) } else { None };
-                        (Some(auth.mode), tok)
-                    }
-                    Ok(_) => (None, None),
-                    Err(err) => {
-                        tracing::warn!("failed to get token for auth status: {err}");
-                        (None, None)
-                    }
-                };
-                codex_protocol::mcp_protocol::GetAuthStatusResponse {
-                    auth_method: reported_auth_method,
-                    preferred_auth_method,
-                    auth_token: token_opt,
-                }
-            }
-            None => codex_protocol::mcp_protocol::GetAuthStatusResponse {
+        // Determine whether auth is required based on the active model provider.
+        // If a custom provider is configured with `requires_openai_auth == false`,
+        // then no auth step is required; otherwise, default to requiring auth.
+        let requires_openai_auth = self.config.model_provider.requires_openai_auth;
+
+        let response = if !requires_openai_auth {
+            codex_protocol::mcp_protocol::GetAuthStatusResponse {
                 auth_method: None,
-                preferred_auth_method,
                 auth_token: None,
-            },
+                requires_openai_auth: Some(false),
+            }
+        } else {
+            match self.auth_manager.auth() {
+                Some(auth) => {
+                    let auth_mode = auth.mode;
+                    let (reported_auth_method, token_opt) = match auth.get_token().await {
+                        Ok(token) if !token.is_empty() => {
+                            let tok = if include_token { Some(token) } else { None };
+                            (Some(auth_mode), tok)
+                        }
+                        Ok(_) => (None, None),
+                        Err(err) => {
+                            tracing::warn!("failed to get token for auth status: {err}");
+                            (None, None)
+                        }
+                    };
+                    codex_protocol::mcp_protocol::GetAuthStatusResponse {
+                        auth_method: reported_auth_method,
+                        auth_token: token_opt,
+                        requires_openai_auth: Some(true),
+                    }
+                }
+                None => codex_protocol::mcp_protocol::GetAuthStatusResponse {
+                    auth_method: None,
+                    auth_token: None,
+                    requires_openai_auth: Some(true),
+                },
+            }
         };
 
         self.outgoing.send_response(request_id, response).await;
@@ -439,6 +504,52 @@ impl CodexMessageProcessor {
         self.outgoing.send_response(request_id, response).await;
     }
 
+    async fn get_user_info(&self, request_id: RequestId) {
+        // Read alleged user email from auth.json (best-effort; not verified).
+        let auth_path = get_auth_file(&self.config.codex_home);
+        let alleged_user_email = match try_read_auth_json(&auth_path) {
+            Ok(auth) => auth.tokens.and_then(|t| t.id_token.email),
+            Err(_) => None,
+        };
+
+        let response = UserInfoResponse { alleged_user_email };
+        self.outgoing.send_response(request_id, response).await;
+    }
+
+    async fn set_default_model(&self, request_id: RequestId, params: SetDefaultModelParams) {
+        let SetDefaultModelParams {
+            model,
+            reasoning_effort,
+        } = params;
+        let effort_str = reasoning_effort.map(|effort| effort.to_string());
+
+        let overrides: [(&[&str], Option<&str>); 2] = [
+            (&[CONFIG_KEY_MODEL], model.as_deref()),
+            (&[CONFIG_KEY_EFFORT], effort_str.as_deref()),
+        ];
+
+        match persist_overrides_and_clear_if_none(
+            &self.config.codex_home,
+            self.config.active_profile.as_deref(),
+            &overrides,
+        )
+        .await
+        {
+            Ok(()) => {
+                let response = SetDefaultModelResponse {};
+                self.outgoing.send_response(request_id, response).await;
+            }
+            Err(err) => {
+                let error = JSONRPCErrorError {
+                    code: INTERNAL_ERROR_CODE,
+                    message: format!("failed to persist overrides: {err}"),
+                    data: None,
+                };
+                self.outgoing.send_error(request_id, error).await;
+            }
+        }
+    }
+
     async fn exec_one_off_command(&self, request_id: RequestId, params: ExecOneOffCommandParams) {
         tracing::debug!("ExecOneOffCommand params: {params:?}");
 
@@ -533,6 +644,7 @@ impl CodexMessageProcessor {
                 let response = NewConversationResponse {
                     conversation_id,
                     model: session_configured.model,
+                    reasoning_effort: session_configured.reasoning_effort,
                     rollout_path: session_configured.rollout_path,
                 };
                 self.outgoing.send_response(request_id, response).await;
@@ -1145,6 +1257,7 @@ fn derive_config_from_params(
     } = params;
     let overrides = ConfigOverrides {
         model,
+        review_model: None,
         config_profile: profile,
         cwd: cwd.map(PathBuf::from),
         approval_policy,

codex-rs/mcp-server/src/codex_tool_config.rs

@@ -152,6 +152,7 @@ impl CodexToolCallParam {
         // Build the `ConfigOverrides` recognized by codex-core.
         let overrides = codex_core::config::ConfigOverrides {
             model,
+            review_model: None,
             config_profile: profile,
             cwd: cwd.map(PathBuf::from),
             approval_policy: approval_policy.map(Into::into),

codex-rs/mcp-server/src/codex_tool_runner.rs

@@ -222,7 +222,7 @@ async fn run_codex_tool_session_inner(
                     }
                     EventMsg::TaskComplete(TaskCompleteEvent { last_agent_message }) => {
                         let text = match last_agent_message {
-                            Some(msg) => msg.clone(),
+                            Some(msg) => msg,
                             None => "".to_string(),
                         };
                         let result = CallToolResult {
@@ -277,9 +277,11 @@ async fn run_codex_tool_session_inner(
                     | EventMsg::GetHistoryEntryResponse(_)
                     | EventMsg::PlanUpdate(_)
                     | EventMsg::TurnAborted(_)
-                    | EventMsg::ConversationHistory(_)
+                    | EventMsg::ConversationPath(_)
                     | EventMsg::UserMessage(_)
-                    | EventMsg::ShutdownComplete => {
+                    | EventMsg::ShutdownComplete
+                    | EventMsg::EnteredReviewMode(_)
+                    | EventMsg::ExitedReviewMode(_) => {
                         // For now, we do not do anything extra for these
                         // events. Note that
                         // send(codex_event_to_notification(&event)) above has

codex-rs/mcp-server/src/message_processor.rs

@@ -56,8 +56,7 @@ impl MessageProcessor {
         config: Arc<Config>,
     ) -> Self {
         let outgoing = Arc::new(outgoing);
-        let auth_manager =
-            AuthManager::shared(config.codex_home.clone(), config.preferred_auth_method);
+        let auth_manager = AuthManager::shared(config.codex_home.clone());
         let conversation_manager = Arc::new(ConversationManager::new(auth_manager.clone()));
         let codex_message_processor = CodexMessageProcessor::new(
             auth_manager,
@@ -234,7 +233,7 @@ impl MessageProcessor {
             },
             instructions: None,
             protocol_version: params.protocol_version.clone(),
-            server_info: mcp_types::McpServerInfo {
+            server_info: mcp_types::Implementation {
                 name: "codex-mcp-server".to_string(),
                 version: env!("CARGO_PKG_VERSION").to_string(),
                 title: Some("Codex".to_string()),
@@ -532,7 +531,6 @@ impl MessageProcessor {
 
         // Spawn the long-running reply handler.
         tokio::spawn({
-            let codex = codex.clone();
             let outgoing = outgoing.clone();
             let prompt = prompt.clone();
             let running_requests_id_to_codex_uuid = running_requests_id_to_codex_uuid.clone();

codex-rs/mcp-server/src/outgoing_message.rs

@@ -258,6 +258,7 @@ pub(crate) struct OutgoingError {
 mod tests {
     use codex_core::protocol::EventMsg;
     use codex_core::protocol::SessionConfiguredEvent;
+    use codex_protocol::config_types::ReasoningEffort;
     use codex_protocol::mcp_protocol::ConversationId;
     use codex_protocol::mcp_protocol::LoginChatGptCompleteNotification;
     use pretty_assertions::assert_eq;
@@ -279,6 +280,7 @@ mod tests {
             msg: EventMsg::SessionConfigured(SessionConfiguredEvent {
                 session_id: conversation_id,
                 model: "gpt-4o".to_string(),
+                reasoning_effort: Some(ReasoningEffort::default()),
                 history_log_id: 1,
                 history_entry_count: 1000,
                 initial_messages: None,
@@ -299,7 +301,7 @@ mod tests {
         let Ok(expected_params) = serde_json::to_value(&event) else {
             panic!("Event must serialize");
         };
-        assert_eq!(params, Some(expected_params.clone()));
+        assert_eq!(params, Some(expected_params));
     }
 
     #[tokio::test]
@@ -312,6 +314,7 @@ mod tests {
         let session_configured_event = SessionConfiguredEvent {
             session_id: conversation_id,
             model: "gpt-4o".to_string(),
+            reasoning_effort: Some(ReasoningEffort::default()),
             history_log_id: 1,
             history_entry_count: 1000,
             initial_messages: None,
@@ -342,6 +345,7 @@ mod tests {
             "msg": {
                 "session_id": session_configured_event.session_id,
                 "model": session_configured_event.model,
+                "reasoning_effort": session_configured_event.reasoning_effort,
                 "history_log_id": session_configured_event.history_log_id,
                 "history_entry_count": session_configured_event.history_entry_count,
                 "type": "session_configured",

codex-rs/mcp-server/tests/common/mcp_process.rs

@@ -18,21 +18,23 @@ use codex_protocol::mcp_protocol::CancelLoginChatGptParams;
 use codex_protocol::mcp_protocol::GetAuthStatusParams;
 use codex_protocol::mcp_protocol::InterruptConversationParams;
 use codex_protocol::mcp_protocol::ListConversationsParams;
+use codex_protocol::mcp_protocol::LoginApiKeyParams;
 use codex_protocol::mcp_protocol::NewConversationParams;
 use codex_protocol::mcp_protocol::RemoveConversationListenerParams;
 use codex_protocol::mcp_protocol::ResumeConversationParams;
 use codex_protocol::mcp_protocol::SendUserMessageParams;
 use codex_protocol::mcp_protocol::SendUserTurnParams;
+use codex_protocol::mcp_protocol::SetDefaultModelParams;
 
 use mcp_types::CallToolRequestParams;
 use mcp_types::ClientCapabilities;
+use mcp_types::Implementation;
 use mcp_types::InitializeRequestParams;
 use mcp_types::JSONRPC_VERSION;
 use mcp_types::JSONRPCMessage;
 use mcp_types::JSONRPCNotification;
 use mcp_types::JSONRPCRequest;
 use mcp_types::JSONRPCResponse;
-use mcp_types::McpClientInfo;
 use mcp_types::ModelContextProtocolNotification;
 use mcp_types::ModelContextProtocolRequest;
 use mcp_types::RequestId;
@@ -134,10 +136,11 @@ impl McpProcess {
                 roots: None,
                 sampling: None,
             },
-            client_info: McpClientInfo {
+            client_info: Implementation {
                 name: "elicitation test".into(),
                 title: Some("Elicitation Test".into()),
                 version: "0.0.0".into(),
+                user_agent: None,
             },
             protocol_version: mcp_types::MCP_SCHEMA_VERSION.into(),
         };
@@ -294,6 +297,20 @@ impl McpProcess {
         self.send_request("getUserAgent", None).await
     }
 
+    /// Send a `userInfo` JSON-RPC request.
+    pub async fn send_user_info_request(&mut self) -> anyhow::Result<i64> {
+        self.send_request("userInfo", None).await
+    }
+
+    /// Send a `setDefaultModel` JSON-RPC request.
+    pub async fn send_set_default_model_request(
+        &mut self,
+        params: SetDefaultModelParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("setDefaultModel", params).await
+    }
+
     /// Send a `listConversations` JSON-RPC request.
     pub async fn send_list_conversations_request(
         &mut self,
@@ -312,6 +329,15 @@ impl McpProcess {
         self.send_request("resumeConversation", params).await
     }
 
+    /// Send a `loginApiKey` JSON-RPC request.
+    pub async fn send_login_api_key_request(
+        &mut self,
+        params: LoginApiKeyParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("loginApiKey", params).await
+    }
+
     /// Send a `loginChatGpt` JSON-RPC request.
     pub async fn send_login_chat_gpt_request(&mut self) -> anyhow::Result<i64> {
         self.send_request("loginChatGpt", None).await

codex-rs/mcp-server/tests/suite/auth.rs

@@ -1,9 +1,10 @@
 use std::path::Path;
 
-use codex_core::auth::login_with_api_key;
 use codex_protocol::mcp_protocol::AuthMode;
 use codex_protocol::mcp_protocol::GetAuthStatusParams;
 use codex_protocol::mcp_protocol::GetAuthStatusResponse;
+use codex_protocol::mcp_protocol::LoginApiKeyParams;
+use codex_protocol::mcp_protocol::LoginApiKeyResponse;
 use mcp_test_support::McpProcess;
 use mcp_test_support::to_response;
 use mcp_types::JSONRPCResponse;
@@ -14,11 +15,17 @@ use tokio::time::timeout;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
-// Helper to create a config.toml; mirrors create_conversation.rs
-fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
+fn create_config_toml_custom_provider(
+    codex_home: &Path,
+    requires_openai_auth: bool,
+) -> std::io::Result<()> {
     let config_toml = codex_home.join("config.toml");
-    std::fs::write(
-        config_toml,
+    let requires_line = if requires_openai_auth {
+        "requires_openai_auth = true\n"
+    } else {
+        ""
+    };
+    let contents = format!(
         r#"
 model = "mock-model"
 approval_policy = "never"
@@ -32,14 +39,47 @@ base_url = "http://127.0.0.1:0/v1"
 wire_api = "chat"
 request_max_retries = 0
 stream_max_retries = 0
+{requires_line}
+"#
+    );
+    std::fs::write(config_toml, contents)
+}
+
+fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        r#"
+model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "danger-full-access"
 "#,
     )
 }
 
+async fn login_with_api_key_via_request(mcp: &mut McpProcess, api_key: &str) {
+    let request_id = mcp
+        .send_login_api_key_request(LoginApiKeyParams {
+            api_key: api_key.to_string(),
+        })
+        .await
+        .unwrap_or_else(|e| panic!("send loginApiKey: {e}"));
+
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await
+    .unwrap_or_else(|e| panic!("loginApiKey timeout: {e}"))
+    .unwrap_or_else(|e| panic!("loginApiKey response: {e}"));
+    let _: LoginApiKeyResponse =
+        to_response(resp).unwrap_or_else(|e| panic!("deserialize login response: {e}"));
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn get_auth_status_no_auth() {
     let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml(codex_home.path()).expect("write config.toml");
+    create_config_toml(codex_home.path()).unwrap_or_else(|err| panic!("write config.toml: {err}"));
 
     let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)])
         .await
@@ -72,8 +112,7 @@ async fn get_auth_status_no_auth() {
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn get_auth_status_with_api_key() {
     let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml(codex_home.path()).expect("write config.toml");
-    login_with_api_key(codex_home.path(), "sk-test-key").expect("seed api key");
+    create_config_toml(codex_home.path()).unwrap_or_else(|err| panic!("write config.toml: {err}"));
 
     let mut mcp = McpProcess::new(codex_home.path())
         .await
@@ -83,6 +122,8 @@ async fn get_auth_status_with_api_key() {
         .expect("init timeout")
         .expect("init failed");
 
+    login_with_api_key_via_request(&mut mcp, "sk-test-key").await;
+
     let request_id = mcp
         .send_get_auth_status_request(GetAuthStatusParams {
             include_token: Some(true),
@@ -101,14 +142,13 @@ async fn get_auth_status_with_api_key() {
     let status: GetAuthStatusResponse = to_response(resp).expect("deserialize status");
     assert_eq!(status.auth_method, Some(AuthMode::ApiKey));
     assert_eq!(status.auth_token, Some("sk-test-key".to_string()));
-    assert_eq!(status.preferred_auth_method, AuthMode::ChatGPT);
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn get_auth_status_with_api_key_no_include_token() {
+async fn get_auth_status_with_api_key_when_auth_not_required() {
     let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml(codex_home.path()).expect("write config.toml");
-    login_with_api_key(codex_home.path(), "sk-test-key").expect("seed api key");
+    create_config_toml_custom_provider(codex_home.path(), false)
+        .unwrap_or_else(|err| panic!("write config.toml: {err}"));
 
     let mut mcp = McpProcess::new(codex_home.path())
         .await
@@ -118,6 +158,48 @@ async fn get_auth_status_with_api_key_no_include_token() {
         .expect("init timeout")
         .expect("init failed");
 
+    login_with_api_key_via_request(&mut mcp, "sk-test-key").await;
+
+    let request_id = mcp
+        .send_get_auth_status_request(GetAuthStatusParams {
+            include_token: Some(true),
+            refresh_token: Some(false),
+        })
+        .await
+        .expect("send getAuthStatus");
+
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await
+    .expect("getAuthStatus timeout")
+    .expect("getAuthStatus response");
+    let status: GetAuthStatusResponse = to_response(resp).expect("deserialize status");
+    assert_eq!(status.auth_method, None, "expected no auth method");
+    assert_eq!(status.auth_token, None, "expected no token");
+    assert_eq!(
+        status.requires_openai_auth,
+        Some(false),
+        "requires_openai_auth should be false",
+    );
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn get_auth_status_with_api_key_no_include_token() {
+    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
+    create_config_toml(codex_home.path()).unwrap_or_else(|err| panic!("write config.toml: {err}"));
+
+    let mut mcp = McpProcess::new(codex_home.path())
+        .await
+        .expect("spawn mcp process");
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
+        .await
+        .expect("init timeout")
+        .expect("init failed");
+
+    login_with_api_key_via_request(&mut mcp, "sk-test-key").await;
+
     // Build params via struct so None field is omitted in wire JSON.
     let params = GetAuthStatusParams {
         include_token: None,
@@ -138,5 +220,4 @@ async fn get_auth_status_with_api_key_no_include_token() {
     let status: GetAuthStatusResponse = to_response(resp).expect("deserialize status");
     assert_eq!(status.auth_method, Some(AuthMode::ApiKey));
     assert!(status.auth_token.is_none(), "token must be omitted");
-    assert_eq!(status.preferred_auth_method, AuthMode::ChatGPT);
 }

codex-rs/mcp-server/tests/suite/codex_message_processor_flow.rs

@@ -90,6 +90,7 @@ async fn test_codex_jsonrpc_conversation_flow() {
     let NewConversationResponse {
         conversation_id,
         model,
+        reasoning_effort: _,
         rollout_path: _,
     } = new_conv_resp;
     assert_eq!(model, "mock-model");
@@ -319,7 +320,7 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
             approval_policy: AskForApproval::Never,
             sandbox_policy: SandboxPolicy::new_read_only_policy(),
             model: "mock-model".to_string(),
-            effort: ReasoningEffort::Medium,
+            effort: Some(ReasoningEffort::Medium),
             summary: ReasoningSummary::Auto,
         })
         .await

codex-rs/mcp-server/tests/suite/create_conversation.rs

@@ -59,6 +59,7 @@ async fn test_conversation_create_and_send_message_ok() {
     let NewConversationResponse {
         conversation_id,
         model,
+        reasoning_effort: _,
         rollout_path: _,
     } = to_response::<NewConversationResponse>(new_conv_resp)
         .expect("deserialize newConversation response");

codex-rs/mcp-server/tests/suite/login.rs

@@ -1,7 +1,7 @@
 use std::path::Path;
 use std::time::Duration;
 
-use codex_core::auth::login_with_api_key;
+use codex_login::login_with_api_key;
 use codex_protocol::mcp_protocol::CancelLoginChatGptParams;
 use codex_protocol::mcp_protocol::CancelLoginChatGptResponse;
 use codex_protocol::mcp_protocol::GetAuthStatusParams;
@@ -95,7 +95,7 @@ async fn logout_chatgpt_removes_auth() {
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn login_and_cancel_chatgpt() {
     let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
-    create_config_toml(codex_home.path()).expect("write config.toml");
+    create_config_toml(codex_home.path()).unwrap_or_else(|err| panic!("write config.toml: {err}"));
 
     let mut mcp = McpProcess::new(codex_home.path())
         .await

codex-rs/mcp-server/tests/suite/mod.rs

@@ -9,4 +9,6 @@ mod interrupt;
 mod list_resume;
 mod login;
 mod send_message;
+mod set_default_model;
 mod user_agent;
+mod user_info;

codex-rs/mcp-server/tests/suite/set_default_model.rs

@@ -0,0 +1,76 @@
+use std::path::Path;
+
+use codex_core::config::ConfigToml;
+use codex_protocol::mcp_protocol::SetDefaultModelParams;
+use codex_protocol::mcp_protocol::SetDefaultModelResponse;
+use mcp_test_support::McpProcess;
+use mcp_test_support::to_response;
+use mcp_types::JSONRPCResponse;
+use mcp_types::RequestId;
+use pretty_assertions::assert_eq;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn set_default_model_persists_overrides() {
+    let codex_home = TempDir::new().expect("create tempdir");
+    create_config_toml(codex_home.path()).expect("write config.toml");
+
+    let mut mcp = McpProcess::new(codex_home.path())
+        .await
+        .expect("spawn mcp process");
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
+        .await
+        .expect("init timeout")
+        .expect("init failed");
+
+    let params = SetDefaultModelParams {
+        model: Some("gpt-4.1".to_string()),
+        reasoning_effort: None,
+    };
+
+    let request_id = mcp
+        .send_set_default_model_request(params)
+        .await
+        .expect("send setDefaultModel");
+
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await
+    .expect("setDefaultModel timeout")
+    .expect("setDefaultModel response");
+
+    let _: SetDefaultModelResponse =
+        to_response(resp).expect("deserialize setDefaultModel response");
+
+    let config_path = codex_home.path().join("config.toml");
+    let config_contents = tokio::fs::read_to_string(&config_path)
+        .await
+        .expect("read config.toml");
+    let config_toml: ConfigToml = toml::from_str(&config_contents).expect("parse config.toml");
+
+    assert_eq!(
+        ConfigToml {
+            model: Some("gpt-4.1".to_string()),
+            model_reasoning_effort: None,
+            ..Default::default()
+        },
+        config_toml,
+    );
+}
+
+// Helper to create a config.toml; mirrors create_conversation.rs
+fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        r#"
+model = "gpt-5"
+model_reasoning_effort = "medium"
+"#,
+    )
+}