fmt + clippy: codex-core deterministic shell tool tests, conflict cleanup

This PR changes get history op to get path. Then, forking will use a
path. This will help us have one unified codepath for resuming/forking
conversations. Will also help in having rollout history in order. It
also fixes a bug where you won't see the UI when resuming after forking.

.github/workflows/rust-ci.yml

@@ -63,6 +63,24 @@ jobs:
       - name: cargo fmt
         run: cargo fmt -- --config imports_granularity=Item --check
 
+  cargo_shear:
+    name: cargo shear
+    runs-on: ubuntu-24.04
+    needs: changed
+    if: ${{ needs.changed.outputs.codex == 'true' || needs.changed.outputs.workflows == 'true' || github.event_name == 'push' }}
+    defaults:
+      run:
+        working-directory: codex-rs
+    steps:
+      - uses: actions/checkout@v5
+      - uses: dtolnay/rust-toolchain@1.89
+      - uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        with:
+          tool: cargo-shear
+          version: 1.5.1
+      - name: cargo shear
+        run: cargo shear
+
   # --- CI to validate on different os/targets --------------------------------
   lint_build_test:
     name: ${{ matrix.runner }} - ${{ matrix.target }}${{ matrix.profile == 'release' && ' (release)' || '' }}
@@ -160,12 +178,17 @@ jobs:
           find . -name Cargo.toml -mindepth 2 -maxdepth 2 -print0 \
             | xargs -0 -n1 -I{} bash -c 'cd "$(dirname "{}")" && cargo check --profile ${{ matrix.profile }}'
 
-      - name: cargo test
+      - uses: taiki-e/install-action@0c5db7f7f897c03b771660e91d065338615679f4 # v2
+        with:
+          tool: nextest
+          version: 0.9.103
+
+      - name: tests
         id: test
-        # `cargo test` takes too long for release builds to run them on every PR
+        # Tests take too long for release builds to run them on every PR.
         if: ${{ matrix.profile != 'release' }}
         continue-on-error: true
-        run: cargo test --all-features --target ${{ matrix.target }} --profile ${{ matrix.profile }}
+        run: cargo nextest run --all-features --no-fail-fast --target ${{ matrix.target }}
         env:
           RUST_BACKTRACE: 1
 
@@ -182,7 +205,7 @@ jobs:
   # --- Gatherer job that you mark as the ONLY required status -----------------
   results:
     name: CI results (required)
-    needs: [changed, general, lint_build_test]
+    needs: [changed, general, cargo_shear, lint_build_test]
     if: always()
     runs-on: ubuntu-24.04
     steps:
@@ -190,6 +213,7 @@ jobs:
         shell: bash
         run: |
           echo "general: ${{ needs.general.result }}"
+          echo "shear  : ${{ needs.cargo_shear.result }}"
           echo "matrix : ${{ needs.lint_build_test.result }}"
 
           # If nothing relevant changed (PR touching only root README, etc.),
@@ -201,4 +225,5 @@ jobs:
 
           # Otherwise require the jobs to have succeeded
           [[ '${{ needs.general.result }}' == 'success' ]] || { echo 'general failed'; exit 1; }
+          [[ '${{ needs.cargo_shear.result }}' == 'success' ]] || { echo 'cargo_shear failed'; exit 1; }
           [[ '${{ needs.lint_build_test.result }}' == 'success' ]] || { echo 'matrix failed'; exit 1; }

codex-rs/README.md

@@ -35,7 +35,7 @@ npx @modelcontextprotocol/inspector codex mcp
 
 You can enable notifications by configuring a script that is run whenever the agent finishes a turn. The [notify documentation](../docs/config.md#notify) includes a detailed example that explains how to get desktop notifications via [terminal-notifier](https://github.com/julienXX/terminal-notifier) on macOS.
 
-### `codex exec` to run Codex programmatially/non-interactively
+### `codex exec` to run Codex programmatically/non-interactively
 
 To run Codex non-interactively, run `codex exec PROMPT` (you can also pass the prompt via `stdin`) and Codex will work on your task until it decides that it is done and exits. Output is printed to the terminal directly. You can set the `RUST_LOG` environment variable to see more about what's going on.
 

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

@@ -726,9 +726,9 @@ fn compute_replacements(
             line_index = start_idx + pattern.len();
         } else {
             return Err(ApplyPatchError::ComputeReplacements(format!(
-                "Failed to find expected lines {:?} in {}",
-                chunk.old_lines,
-                path.display()
+                "Failed to find expected lines in {}:\n{}",
+                path.display(),
+                chunk.old_lines.join("\n"),
             )));
         }
     }

codex-rs/chatgpt/Cargo.toml

@@ -12,7 +12,6 @@ clap = { version = "4", features = ["derive"] }
 codex-common = { path = "../common", features = ["cli"] }
 codex-core = { path = "../core" }
 codex-protocol = { path = "../protocol" }
-reqwest = { version = "0.12", features = ["json", "stream"] }
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 tokio = { version = "1", features = ["full"] }

codex-rs/chatgpt/src/apply_command.rs

@@ -31,7 +31,7 @@ pub async fn run_apply_command(
         ConfigOverrides::default(),
     )?;
 
-    init_chatgpt_token_from_auth(&config.codex_home, &config.responses_originator_header).await?;
+    init_chatgpt_token_from_auth(&config.codex_home).await?;
 
     let task_response = get_task(&config, apply_cli.task_id).await?;
     apply_diff_from_task(task_response, cwd).await

codex-rs/chatgpt/src/chatgpt_client.rs

@@ -13,10 +13,10 @@ pub(crate) async fn chatgpt_get_request<T: DeserializeOwned>(
     path: String,
 ) -> anyhow::Result<T> {
     let chatgpt_base_url = &config.chatgpt_base_url;
-    init_chatgpt_token_from_auth(&config.codex_home, &config.responses_originator_header).await?;
+    init_chatgpt_token_from_auth(&config.codex_home).await?;
 
     // Make direct HTTP request to ChatGPT backend API with the token
-    let client = create_client(&config.responses_originator_header);
+    let client = create_client();
     let url = format!("{chatgpt_base_url}{path}");
 
     let token =

codex-rs/chatgpt/src/chatgpt_token.rs

@@ -19,11 +19,8 @@ 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,
-    originator: &str,
-) -> std::io::Result<()> {
-    let auth = CodexAuth::from_codex_home(codex_home, AuthMode::ChatGPT, originator)?;
+pub async fn init_chatgpt_token_from_auth(codex_home: &Path) -> std::io::Result<()> {
+    let auth = CodexAuth::from_codex_home(codex_home, AuthMode::ChatGPT)?;
     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

@@ -12,8 +12,8 @@ use codex_protocol::mcp_protocol::AuthMode;
 use std::env;
 use std::path::PathBuf;
 
-pub async fn login_with_chatgpt(codex_home: PathBuf, originator: String) -> std::io::Result<()> {
-    let opts = ServerOptions::new(codex_home, CLIENT_ID.to_string(), originator);
+pub async fn login_with_chatgpt(codex_home: PathBuf) -> std::io::Result<()> {
+    let opts = ServerOptions::new(codex_home, CLIENT_ID.to_string());
     let server = run_login_server(opts)?;
 
     eprintln!(
@@ -27,12 +27,7 @@ pub async fn login_with_chatgpt(codex_home: PathBuf, originator: String) -> std:
 pub async fn run_login_with_chatgpt(cli_config_overrides: CliConfigOverrides) -> ! {
     let config = load_config_or_exit(cli_config_overrides);
 
-    match login_with_chatgpt(
-        config.codex_home,
-        config.responses_originator_header.clone(),
-    )
-    .await
-    {
+    match login_with_chatgpt(config.codex_home).await {
         Ok(_) => {
             eprintln!("Successfully logged in");
             std::process::exit(0);
@@ -65,11 +60,7 @@ 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,
-        &config.responses_originator_header,
-    ) {
+    match CodexAuth::from_codex_home(&config.codex_home, config.preferred_auth_method) {
         Ok(Some(auth)) => match auth.mode {
             AuthMode::ApiKey => match auth.get_token().await {
                 Ok(api_key) => {

codex-rs/cli/src/main.rs

@@ -40,18 +40,6 @@ struct MultitoolCli {
     #[clap(flatten)]
     interactive: TuiCli,
 
-    /// Reasoning effort for the model (maps to model_reasoning_effort).
-    #[arg(long = "reasoning", value_parser = ["minimal", "low", "medium", "high"]) ]
-    reasoning: Option<String>,
-
-    /// Reasoning summary verbosity (maps to model_reasoning_summary).
-    #[arg(long = "reasoning-summary", value_parser = ["auto", "concise", "detailed", "none"]) ]
-    reasoning_summary: Option<String>,
-
-    /// Text verbosity for GPT‑5 models (maps to model_verbosity).
-    #[arg(long = "verbosity", value_parser = ["low", "medium", "high"]) ]
-    verbosity: Option<String>,
-
     #[clap(subcommand)]
     subcommand: Option<Subcommand>,
 }
@@ -155,25 +143,7 @@ fn main() -> anyhow::Result<()> {
 }
 
 async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()> {
-    let mut cli = MultitoolCli::parse();
-
-    // Synthesize -c overrides from root-level reasoning flags so they flow to
-    // both interactive and subcommand CLIs via prepend_config_flags.
-    if let Some(ref v) = cli.reasoning {
-        cli.config_overrides
-            .raw_overrides
-            .push(format!("model_reasoning_effort=\"{v}\""));
-    }
-    if let Some(ref v) = cli.reasoning_summary {
-        cli.config_overrides
-            .raw_overrides
-            .push(format!("model_reasoning_summary=\"{v}\""));
-    }
-    if let Some(ref v) = cli.verbosity {
-        cli.config_overrides
-            .raw_overrides
-            .push(format!("model_verbosity=\"{v}\""));
-    }
+    let cli = MultitoolCli::parse();
 
     match cli.subcommand {
         None => {

codex-rs/cli/src/proto.rs

@@ -40,7 +40,6 @@ pub async fn run_main(opts: ProtoCli) -> anyhow::Result<()> {
     let conversation_manager = ConversationManager::new(AuthManager::shared(
         config.codex_home.clone(),
         config.preferred_auth_method,
-        config.responses_originator_header.clone(),
     ));
     let NewConversation {
         conversation_id: _,

codex-rs/core/Cargo.toml

@@ -26,14 +26,12 @@ eventsource-stream = "0.2.3"
 futures = "0.3"
 libc = "0.2.175"
 mcp-types = { path = "../mcp-types" }
-mime_guess = "2.0"
 os_info = "3.12.0"
 portable-pty = "0.9.0"
 rand = "0.9"
 regex-lite = "0.1.7"
 reqwest = { version = "0.12", features = ["json", "stream"] }
 serde = { version = "1", features = ["derive"] }
-serde_bytes = "0.11"
 serde_json = "1"
 sha1 = "0.10.6"
 shlex = "1.3.0"
@@ -56,7 +54,7 @@ tracing = { version = "0.1.41", features = ["log"] }
 tree-sitter = "0.25.9"
 tree-sitter-bash = "0.25.0"
 uuid = { version = "1", features = ["serde", "v4"] }
-whoami = "1.6.1"
+which = "6"
 wildmatch = "2.4.0"
 
 
@@ -72,9 +70,6 @@ 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" }
@@ -85,3 +80,6 @@ tempfile = "3"
 tokio-test = "0.4"
 walkdir = "2.5.0"
 wiremock = "0.6"
+
+[package.metadata.cargo-shear]
+ignored = ["openssl-sys"]

codex-rs/core/src/auth.rs

@@ -75,9 +75,8 @@ impl CodexAuth {
     pub fn from_codex_home(
         codex_home: &Path,
         preferred_auth_method: AuthMode,
-        originator: &str,
     ) -> std::io::Result<Option<CodexAuth>> {
-        load_auth(codex_home, true, preferred_auth_method, originator)
+        load_auth(codex_home, true, preferred_auth_method)
     }
 
     pub async fn get_token_data(&self) -> Result<TokenData, std::io::Error> {
@@ -173,7 +172,7 @@ impl CodexAuth {
             mode: AuthMode::ChatGPT,
             auth_file: PathBuf::new(),
             auth_dot_json,
-            client: crate::default_client::create_client("codex_cli_rs"),
+            client: crate::default_client::create_client(),
         }
     }
 
@@ -188,10 +187,7 @@ impl CodexAuth {
     }
 
     pub fn from_api_key(api_key: &str) -> Self {
-        Self::from_api_key_with_client(
-            api_key,
-            crate::default_client::create_client(crate::default_client::DEFAULT_ORIGINATOR),
-        )
+        Self::from_api_key_with_client(api_key, crate::default_client::create_client())
     }
 }
 
@@ -232,13 +228,12 @@ fn load_auth(
     codex_home: &Path,
     include_env_var: bool,
     preferred_auth_method: AuthMode,
-    originator: &str,
 ) -> 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).
     let auth_file = get_auth_file(codex_home);
-    let client = crate::default_client::create_client(originator);
+    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
@@ -473,7 +468,7 @@ mod tests {
             auth_dot_json,
             auth_file: _,
             ..
-        } = super::load_auth(codex_home.path(), false, AuthMode::ChatGPT, "codex_cli_rs")
+        } = super::load_auth(codex_home.path(), false, AuthMode::ChatGPT)
             .unwrap()
             .unwrap();
         assert_eq!(None, api_key);
@@ -525,7 +520,7 @@ mod tests {
             auth_dot_json,
             auth_file: _,
             ..
-        } = super::load_auth(codex_home.path(), false, AuthMode::ChatGPT, "codex_cli_rs")
+        } = super::load_auth(codex_home.path(), false, AuthMode::ChatGPT)
             .unwrap()
             .unwrap();
         assert_eq!(None, api_key);
@@ -576,7 +571,7 @@ mod tests {
             auth_dot_json,
             auth_file: _,
             ..
-        } = super::load_auth(codex_home.path(), false, AuthMode::ChatGPT, "codex_cli_rs")
+        } = super::load_auth(codex_home.path(), false, AuthMode::ChatGPT)
             .unwrap()
             .unwrap();
         assert_eq!(Some("sk-test-key".to_string()), api_key);
@@ -596,7 +591,7 @@ mod tests {
         )
         .unwrap();
 
-        let auth = super::load_auth(dir.path(), false, AuthMode::ChatGPT, "codex_cli_rs")
+        let auth = super::load_auth(dir.path(), false, AuthMode::ChatGPT)
             .unwrap()
             .unwrap();
         assert_eq!(auth.mode, AuthMode::ApiKey);
@@ -680,7 +675,6 @@ mod tests {
 #[derive(Debug)]
 pub struct AuthManager {
     codex_home: PathBuf,
-    originator: String,
     inner: RwLock<CachedAuth>,
 }
 
@@ -689,13 +683,12 @@ 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, originator: String) -> Self {
-        let auth = CodexAuth::from_codex_home(&codex_home, preferred_auth_mode, &originator)
+    pub fn new(codex_home: PathBuf, preferred_auth_mode: AuthMode) -> Self {
+        let auth = CodexAuth::from_codex_home(&codex_home, preferred_auth_mode)
             .ok()
             .flatten();
         Self {
             codex_home,
-            originator,
             inner: RwLock::new(CachedAuth {
                 preferred_auth_mode,
                 auth,
@@ -712,7 +705,6 @@ impl AuthManager {
         };
         Arc::new(Self {
             codex_home: PathBuf::new(),
-            originator: "codex_cli_rs".to_string(),
             inner: RwLock::new(cached),
         })
     }
@@ -734,7 +726,7 @@ impl AuthManager {
     /// 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, &self.originator)
+        let new_auth = CodexAuth::from_codex_home(&self.codex_home, preferred)
             .ok()
             .flatten();
         if let Ok(mut guard) = self.inner.write() {
@@ -755,12 +747,8 @@ impl AuthManager {
     }
 
     /// Convenience constructor returning an `Arc` wrapper.
-    pub fn shared(
-        codex_home: PathBuf,
-        preferred_auth_mode: AuthMode,
-        originator: String,
-    ) -> Arc<Self> {
-        Arc::new(Self::new(codex_home, preferred_auth_mode, originator))
+    pub fn shared(codex_home: PathBuf, preferred_auth_mode: AuthMode) -> Arc<Self> {
+        Arc::new(Self::new(codex_home, preferred_auth_mode))
     }
 
     /// Attempt to refresh the current auth token (if any). On success, reload

codex-rs/core/src/client.rs

@@ -84,7 +84,7 @@ impl ModelClient {
         summary: ReasoningSummaryConfig,
         conversation_id: ConversationId,
     ) -> Self {
-        let client = create_client(&config.responses_originator_header);
+        let client = create_client();
 
         Self {
             config,

codex-rs/core/src/codex.rs

@@ -9,6 +9,9 @@ use std::sync::atomic::AtomicU64;
 use std::time::Duration;
 
 use crate::AuthManager;
+use crate::config_edit::CONFIG_KEY_EFFORT;
+use crate::config_edit::CONFIG_KEY_MODEL;
+use crate::config_edit::persist_non_null_overrides;
 use crate::event_mapping::map_response_item_to_event_messages;
 use async_channel::Receiver;
 use async_channel::Sender;
@@ -16,12 +19,14 @@ use codex_apply_patch::ApplyPatchAction;
 use codex_apply_patch::MaybeApplyPatchVerified;
 use codex_apply_patch::maybe_parse_apply_patch_verified;
 use codex_protocol::mcp_protocol::ConversationId;
-use codex_protocol::protocol::ConversationHistoryResponseEvent;
+use codex_protocol::protocol::ConversationPathResponseEvent;
+use codex_protocol::protocol::RolloutItem;
 use codex_protocol::protocol::TaskStartedEvent;
 use codex_protocol::protocol::TurnAbortReason;
 use codex_protocol::protocol::TurnAbortedEvent;
 use futures::prelude::*;
 use mcp_types::CallToolResult;
+use serde::Deserialize;
 use serde::Serialize;
 use serde_json;
 use tokio::sync::oneshot;
@@ -44,7 +49,6 @@ use crate::client_common::ResponseEvent;
 use crate::config::Config;
 use crate::config_types::ShellEnvironmentPolicy;
 use crate::conversation_history::ConversationHistory;
-use crate::conversation_manager::InitialHistory;
 use crate::environment_context::EnvironmentContext;
 use crate::error::CodexErr;
 use crate::error::Result as CodexResult;
@@ -109,6 +113,7 @@ use crate::safety::assess_command_safety;
 use crate::safety::assess_safety_for_untrusted_command;
 use crate::shell;
 use crate::turn_diff_tracker::TurnDiffTracker;
+use crate::unified_exec::UnifiedExecSessionManager;
 use crate::user_instructions::UserInstructions;
 use crate::user_notification::UserNotification;
 use crate::util::backoff;
@@ -121,6 +126,7 @@ use codex_protocol::models::LocalShellAction;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::models::ShellToolCallParams;
+use codex_protocol::protocol::InitialHistory;
 
 // A convenience extension trait for acquiring mutex locks where poisoning is
 // unrecoverable and should abort the program. This avoids scattered `.unwrap()`
@@ -203,9 +209,6 @@ impl Codex {
             error!("Failed to create session: {e:#}");
             CodexErr::InternalAgentDied
         })?;
-        session
-            .record_initial_history(&turn_context, conversation_history)
-            .await;
         let conversation_id = session.conversation_id;
 
         // This task will run until Op::Shutdown is received.
@@ -279,6 +282,7 @@ pub(crate) struct Session {
     /// Manager for external MCP servers/tools.
     mcp_connection_manager: McpConnectionManager,
     session_manager: ExecSessionManager,
+    unified_exec_manager: UnifiedExecSessionManager,
 
     /// External notifier command (will be passed as args to exec()). When
     /// `None` this feature is disabled.
@@ -416,6 +420,7 @@ impl Session {
             error!("failed to initialize rollout recorder: {e:#}");
             anyhow::anyhow!("failed to initialize rollout recorder: {e:#}")
         })?;
+        let rollout_path = rollout_recorder.rollout_path.clone();
         // Create the mutable state for the Session.
         let state = State {
             history: ConversationHistory::new(),
@@ -463,12 +468,12 @@ impl Session {
             tools_config: ToolsConfig::new(&ToolsConfigParams {
                 model_family: &config.model_family,
                 approval_policy,
-                sandbox_policy: sandbox_policy.clone(),
                 include_plan_tool: config.include_plan_tool,
                 include_apply_patch_tool: config.include_apply_patch_tool,
                 include_web_search_request: config.tools_web_search_request,
                 use_streamable_shell_tool: config.use_experimental_streamable_shell_tool,
                 include_view_image_tool: config.include_view_image_tool,
+                experimental_unified_exec_tool: config.use_experimental_unified_exec_tool,
             }),
             user_instructions,
             base_instructions,
@@ -482,6 +487,7 @@ impl Session {
             tx_event: tx_event.clone(),
             mcp_connection_manager,
             session_manager: ExecSessionManager::default(),
+            unified_exec_manager: UnifiedExecSessionManager::default(),
             notify,
             state: Mutex::new(state),
             rollout: Mutex::new(Some(rollout_recorder)),
@@ -492,13 +498,9 @@ impl Session {
 
         // Dispatch the SessionConfiguredEvent first and then report any errors.
         // If resuming, include converted initial messages in the payload so UIs can render them immediately.
-        let initial_messages = match &initial_history {
-            InitialHistory::New => None,
-            InitialHistory::Forked(items) => Some(sess.build_initial_messages(items)),
-            InitialHistory::Resumed(resumed_history) => {
-                Some(sess.build_initial_messages(&resumed_history.history))
-            }
-        };
+        let initial_messages = initial_history.get_event_msgs();
+        sess.record_initial_history(&turn_context, initial_history)
+            .await;
 
         let events = std::iter::once(Event {
             id: INITIAL_SUBMIT_ID.to_owned(),
@@ -508,13 +510,12 @@ impl Session {
                 history_log_id,
                 history_entry_count,
                 initial_messages,
+                rollout_path,
             }),
         })
         .chain(post_session_configured_error_events.into_iter());
         for event in events {
-            if let Err(e) = tx_event.send(event).await {
-                error!("failed to send event: {e:?}");
-            }
+            sess.send_event(event).await;
         }
 
         Ok((sess, turn_context))
@@ -544,53 +545,33 @@ impl Session {
     ) {
         match conversation_history {
             InitialHistory::New => {
-                self.record_initial_history_new(turn_context).await;
+                // Build and record initial items (user instructions + environment context)
+                let items = self.build_initial_context(turn_context);
+                self.record_conversation_items(&items).await;
             }
-            InitialHistory::Forked(items) => {
-                self.record_initial_history_from_items(items).await;
-            }
-            InitialHistory::Resumed(resumed_history) => {
-                self.record_initial_history_from_items(resumed_history.history)
-                    .await;
+            InitialHistory::Resumed(_) | InitialHistory::Forked(_) => {
+                let rollout_items = conversation_history.get_rollout_items();
+                let persist = matches!(conversation_history, InitialHistory::Forked(_));
+
+                // Always add response items to conversation history
+                let response_items = conversation_history.get_response_items();
+                if !response_items.is_empty() {
+                    self.record_into_history(&response_items);
+                }
+
+                // If persisting, persist all rollout items as-is (recorder filters)
+                if persist && !rollout_items.is_empty() {
+                    self.persist_rollout_items(&rollout_items).await;
+                }
             }
         }
     }
 
-    async fn record_initial_history_new(&self, turn_context: &TurnContext) {
-        // record the initial user instructions and environment context,
-        // regardless of whether we restored items.
-        // TODO: Those items shouldn't be "user messages" IMO. Maybe developer messages.
-        let mut conversation_items = Vec::<ResponseItem>::with_capacity(2);
-        if let Some(user_instructions) = turn_context.user_instructions.as_deref() {
-            conversation_items.push(UserInstructions::new(user_instructions.to_string()).into());
-        }
-        conversation_items.push(ResponseItem::from(EnvironmentContext::new(
-            Some(turn_context.cwd.clone()),
-            Some(turn_context.approval_policy),
-            Some(turn_context.sandbox_policy.clone()),
-            Some(self.user_shell.clone()),
-        )));
-        self.record_conversation_items(&conversation_items).await;
-    }
-
-    async fn record_initial_history_from_items(&self, items: Vec<ResponseItem>) {
-        self.record_conversation_items_internal(&items, false).await;
-    }
-
-    /// build the initial messages vector for SessionConfigured by converting
-    /// ResponseItems into EventMsg.
-    fn build_initial_messages(&self, items: &[ResponseItem]) -> Vec<EventMsg> {
-        items
-            .iter()
-            .flat_map(|item| {
-                map_response_item_to_event_messages(item, self.show_raw_agent_reasoning)
-            })
-            .collect()
-    }
-
-    /// Sends the given event to the client and swallows the send event, if
-    /// any, logging it as an error.
+    /// Persist the event to rollout and send it to clients.
     pub(crate) async fn send_event(&self, event: Event) {
+        // Persist the event into rollout (recorder filters as needed)
+        let rollout_items = vec![RolloutItem::EventMsg(event.msg.clone())];
+        self.persist_rollout_items(&rollout_items).await;
         if let Err(e) = self.tx_event.send(event).await {
             error!("failed to send tool call event: {e}");
         }
@@ -624,7 +605,7 @@ impl Session {
                 reason,
             }),
         };
-        let _ = self.tx_event.send(event).await;
+        self.send_event(event).await;
         rx_approve
     }
 
@@ -656,7 +637,7 @@ impl Session {
                 grant_root,
             }),
         };
-        let _ = self.tx_event.send(event).await;
+        self.send_event(event).await;
         rx_approve
     }
 
@@ -680,36 +661,76 @@ impl Session {
         state.approved_commands.insert(cmd);
     }
 
-    /// Records items to both the rollout and the chat completions/ZDR
-    /// transcript, if enabled.
+    /// Records input items: always append to conversation history and
+    /// persist these response items to rollout.
     async fn record_conversation_items(&self, items: &[ResponseItem]) {
-        self.record_conversation_items_internal(items, true).await;
+        self.record_into_history(items);
+        self.persist_rollout_response_items(items).await;
     }
 
-    async fn record_conversation_items_internal(&self, items: &[ResponseItem], persist: bool) {
-        debug!("Recording items for conversation: {items:?}");
-        if persist {
-            self.record_state_snapshot(items).await;
+    /// Append ResponseItems to the in-memory conversation history only.
+    fn record_into_history(&self, items: &[ResponseItem]) {
+        self.state
+            .lock_unchecked()
+            .history
+            .record_items(items.iter());
+    }
+
+    async fn persist_rollout_response_items(&self, items: &[ResponseItem]) {
+        let rollout_items: Vec<RolloutItem> = items
+            .iter()
+            .cloned()
+            .map(RolloutItem::ResponseItem)
+            .collect();
+        self.persist_rollout_items(&rollout_items).await;
+    }
+
+    fn build_initial_context(&self, turn_context: &TurnContext) -> Vec<ResponseItem> {
+        let mut items = Vec::<ResponseItem>::with_capacity(2);
+        if let Some(user_instructions) = turn_context.user_instructions.as_deref() {
+            items.push(UserInstructions::new(user_instructions.to_string()).into());
         }
-
-        self.state.lock_unchecked().history.record_items(items);
+        items.push(ResponseItem::from(EnvironmentContext::new(
+            Some(turn_context.cwd.clone()),
+            Some(turn_context.approval_policy),
+            Some(turn_context.sandbox_policy.clone()),
+            Some(self.user_shell.clone()),
+        )));
+        items
     }
 
-    async fn record_state_snapshot(&self, items: &[ResponseItem]) {
-        let snapshot = { crate::rollout::SessionStateSnapshot {} };
-
+    async fn persist_rollout_items(&self, items: &[RolloutItem]) {
         let recorder = {
             let guard = self.rollout.lock_unchecked();
             guard.as_ref().cloned()
         };
+        if let Some(rec) = recorder
+            && let Err(e) = rec.record_items(items).await
+        {
+            error!("failed to record rollout items: {e:#}");
+        }
+    }
 
-        if let Some(rec) = recorder {
-            if let Err(e) = rec.record_state(snapshot).await {
-                error!("failed to record rollout state: {e:#}");
-            }
-            if let Err(e) = rec.record_items(items).await {
-                error!("failed to record rollout items: {e:#}");
-            }
+    /// Record a user input item to conversation history and also persist a
+    /// corresponding UserMessage EventMsg to rollout.
+    async fn record_input_and_rollout_usermsg(&self, response_input: &ResponseInputItem) {
+        let response_item: ResponseItem = response_input.clone().into();
+        // Add to conversation history and persist response item to rollout
+        self.record_conversation_items(std::slice::from_ref(&response_item))
+            .await;
+
+        // Derive user message events and persist only UserMessage to rollout
+        let msgs =
+            map_response_item_to_event_messages(&response_item, self.show_raw_agent_reasoning);
+        let user_msgs: Vec<RolloutItem> = msgs
+            .into_iter()
+            .filter_map(|m| match m {
+                EventMsg::UserMessage(ev) => Some(RolloutItem::EventMsg(EventMsg::UserMessage(ev))),
+                _ => None,
+            })
+            .collect();
+        if !user_msgs.is_empty() {
+            self.persist_rollout_items(&user_msgs).await;
         }
     }
 
@@ -752,7 +773,7 @@ impl Session {
             id: sub_id.to_string(),
             msg,
         };
-        let _ = self.tx_event.send(event).await;
+        self.send_event(event).await;
     }
 
     async fn on_exec_command_end(
@@ -799,7 +820,7 @@ impl Session {
             id: sub_id.to_string(),
             msg,
         };
-        let _ = self.tx_event.send(event).await;
+        self.send_event(event).await;
 
         // If this is an apply_patch, after we emit the end patch, emit a second event
         // with the full turn diff if there is one.
@@ -811,7 +832,7 @@ impl Session {
                     id: sub_id.into(),
                     msg,
                 };
-                let _ = self.tx_event.send(event).await;
+                self.send_event(event).await;
             }
         }
     }
@@ -877,7 +898,7 @@ impl Session {
                 message: message.into(),
             }),
         };
-        let _ = self.tx_event.send(event).await;
+        self.send_event(event).await;
     }
 
     async fn notify_stream_error(&self, sub_id: &str, message: impl Into<String>) {
@@ -887,7 +908,7 @@ impl Session {
                 message: message.into(),
             }),
         };
-        let _ = self.tx_event.send(event).await;
+        self.send_event(event).await;
     }
 
     /// Build the full turn input by concatenating the current conversation
@@ -1050,9 +1071,9 @@ impl AgentTask {
                 id: self.sub_id,
                 msg: EventMsg::TurnAborted(TurnAbortedEvent { reason }),
             };
-            let tx_event = self.sess.tx_event.clone();
+            let sess = self.sess.clone();
             tokio::spawn(async move {
-                tx_event.send(event).await.ok();
+                sess.send_event(event).await;
             });
         }
     }
@@ -1086,10 +1107,10 @@ async fn submission_loop(
                 let provider = prev.client.get_provider();
 
                 // Effective model + family
-                let (effective_model, effective_family) = if let Some(m) = model {
+                let (effective_model, effective_family) = if let Some(ref m) = model {
                     let fam =
-                        find_family_for_model(&m).unwrap_or_else(|| config.model_family.clone());
-                    (m, fam)
+                        find_family_for_model(m).unwrap_or_else(|| config.model_family.clone());
+                    (m.clone(), fam)
                 } else {
                     (prev.client.get_model(), prev.client.get_model_family())
                 };
@@ -1126,12 +1147,12 @@ async fn submission_loop(
                 let tools_config = ToolsConfig::new(&ToolsConfigParams {
                     model_family: &effective_family,
                     approval_policy: new_approval_policy,
-                    sandbox_policy: new_sandbox_policy.clone(),
                     include_plan_tool: config.include_plan_tool,
                     include_apply_patch_tool: config.include_apply_patch_tool,
                     include_web_search_request: config.tools_web_search_request,
                     use_streamable_shell_tool: config.use_experimental_streamable_shell_tool,
                     include_view_image_tool: config.include_view_image_tool,
+                    experimental_unified_exec_tool: config.use_experimental_unified_exec_tool,
                 });
 
                 let new_turn_context = TurnContext {
@@ -1147,25 +1168,34 @@ async fn submission_loop(
 
                 // Install the new persistent context for subsequent tasks/turns.
                 turn_context = Arc::new(new_turn_context);
-                if cwd.is_some() || approval_policy.is_some() || sandbox_policy.is_some() {
-                    sess.record_conversation_items(&[ResponseItem::from(EnvironmentContext::new(
-                        cwd,
-                        approval_policy,
-                        sandbox_policy,
-                        // Shell is not configurable from turn to turn
-                        None,
-                    ))])
-                    .await;
+
+                // Optionally persist changes to model / effort
+                let effort_str = effort.map(|_| effective_effort.to_string());
+
+                if let Err(e) = persist_non_null_overrides(
+                    &config.codex_home,
+                    config.active_profile.as_deref(),
+                    &[
+                        (&[CONFIG_KEY_MODEL], model.as_deref()),
+                        (&[CONFIG_KEY_EFFORT], effort_str.as_deref()),
+                    ],
+                )
+                .await
+                {
+                    warn!("failed to persist overrides: {e:#}");
                 }
             }
             Op::UserInput { items } => {
-                // attempt to inject input into current task
-                if let Err(items) = sess.inject_input(items) {
-                    // no current task, spawn a new one
-                    let task =
-                        AgentTask::spawn(sess.clone(), Arc::clone(&turn_context), sub.id, items);
-                    sess.set_task(task);
-                }
+                submit_user_input(
+                    turn_context.cwd.clone(),
+                    turn_context.approval_policy,
+                    turn_context.sandbox_policy.clone(),
+                    &sess,
+                    &turn_context,
+                    sub.id.clone(),
+                    items,
+                )
+                .await;
             }
             Op::UserTurn {
                 items,
@@ -1210,13 +1240,14 @@ async fn submission_loop(
                         tools_config: ToolsConfig::new(&ToolsConfigParams {
                             model_family: &model_family,
                             approval_policy,
-                            sandbox_policy: sandbox_policy.clone(),
                             include_plan_tool: config.include_plan_tool,
                             include_apply_patch_tool: config.include_apply_patch_tool,
                             include_web_search_request: config.tools_web_search_request,
                             use_streamable_shell_tool: config
                                 .use_experimental_streamable_shell_tool,
                             include_view_image_tool: config.include_view_image_tool,
+                            experimental_unified_exec_tool: config
+                                .use_experimental_unified_exec_tool,
                         }),
                         user_instructions: turn_context.user_instructions.clone(),
                         base_instructions: turn_context.base_instructions.clone(),
@@ -1225,11 +1256,16 @@ async fn submission_loop(
                         shell_environment_policy: turn_context.shell_environment_policy.clone(),
                         cwd,
                     };
-                    // TODO: record the new environment context in the conversation history
-                    // no current task, spawn a new one with the per‑turn context
-                    let task =
-                        AgentTask::spawn(sess.clone(), Arc::new(fresh_turn_context), sub.id, items);
-                    sess.set_task(task);
+                    submit_user_input(
+                        fresh_turn_context.cwd.clone(),
+                        fresh_turn_context.approval_policy,
+                        fresh_turn_context.sandbox_policy.clone(),
+                        &sess,
+                        &Arc::new(fresh_turn_context),
+                        sub.id.clone(),
+                        items,
+                    )
+                    .await;
                 }
             }
             Op::ExecApproval { id, decision } => match decision {
@@ -1257,7 +1293,7 @@ async fn submission_loop(
 
             Op::GetHistoryEntryRequest { offset, log_id } => {
                 let config = config.clone();
-                let tx_event = sess.tx_event.clone();
+                let sess_clone = sess.clone();
                 let sub_id = sub.id.clone();
 
                 tokio::spawn(async move {
@@ -1285,13 +1321,10 @@ async fn submission_loop(
                         ),
                     };
 
-                    if let Err(e) = tx_event.send(event).await {
-                        warn!("failed to send GetHistoryEntryResponse event: {e}");
-                    }
+                    sess_clone.send_event(event).await;
                 });
             }
             Op::ListMcpTools => {
-                let tx_event = sess.tx_event.clone();
                 let sub_id = sub.id.clone();
 
                 // This is a cheap lookup from the connection manager's cache.
@@ -1302,12 +1335,9 @@ async fn submission_loop(
                         crate::protocol::McpListToolsResponseEvent { tools },
                     ),
                 };
-                if let Err(e) = tx_event.send(event).await {
-                    warn!("failed to send McpListToolsResponse event: {e}");
-                }
+                sess.send_event(event).await;
             }
             Op::ListCustomPrompts => {
-                let tx_event = sess.tx_event.clone();
                 let sub_id = sub.id.clone();
 
                 let custom_prompts: Vec<CustomPrompt> =
@@ -1323,9 +1353,7 @@ async fn submission_loop(
                         custom_prompts,
                     }),
                 };
-                if let Err(e) = tx_event.send(event).await {
-                    warn!("failed to send ListCustomPromptsResponse event: {e}");
-                }
+                sess.send_event(event).await;
             }
             Op::Compact => {
                 // Create a summarization request as user input
@@ -1361,34 +1389,42 @@ async fn submission_loop(
                             message: "Failed to shutdown rollout recorder".to_string(),
                         }),
                     };
-                    if let Err(e) = sess.tx_event.send(event).await {
-                        warn!("failed to send error message: {e:?}");
-                    }
+                    sess.send_event(event).await;
                 }
 
                 let event = Event {
                     id: sub.id.clone(),
                     msg: EventMsg::ShutdownComplete,
                 };
-                if let Err(e) = sess.tx_event.send(event).await {
-                    warn!("failed to send Shutdown event: {e}");
-                }
+                sess.send_event(event).await;
                 break;
             }
-            Op::GetHistory => {
-                let tx_event = sess.tx_event.clone();
+            Op::GetPath => {
                 let sub_id = sub.id.clone();
-
+                // Flush rollout writes before returning the path so readers observe a consistent file.
+                let (path, rec_opt) = {
+                    let guard = sess.rollout.lock_unchecked();
+                    match guard.as_ref() {
+                        Some(rec) => (rec.get_rollout_path(), Some(rec.clone())),
+                        None => {
+                            error!("rollout recorder not found");
+                            continue;
+                        }
+                    }
+                };
+                if let Some(rec) = rec_opt
+                    && let Err(e) = rec.flush().await
+                {
+                    warn!("failed to flush rollout recorder before GetHistory: {e}");
+                }
                 let event = Event {
                     id: sub_id.clone(),
-                    msg: EventMsg::ConversationHistory(ConversationHistoryResponseEvent {
+                    msg: EventMsg::ConversationPath(ConversationPathResponseEvent {
                         conversation_id: sess.conversation_id,
-                        entries: sess.state.lock_unchecked().history.contents(),
+                        path,
                     }),
                 };
-                if let Err(e) = tx_event.send(event).await {
-                    warn!("failed to send ConversationHistory event: {e}");
-                }
+                sess.send_event(event).await;
             }
             _ => {
                 // Ignore unknown ops; enum is non_exhaustive to allow extensions.
@@ -1426,12 +1462,10 @@ async fn run_task(
             model_context_window: turn_context.client.get_model_context_window(),
         }),
     };
-    if sess.tx_event.send(event).await.is_err() {
-        return;
-    }
+    sess.send_event(event).await;
 
     let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
-    sess.record_conversation_items(&[initial_input_for_turn.clone().into()])
+    sess.record_input_and_rollout_usermsg(&initial_input_for_turn)
         .await;
 
     let mut last_agent_message: Option<String> = None;
@@ -1600,7 +1634,7 @@ async fn run_task(
                         message: e.to_string(),
                     }),
                 };
-                sess.tx_event.send(event).await.ok();
+                sess.send_event(event).await;
                 // let the user continue the conversation
                 break;
             }
@@ -1611,7 +1645,7 @@ async fn run_task(
         id: sub_id,
         msg: EventMsg::TaskComplete(TaskCompleteEvent { last_agent_message }),
     };
-    sess.tx_event.send(event).await.ok();
+    sess.send_event(event).await;
 }
 
 async fn run_turn(
@@ -1809,13 +1843,12 @@ async fn try_run_turn(
                     st.token_info = info.clone();
                     info
                 };
-                sess.tx_event
-                    .send(Event {
+                let _ = sess
+                    .send_event(Event {
                         id: sub_id.to_string(),
                         msg: EventMsg::TokenCount(crate::protocol::TokenCountEvent { info }),
                     })
-                    .await
-                    .ok();
+                    .await;
 
                 let unified_diff = turn_diff_tracker.get_unified_diff();
                 if let Ok(Some(unified_diff)) = unified_diff {
@@ -1824,7 +1857,7 @@ async fn try_run_turn(
                         id: sub_id.to_string(),
                         msg,
                     };
-                    let _ = sess.tx_event.send(event).await;
+                    sess.send_event(event).await;
                 }
 
                 return Ok(output);
@@ -1834,21 +1867,21 @@ async fn try_run_turn(
                     id: sub_id.to_string(),
                     msg: EventMsg::AgentMessageDelta(AgentMessageDeltaEvent { delta }),
                 };
-                sess.tx_event.send(event).await.ok();
+                sess.send_event(event).await;
             }
             ResponseEvent::ReasoningSummaryDelta(delta) => {
                 let event = Event {
                     id: sub_id.to_string(),
                     msg: EventMsg::AgentReasoningDelta(AgentReasoningDeltaEvent { delta }),
                 };
-                sess.tx_event.send(event).await.ok();
+                sess.send_event(event).await;
             }
             ResponseEvent::ReasoningSummaryPartAdded => {
                 let event = Event {
                     id: sub_id.to_string(),
                     msg: EventMsg::AgentReasoningSectionBreak(AgentReasoningSectionBreakEvent {}),
                 };
-                sess.tx_event.send(event).await.ok();
+                sess.send_event(event).await;
             }
             ResponseEvent::ReasoningContentDelta(delta) => {
                 if sess.show_raw_agent_reasoning {
@@ -1858,7 +1891,7 @@ async fn try_run_turn(
                             AgentReasoningRawContentDeltaEvent { delta },
                         ),
                     };
-                    sess.tx_event.send(event).await.ok();
+                    sess.send_event(event).await;
                 }
             }
         }
@@ -1879,9 +1912,7 @@ async fn run_compact_task(
             model_context_window,
         }),
     };
-    if sess.tx_event.send(start_event).await.is_err() {
-        return;
-    }
+    sess.send_event(start_event).await;
 
     let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
     let turn_input: Vec<ResponseItem> =
@@ -2059,7 +2090,7 @@ async fn handle_response_item(
                     id: sub_id.to_string(),
                     msg,
                 };
-                sess.tx_event.send(event).await.ok();
+                sess.send_event(event).await;
             }
             None
         }
@@ -2068,6 +2099,72 @@ async fn handle_response_item(
     Ok(output)
 }
 
+async fn handle_unified_exec_tool_call(
+    sess: &Session,
+    call_id: String,
+    session_id: Option<String>,
+    arguments: Vec<String>,
+    timeout_ms: Option<u64>,
+) -> ResponseInputItem {
+    let parsed_session_id = if let Some(session_id) = session_id {
+        match session_id.parse::<i32>() {
+            Ok(parsed) => Some(parsed),
+            Err(output) => {
+                return ResponseInputItem::FunctionCallOutput {
+                    call_id: call_id.to_string(),
+                    output: FunctionCallOutputPayload {
+                        content: format!("invalid session_id: {session_id} due to error {output}"),
+                        success: Some(false),
+                    },
+                };
+            }
+        }
+    } else {
+        None
+    };
+
+    let request = crate::unified_exec::UnifiedExecRequest {
+        session_id: parsed_session_id,
+        input_chunks: &arguments,
+        timeout_ms,
+    };
+
+    let result = sess.unified_exec_manager.handle_request(request).await;
+
+    let output_payload = match result {
+        Ok(value) => {
+            #[derive(Serialize)]
+            struct SerializedUnifiedExecResult<'a> {
+                session_id: Option<String>,
+                output: &'a str,
+            }
+
+            match serde_json::to_string(&SerializedUnifiedExecResult {
+                session_id: value.session_id.map(|id| id.to_string()),
+                output: &value.output,
+            }) {
+                Ok(serialized) => FunctionCallOutputPayload {
+                    content: serialized,
+                    success: Some(true),
+                },
+                Err(err) => FunctionCallOutputPayload {
+                    content: format!("failed to serialize unified exec output: {err}"),
+                    success: Some(false),
+                },
+            }
+        }
+        Err(err) => FunctionCallOutputPayload {
+            content: format!("unified exec failed: {err}"),
+            success: Some(false),
+        },
+    };
+
+    ResponseInputItem::FunctionCallOutput {
+        call_id,
+        output: output_payload,
+    }
+}
+
 async fn handle_function_call(
     sess: &Session,
     turn_context: &TurnContext,
@@ -2095,6 +2192,38 @@ async fn handle_function_call(
             )
             .await
         }
+        "unified_exec" => {
+            #[derive(Deserialize)]
+            struct UnifiedExecArgs {
+                input: Vec<String>,
+                #[serde(default)]
+                session_id: Option<String>,
+                #[serde(default)]
+                timeout_ms: Option<u64>,
+            }
+
+            let args = match serde_json::from_str::<UnifiedExecArgs>(&arguments) {
+                Ok(args) => args,
+                Err(err) => {
+                    return ResponseInputItem::FunctionCallOutput {
+                        call_id,
+                        output: FunctionCallOutputPayload {
+                            content: format!("failed to parse function arguments: {err}"),
+                            success: Some(false),
+                        },
+                    };
+                }
+            };
+
+            handle_unified_exec_tool_call(
+                sess,
+                call_id,
+                args.session_id,
+                args.input,
+                args.timeout_ms,
+            )
+            .await
+        }
         "view_image" => {
             #[derive(serde::Deserialize)]
             struct SeeImageArgs {
@@ -2698,6 +2827,30 @@ async fn handle_sandbox_error(
     }
 }
 
+async fn submit_user_input(
+    cwd: PathBuf,
+    approval_policy: AskForApproval,
+    sandbox_policy: SandboxPolicy,
+    sess: &Arc<Session>,
+    turn_context: &Arc<TurnContext>,
+    sub_id: String,
+    items: Vec<InputItem>,
+) {
+    sess.record_conversation_items(&[ResponseItem::from(EnvironmentContext::new(
+        Some(cwd),
+        Some(approval_policy),
+        Some(sandbox_policy),
+        // Shell is not configurable from turn to turn
+        None,
+    ))])
+    .await;
+    if let Err(items) = sess.inject_input(items) {
+        // no current task, spawn a new one
+        let task = AgentTask::spawn(Arc::clone(sess), Arc::clone(turn_context), sub_id, items);
+        sess.set_task(task);
+    }
+}
+
 fn format_exec_output_str(exec_output: &ExecToolCallOutput) -> String {
     let ExecToolCallOutput {
         aggregated_output, ..

codex-rs/core/src/config.rs

@@ -38,9 +38,7 @@ const OPENAI_DEFAULT_MODEL: &str = "gpt-5";
 /// the context window.
 pub(crate) const PROJECT_DOC_MAX_BYTES: usize = 32 * 1024; // 32 KiB
 
-const CONFIG_TOML_FILE: &str = "config.toml";
-
-const DEFAULT_RESPONSES_ORIGINATOR_HEADER: &str = "codex_cli_rs";
+pub(crate) const CONFIG_TOML_FILE: &str = "config.toml";
 
 /// Application configuration loaded from disk and merged with overrides.
 #[derive(Debug, Clone, PartialEq)]
@@ -169,16 +167,20 @@ pub struct Config {
 
     pub tools_web_search_request: bool,
 
-    /// The value for the `originator` header included with Responses API requests.
-    pub responses_originator_header: String,
-
     /// 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,
+
+    /// The active profile name used to derive this `Config` (if any).
+    pub active_profile: Option<String>,
+
     /// When true, disables burst-paste detection for typed input entirely.
     /// All characters are inserted as they are received, and no buffering
     /// or placeholder replacement will occur for fast keypress bursts.
@@ -262,17 +264,7 @@ 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]
@@ -288,14 +280,26 @@ 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 {
@@ -304,12 +308,6 @@ 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())
@@ -328,6 +326,21 @@ 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)?;
@@ -477,9 +490,7 @@ pub struct ConfigToml {
     pub experimental_instructions_file: Option<PathBuf>,
 
     pub experimental_use_exec_command_tool: Option<bool>,
-
-    /// The value for the `originator` header included with Responses API requests.
-    pub responses_originator_header_internal_override: Option<String>,
+    pub experimental_use_unified_exec_tool: Option<bool>,
 
     pub projects: Option<HashMap<String, ProjectConfig>>,
 
@@ -661,7 +672,11 @@ impl Config {
             tools_web_search_request: override_tools_web_search_request,
         } = overrides;
 
-        let config_profile = match config_profile_key.as_ref().or(cfg.profile.as_ref()) {
+        let active_profile_name = config_profile_key
+            .as_ref()
+            .or(cfg.profile.as_ref())
+            .cloned();
+        let config_profile = match active_profile_name.as_ref() {
             Some(key) => cfg
                 .profiles
                 .get(key)
@@ -773,10 +788,6 @@ impl Config {
             Self::get_base_instructions(experimental_instructions_path, &resolved_cwd)?;
         let base_instructions = base_instructions.or(file_base_instructions);
 
-        let responses_originator_header: String = cfg
-            .responses_originator_header_internal_override
-            .unwrap_or(DEFAULT_RESPONSES_ORIGINATOR_HEADER.to_owned());
-
         let config = Self {
             model,
             model_family,
@@ -826,12 +837,15 @@ impl Config {
             include_plan_tool: include_plan_tool.unwrap_or(false),
             include_apply_patch_tool: include_apply_patch_tool.unwrap_or(false),
             tools_web_search_request,
-            responses_originator_header,
             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(true),
             include_view_image_tool,
+            active_profile: active_profile_name,
             disable_paste_burst: cfg.disable_paste_burst.unwrap_or(false),
         };
         Ok(config)
@@ -1203,10 +1217,11 @@ model_verbosity = "high"
                 include_plan_tool: false,
                 include_apply_patch_tool: false,
                 tools_web_search_request: false,
-                responses_originator_header: "codex_cli_rs".to_string(),
                 preferred_auth_method: AuthMode::ChatGPT,
                 use_experimental_streamable_shell_tool: false,
+                use_experimental_unified_exec_tool: true,
                 include_view_image_tool: true,
+                active_profile: Some("o3".to_string()),
                 disable_paste_burst: false,
             },
             o3_profile_config
@@ -1260,10 +1275,11 @@ model_verbosity = "high"
             include_plan_tool: false,
             include_apply_patch_tool: false,
             tools_web_search_request: false,
-            responses_originator_header: "codex_cli_rs".to_string(),
             preferred_auth_method: AuthMode::ChatGPT,
             use_experimental_streamable_shell_tool: false,
+            use_experimental_unified_exec_tool: true,
             include_view_image_tool: true,
+            active_profile: Some("gpt3".to_string()),
             disable_paste_burst: false,
         };
 
@@ -1332,10 +1348,11 @@ model_verbosity = "high"
             include_plan_tool: false,
             include_apply_patch_tool: false,
             tools_web_search_request: false,
-            responses_originator_header: "codex_cli_rs".to_string(),
             preferred_auth_method: AuthMode::ChatGPT,
             use_experimental_streamable_shell_tool: false,
+            use_experimental_unified_exec_tool: true,
             include_view_image_tool: true,
+            active_profile: Some("zdr".to_string()),
             disable_paste_burst: false,
         };
 
@@ -1390,10 +1407,11 @@ model_verbosity = "high"
             include_plan_tool: false,
             include_apply_patch_tool: false,
             tools_web_search_request: false,
-            responses_originator_header: "codex_cli_rs".to_string(),
             preferred_auth_method: AuthMode::ChatGPT,
             use_experimental_streamable_shell_tool: false,
+            use_experimental_unified_exec_tool: true,
             include_view_image_tool: true,
+            active_profile: Some("gpt5".to_string()),
             disable_paste_burst: false,
         };
 
@@ -1404,17 +1422,14 @@ 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 {
@@ -1432,12 +1447,10 @@ 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 {
@@ -1449,13 +1462,12 @@ 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!(
@@ -1470,5 +1482,37 @@ trust_level = "trusted"
         Ok(())
     }
 
-    // No test enforcing the presence of a standalone [projects] header.
+    #[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

@@ -0,0 +1,582 @@
+use crate::config::CONFIG_TOML_FILE;
+use anyhow::Result;
+use std::path::Path;
+use tempfile::NamedTempFile;
+use toml_edit::DocumentMut;
+
+pub const CONFIG_KEY_MODEL: &str = "model";
+pub const CONFIG_KEY_EFFORT: &str = "model_reasoning_effort";
+
+/// 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(
+    codex_home: &Path,
+    profile: Option<&str>,
+    overrides: &[(&[&str], &str)],
+) -> Result<()> {
+    let config_path = codex_home.join(CONFIG_TOML_FILE);
+
+    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 where values may be optional. Any entries with `None`
+/// values are skipped. If all values are `None`, this becomes a no-op and
+/// returns `Ok(())` without touching the file.
+pub async fn persist_non_null_overrides(
+    codex_home: &Path,
+    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();
+
+    if filtered.is_empty() {
+        return Ok(());
+    }
+
+    persist_overrides(codex_home, profile, &filtered).await
+}
+
+/// Apply a single override onto a `toml_edit` document while preserving
+/// existing formatting/comments.
+/// The key is expressed as explicit segments to correctly handle keys that
+/// contain dots or spaces.
+fn apply_toml_edit_override_segments(
+    doc: &mut DocumentMut,
+    segments: &[&str],
+    value: toml_edit::Item,
+) {
+    use toml_edit::Item;
+
+    if segments.is_empty() {
+        return;
+    }
+
+    let mut current = doc.as_table_mut();
+    for seg in &segments[..segments.len() - 1] {
+        if !current.contains_key(seg) {
+            current[*seg] = Item::Table(toml_edit::Table::new());
+            if let Some(t) = current[*seg].as_table_mut() {
+                t.set_implicit(true);
+            }
+        }
+
+        let maybe_item = current.get_mut(seg);
+        let Some(item) = maybe_item else { return };
+
+        if !item.is_table() {
+            *item = Item::Table(toml_edit::Table::new());
+            if let Some(t) = item.as_table_mut() {
+                t.set_implicit(true);
+            }
+        }
+
+        let Some(tbl) = item.as_table_mut() else {
+            return;
+        };
+        current = tbl;
+    }
+
+    let last = segments[segments.len() - 1];
+    current[last] = value;
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use pretty_assertions::assert_eq;
+    use tempfile::tempdir;
+
+    /// Verifies model and effort are written at top-level when no profile is set.
+    #[tokio::test]
+    async fn set_default_model_and_effort_top_level_when_no_profile() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        persist_overrides(
+            codex_home,
+            None,
+            &[
+                (&[CONFIG_KEY_MODEL], "gpt-5"),
+                (&[CONFIG_KEY_EFFORT], "high"),
+            ],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"model = "gpt-5"
+model_reasoning_effort = "high"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies values are written under the active profile when `profile` is set.
+    #[tokio::test]
+    async fn set_defaults_update_profile_when_profile_set() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        // Seed config with a profile selection but without profiles table
+        let seed = "profile = \"o3\"\n";
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        persist_overrides(
+            codex_home,
+            None,
+            &[
+                (&[CONFIG_KEY_MODEL], "o3"),
+                (&[CONFIG_KEY_EFFORT], "minimal"),
+            ],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"profile = "o3"
+
+[profiles.o3]
+model = "o3"
+model_reasoning_effort = "minimal"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies profile names with dots/spaces are preserved via explicit segments.
+    #[tokio::test]
+    async fn set_defaults_update_profile_with_dot_and_space() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        // Seed config with a profile name that contains a dot and a space
+        let seed = "profile = \"my.team name\"\n";
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        persist_overrides(
+            codex_home,
+            None,
+            &[
+                (&[CONFIG_KEY_MODEL], "o3"),
+                (&[CONFIG_KEY_EFFORT], "minimal"),
+            ],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"profile = "my.team name"
+
+[profiles."my.team name"]
+model = "o3"
+model_reasoning_effort = "minimal"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies explicit profile override writes under that profile even without active profile.
+    #[tokio::test]
+    async fn set_defaults_update_when_profile_override_supplied() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        // No profile key in config.toml
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), "")
+            .await
+            .expect("seed write");
+
+        // Persist with an explicit profile override
+        persist_overrides(
+            codex_home,
+            Some("o3"),
+            &[(&[CONFIG_KEY_MODEL], "o3"), (&[CONFIG_KEY_EFFORT], "high")],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"[profiles.o3]
+model = "o3"
+model_reasoning_effort = "high"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies nested tables are created as needed when applying overrides.
+    #[tokio::test]
+    async fn persist_overrides_creates_nested_tables() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        persist_overrides(
+            codex_home,
+            None,
+            &[
+                (&["a", "b", "c"], "v"),
+                (&["x"], "y"),
+                (&["profiles", "p1", CONFIG_KEY_MODEL], "gpt-5"),
+            ],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"x = "y"
+
+[a.b]
+c = "v"
+
+[profiles.p1]
+model = "gpt-5"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies a scalar key becomes a table when nested keys are written.
+    #[tokio::test]
+    async fn persist_overrides_replaces_scalar_with_table() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+        let seed = "foo = \"bar\"\n";
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        persist_overrides(codex_home, None, &[(&["foo", "bar", "baz"], "ok")])
+            .await
+            .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"[foo.bar]
+baz = "ok"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies comments and spacing are preserved when writing under active profile.
+    #[tokio::test]
+    async fn set_defaults_preserve_comments() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        // Seed a config with comments and spacing we expect to preserve
+        let seed = r#"# Global comment
+# Another line
+
+profile = "o3"
+
+# Profile settings
+[profiles.o3]
+# keep me
+existing = "keep"
+"#;
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        // Apply defaults; since profile is set, it should write under [profiles.o3]
+        persist_overrides(
+            codex_home,
+            None,
+            &[(&[CONFIG_KEY_MODEL], "o3"), (&[CONFIG_KEY_EFFORT], "high")],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"# Global comment
+# Another line
+
+profile = "o3"
+
+# Profile settings
+[profiles.o3]
+# keep me
+existing = "keep"
+model = "o3"
+model_reasoning_effort = "high"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies comments and spacing are preserved when writing at top level.
+    #[tokio::test]
+    async fn set_defaults_preserve_global_comments() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        // Seed a config WITHOUT a profile, containing comments and spacing
+        let seed = r#"# Top-level comments
+# should be preserved
+
+existing = "keep"
+"#;
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        // Since there is no profile, the defaults should be written at top-level
+        persist_overrides(
+            codex_home,
+            None,
+            &[
+                (&[CONFIG_KEY_MODEL], "gpt-5"),
+                (&[CONFIG_KEY_EFFORT], "minimal"),
+            ],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"# Top-level comments
+# should be preserved
+
+existing = "keep"
+model = "gpt-5"
+model_reasoning_effort = "minimal"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies errors on invalid TOML propagate and file is not clobbered.
+    #[tokio::test]
+    async fn persist_overrides_errors_on_parse_failure() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        // Write an intentionally invalid TOML file
+        let invalid = "invalid = [unclosed";
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), invalid)
+            .await
+            .expect("seed write");
+
+        // Attempting to persist should return an error and must not clobber the file.
+        let res = persist_overrides(codex_home, None, &[(&["x"], "y")]).await;
+        assert!(res.is_err(), "expected parse error to propagate");
+
+        // File should be unchanged
+        let contents = read_config(codex_home).await;
+        assert_eq!(contents, invalid);
+    }
+
+    /// Verifies changing model only preserves existing effort at top-level.
+    #[tokio::test]
+    async fn changing_only_model_preserves_existing_effort_top_level() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        // Seed with an effort value only
+        let seed = "model_reasoning_effort = \"minimal\"\n";
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        // Change only the model
+        persist_overrides(codex_home, None, &[(&[CONFIG_KEY_MODEL], "o3")])
+            .await
+            .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"model_reasoning_effort = "minimal"
+model = "o3"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies changing effort only preserves existing model at top-level.
+    #[tokio::test]
+    async fn changing_only_effort_preserves_existing_model_top_level() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        // Seed with a model value only
+        let seed = "model = \"gpt-5\"\n";
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        // Change only the effort
+        persist_overrides(codex_home, None, &[(&[CONFIG_KEY_EFFORT], "high")])
+            .await
+            .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"model = "gpt-5"
+model_reasoning_effort = "high"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies changing model only preserves existing effort in active profile.
+    #[tokio::test]
+    async fn changing_only_model_preserves_effort_in_active_profile() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        // Seed with an active profile and an existing effort under that profile
+        let seed = r#"profile = "p1"
+
+[profiles.p1]
+model_reasoning_effort = "low"
+"#;
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        persist_overrides(codex_home, None, &[(&[CONFIG_KEY_MODEL], "o4-mini")])
+            .await
+            .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"profile = "p1"
+
+[profiles.p1]
+model_reasoning_effort = "low"
+model = "o4-mini"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies changing effort only preserves existing model in a profile override.
+    #[tokio::test]
+    async fn changing_only_effort_preserves_model_in_profile_override() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        // No active profile key; we'll target an explicit override
+        let seed = r#"[profiles.team]
+model = "gpt-5"
+"#;
+        tokio::fs::write(codex_home.join(CONFIG_TOML_FILE), seed)
+            .await
+            .expect("seed write");
+
+        persist_overrides(
+            codex_home,
+            Some("team"),
+            &[(&[CONFIG_KEY_EFFORT], "minimal")],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"[profiles.team]
+model = "gpt-5"
+model_reasoning_effort = "minimal"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies `persist_non_null_overrides` skips `None` entries and writes only present values at top-level.
+    #[tokio::test]
+    async fn persist_non_null_skips_none_top_level() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        persist_non_null_overrides(
+            codex_home,
+            None,
+            &[
+                (&[CONFIG_KEY_MODEL], Some("gpt-5")),
+                (&[CONFIG_KEY_EFFORT], None),
+            ],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = "model = \"gpt-5\"\n";
+        assert_eq!(contents, expected);
+    }
+
+    /// Verifies no-op behavior when all provided overrides are `None` (no file created/modified).
+    #[tokio::test]
+    async fn persist_non_null_noop_when_all_none() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        persist_non_null_overrides(
+            codex_home,
+            None,
+            &[(&["a"], None), (&["profiles", "p", "x"], None)],
+        )
+        .await
+        .expect("persist");
+
+        // Should not create config.toml on a pure no-op
+        assert!(!codex_home.join(CONFIG_TOML_FILE).exists());
+    }
+
+    /// Verifies entries are written under the specified profile and `None` entries are skipped.
+    #[tokio::test]
+    async fn persist_non_null_respects_profile_override() {
+        let tmpdir = tempdir().expect("tmp");
+        let codex_home = tmpdir.path();
+
+        persist_non_null_overrides(
+            codex_home,
+            Some("team"),
+            &[
+                (&[CONFIG_KEY_MODEL], Some("o3")),
+                (&[CONFIG_KEY_EFFORT], None),
+            ],
+        )
+        .await
+        .expect("persist");
+
+        let contents = read_config(codex_home).await;
+        let expected = r#"[profiles.team]
+model = "o3"
+"#;
+        assert_eq!(contents, expected);
+    }
+
+    // Test helper moved to bottom per review guidance.
+    async fn read_config(codex_home: &Path) -> String {
+        let p = codex_home.join(CONFIG_TOML_FILE);
+        tokio::fs::read_to_string(p).await.unwrap_or_default()
+    }
+}

codex-rs/core/src/conversation_manager.rs

@@ -13,25 +13,13 @@ use crate::protocol::SessionConfiguredEvent;
 use crate::rollout::RolloutRecorder;
 use codex_protocol::mcp_protocol::ConversationId;
 use codex_protocol::models::ResponseItem;
+use codex_protocol::protocol::InitialHistory;
+use codex_protocol::protocol::RolloutItem;
 use std::collections::HashMap;
 use std::path::PathBuf;
 use std::sync::Arc;
 use tokio::sync::RwLock;
 
-#[derive(Debug, Clone, PartialEq)]
-pub struct ResumedHistory {
-    pub conversation_id: ConversationId,
-    pub history: Vec<ResponseItem>,
-    pub rollout_path: PathBuf,
-}
-
-#[derive(Debug, Clone, PartialEq)]
-pub enum InitialHistory {
-    New,
-    Resumed(ResumedHistory),
-    Forked(Vec<ResponseItem>),
-}
-
 /// Represents a newly created Codex conversation, including the first event
 /// (which is [`EventMsg::SessionConfigured`]).
 pub struct NewConversation {
@@ -145,8 +133,15 @@ impl ConversationManager {
         self.finalize_spawn(codex, conversation_id).await
     }
 
-    pub async fn remove_conversation(&self, conversation_id: ConversationId) {
-        self.conversations.write().await.remove(&conversation_id);
+    /// Removes the conversation from the manager's internal map, though the
+    /// conversation is stored as `Arc<CodexConversation>`, it is possible that
+    /// other references to it exist elsewhere. Returns the conversation if the
+    /// conversation was found and removed.
+    pub async fn remove_conversation(
+        &self,
+        conversation_id: &ConversationId,
+    ) -> Option<Arc<CodexConversation>> {
+        self.conversations.write().await.remove(conversation_id)
     }
 
     /// Fork an existing conversation by dropping the last `drop_last_messages`
@@ -155,13 +150,13 @@ impl ConversationManager {
     /// 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,
         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_dropping_last_messages(history, num_messages_to_drop);
 
         // Spawn a new conversation with the computed initial history.
         let auth_manager = self.auth_manager.clone();
@@ -176,31 +171,37 @@ 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 {
+fn truncate_after_dropping_last_messages(history: InitialHistory, n: usize) -> InitialHistory {
     if n == 0 {
-        return InitialHistory::Forked(items);
+        return InitialHistory::Forked(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
+    // Work directly on rollout items, and cut the vector at the nth-from-last user message input.
+    let items: Vec<RolloutItem> = history.get_rollout_items();
+
+    // 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 n user messages exist, treat as empty.
+    if user_positions.len() < n {
+        return InitialHistory::New;
+    }
+
+    // Cut strictly before the nth-from-last user message (do not keep the nth itself).
+    let cut_idx = user_positions[user_positions.len() - n];
+    let rolled: Vec<RolloutItem> = items.into_iter().take(cut_idx).collect();
+
+    if rolled.is_empty() {
         InitialHistory::New
     } else {
-        InitialHistory::Forked(items.into_iter().take(cut_index).collect())
+        InitialHistory::Forked(rolled)
     }
 }
 
@@ -255,13 +256,30 @@ 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_dropping_last_messages(InitialHistory::Forked(initial), 1);
+        let got_items = truncated.get_rollout_items();
+        let expected_items = vec![
+            RolloutItem::ResponseItem(items[0].clone()),
+            RolloutItem::ResponseItem(items[1].clone()),
+            RolloutItem::ResponseItem(items[2].clone()),
+        ];
         assert_eq!(
-            truncated,
-            InitialHistory::Forked(vec![items[0].clone(), items[1].clone(), items[2].clone(),])
+            serde_json::to_value(&got_items).unwrap(),
+            serde_json::to_value(&expected_items).unwrap()
         );
 
-        let truncated2 = truncate_after_dropping_last_messages(items, 2);
-        assert_eq!(truncated2, InitialHistory::New);
+        let initial2: Vec<RolloutItem> = items
+            .iter()
+            .cloned()
+            .map(RolloutItem::ResponseItem)
+            .collect();
+        let truncated2 = truncate_after_dropping_last_messages(InitialHistory::Forked(initial2), 2);
+        assert!(matches!(truncated2, InitialHistory::New));
     }
 }

codex-rs/core/src/default_client.rs

@@ -1,38 +1,123 @@
-pub const DEFAULT_ORIGINATOR: &str = "codex_cli_rs";
+use reqwest::header::HeaderValue;
+use std::sync::LazyLock;
+use std::sync::Mutex;
 
-pub fn get_codex_user_agent(originator: Option<&str>) -> String {
+/// Set this to add a suffix to the User-Agent string.
+///
+/// It is not ideal that we're using a global singleton for this.
+/// This is primarily designed to differentiate MCP clients from each other.
+/// Because there can only be one MCP server per process, it should be safe for this to be a global static.
+/// However, future users of this should use this with caution as a result.
+/// In addition, we want to be confident that this value is used for ALL clients and doing that requires a
+/// lot of wiring and it's easy to miss code paths by doing so.
+/// See https://github.com/openai/codex/pull/3388/files for an example of what that would look like.
+/// Finally, we want to make sure this is set for ALL mcp clients without needing to know a special env var
+/// or having to set data that they already specified in the mcp initialize request somewhere else.
+///
+/// A space is automatically added between the suffix and the rest of the User-Agent string.
+/// The full user agent string is returned from the mcp initialize response.
+/// Parenthesis will be added by Codex. This should only specify what goes inside of the parenthesis.
+pub static USER_AGENT_SUFFIX: LazyLock<Mutex<Option<String>>> = LazyLock::new(|| Mutex::new(None));
+
+pub const CODEX_INTERNAL_ORIGINATOR_OVERRIDE_ENV_VAR: &str = "CODEX_INTERNAL_ORIGINATOR_OVERRIDE";
+
+#[derive(Debug, Clone)]
+pub struct Originator {
+    pub value: String,
+    pub header_value: HeaderValue,
+}
+
+pub static ORIGINATOR: LazyLock<Originator> = LazyLock::new(|| {
+    let default = "codex_cli_rs";
+    let value = std::env::var(CODEX_INTERNAL_ORIGINATOR_OVERRIDE_ENV_VAR)
+        .unwrap_or_else(|_| default.to_string());
+
+    match HeaderValue::from_str(&value) {
+        Ok(header_value) => Originator {
+            value,
+            header_value,
+        },
+        Err(e) => {
+            tracing::error!("Unable to turn originator override {value} into header value: {e}");
+            Originator {
+                value: default.to_string(),
+                header_value: HeaderValue::from_static(default),
+            }
+        }
+    }
+});
+
+pub fn get_codex_user_agent() -> String {
     let build_version = env!("CARGO_PKG_VERSION");
     let os_info = os_info::get();
-    format!(
+    let prefix = format!(
         "{}/{build_version} ({} {}; {}) {}",
-        originator.unwrap_or(DEFAULT_ORIGINATOR),
+        ORIGINATOR.value.as_str(),
         os_info.os_type(),
         os_info.version(),
         os_info.architecture().unwrap_or("unknown"),
         crate::terminal::user_agent()
-    )
+    );
+    let suffix = USER_AGENT_SUFFIX
+        .lock()
+        .ok()
+        .and_then(|guard| guard.clone());
+    let suffix = suffix
+        .as_deref()
+        .map(str::trim)
+        .filter(|value| !value.is_empty())
+        .map_or_else(String::new, |value| format!(" ({value})"));
+
+    let candidate = format!("{prefix}{suffix}");
+    sanitize_user_agent(candidate, &prefix)
+}
+
+/// Sanitize the user agent string.
+///
+/// Invalid characters are replaced with an underscore.
+///
+/// If the user agent fails to parse, it falls back to fallback and then to ORIGINATOR.
+fn sanitize_user_agent(candidate: String, fallback: &str) -> String {
+    if HeaderValue::from_str(candidate.as_str()).is_ok() {
+        return candidate;
+    }
+
+    let sanitized: String = candidate
+        .chars()
+        .map(|ch| if matches!(ch, ' '..='~') { ch } else { '_' })
+        .collect();
+    if !sanitized.is_empty() && HeaderValue::from_str(sanitized.as_str()).is_ok() {
+        tracing::warn!(
+            "Sanitized Codex user agent because provided suffix contained invalid header characters"
+        );
+        sanitized
+    } else if HeaderValue::from_str(fallback).is_ok() {
+        tracing::warn!(
+            "Falling back to base Codex user agent because provided suffix could not be sanitized"
+        );
+        fallback.to_string()
+    } else {
+        tracing::warn!(
+            "Falling back to default Codex originator because base user agent string is invalid"
+        );
+        ORIGINATOR.value.clone()
+    }
 }
 
 /// Create a reqwest client with default `originator` and `User-Agent` headers set.
-pub fn create_client(originator: &str) -> reqwest::Client {
+pub fn create_client() -> reqwest::Client {
     use reqwest::header::HeaderMap;
-    use reqwest::header::HeaderValue;
 
     let mut headers = HeaderMap::new();
-    let originator_value = HeaderValue::from_str(originator)
-        .unwrap_or_else(|_| HeaderValue::from_static(DEFAULT_ORIGINATOR));
-    headers.insert("originator", originator_value);
-    let ua = get_codex_user_agent(Some(originator));
+    headers.insert("originator", ORIGINATOR.header_value.clone());
+    let ua = get_codex_user_agent();
 
-    match reqwest::Client::builder()
+    reqwest::Client::builder()
         // Set UA via dedicated helper to avoid header validation pitfalls
         .user_agent(ua)
         .default_headers(headers)
         .build()
-    {
-        Ok(client) => client,
-        Err(_) => reqwest::Client::new(),
-    }
+        .unwrap_or_else(|_| reqwest::Client::new())
 }
 
 #[cfg(test)]
@@ -41,7 +126,7 @@ mod tests {
 
     #[test]
     fn test_get_codex_user_agent() {
-        let user_agent = get_codex_user_agent(None);
+        let user_agent = get_codex_user_agent();
         assert!(user_agent.starts_with("codex_cli_rs/"));
     }
 
@@ -53,8 +138,7 @@ mod tests {
         use wiremock::matchers::method;
         use wiremock::matchers::path;
 
-        let originator = "test_originator";
-        let client = create_client(originator);
+        let client = create_client();
 
         // Spin up a local mock server and capture a request.
         let server = MockServer::start().await;
@@ -82,21 +166,43 @@ mod tests {
         let originator_header = headers
             .get("originator")
             .expect("originator header missing");
-        assert_eq!(originator_header.to_str().unwrap(), originator);
+        assert_eq!(originator_header.to_str().unwrap(), "codex_cli_rs");
 
         // User-Agent matches the computed Codex UA for that originator
-        let expected_ua = get_codex_user_agent(Some(originator));
+        let expected_ua = get_codex_user_agent();
         let ua_header = headers
             .get("user-agent")
             .expect("user-agent header missing");
         assert_eq!(ua_header.to_str().unwrap(), expected_ua);
     }
 
+    #[test]
+    fn test_invalid_suffix_is_sanitized() {
+        let prefix = "codex_cli_rs/0.0.0";
+        let suffix = "bad\rsuffix";
+
+        assert_eq!(
+            sanitize_user_agent(format!("{prefix} ({suffix})"), prefix),
+            "codex_cli_rs/0.0.0 (bad_suffix)"
+        );
+    }
+
+    #[test]
+    fn test_invalid_suffix_is_sanitized2() {
+        let prefix = "codex_cli_rs/0.0.0";
+        let suffix = "bad\0suffix";
+
+        assert_eq!(
+            sanitize_user_agent(format!("{prefix} ({suffix})"), prefix),
+            "codex_cli_rs/0.0.0 (bad_suffix)"
+        );
+    }
+
     #[test]
     #[cfg(target_os = "macos")]
     fn test_macos() {
         use regex_lite::Regex;
-        let user_agent = get_codex_user_agent(None);
+        let user_agent = get_codex_user_agent();
         let re = Regex::new(
             r"^codex_cli_rs/\d+\.\d+\.\d+ \(Mac OS \d+\.\d+\.\d+; (x86_64|arm64)\) (\S+)$",
         )

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.clone())
+                    }
+                }
+                _ => 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/event_mapping.rs

@@ -25,31 +25,56 @@ pub(crate) fn map_response_item_to_event_messages(
                 return Vec::new();
             }
 
-            let events: Vec<EventMsg> = content
-                .iter()
-                .filter_map(|content_item| match content_item {
-                    ContentItem::OutputText { text } => {
-                        Some(EventMsg::AgentMessage(AgentMessageEvent {
-                            message: text.clone(),
-                        }))
-                    }
+            let mut events: Vec<EventMsg> = Vec::new();
+            let mut message_parts: Vec<String> = Vec::new();
+            let mut images: Vec<String> = Vec::new();
+            let mut kind: Option<InputMessageKind> = None;
+
+            for content_item in content.iter() {
+                match content_item {
                     ContentItem::InputText { text } => {
-                        let trimmed = text.trim_start();
-                        let kind = if trimmed.starts_with("<environment_context>") {
-                            Some(InputMessageKind::EnvironmentContext)
-                        } else if trimmed.starts_with("<user_instructions>") {
-                            Some(InputMessageKind::UserInstructions)
-                        } else {
-                            Some(InputMessageKind::Plain)
-                        };
-                        Some(EventMsg::UserMessage(UserMessageEvent {
-                            message: text.clone(),
-                            kind,
-                        }))
+                        if kind.is_none() {
+                            let trimmed = text.trim_start();
+                            kind = if trimmed.starts_with("<environment_context>") {
+                                Some(InputMessageKind::EnvironmentContext)
+                            } else if trimmed.starts_with("<user_instructions>") {
+                                Some(InputMessageKind::UserInstructions)
+                            } else {
+                                Some(InputMessageKind::Plain)
+                            };
+                        }
+                        message_parts.push(text.clone());
                     }
-                    _ => None,
-                })
-                .collect();
+                    ContentItem::InputImage { image_url } => {
+                        images.push(image_url.clone());
+                    }
+                    ContentItem::OutputText { text } => {
+                        events.push(EventMsg::AgentMessage(AgentMessageEvent {
+                            message: text.clone(),
+                        }));
+                    }
+                }
+            }
+
+            if !message_parts.is_empty() || !images.is_empty() {
+                let message = if message_parts.is_empty() {
+                    String::new()
+                } else {
+                    message_parts.join("")
+                };
+                let images = if images.is_empty() {
+                    None
+                } else {
+                    Some(images)
+                };
+
+                events.push(EventMsg::UserMessage(UserMessageEvent {
+                    message,
+                    kind,
+                    images,
+                }));
+            }
+
             events
         }
 
@@ -96,3 +121,47 @@ pub(crate) fn map_response_item_to_event_messages(
         | ResponseItem::Other => Vec::new(),
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::map_response_item_to_event_messages;
+    use crate::protocol::EventMsg;
+    use crate::protocol::InputMessageKind;
+    use codex_protocol::models::ContentItem;
+    use codex_protocol::models::ResponseItem;
+    use pretty_assertions::assert_eq;
+
+    #[test]
+    fn maps_user_message_with_text_and_two_images() {
+        let img1 = "https://example.com/one.png".to_string();
+        let img2 = "https://example.com/two.jpg".to_string();
+
+        let item = ResponseItem::Message {
+            id: None,
+            role: "user".to_string(),
+            content: vec![
+                ContentItem::InputText {
+                    text: "Hello world".to_string(),
+                },
+                ContentItem::InputImage {
+                    image_url: img1.clone(),
+                },
+                ContentItem::InputImage {
+                    image_url: img2.clone(),
+                },
+            ],
+        };
+
+        let events = map_response_item_to_event_messages(&item, false);
+        assert_eq!(events.len(), 1, "expected a single user message event");
+
+        match &events[0] {
+            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()]));
+            }
+            other => panic!("expected UserMessage, got {other:?}"),
+        }
+    }
+}

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

@@ -24,6 +24,9 @@ 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 {
@@ -34,6 +37,7 @@ impl ExecCommandSession {
         reader_handle: JoinHandle<()>,
         writer_handle: JoinHandle<()>,
         wait_handle: JoinHandle<()>,
+        exit_status: std::sync::Arc<std::sync::atomic::AtomicBool>,
     ) -> Self {
         Self {
             writer_tx,
@@ -42,6 +46,7 @@ impl ExecCommandSession {
             reader_handle: StdMutex::new(Some(reader_handle)),
             writer_handle: StdMutex::new(Some(writer_handle)),
             wait_handle: StdMutex::new(Some(wait_handle)),
+            exit_status,
         }
     }
 
@@ -52,6 +57,10 @@ impl ExecCommandSession {
     pub(crate) fn output_receiver(&self) -> broadcast::Receiver<Vec<u8>> {
         self.output_tx.subscribe()
     }
+
+    pub(crate) fn has_exited(&self) -> bool {
+        self.exit_status.load(std::sync::atomic::Ordering::SeqCst)
+    }
 }
 
 impl Drop for ExecCommandSession {

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)]
@@ -327,11 +329,14 @@ 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 +348,11 @@ async fn create_exec_command_session(
         reader_handle,
         writer_handle,
         wait_handle,
+        exit_status,
     );
     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 +516,4 @@ 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

@@ -3,6 +3,7 @@ use std::path::Path;
 use std::path::PathBuf;
 
 use codex_protocol::mcp_protocol::GitSha;
+use codex_protocol::protocol::GitInfo;
 use futures::future::join_all;
 use serde::Deserialize;
 use serde::Serialize;
@@ -43,19 +44,6 @@ pub fn get_git_repo_root(base_dir: &Path) -> Option<PathBuf> {
 /// Timeout for git commands to prevent freezing on large repositories
 const GIT_COMMAND_TIMEOUT: TokioDuration = TokioDuration::from_secs(5);
 
-#[derive(Serialize, Deserialize, Clone, Debug)]
-pub struct GitInfo {
-    /// Current commit hash (SHA)
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub commit_hash: Option<String>,
-    /// Current branch name
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub branch: Option<String>,
-    /// Repository URL (if available from remote)
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub repository_url: Option<String>,
-}
-
 #[derive(Serialize, Deserialize, Clone, Debug)]
 pub struct GitDiffToRemote {
     pub sha: GitSha,

codex-rs/core/src/lib.rs

@@ -16,6 +16,7 @@ mod codex_conversation;
 pub mod token_data;
 pub use codex_conversation::CodexConversation;
 pub mod config;
+pub mod config_edit;
 pub mod config_profile;
 pub mod config_types;
 mod conversation_history;
@@ -34,6 +35,8 @@ 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;
@@ -42,6 +45,7 @@ pub use model_provider_info::built_in_model_providers;
 pub use model_provider_info::create_oss_provider_with_base_url;
 mod conversation_manager;
 mod event_mapping;
+pub use codex_protocol::protocol::InitialHistory;
 pub use conversation_manager::ConversationManager;
 pub use conversation_manager::NewConversation;
 // Re-export common auth types for workspace consumers
@@ -61,7 +65,9 @@ pub mod spawn;
 pub mod terminal;
 mod tool_apply_patch;
 pub mod turn_diff_tracker;
+pub use rollout::ARCHIVED_SESSIONS_SUBDIR;
 pub use rollout::RolloutRecorder;
+pub use rollout::SESSIONS_SUBDIR;
 pub use rollout::SessionMeta;
 pub use rollout::list::ConversationItem;
 pub use rollout::list::ConversationsPage;

codex-rs/core/src/mcp_connection_manager.rs

@@ -163,6 +163,10 @@ impl McpConnectionManager {
                                 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

@@ -103,7 +103,7 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
             slug, "gpt-4.1",
             needs_special_apply_patch_instructions: true,
         )
-    } else if slug.starts_with("gpt-oss") {
+    } 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")

codex-rs/core/src/openai_tools.rs

@@ -8,7 +8,6 @@ use std::collections::HashMap;
 use crate::model_family::ModelFamily;
 use crate::plan_tool::PLAN_TOOL;
 use crate::protocol::AskForApproval;
-use crate::protocol::SandboxPolicy;
 use crate::tool_apply_patch::ApplyPatchToolType;
 use crate::tool_apply_patch::create_apply_patch_freeform_tool;
 use crate::tool_apply_patch::create_apply_patch_json_tool;
@@ -58,7 +57,7 @@ pub(crate) enum OpenAiTool {
 #[derive(Debug, Clone)]
 pub enum ConfigShellToolType {
     DefaultShell,
-    ShellWithRequest { sandbox_policy: SandboxPolicy },
+    ShellWithRequest,
     LocalShell,
     StreamableShell,
 }
@@ -70,17 +69,18 @@ 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> {
     pub(crate) model_family: &'a ModelFamily,
     pub(crate) approval_policy: AskForApproval,
-    pub(crate) sandbox_policy: SandboxPolicy,
     pub(crate) include_plan_tool: bool,
     pub(crate) include_apply_patch_tool: bool,
     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 {
@@ -88,12 +88,12 @@ impl ToolsConfig {
         let ToolsConfigParams {
             model_family,
             approval_policy,
-            sandbox_policy,
             include_plan_tool,
             include_apply_patch_tool,
             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
@@ -103,9 +103,7 @@ impl ToolsConfig {
             ConfigShellToolType::DefaultShell
         };
         if matches!(approval_policy, AskForApproval::OnRequest) && !use_streamable_shell_tool {
-            shell_type = ConfigShellToolType::ShellWithRequest {
-                sandbox_policy: sandbox_policy.clone(),
-            }
+            shell_type = ConfigShellToolType::ShellWithRequest;
         }
 
         let apply_patch_tool_type = match model_family.apply_patch_tool_type {
@@ -126,6 +124,7 @@ 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,
         }
     }
 }
@@ -200,7 +199,56 @@ fn create_shell_tool() -> OpenAiTool {
     })
 }
 
-fn create_shell_tool_for_sandbox(sandbox_policy: &SandboxPolicy) -> 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),
+        },
+    })
+}
+
+const SHELL_TOOL_DESCRIPTION: &str = r#"Runs a shell command and returns its output"#;
+
+fn create_shell_tool_for_sandbox() -> OpenAiTool {
     let mut properties = BTreeMap::new();
     properties.insert(
         "command".to_string(),
@@ -212,82 +260,29 @@ fn create_shell_tool_for_sandbox(sandbox_policy: &SandboxPolicy) -> OpenAiTool {
     properties.insert(
         "workdir".to_string(),
         JsonSchema::String {
-            description: Some("The working directory to execute the command in".to_string()),
+            description: Some("Working directory to execute the command in.".to_string()),
         },
     );
     properties.insert(
         "timeout_ms".to_string(),
         JsonSchema::Number {
-            description: Some("The timeout for the command in milliseconds".to_string()),
+            description: Some("Timeout for the command in milliseconds.".to_string()),
         },
     );
-
-    if matches!(sandbox_policy, SandboxPolicy::WorkspaceWrite { .. }) {
-        properties.insert(
+    properties.insert(
         "with_escalated_permissions".to_string(),
         JsonSchema::Boolean {
-            description: Some("Whether to request escalated permissions. Set to true if command needs to be run without sandbox restrictions".to_string()),
+            description: Some("Request escalated permissions, only for when a command would otherwise be blocked by the sandbox.".to_string()),
         },
     );
-        properties.insert(
+    properties.insert(
         "justification".to_string(),
         JsonSchema::String {
-            description: Some("Only set if with_escalated_permissions is true. 1-sentence explanation of why we want to run this command.".to_string()),
+            description: Some("Required if and only if with_escalated_permissions == true. One sentence explaining why escalation is needed (e.g., write outside CWD, network fetch, git commit).".to_string()),
         },
     );
-    }
 
-    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()
-        }
-    };
+    let description = SHELL_TOOL_DESCRIPTION.to_string();
 
     OpenAiTool::Function(ResponsesApiTool {
         name: "shell".to_string(),
@@ -300,7 +295,6 @@ The shell tool is used to execute shell commands.
         },
     })
 }
-
 fn create_view_image_tool() -> OpenAiTool {
     // Support only local filesystem path.
     let mut properties = BTreeMap::new();
@@ -534,23 +528,27 @@ 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 => {
+                tools.push(create_shell_tool_for_sandbox());
+            }
+            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 +575,8 @@ 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));
 
@@ -636,18 +632,18 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
-            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: true,
             include_apply_patch_tool: false,
             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"],
         );
     }
 
@@ -657,18 +653,18 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
-            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: true,
             include_apply_patch_tool: false,
             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"],
         );
     }
 
@@ -678,12 +674,12 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
-            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             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 +722,7 @@ mod tests {
         assert_eq_tool_names(
             &tools,
             &[
-                "shell",
+                "unified_exec",
                 "web_search",
                 "view_image",
                 "test_server/do_something_cool",
@@ -783,12 +779,12 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
-            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             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 +837,11 @@ 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",
@@ -860,12 +856,12 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
-            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             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 +889,7 @@ mod tests {
 
         assert_eq_tool_names(
             &tools,
-            &["shell", "web_search", "view_image", "dash/search"],
+            &["unified_exec", "web_search", "view_image", "dash/search"],
         );
 
         assert_eq!(
@@ -922,12 +918,12 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
-            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             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 +949,7 @@ 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],
@@ -979,12 +975,12 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
-            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             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 +1004,10 @@ 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 {
@@ -1036,12 +1035,12 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
-            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             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 +1064,10 @@ 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 {
@@ -1086,13 +1088,7 @@ mod tests {
 
     #[test]
     fn test_shell_tool_for_sandbox_workspace_write() {
-        let sandbox_policy = SandboxPolicy::WorkspaceWrite {
-            writable_roots: vec!["workspace".into()],
-            network_access: false,
-            exclude_tmpdir_env_var: false,
-            exclude_slash_tmp: false,
-        };
-        let tool = super::create_shell_tool_for_sandbox(&sandbox_policy);
+        let tool = super::create_shell_tool_for_sandbox();
         let OpenAiTool::Function(ResponsesApiTool {
             description, name, ..
         }) = &tool
@@ -1101,29 +1097,13 @@ 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 = super::SHELL_TOOL_DESCRIPTION;
         assert_eq!(description, expected);
     }
 
     #[test]
     fn test_shell_tool_for_sandbox_readonly() {
-        let tool = super::create_shell_tool_for_sandbox(&SandboxPolicy::ReadOnly);
+        let tool = super::create_shell_tool_for_sandbox();
         let OpenAiTool::Function(ResponsesApiTool {
             description, name, ..
         }) = &tool
@@ -1132,27 +1112,13 @@ 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 = super::SHELL_TOOL_DESCRIPTION;
         assert_eq!(description, expected);
     }
 
     #[test]
     fn test_shell_tool_for_sandbox_danger_full_access() {
-        let tool = super::create_shell_tool_for_sandbox(&SandboxPolicy::DangerFullAccess);
+        let tool = super::create_shell_tool_for_sandbox();
         let OpenAiTool::Function(ResponsesApiTool {
             description, name, ..
         }) = &tool
@@ -1161,6 +1127,7 @@ The shell tool is used to execute shell commands.
         };
         assert_eq!(name, "shell");
 
-        assert_eq!(description, "Runs a shell command and returns its output.");
+        let expected = super::SHELL_TOOL_DESCRIPTION;
+        assert_eq!(description, expected);
     }
 }

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

@@ -10,6 +10,9 @@ use time::macros::format_description;
 use uuid::Uuid;
 
 use super::SESSIONS_SUBDIR;
+use crate::protocol::EventMsg;
+use codex_protocol::protocol::RolloutItem;
+use codex_protocol::protocol::RolloutLine;
 
 /// Returned page of conversation summaries.
 #[derive(Debug, Default, PartialEq)]
@@ -34,7 +37,7 @@ pub struct ConversationItem {
 }
 
 /// Hard cap to bound worst‑case work per request.
-const MAX_SCAN_FILES: usize = 10_000;
+const MAX_SCAN_FILES: usize = 100;
 const HEAD_RECORD_LIMIT: usize = 10;
 
 /// Pagination cursor identifying a file by timestamp and UUID.
@@ -167,10 +170,16 @@ async fn traverse_directories_for_paths(
                     if items.len() == page_size {
                         break 'outer;
                     }
-                    let head = read_first_jsonl_records(&path, HEAD_RECORD_LIMIT)
-                        .await
-                        .unwrap_or_default();
-                    items.push(ConversationItem { path, head });
+                    // Read head and simultaneously detect message events within the same
+                    // first N JSONL records to avoid a second file read.
+                    let (head, saw_session_meta, saw_user_event) =
+                        read_head_and_flags(&path, HEAD_RECORD_LIMIT)
+                            .await
+                            .unwrap_or((Vec::new(), false, false));
+                    // Apply filters: must have session meta and at least one user message event
+                    if saw_session_meta && saw_user_event {
+                        items.push(ConversationItem { path, head });
+                    }
                 }
             }
         }
@@ -273,16 +282,19 @@ fn parse_timestamp_uuid_from_filename(name: &str) -> Option<(OffsetDateTime, Uui
     Some((ts, uuid))
 }
 
-async fn read_first_jsonl_records(
+async fn read_head_and_flags(
     path: &Path,
     max_records: usize,
-) -> io::Result<Vec<serde_json::Value>> {
+) -> io::Result<(Vec<serde_json::Value>, bool, bool)> {
     use tokio::io::AsyncBufReadExt;
 
     let file = tokio::fs::File::open(path).await?;
     let reader = tokio::io::BufReader::new(file);
     let mut lines = reader.lines();
     let mut head: Vec<serde_json::Value> = Vec::new();
+    let mut saw_session_meta = false;
+    let mut saw_user_event = false;
+
     while head.len() < max_records {
         let line_opt = lines.next_line().await?;
         let Some(line) = line_opt else { break };
@@ -290,9 +302,29 @@ async fn read_first_jsonl_records(
         if trimmed.is_empty() {
             continue;
         }
-        if let Ok(v) = serde_json::from_str::<serde_json::Value>(trimmed) {
-            head.push(v);
+
+        let parsed: Result<RolloutLine, _> = serde_json::from_str(trimmed);
+        let Ok(rollout_line) = parsed else { continue };
+
+        match rollout_line.item {
+            RolloutItem::SessionMeta(session_meta_line) => {
+                if let Ok(val) = serde_json::to_value(session_meta_line) {
+                    head.push(val);
+                    saw_session_meta = true;
+                }
+            }
+            RolloutItem::ResponseItem(item) => {
+                if let Ok(val) = serde_json::to_value(item) {
+                    head.push(val);
+                }
+            }
+            RolloutItem::EventMsg(ev) => {
+                if matches!(ev, EventMsg::UserMessage(_)) {
+                    saw_user_event = true;
+                }
+            }
         }
     }
-    Ok(head)
+
+    Ok((head, saw_session_meta, saw_user_event))
 }

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

@@ -1,15 +1,15 @@
 //! Rollout module: persistence and discovery of session rollout files.
 
-pub(crate) const SESSIONS_SUBDIR: &str = "sessions";
+pub const SESSIONS_SUBDIR: &str = "sessions";
+pub const ARCHIVED_SESSIONS_SUBDIR: &str = "archived_sessions";
 
 pub mod list;
 pub(crate) mod policy;
 pub mod recorder;
 
+pub use codex_protocol::protocol::SessionMeta;
 pub use recorder::RolloutRecorder;
 pub use recorder::RolloutRecorderParams;
-pub use recorder::SessionMeta;
-pub use recorder::SessionStateSnapshot;
 
 #[cfg(test)]
 pub mod tests;

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

@@ -1,8 +1,21 @@
+use crate::protocol::EventMsg;
+use crate::protocol::RolloutItem;
 use codex_protocol::models::ResponseItem;
 
+/// Whether a rollout `item` should be persisted in rollout files.
+#[inline]
+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,
+    }
+}
+
 /// Whether a `ResponseItem` should be persisted in rollout files.
 #[inline]
-pub(crate) fn is_persisted_response_item(item: &ResponseItem) -> bool {
+pub(crate) fn should_persist_response_item(item: &ResponseItem) -> bool {
     match item {
         ResponseItem::Message { .. }
         | ResponseItem::Reasoning { .. }
@@ -14,3 +27,44 @@ pub(crate) fn is_persisted_response_item(item: &ResponseItem) -> bool {
         ResponseItem::WebSearchCall { .. } | ResponseItem::Other => false,
     }
 }
+
+/// Whether an `EventMsg` should be persisted in rollout files.
+#[inline]
+pub(crate) fn should_persist_event_msg(ev: &EventMsg) -> bool {
+    match ev {
+        EventMsg::UserMessage(_)
+        | EventMsg::AgentMessage(_)
+        | EventMsg::AgentReasoning(_)
+        | EventMsg::AgentReasoningRawContent(_)
+        | EventMsg::TokenCount(_) => true,
+        EventMsg::Error(_)
+        | EventMsg::TaskStarted(_)
+        | EventMsg::TaskComplete(_)
+        | EventMsg::AgentMessageDelta(_)
+        | EventMsg::AgentReasoningDelta(_)
+        | EventMsg::AgentReasoningRawContentDelta(_)
+        | EventMsg::AgentReasoningSectionBreak(_)
+        | EventMsg::SessionConfigured(_)
+        | EventMsg::McpToolCallBegin(_)
+        | EventMsg::McpToolCallEnd(_)
+        | EventMsg::WebSearchBegin(_)
+        | EventMsg::WebSearchEnd(_)
+        | EventMsg::ExecCommandBegin(_)
+        | EventMsg::ExecCommandOutputDelta(_)
+        | EventMsg::ExecCommandEnd(_)
+        | EventMsg::ExecApprovalRequest(_)
+        | EventMsg::ApplyPatchApprovalRequest(_)
+        | EventMsg::BackgroundEvent(_)
+        | EventMsg::StreamError(_)
+        | EventMsg::PatchApplyBegin(_)
+        | EventMsg::PatchApplyEnd(_)
+        | EventMsg::TurnDiff(_)
+        | EventMsg::GetHistoryEntryResponse(_)
+        | EventMsg::McpListToolsResponse(_)
+        | EventMsg::ListCustomPromptsResponse(_)
+        | EventMsg::PlanUpdate(_)
+        | EventMsg::TurnAborted(_)
+        | EventMsg::ShutdownComplete
+        | EventMsg::ConversationPath(_) => false,
+    }
+}

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

@@ -26,26 +26,15 @@ use super::list::Cursor;
 use super::list::get_conversations;
 use super::policy::is_persisted_response_item;
 use crate::config::Config;
-use crate::conversation_manager::InitialHistory;
-use crate::conversation_manager::ResumedHistory;
-use crate::git_info::GitInfo;
+use crate::default_client::ORIGINATOR;
 use crate::git_info::collect_git_info;
 use codex_protocol::models::ResponseItem;
-
-#[derive(Serialize, Deserialize, Clone, Default)]
-pub struct SessionMeta {
-    pub id: ConversationId,
-    pub timestamp: String,
-    pub instructions: Option<String>,
-}
-
-#[derive(Serialize)]
-struct SessionMetaWithGit {
-    #[serde(flatten)]
-    meta: SessionMeta,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    git: Option<GitInfo>,
-}
+use codex_protocol::protocol::InitialHistory;
+use codex_protocol::protocol::ResumedHistory;
+use codex_protocol::protocol::RolloutItem;
+use codex_protocol::protocol::RolloutLine;
+use codex_protocol::protocol::SessionMeta;
+use codex_protocol::protocol::SessionMetaLine;
 
 #[derive(Serialize, Deserialize, Default, Clone)]
 pub struct SessionStateSnapshot {}
@@ -72,6 +61,7 @@ pub struct SavedSession {
 #[derive(Clone)]
 pub struct RolloutRecorder {
     tx: Sender<RolloutCmd>,
+    pub(crate) rollout_path: PathBuf,
 }
 
 #[derive(Clone)]
@@ -86,9 +76,14 @@ pub enum RolloutRecorderParams {
 }
 
 enum RolloutCmd {
-    AddItems(Vec<ResponseItem>),
-    UpdateState(SessionStateSnapshot),
-    Shutdown { ack: oneshot::Sender<()> },
+    AddItems(Vec<RolloutItem>),
+    /// Ensure all prior writes are processed; respond when flushed.
+    Flush {
+        ack: oneshot::Sender<()>,
+    },
+    Shutdown {
+        ack: oneshot::Sender<()>,
+    },
 }
 
 impl RolloutRecorderParams {
@@ -105,7 +100,6 @@ impl RolloutRecorderParams {
 }
 
 impl RolloutRecorder {
-    #[allow(dead_code)]
     /// List conversations (rollout files) under the provided Codex home directory.
     pub async fn list_conversations(
         codex_home: &Path,
@@ -119,13 +113,14 @@ impl RolloutRecorder {
     /// cannot be created or the rollout file cannot be opened we return the
     /// error so the caller can decide whether to disable persistence.
     pub async fn new(config: &Config, params: RolloutRecorderParams) -> std::io::Result<Self> {
-        let (file, meta) = match params {
+        let (file, rollout_path, meta) = match params {
             RolloutRecorderParams::Create {
                 conversation_id,
                 instructions,
             } => {
                 let LogFileInfo {
                     file,
+                    path,
                     conversation_id: session_id,
                     timestamp,
                 } = create_log_file(config, conversation_id)?;
@@ -140,9 +135,13 @@ impl RolloutRecorder {
 
                 (
                     tokio::fs::File::from_std(file),
+                    path,
                     Some(SessionMeta {
-                        timestamp,
                         id: session_id,
+                        timestamp,
+                        cwd: config.cwd.clone(),
+                        originator: ORIGINATOR.value.clone(),
+                        cli_version: env!("CARGO_PKG_VERSION").to_string(),
                         instructions,
                     }),
                 )
@@ -150,8 +149,9 @@ impl RolloutRecorder {
             RolloutRecorderParams::Resume { path } => (
                 tokio::fs::OpenOptions::new()
                     .append(true)
-                    .open(path)
+                    .open(&path)
                     .await?,
+                path,
                 None,
             ),
         };
@@ -169,10 +169,10 @@ impl RolloutRecorder {
         // driver instead of blocking the runtime.
         tokio::task::spawn(rollout_writer(file, rx, meta, cwd));
 
-        Ok(Self { tx })
+        Ok(Self { tx, rollout_path })
     }
 
-    pub(crate) async fn record_items(&self, items: &[ResponseItem]) -> std::io::Result<()> {
+    pub(crate) async fn record_items(&self, items: &[RolloutItem]) -> std::io::Result<()> {
         let mut filtered = Vec::new();
         for item in items {
             // Note that function calls may look a bit strange if they are
@@ -191,60 +191,59 @@ impl RolloutRecorder {
             .map_err(|e| IoError::other(format!("failed to queue rollout items: {e}")))
     }
 
-    pub(crate) async fn record_state(&self, state: SessionStateSnapshot) -> std::io::Result<()> {
+    /// 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::UpdateState(state))
+            .send(RolloutCmd::Flush { ack: tx })
             .await
-            .map_err(|e| IoError::other(format!("failed to queue rollout state: {e}")))
+            .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 async fn get_rollout_history(path: &Path) -> std::io::Result<InitialHistory> {
+    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?;
-        let mut lines = text.lines();
-        let first_line = lines
-            .next()
-            .ok_or_else(|| IoError::other("empty session file"))?;
-        let conversation_id = match serde_json::from_str::<SessionMeta>(first_line) {
-            Ok(rollout_session_meta) => {
-                tracing::error!(
-                    "Parsed conversation ID from rollout file: {:?}",
-                    rollout_session_meta.id
-                );
-                Some(rollout_session_meta.id)
-            }
-            Err(e) => {
-                return Err(IoError::other(format!(
-                    "failed to parse first line of rollout file as SessionMeta: {e}"
-                )));
-            }
-        };
+        if text.trim().is_empty() {
+            return Err(IoError::other("empty session file"));
+        }
 
-        let mut items = Vec::new();
-        for line in lines {
+        let mut items: Vec<RolloutItem> = Vec::new();
+        let mut conversation_id: Option<ConversationId> = None;
+        for line in text.lines() {
             if line.trim().is_empty() {
                 continue;
             }
             let v: Value = match serde_json::from_str(line) {
                 Ok(v) => v,
-                Err(_) => continue,
-            };
-            if v.get("record_type")
-                .and_then(|rt| rt.as_str())
-                .map(|s| s == "state")
-                .unwrap_or(false)
-            {
-                continue;
-            }
-            match serde_json::from_value::<ResponseItem>(v.clone()) {
-                Ok(item) => {
-                    if is_persisted_response_item(&item) {
-                        items.push(item);
-                    }
-                }
                 Err(e) => {
-                    warn!("failed to parse item: {v:?}, error: {e}");
+                    warn!("failed to parse line as JSON: {line:?}, error: {e}");
+                    continue;
+                }
+            };
+
+            // Parse the rollout line structure
+            match serde_json::from_value::<RolloutLine>(v.clone()) {
+                Ok(rollout_line) => match rollout_line.item {
+                    RolloutItem::SessionMeta(session_meta_line) => {
+                        // 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::EventMsg(_ev) => {
+                        items.push(RolloutItem::EventMsg(_ev));
+                    }
+                },
+                Err(e) => {
+                    warn!("failed to parse rollout line: {v:?}, error: {e}");
                 }
             }
         }
@@ -269,6 +268,10 @@ 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 {
@@ -289,6 +292,9 @@ struct LogFileInfo {
     /// Opened file handle to the rollout file.
     file: File,
 
+    /// Full path to the rollout file.
+    path: PathBuf,
+
     /// Session ID (also embedded in filename).
     conversation_id: ConversationId,
 
@@ -328,6 +334,7 @@ fn create_log_file(
 
     Ok(LogFileInfo {
         file,
+        path,
         conversation_id,
         timestamp,
     })
@@ -344,13 +351,15 @@ async fn rollout_writer(
     // If we have a meta, collect git info asynchronously and write meta first
     if let Some(session_meta) = meta.take() {
         let git_info = collect_git_info(&cwd).await;
-        let session_meta_with_git = SessionMetaWithGit {
+        let session_meta_line = SessionMetaLine {
             meta: session_meta,
             git: git_info,
         };
 
-        // Write the SessionMeta as the first item in the file
-        writer.write_line(&session_meta_with_git).await?;
+        // Write the SessionMeta as the first item in the file, wrapped in a rollout line
+        writer
+            .write_rollout_item(RolloutItem::SessionMeta(session_meta_line))
+            .await?;
     }
 
     // Process rollout commands
@@ -359,23 +368,17 @@ async fn rollout_writer(
             RolloutCmd::AddItems(items) => {
                 for item in items {
                     if is_persisted_response_item(&item) {
-                        writer.write_line(&item).await?;
+                        writer.write_rollout_item(item).await?;
                     }
                 }
             }
-            RolloutCmd::UpdateState(state) => {
-                #[derive(Serialize)]
-                struct StateLine<'a> {
-                    record_type: &'static str,
-                    #[serde(flatten)]
-                    state: &'a SessionStateSnapshot,
+            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);
                 }
-                writer
-                    .write_line(&StateLine {
-                        record_type: "state",
-                        state: &state,
-                    })
-                    .await?;
+                let _ = ack.send(());
             }
             RolloutCmd::Shutdown { ack } => {
                 let _ = ack.send(());
@@ -391,6 +394,20 @@ struct JsonlWriter {
 }
 
 impl JsonlWriter {
+    async fn write_rollout_item(&mut self, rollout_item: RolloutItem) -> std::io::Result<()> {
+        let timestamp_format: &[FormatItem] = format_description!(
+            "[year]-[month]-[day]T[hour]:[minute]:[second].[subsecond digits:3]Z"
+        );
+        let timestamp = OffsetDateTime::now_utc()
+            .format(timestamp_format)
+            .map_err(|e| IoError::other(format!("failed to format timestamp: {e}")))?;
+
+        let line = RolloutLine {
+            timestamp,
+            item: rollout_item,
+        };
+        self.write_line(&line).await
+    }
     async fn write_line(&mut self, item: &impl serde::Serialize) -> std::io::Result<()> {
         let mut json = serde_json::to_string(item)?;
         json.push('\n');

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

@@ -42,10 +42,30 @@ fn write_session_file(
 
     let meta = serde_json::json!({
         "timestamp": ts_str,
-        "id": uuid.to_string()
+        "type": "session_meta",
+        "payload": {
+            "id": uuid,
+            "timestamp": ts_str,
+            "instructions": null,
+            "cwd": ".",
+            "originator": "test_originator",
+            "cli_version": "test_version"
+        }
     });
     writeln!(file, "{meta}")?;
 
+    // Include at least one user message event to satisfy listing filters
+    let user_event = serde_json::json!({
+        "timestamp": ts_str,
+        "type": "event_msg",
+        "payload": {
+            "type": "user_message",
+            "message": "Hello from user",
+            "kind": "plain"
+        }
+    });
+    writeln!(file, "{user_event}")?;
+
     for i in 0..num_records {
         let rec = serde_json::json!({
             "record_type": "response",
@@ -93,24 +113,30 @@ async fn test_list_conversations_latest_first() {
         .join("01")
         .join(format!("rollout-2025-01-01T12-00-00-{u1}.jsonl"));
 
-    let head_3 = vec![
-        serde_json::json!({"timestamp": "2025-01-03T12-00-00", "id": u3.to_string()}),
-        serde_json::json!({"record_type": "response", "index": 0}),
-        serde_json::json!({"record_type": "response", "index": 1}),
-        serde_json::json!({"record_type": "response", "index": 2}),
-    ];
-    let head_2 = vec![
-        serde_json::json!({"timestamp": "2025-01-02T12-00-00", "id": u2.to_string()}),
-        serde_json::json!({"record_type": "response", "index": 0}),
-        serde_json::json!({"record_type": "response", "index": 1}),
-        serde_json::json!({"record_type": "response", "index": 2}),
-    ];
-    let head_1 = vec![
-        serde_json::json!({"timestamp": "2025-01-01T12-00-00", "id": u1.to_string()}),
-        serde_json::json!({"record_type": "response", "index": 0}),
-        serde_json::json!({"record_type": "response", "index": 1}),
-        serde_json::json!({"record_type": "response", "index": 2}),
-    ];
+    let head_3 = vec![serde_json::json!({
+        "id": u3,
+        "timestamp": "2025-01-03T12-00-00",
+        "instructions": null,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version"
+    })];
+    let head_2 = vec![serde_json::json!({
+        "id": u2,
+        "timestamp": "2025-01-02T12-00-00",
+        "instructions": null,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version"
+    })];
+    let head_1 = vec![serde_json::json!({
+        "id": u1,
+        "timestamp": "2025-01-01T12-00-00",
+        "instructions": null,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version"
+    })];
 
     let expected_cursor: Cursor =
         serde_json::from_str(&format!("\"2025-01-01T12-00-00|{u1}\"")).unwrap();
@@ -170,14 +196,22 @@ async fn test_pagination_cursor() {
         .join("03")
         .join("04")
         .join(format!("rollout-2025-03-04T09-00-00-{u4}.jsonl"));
-    let head_5 = vec![
-        serde_json::json!({"timestamp": "2025-03-05T09-00-00", "id": u5.to_string()}),
-        serde_json::json!({"record_type": "response", "index": 0}),
-    ];
-    let head_4 = vec![
-        serde_json::json!({"timestamp": "2025-03-04T09-00-00", "id": u4.to_string()}),
-        serde_json::json!({"record_type": "response", "index": 0}),
-    ];
+    let head_5 = vec![serde_json::json!({
+        "id": u5,
+        "timestamp": "2025-03-05T09-00-00",
+        "instructions": null,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version"
+    })];
+    let head_4 = vec![serde_json::json!({
+        "id": u4,
+        "timestamp": "2025-03-04T09-00-00",
+        "instructions": null,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version"
+    })];
     let expected_cursor1: Cursor =
         serde_json::from_str(&format!("\"2025-03-04T09-00-00|{u4}\"")).unwrap();
     let expected_page1 = ConversationsPage {
@@ -212,14 +246,22 @@ async fn test_pagination_cursor() {
         .join("03")
         .join("02")
         .join(format!("rollout-2025-03-02T09-00-00-{u2}.jsonl"));
-    let head_3 = vec![
-        serde_json::json!({"timestamp": "2025-03-03T09-00-00", "id": u3.to_string()}),
-        serde_json::json!({"record_type": "response", "index": 0}),
-    ];
-    let head_2 = vec![
-        serde_json::json!({"timestamp": "2025-03-02T09-00-00", "id": u2.to_string()}),
-        serde_json::json!({"record_type": "response", "index": 0}),
-    ];
+    let head_3 = vec![serde_json::json!({
+        "id": u3,
+        "timestamp": "2025-03-03T09-00-00",
+        "instructions": null,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version"
+    })];
+    let head_2 = vec![serde_json::json!({
+        "id": u2,
+        "timestamp": "2025-03-02T09-00-00",
+        "instructions": null,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version"
+    })];
     let expected_cursor2: Cursor =
         serde_json::from_str(&format!("\"2025-03-02T09-00-00|{u2}\"")).unwrap();
     let expected_page2 = ConversationsPage {
@@ -248,10 +290,14 @@ async fn test_pagination_cursor() {
         .join("03")
         .join("01")
         .join(format!("rollout-2025-03-01T09-00-00-{u1}.jsonl"));
-    let head_1 = vec![
-        serde_json::json!({"timestamp": "2025-03-01T09-00-00", "id": u1.to_string()}),
-        serde_json::json!({"record_type": "response", "index": 0}),
-    ];
+    let head_1 = vec![serde_json::json!({
+        "id": u1,
+        "timestamp": "2025-03-01T09-00-00",
+        "instructions": null,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version"
+    })];
     let expected_cursor3: Cursor =
         serde_json::from_str(&format!("\"2025-03-01T09-00-00|{u1}\"")).unwrap();
     let expected_page3 = ConversationsPage {
@@ -287,11 +333,14 @@ async fn test_get_conversation_contents() {
         .join("04")
         .join("01")
         .join(format!("rollout-2025-04-01T10-30-00-{uuid}.jsonl"));
-    let expected_head = vec![
-        serde_json::json!({"timestamp": ts, "id": uuid.to_string()}),
-        serde_json::json!({"record_type": "response", "index": 0}),
-        serde_json::json!({"record_type": "response", "index": 1}),
-    ];
+    let expected_head = vec![serde_json::json!({
+        "id": uuid,
+        "timestamp": ts,
+        "instructions": null,
+        "cwd": ".",
+        "originator": "test_originator",
+        "cli_version": "test_version"
+    })];
     let expected_cursor: Cursor = serde_json::from_str(&format!("\"{ts}|{uuid}\"")).unwrap();
     let expected_page = ConversationsPage {
         items: vec![ConversationItem {
@@ -305,10 +354,15 @@ async fn test_get_conversation_contents() {
     assert_eq!(page, expected_page);
 
     // Entire file contents equality
-    let meta = serde_json::json!({"timestamp": ts, "id": uuid.to_string()});
+    let meta = serde_json::json!({"timestamp": ts, "type": "session_meta", "payload": {"id": uuid, "timestamp": ts, "instructions": null, "cwd": ".", "originator": "test_originator", "cli_version": "test_version"}});
+    let user_event = serde_json::json!({
+        "timestamp": ts,
+        "type": "event_msg",
+        "payload": {"type": "user_message", "message": "Hello from user", "kind": "plain"}
+    });
     let rec0 = serde_json::json!({"record_type": "response", "index": 0});
     let rec1 = serde_json::json!({"record_type": "response", "index": 1});
-    let expected_content = format!("{meta}\n{rec0}\n{rec1}\n");
+    let expected_content = format!("{meta}\n{user_event}\n{rec0}\n{rec1}\n");
     assert_eq!(content, expected_content);
 }
 
@@ -341,7 +395,14 @@ async fn test_stable_ordering_same_second_pagination() {
         .join("01")
         .join(format!("rollout-2025-07-01T00-00-00-{u2}.jsonl"));
     let head = |u: Uuid| -> Vec<serde_json::Value> {
-        vec![serde_json::json!({"timestamp": ts, "id": u.to_string()})]
+        vec![serde_json::json!({
+            "id": u,
+            "timestamp": ts,
+            "instructions": null,
+            "cwd": ".",
+            "originator": "test_originator",
+            "cli_version": "test_version"
+        })]
     };
     let expected_cursor1: Cursor = serde_json::from_str(&format!("\"{ts}|{u2}\"")).unwrap();
     let expected_page1 = ConversationsPage {

codex-rs/core/src/seatbelt_base_policy.sbpl

@@ -69,3 +69,8 @@
 ; 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")
+)

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/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 +1,653 @@
+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 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);
+    });
+
+    Ok(ExecCommandSession::new(
+        writer_tx,
+        output_tx,
+        killer,
+        reader_handle,
+        writer_handle,
+        wait_handle,
+        exit_status,
+    ))
+}
+
+#[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(1_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 $CODEX_INTERACTIVE_SHELL_VAR\n".to_string()],
+                timeout_ms: Some(1_500),
+            })
+            .await?;
+        assert!(out_2.output.contains("codex"));
+
+        Ok(())
+    }
+
+    #[cfg(unix)]
+    #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+    async fn multi_unified_exec_sessions() -> Result<(), UnifiedExecError> {
+        let manager = UnifiedExecSessionManager::default();
+
+        let shell_a = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: None,
+                input_chunks: &["/bin/bash".to_string(), "-i".to_string()],
+                timeout_ms: Some(1_500),
+            })
+            .await?;
+        let session_a = shell_a.session_id.expect("expected session id");
+
+        manager
+            .handle_request(UnifiedExecRequest {
+                session_id: Some(session_a),
+                input_chunks: &["export CODEX_INTERACTIVE_SHELL_VAR=codex\n".to_string()],
+                timeout_ms: Some(1_500),
+            })
+            .await?;
+
+        let out_2 = manager
+            .handle_request(UnifiedExecRequest {
+                session_id: None,
+                input_chunks: &[
+                    "echo".to_string(),
+                    "$CODEX_INTERACTIVE_SHELL_VAR\n".to_string(),
+                ],
+                timeout_ms: Some(1_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(1_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(1_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(1_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(1_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]
+    async fn correct_path_resolution() -> 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(1_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(1_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(1_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/tests/suite/cli_stream.rs

@@ -1,4 +1,6 @@
 use assert_cmd::Command as AssertCommand;
+use codex_core::RolloutRecorder;
+use codex_core::protocol::GitInfo;
 use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use std::time::Duration;
 use std::time::Instant;
@@ -77,6 +79,22 @@ async fn chat_mode_stream_cli() {
     assert_eq!(hi_lines, 1, "Expected exactly one line with 'hi'");
 
     server.verify().await;
+
+    // Verify a new session rollout was created and is discoverable via list_conversations
+    let page = RolloutRecorder::list_conversations(home.path(), 10, None)
+        .await
+        .expect("list conversations");
+    assert!(
+        !page.items.is_empty(),
+        "expected at least one session to be listed"
+    );
+    // First line of head must be the SessionMeta payload (id/timestamp)
+    let head0 = page.items[0].head.first().expect("missing head record");
+    assert!(head0.get("id").is_some(), "head[0] missing id");
+    assert!(
+        head0.get("timestamp").is_some(),
+        "head[0] missing timestamp"
+    );
 }
 
 /// Verify that passing `-c experimental_instructions_file=...` to the CLI
@@ -297,8 +315,10 @@ async fn integration_creates_and_checks_session_file() {
                     Ok(v) => v,
                     Err(_) => continue,
                 };
-                if item.get("type").and_then(|t| t.as_str()) == Some("message")
-                    && let Some(c) = item.get("content")
+                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")
+                    && let Some(c) = payload.get("content")
                     && c.to_string().contains(&marker)
                 {
                     matching_path = Some(path.to_path_buf());
@@ -361,9 +381,16 @@ async fn integration_creates_and_checks_session_file() {
         .unwrap_or_else(|_| panic!("missing session meta line"));
     let meta: serde_json::Value = serde_json::from_str(meta_line)
         .unwrap_or_else(|_| panic!("Failed to parse session meta line as JSON"));
-    assert!(meta.get("id").is_some(), "SessionMeta missing id");
+    assert_eq!(
+        meta.get("type").and_then(|v| v.as_str()),
+        Some("session_meta")
+    );
+    let payload = meta
+        .get("payload")
+        .unwrap_or_else(|| panic!("Missing payload in meta line"));
+    assert!(payload.get("id").is_some(), "SessionMeta missing id");
     assert!(
-        meta.get("timestamp").is_some(),
+        payload.get("timestamp").is_some(),
         "SessionMeta missing timestamp"
     );
 
@@ -375,8 +402,10 @@ async fn integration_creates_and_checks_session_file() {
         let Ok(item) = serde_json::from_str::<serde_json::Value>(line) else {
             continue;
         };
-        if item.get("type").and_then(|t| t.as_str()) == Some("message")
-            && let Some(c) = item.get("content")
+        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")
+            && let Some(c) = payload.get("content")
             && c.to_string().contains(&marker)
         {
             found_message = true;
@@ -589,7 +618,7 @@ async fn integration_git_info_unit_test() {
 
     // 5. Test serialization to ensure it works in SessionMeta
     let serialized = serde_json::to_string(&git_info).unwrap();
-    let deserialized: codex_core::git_info::GitInfo = serde_json::from_str(&serialized).unwrap();
+    let deserialized: GitInfo = serde_json::from_str(&serialized).unwrap();
 
     assert_eq!(git_info.commit_hash, deserialized.commit_hash);
     assert_eq!(git_info.branch, deserialized.branch);

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

@@ -123,10 +123,22 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
     let tmpdir = TempDir::new().unwrap();
     let session_path = tmpdir.path().join("resume-session.jsonl");
     let mut f = std::fs::File::create(&session_path).unwrap();
+    let convo_id = Uuid::new_v4();
     writeln!(
         f,
         "{}",
-        json!({"meta":"test","instructions":"be nice", "id": Uuid::new_v4(), "timestamp": "2024-01-01T00:00:00Z"})
+        json!({
+            "timestamp": "2024-01-01T00:00:00.000Z",
+            "type": "session_meta",
+            "payload": {
+                "id": convo_id,
+                "timestamp": "2024-01-01T00:00:00Z",
+                "instructions": "be nice",
+                "cwd": ".",
+                "originator": "test_originator",
+                "cli_version": "test_version"
+            }
+        })
     )
     .unwrap();
 
@@ -138,7 +150,17 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
             text: "resumed user message".to_string(),
         }],
     };
-    writeln!(f, "{}", serde_json::to_string(&prior_user).unwrap()).unwrap();
+    let prior_user_json = serde_json::to_value(&prior_user).unwrap();
+    writeln!(
+        f,
+        "{}",
+        json!({
+            "timestamp": "2024-01-01T00:00:01.000Z",
+            "type": "response_item",
+            "payload": prior_user_json
+        })
+    )
+    .unwrap();
 
     // Prior item: system message (excluded from API history)
     let prior_system = codex_protocol::models::ResponseItem::Message {
@@ -148,7 +170,17 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
             text: "resumed system instruction".to_string(),
         }],
     };
-    writeln!(f, "{}", serde_json::to_string(&prior_system).unwrap()).unwrap();
+    let prior_system_json = serde_json::to_value(&prior_system).unwrap();
+    writeln!(
+        f,
+        "{}",
+        json!({
+            "timestamp": "2024-01-01T00:00:02.000Z",
+            "type": "response_item",
+            "payload": prior_system_json
+        })
+    )
+    .unwrap();
 
     // Prior item: assistant message
     let prior_item = codex_protocol::models::ResponseItem::Message {
@@ -158,7 +190,17 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
             text: "resumed assistant message".to_string(),
         }],
     };
-    writeln!(f, "{}", serde_json::to_string(&prior_item).unwrap()).unwrap();
+    let prior_item_json = serde_json::to_value(&prior_item).unwrap();
+    writeln!(
+        f,
+        "{}",
+        json!({
+            "timestamp": "2024-01-01T00:00:03.000Z",
+            "type": "response_item",
+            "payload": prior_item_json
+        })
+    )
+    .unwrap();
     drop(f);
 
     // Mock server that will receive the resumed request
@@ -196,16 +238,13 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
         .await
         .expect("create new conversation");
 
-    // 1) Assert initial_messages contains the prior user + assistant messages as EventMsg entries
+    // 1) Assert initial_messages only includes existing EventMsg entries; response items are not converted
     let initial_msgs = session_configured
         .initial_messages
         .clone()
-        .expect("expected initial messages for resumed session");
+        .expect("expected initial messages option for resumed session");
     let initial_json = serde_json::to_value(&initial_msgs).unwrap();
-    let expected_initial_json = json!([
-        { "type": "user_message", "message": "resumed user message", "kind": "plain" },
-        { "type": "agent_message", "message": "resumed assistant message" }
-    ]);
+    let expected_initial_json = json!([]);
     assert_eq!(initial_json, expected_initial_json);
 
     // 2) Submit new input; the request body must include the prior item followed by the new user input.
@@ -238,7 +277,25 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
             "content": [{ "type": "input_text", "text": "hello" }]
         }
     ]);
-    assert_eq!(request_body["input"], expected_input);
+    let input_array = request_body
+        .get("input")
+        .and_then(|v| v.as_array())
+        .cloned()
+        .expect("input array in request body");
+    let filtered: Vec<serde_json::Value> = input_array
+        .into_iter()
+        .filter(|item| {
+            let text = item
+                .get("content")
+                .and_then(|c| c.as_array())
+                .and_then(|a| a.first())
+                .and_then(|o| o.get("text"))
+                .and_then(|t| t.as_str())
+                .unwrap_or("");
+            !text.contains("<environment_context>")
+        })
+        .collect();
+    assert_eq!(serde_json::json!(filtered), expected_input);
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
@@ -371,56 +428,6 @@ async fn includes_base_instructions_override_in_request() {
     );
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn originator_config_override_is_used() {
-    // 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");
-
-    Mock::given(method("POST"))
-        .and(path("/v1/responses"))
-        .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()
-    };
-
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home);
-    config.model_provider = model_provider;
-    config.responses_originator_header = "my_override".to_owned();
-
-    let conversation_manager =
-        ConversationManager::with_auth(CodexAuth::from_api_key("Test API Key"));
-    let codex = conversation_manager
-        .new_conversation(config)
-        .await
-        .expect("create new conversation")
-        .conversation;
-
-    codex
-        .submit(Op::UserInput {
-            items: vec![InputItem::Text {
-                text: "hello".into(),
-            }],
-        })
-        .await
-        .unwrap();
-
-    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
-
-    let request = &server.received_requests().await.unwrap()[0];
-    let request_originator = request.headers.get("originator").unwrap();
-    assert_eq!(request_originator.to_str().unwrap(), "my_override");
-}
-
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn chatgpt_auth_sends_correct_request() {
     if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
@@ -546,15 +553,12 @@ async fn prefers_chatgpt_token_when_config_prefers_chatgpt() {
     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,
-        &config.responses_originator_header,
-    ) {
-        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(), 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,
@@ -622,15 +626,12 @@ async fn prefers_apikey_when_config_prefers_apikey_even_with_chatgpt_tokens() {
     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,
-        &config.responses_originator_header,
-    ) {
-        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(), 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,
@@ -967,34 +968,6 @@ async fn history_dedupes_streamed_and_final_messages_across_turns() {
     assert_eq!(requests.len(), 3, "expected 3 requests (one per turn)");
 
     // Replace full-array compare with tail-only raw JSON compare using a single hard-coded value.
-    let r3_tail_expected = json!([
-        {
-            "type": "message",
-            "role": "user",
-            "content": [{"type":"input_text","text":"U1"}]
-        },
-        {
-            "type": "message",
-            "role": "assistant",
-            "content": [{"type":"output_text","text":"Hey there!\n"}]
-        },
-        {
-            "type": "message",
-            "role": "user",
-            "content": [{"type":"input_text","text":"U2"}]
-        },
-        {
-            "type": "message",
-            "role": "assistant",
-            "content": [{"type":"output_text","text":"Hey there!\n"}]
-        },
-        {
-            "type": "message",
-            "role": "user",
-            "content": [{"type":"input_text","text":"U3"}]
-        }
-    ]);
-
     let r3_input_array = requests[2]
         .body_json::<serde_json::Value>()
         .unwrap()
@@ -1002,12 +975,60 @@ async fn history_dedupes_streamed_and_final_messages_across_turns() {
         .and_then(|v| v.as_array())
         .cloned()
         .expect("r3 missing input array");
-    // skipping earlier context and developer messages
-    let tail_len = r3_tail_expected.as_array().unwrap().len();
-    let actual_tail = &r3_input_array[r3_input_array.len() - tail_len..];
+    // We only assert on the last 5 items of the input history for request 3.
+    // With per-turn environment context injected, the last 5 should be:
+    // [env_ctx, U2, assistant("Hey there!\n"), env_ctx, U3]
+    let actual_tail = &r3_input_array[r3_input_array.len() - 5..];
+
+    // env_ctx 1
+    assert_eq!(actual_tail[0]["type"], serde_json::json!("message"));
+    assert_eq!(actual_tail[0]["role"], serde_json::json!("user"));
+    let env_text_1 = &actual_tail[0]["content"][0]["text"];
+    assert!(
+        env_text_1
+            .as_str()
+            .expect("env text should be string")
+            .contains("<environment_context>")
+    );
+
+    // U2
     assert_eq!(
-        serde_json::Value::Array(actual_tail.to_vec()),
-        r3_tail_expected,
-        "request 3 tail mismatch",
+        actual_tail[1],
+        serde_json::json!({
+            "type": "message",
+            "role": "user",
+            "content": [ { "type": "input_text", "text": "U2" } ]
+        })
+    );
+
+    // assistant response
+    assert_eq!(
+        actual_tail[2],
+        serde_json::json!({
+            "type": "message",
+            "role": "assistant",
+            "content": [ { "type": "output_text", "text": "Hey there!\n" } ]
+        })
+    );
+
+    // env_ctx 2
+    assert_eq!(actual_tail[3]["type"], serde_json::json!("message"));
+    assert_eq!(actual_tail[3]["role"], serde_json::json!("user"));
+    let env_text_2 = &actual_tail[3]["content"][0]["text"];
+    assert!(
+        env_text_2
+            .as_str()
+            .expect("env text should be string")
+            .contains("<environment_context>")
+    );
+
+    // U3
+    assert_eq!(
+        actual_tail[4],
+        serde_json::json!({
+            "type": "message",
+            "role": "user",
+            "content": [ { "type": "input_text", "text": "U3" } ]
+        })
     );
 }

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 +75,121 @@ 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 at nth-from-last user input.
+    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 dropping last user input (n=1), cut strictly before that input if present, else empty.
+    let cut1 = user_inputs
+        .get(user_inputs.len().saturating_sub(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"),
     };
 
+    // 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=1 → 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(1, 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/prompt_caching.rs

@@ -191,7 +191,8 @@ async fn prompt_tools_are_consistent_across_requests() {
     let expected_instructions: &str = include_str!("../../prompt.md");
     // our internal implementation is responsible for keeping tools in sync
     // with the OpenAI schema, so we just verify the tool presence here
-    let expected_tools_names: &[&str] = &["shell", "update_plan", "apply_patch", "view_image"];
+    let expected_tools_names: &[&str] =
+        &["unified_exec", "update_plan", "apply_patch", "view_image"];
     let body0 = requests[0].body_json::<serde_json::Value>().unwrap();
     assert_eq!(
         body0["instructions"],
@@ -271,7 +272,7 @@ async fn prefixes_context_and_instructions_once_and_consistently_across_requests
 
     let shell = default_user_shell().await;
 
-    let expected_env_text = format!(
+    let expected_env_text_init = format!(
         r#"<environment_context>
   <cwd>{}</cwd>
   <approval_policy>on-request</approval_policy>
@@ -284,13 +285,28 @@ async fn prefixes_context_and_instructions_once_and_consistently_across_requests
             None => String::new(),
         }
     );
+    // Per-turn environment context omits the shell tag.
+    let expected_env_text_turn = format!(
+        r#"<environment_context>
+  <cwd>{}</cwd>
+  <approval_policy>on-request</approval_policy>
+  <sandbox_mode>read-only</sandbox_mode>
+  <network_access>restricted</network_access>
+</environment_context>"#,
+        cwd.path().to_string_lossy(),
+    );
     let expected_ui_text =
         "<user_instructions>\n\nbe consistent and helpful\n\n</user_instructions>";
 
-    let expected_env_msg = serde_json::json!({
+    let expected_env_msg_init = serde_json::json!({
         "type": "message",
         "role": "user",
-        "content": [ { "type": "input_text", "text": expected_env_text } ]
+        "content": [ { "type": "input_text", "text": expected_env_text_init } ]
+    });
+    let expected_env_msg_turn = serde_json::json!({
+        "type": "message",
+        "role": "user",
+        "content": [ { "type": "input_text", "text": expected_env_text_turn } ]
     });
     let expected_ui_msg = serde_json::json!({
         "type": "message",
@@ -306,7 +322,12 @@ async fn prefixes_context_and_instructions_once_and_consistently_across_requests
     let body1 = requests[0].body_json::<serde_json::Value>().unwrap();
     assert_eq!(
         body1["input"],
-        serde_json::json!([expected_ui_msg, expected_env_msg, expected_user_message_1])
+        serde_json::json!([
+            expected_ui_msg,
+            expected_env_msg_init,
+            expected_env_msg_turn,
+            expected_user_message_1
+        ])
     );
 
     let expected_user_message_2 = serde_json::json!({
@@ -318,7 +339,7 @@ async fn prefixes_context_and_instructions_once_and_consistently_across_requests
     let expected_body2 = serde_json::json!(
         [
             body1["input"].as_array().unwrap().as_slice(),
-            [expected_user_message_2].as_slice(),
+            [expected_env_msg_turn, expected_user_message_2].as_slice(),
         ]
         .concat()
     );
@@ -426,11 +447,17 @@ 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",
@@ -540,10 +567,24 @@ async fn per_turn_overrides_keep_cached_prefix_and_key_constant() {
         "role": "user",
         "content": [ { "type": "input_text", "text": "hello 2" } ]
     });
+    let expected_env_text_2 = format!(
+        r#"<environment_context>
+  <cwd>{}</cwd>
+  <approval_policy>never</approval_policy>
+  <sandbox_mode>workspace-write</sandbox_mode>
+  <network_access>enabled</network_access>
+</environment_context>"#,
+        new_cwd.path().to_string_lossy()
+    );
+    let expected_env_msg_2 = serde_json::json!({
+        "type": "message",
+        "role": "user",
+        "content": [ { "type": "input_text", "text": expected_env_text_2 } ]
+    });
     let expected_body2 = serde_json::json!(
         [
             body1["input"].as_array().unwrap().as_slice(),
-            [expected_user_message_2].as_slice(),
+            [expected_env_msg_2, expected_user_message_2].as_slice(),
         ]
         .concat()
     );

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

@@ -159,6 +159,41 @@ async fn read_only_forbids_all_writes() {
         .await;
 }
 
+/// Verify that user lookups via `pwd.getpwuid(os.getuid())` work under the
+/// seatbelt sandbox. Prior to allowing the necessary mach‑lookup for
+/// OpenDirectory libinfo, this would fail with `KeyError: getpwuid(): uid not found`.
+#[tokio::test]
+async fn python_getpwuid_works_under_seatbelt() {
+    if std::env::var(CODEX_SANDBOX_ENV_VAR) == Ok("seatbelt".to_string()) {
+        eprintln!("{CODEX_SANDBOX_ENV_VAR} is set to 'seatbelt', skipping test.");
+        return;
+    }
+
+    // ReadOnly is sufficient here since we are only exercising user lookup.
+    let policy = SandboxPolicy::ReadOnly;
+
+    let mut child = spawn_command_under_seatbelt(
+        vec![
+            "python3".to_string(),
+            "-c".to_string(),
+            // Print the passwd struct; success implies lookup worked.
+            "import pwd, os; print(pwd.getpwuid(os.getuid()))".to_string(),
+        ],
+        &policy,
+        std::env::current_dir().expect("should be able to get current dir"),
+        StdioPolicy::RedirectForShellTool,
+        HashMap::new(),
+    )
+    .await
+    .expect("should be able to spawn python under seatbelt");
+
+    let status = child
+        .wait()
+        .await
+        .expect("should be able to wait for child process");
+    assert!(status.success(), "python exited with {status:?}");
+}
+
 #[expect(clippy::expect_used)]
 fn create_test_scenario(tmp: &TempDir) -> TestScenario {
     let repo_parent = tmp.path().to_path_buf();

codex-rs/exec/Cargo.toml

@@ -25,7 +25,6 @@ codex-common = { path = "../common", features = [
     "sandbox_summary",
 ] }
 codex-core = { path = "../core" }
-codex-login = { path = "../login" }
 codex-ollama = { path = "../ollama" }
 codex-protocol = { path = "../protocol" }
 owo-colors = "4.2.0"

codex-rs/exec/src/event_processor_with_human_output.rs

@@ -523,6 +523,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                     history_log_id: _,
                     history_entry_count: _,
                     initial_messages: _,
+                    rollout_path: _,
                 } = session_configured_event;
 
                 ts_println!(
@@ -558,7 +559,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                 }
             },
             EventMsg::ShutdownComplete => return CodexStatus::Shutdown,
-            EventMsg::ConversationHistory(_) => {}
+            EventMsg::ConversationPath(_) => {}
             EventMsg::UserMessage(_) => {}
         }
         CodexStatus::Running

codex-rs/exec/src/lib.rs

@@ -190,7 +190,6 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
     let conversation_manager = ConversationManager::new(AuthManager::shared(
         config.codex_home.clone(),
         config.preferred_auth_method,
-        config.responses_originator_header.clone(),
     ));
     let NewConversation {
         conversation_id: _,

codex-rs/linux-sandbox/Cargo.toml

@@ -15,9 +15,7 @@ path = "src/lib.rs"
 workspace = true
 
 [target.'cfg(target_os = "linux")'.dependencies]
-anyhow = "1"
 clap = { version = "4", features = ["derive"] }
-codex-common = { path = "../common", features = ["cli"] }
 codex-core = { path = "../core" }
 landlock = "0.4.1"
 libc = "0.2.175"

codex-rs/login/Cargo.toml

@@ -17,7 +17,6 @@ serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 sha2 = "0.10"
 tempfile = "3"
-thiserror = "2.0.16"
 tiny_http = "0.12"
 tokio = { version = "1", features = [
     "io-std",
@@ -31,5 +30,4 @@ urlencoding = "2.1"
 webbrowser = "1.0"
 
 [dev-dependencies]
-pretty_assertions = "1.4.1"
 tempfile = "3"

codex-rs/login/src/server.rs

@@ -16,6 +16,7 @@ use base64::Engine;
 use chrono::Utc;
 use codex_core::auth::AuthDotJson;
 use codex_core::auth::get_auth_file;
+use codex_core::default_client::ORIGINATOR;
 use codex_core::token_data::TokenData;
 use codex_core::token_data::parse_id_token;
 use rand::RngCore;
@@ -35,11 +36,10 @@ pub struct ServerOptions {
     pub port: u16,
     pub open_browser: bool,
     pub force_state: Option<String>,
-    pub originator: String,
 }
 
 impl ServerOptions {
-    pub fn new(codex_home: PathBuf, client_id: String, originator: String) -> Self {
+    pub fn new(codex_home: PathBuf, client_id: String) -> Self {
         Self {
             codex_home,
             client_id: client_id.to_string(),
@@ -47,7 +47,6 @@ impl ServerOptions {
             port: DEFAULT_PORT,
             open_browser: true,
             force_state: None,
-            originator,
         }
     }
 }
@@ -103,14 +102,7 @@ pub fn run_login_server(opts: ServerOptions) -> io::Result<LoginServer> {
     let server = Arc::new(server);
 
     let redirect_uri = format!("http://localhost:{actual_port}/auth/callback");
-    let auth_url = build_authorize_url(
-        &opts.issuer,
-        &opts.client_id,
-        &redirect_uri,
-        &pkce,
-        &state,
-        &opts.originator,
-    );
+    let auth_url = build_authorize_url(&opts.issuer, &opts.client_id, &redirect_uri, &pkce, &state);
 
     if opts.open_browser {
         let _ = webbrowser::open(&auth_url);
@@ -311,7 +303,6 @@ fn build_authorize_url(
     redirect_uri: &str,
     pkce: &PkceCodes,
     state: &str,
-    originator: &str,
 ) -> String {
     let query = vec![
         ("response_type", "code"),
@@ -323,7 +314,7 @@ fn build_authorize_url(
         ("id_token_add_organizations", "true"),
         ("codex_cli_simplified_flow", "true"),
         ("state", state),
-        ("originator", originator),
+        ("originator", ORIGINATOR.value.as_str()),
     ];
     let qs = query
         .into_iter()

codex-rs/login/tests/suite/login_server_e2e.rs

@@ -102,7 +102,6 @@ async fn end_to_end_login_flow_persists_auth_json() {
         port: 0,
         open_browser: false,
         force_state: Some(state),
-        originator: "test_originator".to_string(),
     };
     let server = run_login_server(opts).unwrap();
     let login_port = server.actual_port;
@@ -161,7 +160,6 @@ async fn creates_missing_codex_home_dir() {
         port: 0,
         open_browser: false,
         force_state: Some(state),
-        originator: "test_originator".to_string(),
     };
     let server = run_login_server(opts).unwrap();
     let login_port = server.actual_port;
@@ -202,7 +200,6 @@ async fn cancels_previous_login_server_when_port_is_in_use() {
         port: 0,
         open_browser: false,
         force_state: Some("cancel_state".to_string()),
-        originator: "test_originator".to_string(),
     };
 
     let first_server = run_login_server(first_opts).unwrap();
@@ -221,7 +218,6 @@ async fn cancels_previous_login_server_when_port_is_in_use() {
         port: login_port,
         open_browser: false,
         force_state: Some("cancel_state_2".to_string()),
-        originator: "test_originator".to_string(),
     };
 
     let second_server = run_login_server(second_opts).unwrap();

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

@@ -64,6 +64,9 @@ async fn main() -> Result<()> {
             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

@@ -26,7 +26,6 @@ schemars = "0.8.22"
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 shlex = "1.3.0"
-strum_macros = "0.27.2"
 tokio = { version = "1", features = [
     "io-std",
     "macros",
@@ -41,8 +40,9 @@ 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"
 tempfile = "3"
-tokio-test = "0.4"
 wiremock = "0.6"

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

@@ -11,6 +11,8 @@ 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::try_read_auth_json;
 use codex_core::config::Config;
 use codex_core::config::ConfigOverrides;
 use codex_core::config::ConfigToml;
@@ -35,6 +37,8 @@ use codex_protocol::mcp_protocol::AddConversationListenerParams;
 use codex_protocol::mcp_protocol::AddConversationSubscriptionResponse;
 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;
@@ -65,6 +69,7 @@ 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::UserInfoResponse;
 use codex_protocol::mcp_protocol::UserSavedConfig;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseItem;
@@ -73,12 +78,16 @@ use codex_protocol::protocol::USER_MESSAGE_BEGIN;
 use mcp_types::JSONRPCErrorError;
 use mcp_types::RequestId;
 use std::collections::HashMap;
+use std::ffi::OsStr;
 use std::path::PathBuf;
 use std::sync::Arc;
 use std::time::Duration;
+use tokio::select;
 use tokio::sync::Mutex;
 use tokio::sync::oneshot;
 use tracing::error;
+use tracing::info;
+use tracing::warn;
 use uuid::Uuid;
 
 // Duration before a ChatGPT login attempt is abandoned.
@@ -142,6 +151,9 @@ impl CodexMessageProcessor {
             ClientRequest::ResumeConversation { request_id, params } => {
                 self.handle_resume_conversation(request_id, params).await;
             }
+            ClientRequest::ArchiveConversation { request_id, params } => {
+                self.archive_conversation(request_id, params).await;
+            }
             ClientRequest::SendUserMessage { request_id, params } => {
                 self.send_user_message(request_id, params).await;
             }
@@ -178,6 +190,9 @@ impl CodexMessageProcessor {
             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;
             }
@@ -189,11 +204,7 @@ impl CodexMessageProcessor {
 
         let opts = LoginServerOptions {
             open_browser: false,
-            ..LoginServerOptions::new(
-                config.codex_home.clone(),
-                CLIENT_ID.to_string(),
-                config.responses_originator_header.clone(),
-            )
+            ..LoginServerOptions::new(config.codex_home.clone(), CLIENT_ID.to_string())
         };
 
         enum LoginChatGptReply {
@@ -394,7 +405,7 @@ impl CodexMessageProcessor {
     }
 
     async fn get_user_agent(&self, request_id: RequestId) {
-        let user_agent = get_codex_user_agent(Some(&self.config.responses_originator_header));
+        let user_agent = get_codex_user_agent();
         let response = GetUserAgentResponse { user_agent };
         self.outgoing.send_response(request_id, response).await;
     }
@@ -434,6 +445,18 @@ 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 exec_one_off_command(&self, request_id: RequestId, params: ExecOneOffCommandParams) {
         tracing::debug!("ExecOneOffCommand params: {params:?}");
 
@@ -528,6 +551,7 @@ impl CodexMessageProcessor {
                 let response = NewConversationResponse {
                     conversation_id,
                     model: session_configured.model,
+                    rollout_path: session_configured.rollout_path,
                 };
                 self.outgoing.send_response(request_id, response).await;
             }
@@ -669,6 +693,141 @@ impl CodexMessageProcessor {
         }
     }
 
+    async fn archive_conversation(&self, request_id: RequestId, params: ArchiveConversationParams) {
+        let ArchiveConversationParams {
+            conversation_id,
+            rollout_path,
+        } = params;
+
+        // Verify that the rollout path is in the sessions directory or else
+        // a malicious client could specify an arbitrary path.
+        let rollout_folder = self.config.codex_home.join(codex_core::SESSIONS_SUBDIR);
+        let canonical_rollout_path = tokio::fs::canonicalize(&rollout_path).await;
+        let canonical_rollout_path = if let Ok(path) = canonical_rollout_path
+            && path.starts_with(&rollout_folder)
+        {
+            path
+        } else {
+            let error = JSONRPCErrorError {
+                code: INVALID_REQUEST_ERROR_CODE,
+                message: format!(
+                    "rollout path `{}` must be in sessions directory",
+                    rollout_path.display()
+                ),
+                data: None,
+            };
+            self.outgoing.send_error(request_id, error).await;
+            return;
+        };
+
+        let required_suffix = format!("{}.jsonl", conversation_id.0);
+        let Some(file_name) = canonical_rollout_path.file_name().map(OsStr::to_owned) else {
+            let error = JSONRPCErrorError {
+                code: INVALID_REQUEST_ERROR_CODE,
+                message: format!(
+                    "rollout path `{}` missing file name",
+                    rollout_path.display()
+                ),
+                data: None,
+            };
+            self.outgoing.send_error(request_id, error).await;
+            return;
+        };
+
+        if !file_name
+            .to_string_lossy()
+            .ends_with(required_suffix.as_str())
+        {
+            let error = JSONRPCErrorError {
+                code: INVALID_REQUEST_ERROR_CODE,
+                message: format!(
+                    "rollout path `{}` does not match conversation id {conversation_id}",
+                    rollout_path.display()
+                ),
+                data: None,
+            };
+            self.outgoing.send_error(request_id, error).await;
+            return;
+        }
+
+        let removed_conversation = self
+            .conversation_manager
+            .remove_conversation(&conversation_id)
+            .await;
+        if let Some(conversation) = removed_conversation {
+            info!("conversation {conversation_id} was active; shutting down");
+            let conversation_clone = conversation.clone();
+            let notify = Arc::new(tokio::sync::Notify::new());
+            let notify_clone = notify.clone();
+
+            // Establish the listener for ShutdownComplete before submitting
+            // Shutdown so it is not missed.
+            let is_shutdown = tokio::spawn(async move {
+                loop {
+                    select! {
+                        _ = notify_clone.notified() => {
+                            break;
+                        }
+                        event = conversation_clone.next_event() => {
+                            if let Ok(event) = event && matches!(event.msg, EventMsg::ShutdownComplete) {
+                                break;
+                            }
+                        }
+                    }
+                }
+            });
+
+            // Request shutdown.
+            match conversation.submit(Op::Shutdown).await {
+                Ok(_) => {
+                    // Successfully submitted Shutdown; wait before proceeding.
+                    select! {
+                        _ = is_shutdown => {
+                            // Normal shutdown: proceed with archive.
+                        }
+                        _ = tokio::time::sleep(Duration::from_secs(10)) => {
+                            warn!("conversation {conversation_id} shutdown timed out; proceeding with archive");
+                            notify.notify_one();
+                        }
+                    }
+                }
+                Err(err) => {
+                    error!("failed to submit Shutdown to conversation {conversation_id}: {err}");
+                    notify.notify_one();
+                    // Perhaps we lost a shutdown race, so let's continue to
+                    // clean up the .jsonl file.
+                }
+            }
+        }
+
+        // Move the .jsonl file to the archived sessions subdir.
+        let result: std::io::Result<()> = async {
+            let archive_folder = self
+                .config
+                .codex_home
+                .join(codex_core::ARCHIVED_SESSIONS_SUBDIR);
+            tokio::fs::create_dir_all(&archive_folder).await?;
+            tokio::fs::rename(&canonical_rollout_path, &archive_folder.join(&file_name)).await?;
+            Ok(())
+        }
+        .await;
+
+        match result {
+            Ok(()) => {
+                let response = ArchiveConversationResponse {};
+                self.outgoing.send_response(request_id, response).await;
+            }
+            Err(err) => {
+                let error = JSONRPCErrorError {
+                    code: INTERNAL_ERROR_CODE,
+                    message: format!("failed to archive conversation: {err}"),
+                    data: None,
+                };
+                self.outgoing.send_error(request_id, error).await;
+            }
+        }
+    }
+
     async fn send_user_message(&self, request_id: RequestId, params: SendUserMessageParams) {
         let SendUserMessageParams {
             conversation_id,
@@ -1110,10 +1269,7 @@ fn extract_conversation_summary(
     head: &[serde_json::Value],
 ) -> Option<ConversationSummary> {
     let session_meta = match head.first() {
-        Some(first_line) => match serde_json::from_value::<SessionMeta>(first_line.clone()) {
-            Ok(session_meta) => session_meta,
-            Err(..) => return None,
-        },
+        Some(first_line) => serde_json::from_value::<SessionMeta>(first_line.clone()).ok()?,
         None => return None,
     };
 
@@ -1171,6 +1327,10 @@ mod tests {
             json!({
                 "id": conversation_id.0,
                 "timestamp": timestamp,
+                "cwd": "/",
+                "originator": "codex",
+                "cli_version": "0.0.0",
+                "instructions": null
             }),
             json!({
                 "type": "message",

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

@@ -277,7 +277,7 @@ async fn run_codex_tool_session_inner(
                     | EventMsg::GetHistoryEntryResponse(_)
                     | EventMsg::PlanUpdate(_)
                     | EventMsg::TurnAborted(_)
-                    | EventMsg::ConversationHistory(_)
+                    | EventMsg::ConversationPath(_)
                     | EventMsg::UserMessage(_)
                     | EventMsg::ShutdownComplete => {
                         // For now, we do not do anything extra for these

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

@@ -14,6 +14,8 @@ use codex_protocol::mcp_protocol::ConversationId;
 use codex_core::AuthManager;
 use codex_core::ConversationManager;
 use codex_core::config::Config;
+use codex_core::default_client::USER_AGENT_SUFFIX;
+use codex_core::default_client::get_codex_user_agent;
 use codex_core::protocol::Submission;
 use mcp_types::CallToolRequestParams;
 use mcp_types::CallToolResult;
@@ -54,11 +56,8 @@ impl MessageProcessor {
         config: Arc<Config>,
     ) -> Self {
         let outgoing = Arc::new(outgoing);
-        let auth_manager = AuthManager::shared(
-            config.codex_home.clone(),
-            config.preferred_auth_method,
-            config.responses_originator_header.clone(),
-        );
+        let auth_manager =
+            AuthManager::shared(config.codex_home.clone(), config.preferred_auth_method);
         let conversation_manager = Arc::new(ConversationManager::new(auth_manager.clone()));
         let codex_message_processor = CodexMessageProcessor::new(
             auth_manager,
@@ -211,6 +210,14 @@ impl MessageProcessor {
             return;
         }
 
+        let client_info = params.client_info;
+        let name = client_info.name;
+        let version = client_info.version;
+        let user_agent_suffix = format!("{name}; {version}");
+        if let Ok(mut suffix) = USER_AGENT_SUFFIX.lock() {
+            *suffix = Some(user_agent_suffix);
+        }
+
         self.initialized = true;
 
         // Build a minimal InitializeResult. Fill with placeholders.
@@ -231,6 +238,7 @@ impl MessageProcessor {
                 name: "codex-mcp-server".to_string(),
                 version: env!("CARGO_PKG_VERSION").to_string(),
                 title: Some("Codex".to_string()),
+                user_agent: Some(get_codex_user_agent()),
             },
         };
 

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

@@ -262,6 +262,7 @@ mod tests {
     use codex_protocol::mcp_protocol::LoginChatGptCompleteNotification;
     use pretty_assertions::assert_eq;
     use serde_json::json;
+    use tempfile::NamedTempFile;
     use uuid::Uuid;
 
     use super::*;
@@ -272,6 +273,7 @@ mod tests {
         let outgoing_message_sender = OutgoingMessageSender::new(outgoing_tx);
 
         let conversation_id = ConversationId::new();
+        let rollout_file = NamedTempFile::new().unwrap();
         let event = Event {
             id: "1".to_string(),
             msg: EventMsg::SessionConfigured(SessionConfiguredEvent {
@@ -280,6 +282,7 @@ mod tests {
                 history_log_id: 1,
                 history_entry_count: 1000,
                 initial_messages: None,
+                rollout_path: rollout_file.path().to_path_buf(),
             }),
         };
 
@@ -305,12 +308,14 @@ mod tests {
         let outgoing_message_sender = OutgoingMessageSender::new(outgoing_tx);
 
         let conversation_id = ConversationId::new();
+        let rollout_file = NamedTempFile::new().unwrap();
         let session_configured_event = SessionConfiguredEvent {
             session_id: conversation_id,
             model: "gpt-4o".to_string(),
             history_log_id: 1,
             history_entry_count: 1000,
             initial_messages: None,
+            rollout_path: rollout_file.path().to_path_buf(),
         };
         let event = Event {
             id: "1".to_string(),
@@ -340,6 +345,7 @@ mod tests {
                 "history_log_id": session_configured_event.history_log_id,
                 "history_entry_count": session_configured_event.history_entry_count,
                 "type": "session_configured",
+                "rollout_path": rollout_file.path().to_path_buf(),
             }
         });
         assert_eq!(params.unwrap(), expected_params);

codex-rs/mcp-server/tests/common/Cargo.toml

@@ -13,16 +13,14 @@ codex-core = { path = "../../../core" }
 codex-mcp-server = { path = "../.." }
 codex-protocol = { path = "../../../protocol" }
 mcp-types = { path = "../../../mcp-types" }
+os_info = "3.12.0"
 pretty_assertions = "1.4.1"
 serde = { version = "1" }
 serde_json = "1"
-shlex = "1.3.0"
-tempfile = "3"
 tokio = { version = "1", features = [
     "io-std",
     "macros",
     "process",
     "rt-multi-thread",
 ] }
-uuid = { version = "1", features = ["serde", "v4"] }
 wiremock = "0.6"

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

@@ -13,6 +13,7 @@ use anyhow::Context;
 use assert_cmd::prelude::*;
 use codex_mcp_server::CodexToolCallParam;
 use codex_protocol::mcp_protocol::AddConversationListenerParams;
+use codex_protocol::mcp_protocol::ArchiveConversationParams;
 use codex_protocol::mcp_protocol::CancelLoginChatGptParams;
 use codex_protocol::mcp_protocol::GetAuthStatusParams;
 use codex_protocol::mcp_protocol::InterruptConversationParams;
@@ -53,6 +54,18 @@ pub struct McpProcess {
 
 impl McpProcess {
     pub async fn new(codex_home: &Path) -> anyhow::Result<Self> {
+        Self::new_with_env(codex_home, &[]).await
+    }
+
+    /// Creates a new MCP process, allowing tests to override or remove
+    /// specific environment variables for the child process only.
+    ///
+    /// Pass a tuple of (key, Some(value)) to set/override, or (key, None) to
+    /// remove a variable from the child's environment.
+    pub async fn new_with_env(
+        codex_home: &Path,
+        env_overrides: &[(&str, Option<&str>)],
+    ) -> anyhow::Result<Self> {
         // Use assert_cmd to locate the binary path and then switch to tokio::process::Command
         let std_cmd = StdCommand::cargo_bin("codex-mcp-server")
             .context("should find binary for codex-mcp-server")?;
@@ -67,6 +80,17 @@ impl McpProcess {
         cmd.env("CODEX_HOME", codex_home);
         cmd.env("RUST_LOG", "debug");
 
+        for (k, v) in env_overrides {
+            match v {
+                Some(val) => {
+                    cmd.env(k, val);
+                }
+                None => {
+                    cmd.env_remove(k);
+                }
+            }
+        }
+
         let mut process = cmd
             .kill_on_drop(true)
             .spawn()
@@ -114,6 +138,7 @@ impl McpProcess {
                 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(),
         };
@@ -128,6 +153,14 @@ impl McpProcess {
         .await?;
 
         let initialized = self.read_jsonrpc_message().await?;
+        let os_info = os_info::get();
+        let user_agent = format!(
+            "codex_cli_rs/0.0.0 ({} {}; {}) {} (elicitation test; 0.0.0)",
+            os_info.os_type(),
+            os_info.version(),
+            os_info.architecture().unwrap_or("unknown"),
+            codex_core::terminal::user_agent()
+        );
         assert_eq!(
             JSONRPCMessage::Response(JSONRPCResponse {
                 jsonrpc: JSONRPC_VERSION.into(),
@@ -141,7 +174,8 @@ impl McpProcess {
                     "serverInfo": {
                         "name": "codex-mcp-server",
                         "title": "Codex",
-                        "version": "0.0.0"
+                        "version": "0.0.0",
+                        "user_agent": user_agent
                     },
                     "protocolVersion": mcp_types::MCP_SCHEMA_VERSION
                 })
@@ -186,6 +220,15 @@ impl McpProcess {
         self.send_request("newConversation", params).await
     }
 
+    /// Send an `archiveConversation` JSON-RPC request.
+    pub async fn send_archive_conversation_request(
+        &mut self,
+        params: ArchiveConversationParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("archiveConversation", params).await
+    }
+
     /// Send an `addConversationListener` JSON-RPC request.
     pub async fn send_add_conversation_listener_request(
         &mut self,
@@ -252,6 +295,11 @@ 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 `listConversations` JSON-RPC request.
     pub async fn send_list_conversations_request(
         &mut self,

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

@@ -0,0 +1,105 @@
+use std::path::Path;
+
+use codex_core::ARCHIVED_SESSIONS_SUBDIR;
+use codex_protocol::mcp_protocol::ArchiveConversationParams;
+use codex_protocol::mcp_protocol::ArchiveConversationResponse;
+use codex_protocol::mcp_protocol::NewConversationParams;
+use codex_protocol::mcp_protocol::NewConversationResponse;
+use mcp_test_support::McpProcess;
+use mcp_test_support::to_response;
+use mcp_types::JSONRPCResponse;
+use mcp_types::RequestId;
+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 archive_conversation_moves_rollout_into_archived_directory() {
+    let codex_home = TempDir::new().expect("create temp dir");
+    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("initialize timeout")
+        .expect("initialize request");
+
+    let new_request_id = mcp
+        .send_new_conversation_request(NewConversationParams {
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await
+        .expect("send newConversation");
+    let new_response: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(new_request_id)),
+    )
+    .await
+    .expect("newConversation timeout")
+    .expect("newConversation response");
+
+    let NewConversationResponse {
+        conversation_id,
+        rollout_path,
+        ..
+    } = to_response::<NewConversationResponse>(new_response)
+        .expect("deserialize newConversation response");
+
+    assert!(
+        rollout_path.exists(),
+        "expected rollout path {} to exist",
+        rollout_path.display()
+    );
+
+    let archive_request_id = mcp
+        .send_archive_conversation_request(ArchiveConversationParams {
+            conversation_id,
+            rollout_path: rollout_path.clone(),
+        })
+        .await
+        .expect("send archiveConversation");
+    let archive_response: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(archive_request_id)),
+    )
+    .await
+    .expect("archiveConversation timeout")
+    .expect("archiveConversation response");
+
+    let _: ArchiveConversationResponse =
+        to_response::<ArchiveConversationResponse>(archive_response)
+            .expect("deserialize archiveConversation response");
+
+    let archived_directory = codex_home.path().join(ARCHIVED_SESSIONS_SUBDIR);
+    let archived_rollout_path =
+        archived_directory.join(rollout_path.file_name().unwrap_or_else(|| {
+            panic!("rollout path {} missing file name", rollout_path.display())
+        }));
+
+    assert!(
+        !rollout_path.exists(),
+        "expected rollout path {} to be moved",
+        rollout_path.display()
+    );
+    assert!(
+        archived_rollout_path.exists(),
+        "expected archived rollout path {} to exist",
+        archived_rollout_path.display()
+    );
+}
+
+fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(config_toml, config_contents())
+}
+
+fn config_contents() -> &'static str {
+    r#"model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "read-only"
+"#
+}

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

@@ -41,7 +41,7 @@ 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");
 
-    let mut mcp = McpProcess::new(codex_home.path())
+    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)])
         .await
         .expect("spawn mcp process");
     timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())

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,
+        rollout_path: _,
     } = new_conv_resp;
     assert_eq!(model, "mock-model");
 

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,
+        rollout_path: _,
     } = to_response::<NewConversationResponse>(new_conv_resp)
         .expect("deserialize newConversation response");
     assert_eq!(model, "o3");

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

@@ -156,14 +156,45 @@ fn create_fake_rollout(codex_home: &Path, filename_ts: &str, meta_rfc3339: &str,
 
     let file_path = dir.join(format!("rollout-{filename_ts}-{uuid}.jsonl"));
     let mut lines = Vec::new();
-    // Meta line with timestamp
-    lines.push(json!({"timestamp": meta_rfc3339, "id": uuid}).to_string());
-    // Minimal user message entry as a persisted response item
+    // Meta line with timestamp (flattened meta in payload for new schema)
     lines.push(
         json!({
-            "type":"message",
-            "role":"user",
-            "content":[{"type":"input_text","text": preview}]
+            "timestamp": meta_rfc3339,
+            "type": "session_meta",
+            "payload": {
+                "id": uuid,
+                "timestamp": meta_rfc3339,
+                "cwd": "/",
+                "originator": "codex",
+                "cli_version": "0.0.0",
+                "instructions": null
+            }
+        })
+        .to_string(),
+    );
+    // Minimal user message entry as a persisted response item (with envelope timestamp)
+    lines.push(
+        json!({
+            "timestamp": meta_rfc3339,
+            "type":"response_item",
+            "payload": {
+                "type":"message",
+                "role":"user",
+                "content":[{"type":"input_text","text": preview}]
+            }
+        })
+        .to_string(),
+    );
+    // Add a matching user message event line to satisfy filters
+    lines.push(
+        json!({
+            "timestamp": meta_rfc3339,
+            "type":"event_msg",
+            "payload": {
+                "type":"user_message",
+                "message": preview,
+                "kind": "plain"
+            }
         })
         .to_string(),
     );

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

@@ -46,7 +46,7 @@ async fn logout_chatgpt_removes_auth() {
     login_with_api_key(codex_home.path(), "sk-test-key").expect("seed api key");
     assert!(codex_home.path().join("auth.json").exists());
 
-    let mut mcp = McpProcess::new(codex_home.path())
+    let mut mcp = McpProcess::new_with_env(codex_home.path(), &[("OPENAI_API_KEY", None)])
         .await
         .expect("spawn mcp process");
     timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())

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

@@ -1,4 +1,5 @@
 // Aggregates all former standalone integration tests as modules.
+mod archive_conversation;
 mod auth;
 mod codex_message_processor_flow;
 mod codex_tool;
@@ -9,3 +10,4 @@ mod list_resume;
 mod login;
 mod send_message;
 mod user_agent;
+mod user_info;

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

@@ -1,5 +1,3 @@
-use codex_core::default_client::DEFAULT_ORIGINATOR;
-use codex_core::default_client::get_codex_user_agent;
 use codex_protocol::mcp_protocol::GetUserAgentResponse;
 use mcp_test_support::McpProcess;
 use mcp_test_support::to_response;
@@ -35,11 +33,18 @@ async fn get_user_agent_returns_current_codex_user_agent() {
     .expect("getUserAgent timeout")
     .expect("getUserAgent response");
 
+    let os_info = os_info::get();
+    let user_agent = format!(
+        "codex_cli_rs/0.0.0 ({} {}; {}) {} (elicitation test; 0.0.0)",
+        os_info.os_type(),
+        os_info.version(),
+        os_info.architecture().unwrap_or("unknown"),
+        codex_core::terminal::user_agent()
+    );
+
     let received: GetUserAgentResponse =
         to_response(response).expect("deserialize getUserAgent response");
-    let expected = GetUserAgentResponse {
-        user_agent: get_codex_user_agent(Some(DEFAULT_ORIGINATOR)),
-    };
+    let expected = GetUserAgentResponse { user_agent };
 
     assert_eq!(received, expected);
 }

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

@@ -0,0 +1,78 @@
+use std::time::Duration;
+
+use anyhow::Context;
+use base64::Engine;
+use base64::engine::general_purpose::URL_SAFE_NO_PAD;
+use codex_core::auth::AuthDotJson;
+use codex_core::auth::get_auth_file;
+use codex_core::auth::write_auth_json;
+use codex_core::token_data::IdTokenInfo;
+use codex_core::token_data::TokenData;
+use codex_protocol::mcp_protocol::UserInfoResponse;
+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 serde_json::json;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: Duration = Duration::from_secs(10);
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn user_info_returns_email_from_auth_json() {
+    let codex_home = TempDir::new().expect("create tempdir");
+
+    let auth_path = get_auth_file(codex_home.path());
+    let mut id_token = IdTokenInfo::default();
+    id_token.email = Some("user@example.com".to_string());
+    id_token.raw_jwt = encode_id_token_with_email("user@example.com").expect("encode id token");
+
+    let auth = AuthDotJson {
+        openai_api_key: None,
+        tokens: Some(TokenData {
+            id_token,
+            access_token: "access".to_string(),
+            refresh_token: "refresh".to_string(),
+            account_id: None,
+        }),
+        last_refresh: None,
+    };
+    write_auth_json(&auth_path, &auth).expect("write auth.json");
+
+    let mut mcp = McpProcess::new(codex_home.path())
+        .await
+        .expect("spawn mcp process");
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
+        .await
+        .expect("initialize timeout")
+        .expect("initialize request");
+
+    let request_id = mcp.send_user_info_request().await.expect("send userInfo");
+    let response: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(request_id)),
+    )
+    .await
+    .expect("userInfo timeout")
+    .expect("userInfo response");
+
+    let received: UserInfoResponse = to_response(response).expect("deserialize userInfo response");
+    let expected = UserInfoResponse {
+        alleged_user_email: Some("user@example.com".to_string()),
+    };
+
+    assert_eq!(received, expected);
+}
+
+fn encode_id_token_with_email(email: &str) -> anyhow::Result<String> {
+    let header_b64 = URL_SAFE_NO_PAD.encode(
+        serde_json::to_vec(&json!({ "alg": "none", "typ": "JWT" }))
+            .context("serialize jwt header")?,
+    );
+    let payload =
+        serde_json::to_vec(&json!({ "email": email })).context("serialize jwt payload")?;
+    let payload_b64 = URL_SAFE_NO_PAD.encode(payload);
+    Ok(format!("{header_b64}.{payload_b64}.signature"))
+}

codex-rs/mcp-types/generate_mcp_types.py

@@ -265,8 +265,11 @@ class StructField:
     name: str
     type_name: str
     serde: str | None = None
+    comment: str | None = None
 
     def append(self, out: list[str], supports_const: bool) -> None:
+        if self.comment:
+            out.append(f"    // {self.comment}\n")
         if self.serde:
             out.append(f"    {self.serde}\n")
         if self.viz == "const":
@@ -312,6 +315,18 @@ def define_struct(
         else:
             fields.append(StructField("pub", rs_prop.name, prop_type, rs_prop.serde))
 
+    # Special-case: add Codex-specific user_agent to Implementation
+    if name == "Implementation":
+        fields.append(
+            StructField(
+                "pub",
+                "user_agent",
+                "Option<String>",
+                '#[serde(default, skip_serializing_if = "Option::is_none")]',
+                "This is an extra field that the Codex MCP server sends as part of InitializeResult.",
+            )
+        )
+
     if implements_request_trait(name):
         add_trait_impl(name, "ModelContextProtocolRequest", fields, out)
     elif implements_notification_trait(name):

codex-rs/mcp-types/src/lib.rs

@@ -487,6 +487,9 @@ pub struct Implementation {
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub title: Option<String>,
     pub version: String,
+    // This is an extra field that the Codex MCP server sends as part of InitializeResult.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub user_agent: Option<String>,
 }
 
 #[derive(Debug, Clone, PartialEq, Deserialize, Serialize, TS)]

codex-rs/mcp-types/tests/suite/initialize.rs

@@ -62,6 +62,7 @@ fn deserialize_initialize_request() {
                 name: "acme-client".into(),
                 title: Some("Acme".to_string()),
                 version: "1.2.3".into(),
+                user_agent: None,
             },
             protocol_version: "2025-06-18".into(),
         }

codex-rs/ollama/Cargo.toml

@@ -24,9 +24,7 @@ tokio = { version = "1", features = [
     "rt-multi-thread",
     "signal",
 ] }
-toml = "0.9.5"
 tracing = { version = "0.1.41", features = ["log"] }
 wiremock = "0.6"
 
 [dev-dependencies]
-tempfile = "3"

codex-rs/protocol-ts/Cargo.toml

@@ -16,6 +16,7 @@ path = "src/main.rs"
 
 [dependencies]
 anyhow = "1"
+mcp-types = { path = "../mcp-types" }
 codex-protocol = { path = "../protocol" }
 ts-rs = "11"
 clap = { version = "4", features = ["derive"] }

codex-rs/protocol-ts/src/lib.rs

@@ -16,11 +16,15 @@ pub fn generate_ts(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
     ensure_dir(out_dir)?;
 
     // Generate TS bindings
+    mcp_types::InitializeResult::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::ConversationId::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::InputItem::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::ClientRequest::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::ServerRequest::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::NewConversationResponse::export_all_to(out_dir)?;
+    codex_protocol::mcp_protocol::ListConversationsResponse::export_all_to(out_dir)?;
+    codex_protocol::mcp_protocol::ResumeConversationResponse::export_all_to(out_dir)?;
+    codex_protocol::mcp_protocol::ArchiveConversationResponse::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::AddConversationSubscriptionResponse::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::RemoveConversationSubscriptionResponse::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::SendUserMessageResponse::export_all_to(out_dir)?;
@@ -28,7 +32,6 @@ pub fn generate_ts(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
     codex_protocol::mcp_protocol::InterruptConversationResponse::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::GitDiffToRemoteResponse::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::LoginChatGptResponse::export_all_to(out_dir)?;
-    codex_protocol::mcp_protocol::LoginChatGptCompleteNotification::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::CancelLoginChatGptResponse::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::LogoutChatGptResponse::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::GetAuthStatusResponse::export_all_to(out_dir)?;
@@ -36,9 +39,11 @@ pub fn generate_ts(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
     codex_protocol::mcp_protocol::ExecCommandApprovalResponse::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::GetUserSavedConfigResponse::export_all_to(out_dir)?;
     codex_protocol::mcp_protocol::GetUserAgentResponse::export_all_to(out_dir)?;
+    codex_protocol::mcp_protocol::UserInfoResponse::export_all_to(out_dir)?;
+
+    // All notification types reachable from this enum will be generated by
+    // induction, so they do not need to be listed individually.
     codex_protocol::mcp_protocol::ServerNotification::export_all_to(out_dir)?;
-    codex_protocol::mcp_protocol::ListConversationsResponse::export_all_to(out_dir)?;
-    codex_protocol::mcp_protocol::ResumeConversationResponse::export_all_to(out_dir)?;
 
     generate_index_ts(out_dir)?;
 

codex-rs/protocol/Cargo.toml

@@ -17,7 +17,6 @@ icu_locale_core = "2.0.0"
 mcp-types = { path = "../mcp-types" }
 mime_guess = "2.0.5"
 serde = { version = "1", features = ["derive"] }
-serde_bytes = "0.11"
 serde_json = "1"
 serde_with = { version = "3.14.0", features = ["macros", "base64"] }
 strum = "0.27.2"
@@ -29,3 +28,8 @@ uuid = { version = "1", features = ["serde", "v4"] }
 
 [dev-dependencies]
 pretty_assertions = "1.4.1"
+tempfile = "3"
+
+[package.metadata.cargo-shear]
+# Required because the not imported as strum_macros in non-nightly builds.
+ignored = ["strum"]

codex-rs/protocol/src/mcp_protocol.rs

@@ -91,6 +91,11 @@ pub enum ClientRequest {
         request_id: RequestId,
         params: ResumeConversationParams,
     },
+    ArchiveConversation {
+        #[serde(rename = "id")]
+        request_id: RequestId,
+        params: ArchiveConversationParams,
+    },
     SendUserMessage {
         #[serde(rename = "id")]
         request_id: RequestId,
@@ -147,6 +152,10 @@ pub enum ClientRequest {
         #[serde(rename = "id")]
         request_id: RequestId,
     },
+    UserInfo {
+        #[serde(rename = "id")]
+        request_id: RequestId,
+    },
     /// Execute a command (argv vector) under the server's sandbox.
     ExecOneOffCommand {
         #[serde(rename = "id")]
@@ -203,6 +212,7 @@ pub struct NewConversationParams {
 pub struct NewConversationResponse {
     pub conversation_id: ConversationId,
     pub model: String,
+    pub rollout_path: PathBuf,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, TS)]
@@ -262,6 +272,18 @@ pub struct AddConversationSubscriptionResponse {
     pub subscription_id: Uuid,
 }
 
+/// The [`ConversationId`] must match the `rollout_path`.
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
+#[serde(rename_all = "camelCase")]
+pub struct ArchiveConversationParams {
+    pub conversation_id: ConversationId,
+    pub rollout_path: PathBuf,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
+#[serde(rename_all = "camelCase")]
+pub struct ArchiveConversationResponse {}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
 #[serde(rename_all = "camelCase")]
 pub struct RemoveConversationSubscriptionResponse {}
@@ -356,6 +378,16 @@ pub struct GetUserAgentResponse {
     pub user_agent: String,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
+#[serde(rename_all = "camelCase")]
+pub struct UserInfoResponse {
+    /// Note: `alleged_user_email` is not currently verified. We read it from
+    /// the local auth.json, which the user could theoretically modify. In the
+    /// future, we may add logic to verify the email against the server before
+    /// returning it.
+    pub alleged_user_email: Option<String>,
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
 #[serde(rename_all = "camelCase")]
 pub struct GetUserSavedConfigResponse {

codex-rs/protocol/src/models.rs

@@ -49,7 +49,7 @@ pub enum ResponseItem {
         content: Vec<ContentItem>,
     },
     Reasoning {
-        #[serde(default)]
+        #[serde(default, skip_serializing)]
         id: String,
         summary: Vec<ReasoningItemReasoningSummary>,
         #[serde(default, skip_serializing_if = "should_serialize_reasoning_content")]
@@ -115,7 +115,6 @@ pub enum ResponseItem {
         status: Option<String>,
         action: WebSearchAction,
     },
-
     #[serde(other)]
     Other,
 }

codex-rs/protocol/src/protocol.rs

@@ -149,7 +149,7 @@ pub enum Op {
 
     /// Request the full in-memory conversation transcript for the current session.
     /// Reply is delivered via `EventMsg::ConversationHistory`.
-    GetHistory,
+    GetPath,
 
     /// Request the list of MCP tools available across all configured servers.
     /// Reply is delivered via `EventMsg::McpListToolsResponse`.
@@ -499,7 +499,7 @@ pub enum EventMsg {
     /// Notification that the agent is shutting down.
     ShutdownComplete,
 
-    ConversationHistory(ConversationHistoryResponseEvent),
+    ConversationPath(ConversationPathResponseEvent),
 }
 
 // Individual event payload types matching each `EventMsg` variant.
@@ -695,6 +695,8 @@ pub struct UserMessageEvent {
     pub message: String,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub kind: Option<InputMessageKind>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub images: Option<Vec<String>>,
 }
 
 impl<T, U> From<(T, U)> for InputMessageKind
@@ -799,9 +801,123 @@ pub struct WebSearchEndEvent {
 /// Response payload for `Op::GetHistory` containing the current session's
 /// in-memory transcript.
 #[derive(Debug, Clone, Deserialize, Serialize, TS)]
-pub struct ConversationHistoryResponseEvent {
+pub struct ConversationPathResponseEvent {
     pub conversation_id: ConversationId,
-    pub entries: Vec<ResponseItem>,
+    pub path: PathBuf,
+}
+
+#[derive(Debug, Clone, Deserialize, Serialize, TS)]
+pub struct ResumedHistory {
+    pub conversation_id: ConversationId,
+    pub history: Vec<RolloutItem>,
+    pub rollout_path: PathBuf,
+}
+
+#[derive(Debug, Clone, Deserialize, Serialize, TS)]
+pub enum InitialHistory {
+    New,
+    Resumed(ResumedHistory),
+    Forked(Vec<RolloutItem>),
+}
+
+impl InitialHistory {
+    pub fn get_rollout_items(&self) -> Vec<RolloutItem> {
+        match self {
+            InitialHistory::New => Vec::new(),
+            InitialHistory::Resumed(resumed) => resumed.history.clone(),
+            InitialHistory::Forked(items) => items.clone(),
+        }
+    }
+    pub fn get_response_items(&self) -> Vec<ResponseItem> {
+        match self {
+            InitialHistory::New => Vec::new(),
+            InitialHistory::Resumed(resumed) => resumed
+                .history
+                .iter()
+                .filter_map(|ri| match ri {
+                    RolloutItem::ResponseItem(item) => Some(item.clone()),
+                    _ => None,
+                })
+                .collect(),
+            InitialHistory::Forked(items) => items
+                .iter()
+                .filter_map(|ri| match ri {
+                    RolloutItem::ResponseItem(item) => Some(item.clone()),
+                    _ => None,
+                })
+                .collect(),
+        }
+    }
+    pub fn get_event_msgs(&self) -> Option<Vec<EventMsg>> {
+        match self {
+            InitialHistory::New => None,
+            InitialHistory::Resumed(resumed) => Some(
+                resumed
+                    .history
+                    .iter()
+                    .filter_map(|ri| match ri {
+                        RolloutItem::EventMsg(ev) => Some(ev.clone()),
+                        _ => None,
+                    })
+                    .collect(),
+            ),
+            InitialHistory::Forked(items) => Some(
+                items
+                    .iter()
+                    .filter_map(|ri| match ri {
+                        RolloutItem::EventMsg(ev) => Some(ev.clone()),
+                        _ => None,
+                    })
+                    .collect(),
+            ),
+        }
+    }
+}
+
+#[derive(Serialize, Deserialize, Clone, Default, Debug, TS)]
+pub struct SessionMeta {
+    pub id: ConversationId,
+    pub timestamp: String,
+    pub cwd: PathBuf,
+    pub originator: String,
+    pub cli_version: String,
+    pub instructions: Option<String>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, TS)]
+pub struct SessionMetaLine {
+    #[serde(flatten)]
+    pub meta: SessionMeta,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub git: Option<GitInfo>,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, TS)]
+#[serde(tag = "type", content = "payload", rename_all = "snake_case")]
+pub enum RolloutItem {
+    SessionMeta(SessionMetaLine),
+    ResponseItem(ResponseItem),
+    EventMsg(EventMsg),
+}
+
+#[derive(Serialize, Deserialize, Clone)]
+pub struct RolloutLine {
+    pub timestamp: String,
+    #[serde(flatten)]
+    pub item: RolloutItem,
+}
+
+#[derive(Serialize, Deserialize, Clone, Debug, TS)]
+pub struct GitInfo {
+    /// Current commit hash (SHA)
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub commit_hash: Option<String>,
+    /// Current branch name
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub branch: Option<String>,
+    /// Repository URL (if available from remote)
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub repository_url: Option<String>,
 }
 
 #[derive(Debug, Clone, Deserialize, Serialize, TS)]
@@ -958,6 +1074,8 @@ pub struct SessionConfiguredEvent {
     /// When present, UIs can use these to seed the history.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub initial_messages: Option<Vec<EventMsg>>,
+
+    pub rollout_path: PathBuf,
 }
 
 /// User's decision in response to an ExecApprovalRequest.
@@ -1020,12 +1138,15 @@ pub enum TurnAbortReason {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use serde_json::json;
+    use tempfile::NamedTempFile;
 
     /// Serialize Event to verify that its JSON representation has the expected
     /// amount of nesting.
     #[test]
     fn serialize_event() {
         let conversation_id = ConversationId(uuid::uuid!("67e55044-10b1-426f-9247-bb680e5fe0c8"));
+        let rollout_file = NamedTempFile::new().unwrap();
         let event = Event {
             id: "1234".to_string(),
             msg: EventMsg::SessionConfigured(SessionConfiguredEvent {
@@ -1034,13 +1155,22 @@ mod tests {
                 history_log_id: 0,
                 history_entry_count: 0,
                 initial_messages: None,
+                rollout_path: rollout_file.path().to_path_buf(),
             }),
         };
-        let serialized = serde_json::to_string(&event).unwrap();
-        assert_eq!(
-            serialized,
-            r#"{"id":"1234","msg":{"type":"session_configured","session_id":"67e55044-10b1-426f-9247-bb680e5fe0c8","model":"codex-mini-latest","history_log_id":0,"history_entry_count":0}}"#
-        );
+
+        let expected = json!({
+            "id": "1234",
+            "msg": {
+                "type": "session_configured",
+                "session_id": "67e55044-10b1-426f-9247-bb680e5fe0c8",
+                "model": "codex-mini-latest",
+                "history_log_id": 0,
+                "history_entry_count": 0,
+                "rollout_path": format!("{}", rollout_file.path().display()),
+            }
+        });
+        assert_eq!(expected, serde_json::to_value(&event).unwrap());
     }
 
     #[test]

codex-rs/tui/Cargo.toml

@@ -59,9 +59,7 @@ ratatui = { version = "0.29.0", features = [
     "unstable-rendered-line-info",
     "unstable-widget-ref",
 ] }
-ratatui-image = "8.0.0"
 regex-lite = "0.1"
-reqwest = { version = "0.12", features = ["json"] }
 serde = { version = "1", features = ["derive"] }
 serde_json = { version = "1", features = ["preserve_order"] }
 shlex = "1.3.0"
@@ -81,12 +79,10 @@ tokio-stream = "0.1.17"
 tracing = { version = "0.1.41", features = ["log"] }
 tracing-appender = "0.2.3"
 tracing-subscriber = { version = "0.3.19", features = ["env-filter"] }
-tui-input = "0.14.0"
-tui-markdown = "0.3.3"
+pulldown-cmark = "0.10"
 unicode-segmentation = "1.12.0"
 unicode-width = "0.1"
 url = "2"
-uuid = "1"
 pathdiff = "0.2"
 
 [target.'cfg(unix)'.dependencies]

codex-rs/tui/src/app_backtrack.rs

@@ -1,9 +1,11 @@
+use std::path::PathBuf;
+
 use crate::app::App;
 use crate::backtrack_helpers;
 use crate::pager_overlay::Overlay;
 use crate::tui;
 use crate::tui::TuiEvent;
-use codex_core::protocol::ConversationHistoryResponseEvent;
+use codex_core::protocol::ConversationPathResponseEvent;
 use codex_protocol::mcp_protocol::ConversationId;
 use color_eyre::eyre::Result;
 use crossterm::event::KeyCode;
@@ -98,7 +100,7 @@ impl App {
     ) {
         self.backtrack.pending = Some((base_id, drop_last_messages, prefill));
         self.app_event_tx.send(crate::app_event::AppEvent::CodexOp(
-            codex_core::protocol::Op::GetHistory,
+            codex_core::protocol::Op::GetPath,
         ));
     }
 
@@ -265,7 +267,7 @@ impl App {
     pub(crate) async fn on_conversation_history_for_backtrack(
         &mut self,
         tui: &mut tui::Tui,
-        ev: ConversationHistoryResponseEvent,
+        ev: ConversationPathResponseEvent,
     ) -> Result<()> {
         if let Some((base_id, _, _)) = self.backtrack.pending.as_ref()
             && ev.conversation_id == *base_id
@@ -281,14 +283,14 @@ impl App {
     async fn fork_and_switch_to_new_conversation(
         &mut self,
         tui: &mut tui::Tui,
-        ev: ConversationHistoryResponseEvent,
+        ev: ConversationPathResponseEvent,
         drop_count: usize,
         prefill: String,
     ) {
         let cfg = self.chat_widget.config_ref().clone();
         // Perform the fork via a thin wrapper for clarity/testability.
         let result = self
-            .perform_fork(ev.entries.clone(), drop_count, cfg.clone())
+            .perform_fork(ev.path.clone(), drop_count, cfg.clone())
             .await;
         match result {
             Ok(new_conv) => {
@@ -301,13 +303,11 @@ impl App {
     /// Thin wrapper around ConversationManager::fork_conversation.
     async fn perform_fork(
         &self,
-        entries: Vec<codex_protocol::models::ResponseItem>,
+        path: PathBuf,
         drop_count: usize,
         cfg: codex_core::config::Config,
     ) -> codex_core::error::Result<codex_core::NewConversation> {
-        self.server
-            .fork_conversation(entries, drop_count, cfg)
-            .await
+        self.server.fork_conversation(drop_count, cfg, path).await
     }
 
     /// Install a forked conversation into the ChatWidget and update UI to reflect selection.

codex-rs/tui/src/app_event.rs

@@ -1,4 +1,4 @@
-use codex_core::protocol::ConversationHistoryResponseEvent;
+use codex_core::protocol::ConversationPathResponseEvent;
 use codex_core::protocol::Event;
 use codex_file_search::FileMatch;
 
@@ -58,5 +58,5 @@ pub(crate) enum AppEvent {
     UpdateSandboxPolicy(SandboxPolicy),
 
     /// Forwarded conversation history snapshot from the current conversation.
-    ConversationHistory(ConversationHistoryResponseEvent),
+    ConversationHistory(ConversationPathResponseEvent),
 }

codex-rs/tui/src/bin/md-events.rs

@@ -0,0 +1,15 @@
+use std::io::Read;
+use std::io::{self};
+
+fn main() {
+    let mut input = String::new();
+    if let Err(err) = io::stdin().read_to_string(&mut input) {
+        eprintln!("failed to read stdin: {err}");
+        std::process::exit(1);
+    }
+
+    let parser = pulldown_cmark::Parser::new(&input);
+    for event in parser {
+        println!("{event:?}");
+    }
+}

codex-rs/tui/src/bottom_pane/textarea.rs

@@ -252,6 +252,11 @@ impl TextArea {
                 modifiers: KeyModifiers::CONTROL,
                 ..
             } => self.delete_backward(1),
+            KeyEvent {
+                code: KeyCode::Delete,
+                modifiers: KeyModifiers::ALT,
+                ..
+            }  => self.delete_forward_word(),
             KeyEvent {
                 code: KeyCode::Delete,
                 ..
@@ -435,6 +440,18 @@ impl TextArea {
         self.replace_range(start..self.cursor_pos, "");
     }
 
+    /// Delete text to the right of the cursor using "word" semantics.
+    ///
+    /// Deletes from the current cursor position through the end of the next word as determined
+    /// by `end_of_next_word()`. Any whitespace (including newlines) between the cursor and that
+    /// word is included in the deletion.
+    pub fn delete_forward_word(&mut self) {
+        let end = self.end_of_next_word();
+        if end > self.cursor_pos {
+            self.replace_range(self.cursor_pos..end, "");
+        }
+    }
+
     pub fn kill_to_end_of_line(&mut self) {
         let eol = self.end_of_current_line();
         if self.cursor_pos == eol {
@@ -1104,6 +1121,79 @@ mod tests {
         assert_eq!(t.cursor(), 3);
     }
 
+    #[test]
+    fn delete_forward_word_variants() {
+        let mut t = ta_with("hello   world ");
+        t.set_cursor(0);
+        t.delete_forward_word();
+        assert_eq!(t.text(), "   world ");
+        assert_eq!(t.cursor(), 0);
+
+        let mut t = ta_with("hello   world ");
+        t.set_cursor(1);
+        t.delete_forward_word();
+        assert_eq!(t.text(), "h   world ");
+        assert_eq!(t.cursor(), 1);
+
+        let mut t = ta_with("hello   world");
+        t.set_cursor(t.text().len());
+        t.delete_forward_word();
+        assert_eq!(t.text(), "hello   world");
+        assert_eq!(t.cursor(), t.text().len());
+
+        let mut t = ta_with("foo   \nbar");
+        t.set_cursor(3);
+        t.delete_forward_word();
+        assert_eq!(t.text(), "foo");
+        assert_eq!(t.cursor(), 3);
+
+        let mut t = ta_with("foo\nbar");
+        t.set_cursor(3);
+        t.delete_forward_word();
+        assert_eq!(t.text(), "foo");
+        assert_eq!(t.cursor(), 3);
+
+        let mut t = ta_with("hello   world ");
+        t.set_cursor(t.text().len() + 10);
+        t.delete_forward_word();
+        assert_eq!(t.text(), "hello   world ");
+        assert_eq!(t.cursor(), t.text().len());
+    }
+
+    #[test]
+    fn delete_forward_word_handles_atomic_elements() {
+        let mut t = TextArea::new();
+        t.insert_element("<element>");
+        t.insert_str(" tail");
+
+        t.set_cursor(0);
+        t.delete_forward_word();
+        assert_eq!(t.text(), " tail");
+        assert_eq!(t.cursor(), 0);
+
+        let mut t = TextArea::new();
+        t.insert_str("   ");
+        t.insert_element("<element>");
+        t.insert_str(" tail");
+
+        t.set_cursor(0);
+        t.delete_forward_word();
+        assert_eq!(t.text(), " tail");
+        assert_eq!(t.cursor(), 0);
+
+        let mut t = TextArea::new();
+        t.insert_str("prefix ");
+        t.insert_element("<element>");
+        t.insert_str(" tail");
+
+        // cursor in the middle of the element, delete_forward_word deletes the element
+        let elem_range = t.elements[0].range.clone();
+        t.cursor_pos = elem_range.start + (elem_range.len() / 2);
+        t.delete_forward_word();
+        assert_eq!(t.text(), "prefix  tail");
+        assert_eq!(t.cursor(), elem_range.start);
+    }
+
     #[test]
     fn cursor_left_and_right_handle_graphemes() {
         let mut t = ta_with("a👍b");
@@ -1174,6 +1264,21 @@ mod tests {
         assert_eq!(t.cursor(), 6);
     }
 
+    #[test]
+    fn delete_forward_word_with_without_alt_modifier() {
+        let mut t = ta_with("hello world");
+        t.set_cursor(0);
+        t.input(KeyEvent::new(KeyCode::Delete, KeyModifiers::ALT));
+        assert_eq!(t.text(), " world");
+        assert_eq!(t.cursor(), 0);
+
+        let mut t = ta_with("hello");
+        t.set_cursor(0);
+        t.input(KeyEvent::new(KeyCode::Delete, KeyModifiers::NONE));
+        assert_eq!(t.text(), "ello");
+        assert_eq!(t.cursor(), 0);
+    }
+
     #[test]
     fn control_h_backspace() {
         // Test Ctrl+H as backspace

codex-rs/tui/src/chatwidget.rs

@@ -1083,7 +1083,7 @@ impl ChatWidget {
                     self.on_user_message_event(ev);
                 }
             }
-            EventMsg::ConversationHistory(ev) => {
+            EventMsg::ConversationPath(ev) => {
                 self.app_event_tx
                     .send(crate::app_event::AppEvent::ConversationHistory(ev));
             }

codex-rs/tui/src/chatwidget/snapshots/codex_tui__chatwidget__tests__chatwidget_markdown_code_blocks_vt100_snapshot.snap

@@ -0,0 +1,18 @@
+---
+source: tui/src/chatwidget/tests.rs
+expression: visual
+---
+>     -- Indented code block (4 spaces)
+      SELECT *
+      FROM "users"
+      WHERE "email" LIKE '%@example.com';
+
+  ```sh
+  printf 'fenced within fenced\n'
+  ```
+
+  {
+    // comment allowed in jsonc
+    "path": "C:\\Program Files\\App",
+    "regex": "^foo.*(bar)?$"
+  }

codex-rs/tui/src/chatwidget/tests.rs

@@ -35,6 +35,7 @@ use std::fs::File;
 use std::io::BufRead;
 use std::io::BufReader;
 use std::path::PathBuf;
+use tempfile::NamedTempFile;
 use tokio::sync::mpsc::unbounded_channel;
 
 fn test_config() -> Config {
@@ -133,7 +134,7 @@ fn resumed_initial_messages_render_history() {
     let (mut chat, mut rx, _ops) = make_chatwidget_manual();
 
     let conversation_id = ConversationId::new();
-
+    let rollout_file = NamedTempFile::new().unwrap();
     let configured = codex_core::protocol::SessionConfiguredEvent {
         session_id: conversation_id,
         model: "test-model".to_string(),
@@ -143,11 +144,13 @@ fn resumed_initial_messages_render_history() {
             EventMsg::UserMessage(UserMessageEvent {
                 message: "hello from user".to_string(),
                 kind: Some(InputMessageKind::Plain),
+                images: None,
             }),
             EventMsg::AgentMessage(AgentMessageEvent {
                 message: "assistant reply".to_string(),
             }),
         ]),
+        rollout_path: rollout_file.path().to_path_buf(),
     };
 
     chat.handle_codex_event(Event {
@@ -1753,3 +1756,123 @@ fn chatwidget_exec_and_status_layout_vt100_snapshot() {
     let visual = vt_lines.join("\n");
     assert_snapshot!(visual);
 }
+
+// E2E vt100 snapshot for complex markdown with indented and nested fenced code blocks
+#[test]
+fn chatwidget_markdown_code_blocks_vt100_snapshot() {
+    let (mut chat, mut rx, _op_rx) = make_chatwidget_manual();
+
+    // Simulate a final agent message via streaming deltas instead of a single message
+
+    chat.handle_codex_event(Event {
+        id: "t1".into(),
+        msg: EventMsg::TaskStarted(TaskStartedEvent {
+            model_context_window: None,
+        }),
+    });
+    // Build a vt100 visual from the history insertions only (no UI overlay)
+    let width: u16 = 80;
+    let height: u16 = 50;
+    let backend = ratatui::backend::TestBackend::new(width, height);
+    let mut term = crate::custom_terminal::Terminal::with_options(backend).expect("terminal");
+    // Place viewport at the last line so that history lines insert above it
+    term.set_viewport_area(Rect::new(0, height - 1, width, 1));
+
+    let mut ansi: Vec<u8> = Vec::new();
+
+    // Simulate streaming via AgentMessageDelta in 2-character chunks (no final AgentMessage).
+    let source: &str = r#"
+
+    -- Indented code block (4 spaces)
+    SELECT *
+    FROM "users"
+    WHERE "email" LIKE '%@example.com';
+
+````markdown
+```sh
+printf 'fenced within fenced\n'
+```
+````
+
+```jsonc
+{
+  // comment allowed in jsonc
+  "path": "C:\\Program Files\\App",
+  "regex": "^foo.*(bar)?$"
+}
+```
+"#;
+
+    let mut it = source.chars();
+    loop {
+        let mut delta = String::new();
+        match it.next() {
+            Some(c) => delta.push(c),
+            None => break,
+        }
+        if let Some(c2) = it.next() {
+            delta.push(c2);
+        }
+
+        chat.handle_codex_event(Event {
+            id: "t1".into(),
+            msg: EventMsg::AgentMessageDelta(AgentMessageDeltaEvent { delta }),
+        });
+        // Drive commit ticks and drain emitted history lines into the vt100 buffer.
+        loop {
+            chat.on_commit_tick();
+            let mut inserted_any = false;
+            while let Ok(app_ev) = rx.try_recv() {
+                if let AppEvent::InsertHistoryCell(cell) = app_ev {
+                    let lines = cell.display_lines(width);
+                    crate::insert_history::insert_history_lines_to_writer(
+                        &mut term, &mut ansi, lines,
+                    );
+                    inserted_any = true;
+                }
+            }
+            if !inserted_any {
+                break;
+            }
+        }
+    }
+
+    // Finalize the stream without sending a final AgentMessage, to flush any tail.
+    chat.handle_codex_event(Event {
+        id: "t1".into(),
+        msg: EventMsg::TaskComplete(TaskCompleteEvent {
+            last_agent_message: None,
+        }),
+    });
+    for lines in drain_insert_history(&mut rx) {
+        crate::insert_history::insert_history_lines_to_writer(&mut term, &mut ansi, lines);
+    }
+
+    let mut parser = vt100::Parser::new(height, width, 0);
+    parser.process(&ansi);
+
+    let mut vt_lines: Vec<String> = (0..height)
+        .map(|row| {
+            let mut s = String::with_capacity(width as usize);
+            for col in 0..width {
+                if let Some(cell) = parser.screen().cell(row, col) {
+                    if let Some(ch) = cell.contents().chars().next() {
+                        s.push(ch);
+                    } else {
+                        s.push(' ');
+                    }
+                } else {
+                    s.push(' ');
+                }
+            }
+            s.trim_end().to_string()
+        })
+        .collect();
+
+    // Compact trailing blank rows for a stable snapshot
+    while matches!(vt_lines.last(), Some(l) if l.trim().is_empty()) {
+        vt_lines.pop();
+    }
+    let visual = vt_lines.join("\n");
+    assert_snapshot!(visual);
+}

codex-rs/tui/src/chatwidget_stream_tests.rs

@@ -1 +0,0 @@
-

codex-rs/tui/src/history_cell.rs

@@ -605,6 +605,7 @@ pub(crate) fn new_session_info(
         history_log_id: _,
         history_entry_count: _,
         initial_messages: _,
+        rollout_path: _,
     } = event;
     if is_first_event {
         let cwd_str = match relativize_to_home(&config.cwd) {

codex-rs/tui/src/insert_history.rs

@@ -97,7 +97,17 @@ pub fn insert_history_lines_to_writer<B, W>(
 
     for line in wrapped {
         queue!(writer, Print("\r\n")).ok();
-        write_spans(writer, &line).ok();
+        // Merge line-level style into each span so that ANSI colors reflect
+        // line styles (e.g., blockquotes with green fg).
+        let merged_spans: Vec<Span> = line
+            .spans
+            .iter()
+            .map(|s| Span {
+                style: s.style.patch(line.style),
+                content: s.content.clone(),
+            })
+            .collect();
+        write_spans(writer, merged_spans.iter()).ok();
     }
 
     queue!(writer, ResetScrollRegion).ok();
@@ -264,6 +274,10 @@ where
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::markdown_render::render_markdown_text;
+    use ratatui::layout::Rect;
+    use ratatui::style::Color;
+    use vt100::Parser;
 
     #[test]
     fn writes_bold_then_regular_spans() {
@@ -292,4 +306,240 @@ mod tests {
             String::from_utf8(expected).unwrap()
         );
     }
+
+    #[test]
+    fn vt100_blockquote_line_emits_green_fg() {
+        // Set up a small off-screen terminal
+        let width: u16 = 40;
+        let height: u16 = 10;
+        let backend = ratatui::backend::TestBackend::new(width, height);
+        let mut term = crate::custom_terminal::Terminal::with_options(backend).expect("terminal");
+        // Place viewport on the last line so history inserts scroll upward
+        let viewport = Rect::new(0, height - 1, width, 1);
+        term.set_viewport_area(viewport);
+
+        // Build a blockquote-like line: apply line-level green style and prefix "> "
+        let mut line: Line<'static> = Line::from(vec!["> ".into(), "Hello world".into()]);
+        line = line.style(Color::Green);
+        let mut ansi: Vec<u8> = Vec::new();
+        insert_history_lines_to_writer(&mut term, &mut ansi, vec![line]);
+
+        // Parse ANSI using vt100 and assert at least one non-default fg color appears
+        let mut parser = Parser::new(height, width, 0);
+        parser.process(&ansi);
+
+        let mut saw_colored = false;
+        'outer: for row in 0..height {
+            for col in 0..width {
+                if let Some(cell) = parser.screen().cell(row, col)
+                    && cell.has_contents()
+                    && cell.fgcolor() != vt100::Color::Default
+                {
+                    saw_colored = true;
+                    break 'outer;
+                }
+            }
+        }
+        assert!(
+            saw_colored,
+            "expected at least one colored cell in vt100 output"
+        );
+    }
+
+    #[test]
+    fn vt100_blockquote_wrap_preserves_color_on_all_wrapped_lines() {
+        // Force wrapping by using a narrow viewport width and a long blockquote line.
+        let width: u16 = 20;
+        let height: u16 = 8;
+        let backend = ratatui::backend::TestBackend::new(width, height);
+        let mut term = crate::custom_terminal::Terminal::with_options(backend).expect("terminal");
+        // Viewport is the last line so history goes directly above it.
+        let viewport = Rect::new(0, height - 1, width, 1);
+        term.set_viewport_area(viewport);
+
+        // Create a long blockquote with a distinct prefix and enough text to wrap.
+        let mut line: Line<'static> = Line::from(vec![
+            "> ".into(),
+            "This is a long quoted line that should wrap".into(),
+        ]);
+        line = line.style(Color::Green);
+
+        let mut ansi: Vec<u8> = Vec::new();
+        insert_history_lines_to_writer(&mut term, &mut ansi, vec![line]);
+
+        // Parse and inspect the final screen buffer.
+        let mut parser = Parser::new(height, width, 0);
+        parser.process(&ansi);
+        let screen = parser.screen();
+
+        // Collect rows that are non-empty; these should correspond to our wrapped lines.
+        let mut non_empty_rows: Vec<u16> = Vec::new();
+        for row in 0..height {
+            let mut any = false;
+            for col in 0..width {
+                if let Some(cell) = screen.cell(row, col)
+                    && cell.has_contents()
+                    && cell.contents() != "\0"
+                    && cell.contents() != " "
+                {
+                    any = true;
+                    break;
+                }
+            }
+            if any {
+                non_empty_rows.push(row);
+            }
+        }
+
+        // Expect at least two rows due to wrapping.
+        assert!(
+            non_empty_rows.len() >= 2,
+            "expected wrapped output to span >=2 rows, got {non_empty_rows:?}",
+        );
+
+        // For each non-empty row, ensure all non-space cells are using a non-default fg color.
+        for row in non_empty_rows {
+            for col in 0..width {
+                if let Some(cell) = screen.cell(row, col) {
+                    let contents = cell.contents();
+                    if !contents.is_empty() && contents != " " {
+                        assert!(
+                            cell.fgcolor() != vt100::Color::Default,
+                            "expected non-default fg on row {row} col {col}, got {:?}",
+                            cell.fgcolor()
+                        );
+                    }
+                }
+            }
+        }
+    }
+
+    #[test]
+    fn vt100_colored_prefix_then_plain_text_resets_color() {
+        let width: u16 = 40;
+        let height: u16 = 6;
+        let backend = ratatui::backend::TestBackend::new(width, height);
+        let mut term = crate::custom_terminal::Terminal::with_options(backend).expect("terminal");
+        let viewport = Rect::new(0, height - 1, width, 1);
+        term.set_viewport_area(viewport);
+
+        // First span colored, rest plain.
+        let line: Line<'static> = Line::from(vec![
+            Span::styled("1. ", ratatui::style::Style::default().fg(Color::LightBlue)),
+            Span::raw("Hello world"),
+        ]);
+
+        let mut ansi: Vec<u8> = Vec::new();
+        insert_history_lines_to_writer(&mut term, &mut ansi, vec![line]);
+
+        let mut parser = Parser::new(height, width, 0);
+        parser.process(&ansi);
+        let screen = parser.screen();
+
+        // Find the first non-empty row; verify first three cells are colored, following cells default.
+        'rows: for row in 0..height {
+            let mut has_text = false;
+            for col in 0..width {
+                if let Some(cell) = screen.cell(row, col)
+                    && cell.has_contents()
+                    && cell.contents() != " "
+                {
+                    has_text = true;
+                    break;
+                }
+            }
+            if !has_text {
+                continue;
+            }
+
+            // Expect "1. Hello world" starting at col 0.
+            for col in 0..3 {
+                let cell = screen.cell(row, col).unwrap();
+                assert!(
+                    cell.fgcolor() != vt100::Color::Default,
+                    "expected colored prefix at col {col}, got {:?}",
+                    cell.fgcolor()
+                );
+            }
+            for col in 3..(3 + "Hello world".len() as u16) {
+                let cell = screen.cell(row, col).unwrap();
+                assert_eq!(
+                    cell.fgcolor(),
+                    vt100::Color::Default,
+                    "expected default color for plain text at col {col}, got {:?}",
+                    cell.fgcolor()
+                );
+            }
+            break 'rows;
+        }
+    }
+
+    #[test]
+    fn vt100_deep_nested_mixed_list_third_level_marker_is_colored() {
+        // Markdown with five levels (ordered → unordered → ordered → unordered → unordered).
+        let md = "1. First\n   - Second level\n     1. Third level (ordered)\n        - Fourth level (bullet)\n          - Fifth level to test indent consistency\n";
+        let text = render_markdown_text(md);
+        let lines: Vec<Line<'static>> = text.lines.clone();
+
+        let width: u16 = 60;
+        let height: u16 = 12;
+        let backend = ratatui::backend::TestBackend::new(width, height);
+        let mut term = crate::custom_terminal::Terminal::with_options(backend).expect("terminal");
+        let viewport = ratatui::layout::Rect::new(0, height - 1, width, 1);
+        term.set_viewport_area(viewport);
+
+        let mut ansi: Vec<u8> = Vec::new();
+        insert_history_lines_to_writer(&mut term, &mut ansi, lines);
+
+        let mut parser = Parser::new(height, width, 0);
+        parser.process(&ansi);
+        let screen = parser.screen();
+
+        // Reconstruct screen rows as strings to locate the 3rd level line.
+        let mut rows: Vec<String> = Vec::with_capacity(height as usize);
+        for row in 0..height {
+            let mut s = String::with_capacity(width as usize);
+            for col in 0..width {
+                if let Some(cell) = screen.cell(row, col) {
+                    if let Some(ch) = cell.contents().chars().next() {
+                        s.push(ch);
+                    } else {
+                        s.push(' ');
+                    }
+                } else {
+                    s.push(' ');
+                }
+            }
+            rows.push(s.trim_end().to_string());
+        }
+
+        let needle = "1. Third level (ordered)";
+        let row_idx = rows
+            .iter()
+            .position(|r| r.contains(needle))
+            .unwrap_or_else(|| {
+                panic!("expected to find row containing {needle:?}, have rows: {rows:?}")
+            });
+        let col_start = rows[row_idx].find(needle).unwrap() as u16; // column where '1' starts
+
+        // Verify that the numeric marker ("1.") at the third level is colored
+        // (non-default fg) and the content after the following space resets to default.
+        for c in [col_start, col_start + 1] {
+            let cell = screen.cell(row_idx as u16, c).unwrap();
+            assert!(
+                cell.fgcolor() != vt100::Color::Default,
+                "expected colored 3rd-level marker at row {row_idx} col {c}, got {:?}",
+                cell.fgcolor()
+            );
+        }
+        let content_col = col_start + 3; // skip '1', '.', and the space
+        if let Some(cell) = screen.cell(row_idx as u16, content_col) {
+            assert_eq!(
+                cell.fgcolor(),
+                vt100::Color::Default,
+                "expected default color for 3rd-level content at row {row_idx} col {content_col}, got {:?}",
+                cell.fgcolor()
+            );
+        }
+    }
 }

codex-rs/tui/src/lib.rs

@@ -45,6 +45,7 @@ pub mod insert_history;
 mod key_hint;
 pub mod live_wrap;
 mod markdown;
+mod markdown_render;
 mod markdown_stream;
 pub mod onboarding;
 mod pager_overlay;
@@ -61,11 +62,6 @@ mod user_approval_widget;
 mod version;
 mod wrapping;
 
-// Internal vt100-based replay tests live as a separate source file to keep them
-// close to the widget code. Include them in unit tests.
-#[cfg(test)]
-mod chatwidget_stream_tests;
-
 #[cfg(not(debug_assertions))]
 mod updates;
 
@@ -312,11 +308,7 @@ async fn run_ratatui_app(
         ..
     } = cli;
 
-    let auth_manager = AuthManager::shared(
-        config.codex_home.clone(),
-        config.preferred_auth_method,
-        config.responses_originator_header.clone(),
-    );
+    let auth_manager = AuthManager::shared(config.codex_home.clone(), config.preferred_auth_method);
     let login_status = get_login_status(&config);
     let should_show_onboarding =
         should_show_onboarding(login_status, &config, should_show_trust_screen);
@@ -400,11 +392,7 @@ fn get_login_status(config: &Config) -> LoginStatus {
         // Reading the OpenAI API key is an async operation because it may need
         // to refresh the token. Block on it.
         let codex_home = config.codex_home.clone();
-        match CodexAuth::from_codex_home(
-            &codex_home,
-            config.preferred_auth_method,
-            &config.responses_originator_header,
-        ) {
+        match CodexAuth::from_codex_home(&codex_home, config.preferred_auth_method) {
             Ok(Some(auth)) => LoginStatus::AuthMode(auth.mode),
             Ok(None) => LoginStatus::NotAuthenticated,
             Err(err) => {

codex-rs/tui/src/markdown.rs

@@ -1,8 +1,6 @@
-use crate::citation_regex::CITATION_REGEX;
 use codex_core::config::Config;
 use codex_core::config_types::UriBasedFileOpener;
 use ratatui::text::Line;
-use std::borrow::Cow;
 use std::path::Path;
 
 pub(crate) fn append_markdown(
@@ -19,238 +17,13 @@ fn append_markdown_with_opener_and_cwd(
     file_opener: UriBasedFileOpener,
     cwd: &Path,
 ) {
-    // Historically, we fed the entire `markdown_source` into the renderer in
-    // one pass. However, fenced code blocks sometimes lost leading whitespace
-    // when formatted by the markdown renderer/highlighter. To preserve code
-    // block content exactly, split the source into "text" and "code" segments:
-    // - Render non-code text through `tui_markdown` (with citation rewrite).
-    // - Render code block content verbatim as plain lines without additional
-    //   formatting, preserving leading spaces.
-    for seg in split_text_and_fences(markdown_source) {
-        match seg {
-            Segment::Text(s) => {
-                let processed = rewrite_file_citations(&s, file_opener, cwd);
-                let rendered = tui_markdown::from_str(&processed);
-                crate::render::line_utils::push_owned_lines(&rendered.lines, lines);
-            }
-            Segment::Code { content, .. } => {
-                // Emit the code content exactly as-is, line by line.
-                // We don't attempt syntax highlighting to avoid whitespace bugs.
-                for line in content.split_inclusive('\n') {
-                    // split_inclusive keeps the trailing \n; we want lines without it.
-                    let line = if let Some(stripped) = line.strip_suffix('\n') {
-                        stripped
-                    } else {
-                        line
-                    };
-                    let owned_line: Line<'static> = line.to_string().into();
-                    lines.push(owned_line);
-                }
-            }
-        }
-    }
-}
-
-/// Rewrites file citations in `src` into markdown hyperlinks using the
-/// provided `scheme` (`vscode`, `cursor`, etc.). The resulting URI follows the
-/// format expected by VS Code-compatible file openers:
-///
-/// ```text
-/// <scheme>://file<ABS_PATH>:<LINE>
-/// ```
-fn rewrite_file_citations<'a>(
-    src: &'a str,
-    file_opener: UriBasedFileOpener,
-    cwd: &Path,
-) -> Cow<'a, str> {
-    // Map enum values to the corresponding URI scheme strings.
-    let scheme: &str = match file_opener.get_scheme() {
-        Some(scheme) => scheme,
-        None => return Cow::Borrowed(src),
-    };
-
-    CITATION_REGEX.replace_all(src, |caps: &regex_lite::Captures<'_>| {
-        let file = &caps[1];
-        let start_line = &caps[2];
-
-        // Resolve the path against `cwd` when it is relative.
-        let absolute_path = {
-            let p = Path::new(file);
-            let absolute_path = if p.is_absolute() {
-                path_clean::clean(p)
-            } else {
-                path_clean::clean(cwd.join(p))
-            };
-            // VS Code expects forward slashes even on Windows because URIs use
-            // `/` as the path separator.
-            absolute_path.to_string_lossy().replace('\\', "/")
-        };
-
-        // Render as a normal markdown link so the downstream renderer emits
-        // the hyperlink escape sequence (when supported by the terminal).
-        //
-        // In practice, sometimes multiple citations for the same file, but with a
-        // different line number, are shown sequentially, so we:
-        // - include the line number in the label to disambiguate them
-        // - add a space after the link to make it easier to read
-        format!("[{file}:{start_line}]({scheme}://file{absolute_path}:{start_line}) ")
-    })
-}
-
-// use shared helper from `line_utils`
-
-// Minimal code block splitting.
-// - Recognizes fenced blocks opened by ``` or ~~~ (allowing leading whitespace).
-//   The opening fence may include a language string which we ignore.
-//   The closing fence must be on its own line (ignoring surrounding whitespace).
-// - Additionally recognizes indented code blocks that begin after a blank line
-//   with a line starting with at least 4 spaces or a tab, and continue for
-//   consecutive lines that are blank or also indented by >= 4 spaces or a tab.
-enum Segment {
-    Text(String),
-    Code {
-        _lang: Option<String>,
-        content: String,
-    },
-}
-
-fn split_text_and_fences(src: &str) -> Vec<Segment> {
-    let mut segments = Vec::new();
-    let mut curr_text = String::new();
-    #[derive(Copy, Clone, PartialEq)]
-    enum CodeMode {
-        None,
-        Fenced,
-        Indented,
-    }
-    let mut code_mode = CodeMode::None;
-    let mut fence_token = "";
-    let mut code_lang: Option<String> = None;
-    let mut code_content = String::new();
-    // We intentionally do not require a preceding blank line for indented code blocks,
-    // since streamed model output often omits it. This favors preserving indentation.
-
-    for line in src.split_inclusive('\n') {
-        let line_no_nl = line.strip_suffix('\n');
-        let trimmed_start = match line_no_nl {
-            Some(l) => l.trim_start(),
-            None => line.trim_start(),
-        };
-        if code_mode == CodeMode::None {
-            let open = if trimmed_start.starts_with("```") {
-                Some("```")
-            } else if trimmed_start.starts_with("~~~") {
-                Some("~~~")
-            } else {
-                None
-            };
-            if let Some(tok) = open {
-                // Flush pending text segment.
-                if !curr_text.is_empty() {
-                    segments.push(Segment::Text(curr_text.clone()));
-                    curr_text.clear();
-                }
-                fence_token = tok;
-                // Capture language after the token on this line (before newline).
-                let after = &trimmed_start[tok.len()..];
-                let lang = after.trim();
-                code_lang = if lang.is_empty() {
-                    None
-                } else {
-                    Some(lang.to_string())
-                };
-                code_mode = CodeMode::Fenced;
-                code_content.clear();
-                // Do not include the opening fence line in output.
-                continue;
-            }
-            // Check for start of an indented code block: only after a blank line
-            // (or at the beginning), and the line must start with >=4 spaces or a tab.
-            let raw_line = match line_no_nl {
-                Some(l) => l,
-                None => line,
-            };
-            let leading_spaces = raw_line.chars().take_while(|c| *c == ' ').count();
-            let starts_with_tab = raw_line.starts_with('\t');
-            // Consider any line that begins with >=4 spaces or a tab to start an
-            // indented code block. This favors preserving indentation even when a
-            // preceding blank line is omitted (common in streamed model output).
-            let starts_indented_code = (leading_spaces >= 4) || starts_with_tab;
-            if starts_indented_code {
-                // Flush pending text and begin an indented code block.
-                if !curr_text.is_empty() {
-                    segments.push(Segment::Text(curr_text.clone()));
-                    curr_text.clear();
-                }
-                code_mode = CodeMode::Indented;
-                code_content.clear();
-                code_content.push_str(line);
-                // Inside code now; do not treat this line as normal text.
-                continue;
-            }
-            // Normal text line.
-            curr_text.push_str(line);
-        } else {
-            match code_mode {
-                CodeMode::Fenced => {
-                    // inside fenced code: check for closing fence on its own line
-                    let trimmed = match line_no_nl {
-                        Some(l) => l.trim(),
-                        None => line.trim(),
-                    };
-                    if trimmed == fence_token {
-                        // End code block: emit segment without fences
-                        segments.push(Segment::Code {
-                            _lang: code_lang.take(),
-                            content: code_content.clone(),
-                        });
-                        code_content.clear();
-                        code_mode = CodeMode::None;
-                        fence_token = "";
-                        continue;
-                    }
-                    // Accumulate code content exactly as-is.
-                    code_content.push_str(line);
-                }
-                CodeMode::Indented => {
-                    // Continue while the line is blank, or starts with >=4 spaces, or a tab.
-                    let raw_line = match line_no_nl {
-                        Some(l) => l,
-                        None => line,
-                    };
-                    let is_blank = raw_line.trim().is_empty();
-                    let leading_spaces = raw_line.chars().take_while(|c| *c == ' ').count();
-                    let starts_with_tab = raw_line.starts_with('\t');
-                    if is_blank || leading_spaces >= 4 || starts_with_tab {
-                        code_content.push_str(line);
-                    } else {
-                        // Close the indented code block and reprocess this line as normal text.
-                        segments.push(Segment::Code {
-                            _lang: None,
-                            content: code_content.clone(),
-                        });
-                        code_content.clear();
-                        code_mode = CodeMode::None;
-                        // Now handle current line as text.
-                        curr_text.push_str(line);
-                    }
-                }
-                CodeMode::None => unreachable!(),
-            }
-        }
-    }
-
-    if code_mode != CodeMode::None {
-        // Unterminated code fence: treat accumulated content as a code segment.
-        segments.push(Segment::Code {
-            _lang: code_lang.take(),
-            content: code_content.clone(),
-        });
-    } else if !curr_text.is_empty() {
-        segments.push(Segment::Text(curr_text.clone()));
-    }
-
-    segments
+    // Render via pulldown-cmark and rewrite citations during traversal (outside code blocks).
+    let rendered = crate::markdown_render::render_markdown_text_with_citations(
+        markdown_source,
+        file_opener.get_scheme(),
+        cwd,
+    );
+    crate::render::line_utils::push_owned_lines(&rendered.lines, lines);
 }
 
 #[cfg(test)]
@@ -258,88 +31,6 @@ mod tests {
     use super::*;
     use pretty_assertions::assert_eq;
 
-    #[test]
-    fn citation_is_rewritten_with_absolute_path() {
-        let markdown = "See 【F:/src/main.rs†L42-L50】 for details.";
-        let cwd = Path::new("/workspace");
-        let result = rewrite_file_citations(markdown, UriBasedFileOpener::VsCode, cwd);
-
-        assert_eq!(
-            "See [/src/main.rs:42](vscode://file/src/main.rs:42)  for details.",
-            result
-        );
-    }
-
-    #[test]
-    fn citation_is_rewritten_with_relative_path() {
-        let markdown = "Refer to 【F:lib/mod.rs†L5】 here.";
-        let cwd = Path::new("/home/user/project");
-        let result = rewrite_file_citations(markdown, UriBasedFileOpener::Windsurf, cwd);
-
-        assert_eq!(
-            "Refer to [lib/mod.rs:5](windsurf://file/home/user/project/lib/mod.rs:5)  here.",
-            result
-        );
-    }
-
-    #[test]
-    fn citation_followed_by_space_so_they_do_not_run_together() {
-        let markdown = "References on lines 【F:src/foo.rs†L24】【F:src/foo.rs†L42】";
-        let cwd = Path::new("/home/user/project");
-        let result = rewrite_file_citations(markdown, UriBasedFileOpener::VsCode, cwd);
-
-        assert_eq!(
-            "References on lines [src/foo.rs:24](vscode://file/home/user/project/src/foo.rs:24) [src/foo.rs:42](vscode://file/home/user/project/src/foo.rs:42) ",
-            result
-        );
-    }
-
-    #[test]
-    fn citation_unchanged_without_file_opener() {
-        let markdown = "Look at 【F:file.rs†L1】.";
-        let cwd = Path::new("/");
-        let unchanged = rewrite_file_citations(markdown, UriBasedFileOpener::VsCode, cwd);
-        // The helper itself always rewrites – this test validates behaviour of
-        // append_markdown when `file_opener` is None.
-        let mut out = Vec::new();
-        append_markdown_with_opener_and_cwd(markdown, &mut out, UriBasedFileOpener::None, cwd);
-        // Convert lines back to string for comparison.
-        let rendered: String = out
-            .iter()
-            .flat_map(|l| l.spans.iter())
-            .map(|s| s.content.clone())
-            .collect::<Vec<_>>()
-            .join("");
-        assert_eq!(markdown, rendered);
-        // Ensure helper rewrites.
-        assert_ne!(markdown, unchanged);
-    }
-
-    #[test]
-    fn fenced_code_blocks_preserve_leading_whitespace() {
-        let src = "```\n  indented\n\t\twith tabs\n    four spaces\n```\n";
-        let cwd = Path::new("/");
-        let mut out = Vec::new();
-        append_markdown_with_opener_and_cwd(src, &mut out, UriBasedFileOpener::None, cwd);
-        let rendered: Vec<String> = out
-            .iter()
-            .map(|l| {
-                l.spans
-                    .iter()
-                    .map(|s| s.content.clone())
-                    .collect::<String>()
-            })
-            .collect();
-        assert_eq!(
-            rendered,
-            vec![
-                "  indented".to_string(),
-                "\t\twith tabs".to_string(),
-                "    four spaces".to_string()
-            ]
-        );
-    }
-
     #[test]
     fn citations_not_rewritten_inside_code_blocks() {
         let src = "Before 【F:/x.rs†L1】\n```\nInside 【F:/x.rs†L2】\n```\nAfter 【F:/x.rs†L3】\n";
@@ -355,19 +46,31 @@ mod tests {
                     .collect::<String>()
             })
             .collect();
-        // Expect first and last lines rewritten, middle line unchanged.
-        assert!(rendered[0].contains("vscode://file"));
-        assert_eq!(rendered[1], "Inside 【F:/x.rs†L2】");
-        assert!(matches!(rendered.last(), Some(s) if s.contains("vscode://file")));
+        // Expect a line containing the inside text unchanged.
+        assert!(rendered.iter().any(|s| s.contains("Inside 【F:/x.rs†L2】")));
+        // And first/last sections rewritten.
+        assert!(
+            rendered
+                .first()
+                .map(|s| s.contains("vscode://file"))
+                .unwrap_or(false)
+        );
+        assert!(
+            rendered
+                .last()
+                .map(|s| s.contains("vscode://file"))
+                .unwrap_or(false)
+        );
     }
 
     #[test]
     fn indented_code_blocks_preserve_leading_whitespace() {
-        let src = "Before\n    code 1\n\tcode with tab\n        code 2\nAfter\n";
+        // Basic sanity: indented code with surrounding blank lines should produce the indented line.
+        let src = "Before\n\n    code 1\n\nAfter\n";
         let cwd = Path::new("/");
         let mut out = Vec::new();
         append_markdown_with_opener_and_cwd(src, &mut out, UriBasedFileOpener::None, cwd);
-        let rendered: Vec<String> = out
+        let lines: Vec<String> = out
             .iter()
             .map(|l| {
                 l.spans
@@ -376,16 +79,7 @@ mod tests {
                     .collect::<String>()
             })
             .collect();
-        assert_eq!(
-            rendered,
-            vec![
-                "Before".to_string(),
-                "    code 1".to_string(),
-                "\tcode with tab".to_string(),
-                "        code 2".to_string(),
-                "After".to_string()
-            ]
-        );
+        assert_eq!(lines, vec!["Before", "", "    code 1", "", "After"]);
     }
 
     #[test]
@@ -403,11 +97,17 @@ mod tests {
                     .collect::<String>()
             })
             .collect();
-        // Expect first and last lines rewritten, and the indented code line present
-        // unchanged (citations inside not rewritten). We do not assert on blank
-        // separator lines since the markdown renderer may normalize them.
-        assert!(rendered.iter().any(|s| s.contains("vscode://file")));
-        assert!(rendered.iter().any(|s| s == "    Inside 【F:/x.rs†L2】"));
+        assert!(
+            rendered
+                .iter()
+                .any(|s| s.contains("Start") && s.contains("vscode://file"))
+        );
+        assert!(
+            rendered
+                .iter()
+                .any(|s| s.contains("End") && s.contains("vscode://file"))
+        );
+        assert!(rendered.iter().any(|s| s.contains("Inside 【F:/x.rs†L2】")));
     }
 
     #[test]
@@ -435,27 +135,6 @@ mod tests {
         );
     }
 
-    #[test]
-    fn tui_markdown_splits_ordered_marker_and_text() {
-        // With marker and content on the same line, tui_markdown keeps it as one line
-        // even in the surrounding section context.
-        let rendered = tui_markdown::from_str("Loose vs. tight list items:\n1. Tight item\n");
-        let lines: Vec<String> = rendered
-            .lines
-            .iter()
-            .map(|l| {
-                l.spans
-                    .iter()
-                    .map(|s| s.content.clone())
-                    .collect::<String>()
-            })
-            .collect();
-        assert!(
-            lines.iter().any(|w| w == "1. Tight item"),
-            "expected single line '1. Tight item' in context: {lines:?}"
-        );
-    }
-
     #[test]
     fn append_markdown_matches_tui_markdown_for_ordered_item() {
         use codex_core::config_types::UriBasedFileOpener;
@@ -480,72 +159,6 @@ mod tests {
         assert_eq!(lines, vec!["1. Tight item".to_string()]);
     }
 
-    #[test]
-    fn tui_markdown_shape_for_loose_tight_section() {
-        // Use the exact source from the session deltas used in tests.
-        let source = r#"
-Loose vs. tight list items:
-1. Tight item
-2. Another tight item
-
-3.
-   Loose item
-"#;
-
-        let rendered = tui_markdown::from_str(source);
-        let lines: Vec<String> = rendered
-            .lines
-            .iter()
-            .map(|l| {
-                l.spans
-                    .iter()
-                    .map(|s| s.content.clone())
-                    .collect::<String>()
-            })
-            .collect();
-        // Join into a single string and assert the exact shape we observe
-        // from tui_markdown in this larger context (marker and content split).
-        let joined = {
-            let mut s = String::new();
-            for (i, l) in lines.iter().enumerate() {
-                s.push_str(l);
-                if i + 1 < lines.len() {
-                    s.push('\n');
-                }
-            }
-            s
-        };
-        let expected = r#"Loose vs. tight list items:
-
-1. 
-Tight item
-2. 
-Another tight item
-3. 
-Loose item"#;
-        assert_eq!(
-            joined, expected,
-            "unexpected tui_markdown shape: {joined:?}"
-        );
-    }
-
-    #[test]
-    fn split_text_and_fences_keeps_ordered_list_line_as_text() {
-        // No fences here; expect a single Text segment containing the full input.
-        let src = "Loose vs. tight list items:\n1. Tight item\n";
-        let segs = super::split_text_and_fences(src);
-        assert_eq!(
-            segs.len(),
-            1,
-            "expected single text segment, got {}",
-            segs.len()
-        );
-        match &segs[0] {
-            super::Segment::Text(s) => assert_eq!(s, src),
-            _ => panic!("expected Text segment for non-fence input"),
-        }
-    }
-
     #[test]
     fn append_markdown_keeps_ordered_list_line_unsplit_in_context() {
         use codex_core::config_types::UriBasedFileOpener;

codex-rs/tui/src/markdown_render.rs

@@ -0,0 +1,566 @@
+use crate::citation_regex::CITATION_REGEX;
+use pulldown_cmark::CodeBlockKind;
+use pulldown_cmark::CowStr;
+use pulldown_cmark::Event;
+use pulldown_cmark::HeadingLevel;
+use pulldown_cmark::Options;
+use pulldown_cmark::Parser;
+use pulldown_cmark::Tag;
+use pulldown_cmark::TagEnd;
+use ratatui::style::Style;
+use ratatui::style::Stylize;
+use ratatui::text::Line;
+use ratatui::text::Span;
+use ratatui::text::Text;
+use std::borrow::Cow;
+use std::path::Path;
+
+#[derive(Clone, Debug)]
+struct IndentContext {
+    prefix: Vec<Span<'static>>,
+    marker: Option<Vec<Span<'static>>>,
+    is_list: bool,
+}
+
+impl IndentContext {
+    fn new(prefix: Vec<Span<'static>>, marker: Option<Vec<Span<'static>>>, is_list: bool) -> Self {
+        Self {
+            prefix,
+            marker,
+            is_list,
+        }
+    }
+}
+
+#[allow(dead_code)]
+pub(crate) fn render_markdown_text(input: &str) -> Text<'static> {
+    let mut options = Options::empty();
+    options.insert(Options::ENABLE_STRIKETHROUGH);
+    let parser = Parser::new_ext(input, options);
+    let mut w = Writer::new(parser, None, None);
+    w.run();
+    w.text
+}
+
+pub(crate) fn render_markdown_text_with_citations(
+    input: &str,
+    scheme: Option<&str>,
+    cwd: &Path,
+) -> Text<'static> {
+    let mut options = Options::empty();
+    options.insert(Options::ENABLE_STRIKETHROUGH);
+    let parser = Parser::new_ext(input, options);
+    let mut w = Writer::new(
+        parser,
+        scheme.map(|s| s.to_string()),
+        Some(cwd.to_path_buf()),
+    );
+    w.run();
+    w.text
+}
+
+struct Writer<'a, I>
+where
+    I: Iterator<Item = Event<'a>>,
+{
+    iter: I,
+    text: Text<'static>,
+    inline_styles: Vec<Style>,
+    indent_stack: Vec<IndentContext>,
+    list_indices: Vec<Option<u64>>,
+    link: Option<String>,
+    needs_newline: bool,
+    pending_marker_line: bool,
+    in_paragraph: bool,
+    scheme: Option<String>,
+    cwd: Option<std::path::PathBuf>,
+    in_code_block: bool,
+}
+
+impl<'a, I> Writer<'a, I>
+where
+    I: Iterator<Item = Event<'a>>,
+{
+    fn new(iter: I, scheme: Option<String>, cwd: Option<std::path::PathBuf>) -> Self {
+        Self {
+            iter,
+            text: Text::default(),
+            inline_styles: Vec::new(),
+            indent_stack: Vec::new(),
+            list_indices: Vec::new(),
+            link: None,
+            needs_newline: false,
+            pending_marker_line: false,
+            in_paragraph: false,
+            scheme,
+            cwd,
+            in_code_block: false,
+        }
+    }
+
+    fn run(&mut self) {
+        while let Some(ev) = self.iter.next() {
+            self.handle_event(ev);
+        }
+    }
+
+    fn handle_event(&mut self, event: Event<'a>) {
+        match event {
+            Event::Start(tag) => self.start_tag(tag),
+            Event::End(tag) => self.end_tag(tag),
+            Event::Text(text) => self.text(text),
+            Event::Code(code) => self.code(code),
+            Event::SoftBreak => self.soft_break(),
+            Event::HardBreak => self.hard_break(),
+            Event::Rule => {
+                if !self.text.lines.is_empty() {
+                    self.push_blank_line();
+                }
+                self.push_line(Line::from("———"));
+                self.needs_newline = true;
+            }
+            Event::Html(html) => self.html(html, false),
+            Event::InlineHtml(html) => self.html(html, true),
+            Event::FootnoteReference(_) => {}
+            Event::TaskListMarker(_) => {}
+        }
+    }
+
+    fn start_tag(&mut self, tag: Tag<'a>) {
+        match tag {
+            Tag::Paragraph => self.start_paragraph(),
+            Tag::Heading { level, .. } => self.start_heading(level),
+            Tag::BlockQuote => self.start_blockquote(),
+            Tag::CodeBlock(kind) => {
+                let indent = match kind {
+                    CodeBlockKind::Fenced(_) => None,
+                    CodeBlockKind::Indented => Some(Span::from(" ".repeat(4))),
+                };
+                let lang = match kind {
+                    CodeBlockKind::Fenced(lang) => Some(lang.to_string()),
+                    CodeBlockKind::Indented => None,
+                };
+                self.start_codeblock(lang, indent)
+            }
+            Tag::List(start) => self.start_list(start),
+            Tag::Item => self.start_item(),
+            Tag::Emphasis => self.push_inline_style(Style::new().italic()),
+            Tag::Strong => self.push_inline_style(Style::new().bold()),
+            Tag::Strikethrough => self.push_inline_style(Style::new().crossed_out()),
+            Tag::Link { dest_url, .. } => self.push_link(dest_url.to_string()),
+            Tag::HtmlBlock
+            | Tag::FootnoteDefinition(_)
+            | Tag::Table(_)
+            | Tag::TableHead
+            | Tag::TableRow
+            | Tag::TableCell
+            | Tag::Image { .. }
+            | Tag::MetadataBlock(_) => {}
+        }
+    }
+
+    fn end_tag(&mut self, tag: TagEnd) {
+        match tag {
+            TagEnd::Paragraph => self.end_paragraph(),
+            TagEnd::Heading(_) => self.end_heading(),
+            TagEnd::BlockQuote => self.end_blockquote(),
+            TagEnd::CodeBlock => self.end_codeblock(),
+            TagEnd::List(_) => self.end_list(),
+            TagEnd::Item => {
+                self.indent_stack.pop();
+                self.pending_marker_line = false;
+            }
+            TagEnd::Emphasis | TagEnd::Strong | TagEnd::Strikethrough => self.pop_inline_style(),
+            TagEnd::Link => self.pop_link(),
+            TagEnd::HtmlBlock
+            | TagEnd::FootnoteDefinition
+            | TagEnd::Table
+            | TagEnd::TableHead
+            | TagEnd::TableRow
+            | TagEnd::TableCell
+            | TagEnd::Image
+            | TagEnd::MetadataBlock(_) => {}
+        }
+    }
+
+    fn start_paragraph(&mut self) {
+        if self.needs_newline {
+            self.push_blank_line();
+        }
+        self.push_line(Line::default());
+        self.needs_newline = false;
+        self.in_paragraph = true;
+    }
+
+    fn end_paragraph(&mut self) {
+        self.needs_newline = true;
+        self.in_paragraph = false;
+        self.pending_marker_line = false;
+    }
+
+    fn start_heading(&mut self, level: HeadingLevel) {
+        if self.needs_newline {
+            self.push_line(Line::default());
+            self.needs_newline = false;
+        }
+        let heading_style = match level {
+            HeadingLevel::H1 => Style::new().bold().underlined(),
+            HeadingLevel::H2 => Style::new().bold(),
+            HeadingLevel::H3 => Style::new().bold().italic(),
+            HeadingLevel::H4 => Style::new().italic(),
+            HeadingLevel::H5 => Style::new().italic(),
+            HeadingLevel::H6 => Style::new().italic(),
+        };
+        let content = format!("{} ", "#".repeat(level as usize));
+        self.push_line(Line::from(vec![Span::styled(content, heading_style)]));
+        self.push_inline_style(heading_style);
+        self.needs_newline = false;
+    }
+
+    fn end_heading(&mut self) {
+        self.needs_newline = true;
+        self.pop_inline_style();
+    }
+
+    fn start_blockquote(&mut self) {
+        if self.needs_newline {
+            self.push_blank_line();
+            self.needs_newline = false;
+        }
+        self.indent_stack
+            .push(IndentContext::new(vec![Span::from("> ")], None, false));
+    }
+
+    fn end_blockquote(&mut self) {
+        self.indent_stack.pop();
+        self.needs_newline = true;
+    }
+
+    fn text(&mut self, text: CowStr<'a>) {
+        if self.pending_marker_line {
+            self.push_line(Line::default());
+        }
+        self.pending_marker_line = false;
+        if self.in_code_block
+            && !self.needs_newline
+            && self
+                .text
+                .lines
+                .last()
+                .map(|line| !line.spans.is_empty())
+                .unwrap_or(false)
+        {
+            self.push_line(Line::default());
+        }
+        for (i, line) in text.lines().enumerate() {
+            if self.needs_newline {
+                self.push_line(Line::default());
+                self.needs_newline = false;
+            }
+            if i > 0 {
+                self.push_line(Line::default());
+            }
+            let mut content = line.to_string();
+            if !self.in_code_block
+                && let (Some(scheme), Some(cwd)) = (&self.scheme, &self.cwd)
+            {
+                let cow = rewrite_file_citations_with_scheme(&content, Some(scheme.as_str()), cwd);
+                if let std::borrow::Cow::Owned(s) = cow {
+                    content = s;
+                }
+            }
+            let span = Span::styled(
+                content,
+                self.inline_styles.last().copied().unwrap_or_default(),
+            );
+            self.push_span(span);
+        }
+        self.needs_newline = false;
+    }
+
+    fn code(&mut self, code: CowStr<'a>) {
+        if self.pending_marker_line {
+            self.push_line(Line::default());
+            self.pending_marker_line = false;
+        }
+        let span = Span::from(code.into_string()).dim();
+        self.push_span(span);
+    }
+
+    fn html(&mut self, html: CowStr<'a>, inline: bool) {
+        self.pending_marker_line = false;
+        for (i, line) in html.lines().enumerate() {
+            if self.needs_newline {
+                self.push_line(Line::default());
+                self.needs_newline = false;
+            }
+            if i > 0 {
+                self.push_line(Line::default());
+            }
+            let style = self.inline_styles.last().copied().unwrap_or_default();
+            self.push_span(Span::styled(line.to_string(), style));
+        }
+        self.needs_newline = !inline;
+    }
+
+    fn hard_break(&mut self) {
+        self.push_line(Line::default());
+    }
+
+    fn soft_break(&mut self) {
+        self.push_line(Line::default());
+    }
+
+    fn start_list(&mut self, index: Option<u64>) {
+        if self.list_indices.is_empty() && self.needs_newline {
+            self.push_line(Line::default());
+        }
+        self.list_indices.push(index);
+    }
+
+    fn end_list(&mut self) {
+        self.list_indices.pop();
+        self.needs_newline = true;
+    }
+
+    fn start_item(&mut self) {
+        self.pending_marker_line = true;
+        let depth = self.list_indices.len();
+        let is_ordered = self
+            .list_indices
+            .last()
+            .map(|index| index.is_some())
+            .unwrap_or(false);
+        let width = depth * 4 - 3;
+        let marker = if let Some(last_index) = self.list_indices.last_mut() {
+            match last_index {
+                None => Some(vec![Span::from(" ".repeat(width - 1) + "- ")]),
+                Some(index) => {
+                    *index += 1;
+                    Some(vec![format!("{:width$}. ", *index - 1).light_blue()])
+                }
+            }
+        } else {
+            None
+        };
+        let indent_prefix = if depth == 0 {
+            Vec::new()
+        } else {
+            let indent_len = if is_ordered { width + 2 } else { width + 1 };
+            vec![Span::from(" ".repeat(indent_len))]
+        };
+        self.indent_stack
+            .push(IndentContext::new(indent_prefix, marker, true));
+        self.needs_newline = false;
+    }
+
+    fn start_codeblock(&mut self, _lang: Option<String>, indent: Option<Span<'static>>) {
+        if !self.text.lines.is_empty() {
+            self.push_blank_line();
+        }
+        self.in_code_block = true;
+        self.indent_stack.push(IndentContext::new(
+            vec![indent.unwrap_or_default()],
+            None,
+            false,
+        ));
+        // let opener = match lang {
+        //     Some(l) if !l.is_empty() => format!("```{l}"),
+        //     _ => "```".to_string(),
+        // };
+        // self.push_line(opener.into());
+        self.needs_newline = true;
+    }
+
+    fn end_codeblock(&mut self) {
+        // self.push_line("```".into());
+        self.needs_newline = true;
+        self.in_code_block = false;
+        self.indent_stack.pop();
+    }
+
+    fn push_inline_style(&mut self, style: Style) {
+        let current = self.inline_styles.last().copied().unwrap_or_default();
+        let merged = current.patch(style);
+        self.inline_styles.push(merged);
+    }
+
+    fn pop_inline_style(&mut self) {
+        self.inline_styles.pop();
+    }
+
+    fn push_link(&mut self, dest_url: String) {
+        self.link = Some(dest_url);
+    }
+
+    fn pop_link(&mut self) {
+        if let Some(link) = self.link.take() {
+            self.push_span(" (".into());
+            self.push_span(link.cyan().underlined());
+            self.push_span(")".into());
+        }
+    }
+
+    fn push_line(&mut self, line: Line<'static>) {
+        let mut line = line;
+        let was_pending = self.pending_marker_line;
+        let mut spans = self.current_prefix_spans();
+        spans.append(&mut line.spans);
+        let blockquote_active = self
+            .indent_stack
+            .iter()
+            .any(|ctx| ctx.prefix.iter().any(|s| s.content.contains('>')));
+        let style = if blockquote_active {
+            Style::new().green()
+        } else {
+            line.style
+        };
+        self.text.lines.push(Line::from_iter(spans).style(style));
+        if was_pending {
+            self.pending_marker_line = false;
+        }
+    }
+
+    fn push_span(&mut self, span: Span<'static>) {
+        if let Some(last) = self.text.lines.last_mut() {
+            last.push_span(span);
+        } else {
+            self.push_line(Line::from(vec![span]));
+        }
+    }
+
+    fn push_blank_line(&mut self) {
+        if self.indent_stack.iter().all(|ctx| ctx.is_list) {
+            self.text.lines.push(Line::default());
+        } else {
+            self.push_line(Line::default());
+        }
+    }
+
+    fn current_prefix_spans(&self) -> Vec<Span<'static>> {
+        let mut prefix: Vec<Span<'static>> = Vec::new();
+        let last_marker_index = if self.pending_marker_line {
+            self.indent_stack
+                .iter()
+                .enumerate()
+                .rev()
+                .find_map(|(i, ctx)| if ctx.marker.is_some() { Some(i) } else { None })
+        } else {
+            None
+        };
+        let last_list_index = self.indent_stack.iter().rposition(|ctx| ctx.is_list);
+
+        for (i, ctx) in self.indent_stack.iter().enumerate() {
+            if self.pending_marker_line {
+                if Some(i) == last_marker_index
+                    && let Some(marker) = &ctx.marker
+                {
+                    prefix.extend(marker.iter().cloned());
+                    continue;
+                }
+                if ctx.is_list && last_marker_index.is_some_and(|idx| idx > i) {
+                    continue;
+                }
+            } else if ctx.is_list && Some(i) != last_list_index {
+                continue;
+            }
+            prefix.extend(ctx.prefix.iter().cloned());
+        }
+
+        prefix
+    }
+}
+
+pub(crate) fn rewrite_file_citations_with_scheme<'a>(
+    src: &'a str,
+    scheme_opt: Option<&str>,
+    cwd: &Path,
+) -> Cow<'a, str> {
+    let scheme: &str = match scheme_opt {
+        Some(s) => s,
+        None => return Cow::Borrowed(src),
+    };
+
+    CITATION_REGEX.replace_all(src, |caps: &regex_lite::Captures<'_>| {
+        let file = &caps[1];
+        let start_line = &caps[2];
+
+        // Resolve the path against `cwd` when it is relative.
+        let absolute_path = {
+            let p = Path::new(file);
+            let absolute_path = if p.is_absolute() {
+                path_clean::clean(p)
+            } else {
+                path_clean::clean(cwd.join(p))
+            };
+            // VS Code expects forward slashes even on Windows because URIs use
+            // `/` as the path separator.
+            absolute_path.to_string_lossy().replace('\\', "/")
+        };
+
+        // Render as a normal markdown link so the downstream renderer emits
+        // the hyperlink escape sequence (when supported by the terminal).
+        //
+        // In practice, sometimes multiple citations for the same file, but with a
+        // different line number, are shown sequentially, so we:
+        // - include the line number in the label to disambiguate them
+        // - add a space after the link to make it easier to read
+        format!("[{file}:{start_line}]({scheme}://file{absolute_path}:{start_line}) ")
+    })
+}
+
+#[cfg(test)]
+mod markdown_render_tests {
+    include!("markdown_render_tests.rs");
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use pretty_assertions::assert_eq;
+
+    #[test]
+    fn citation_is_rewritten_with_absolute_path() {
+        let markdown = "See 【F:/src/main.rs†L42-L50】 for details.";
+        let cwd = Path::new("/workspace");
+        let result = rewrite_file_citations_with_scheme(markdown, Some("vscode"), cwd);
+
+        assert_eq!(
+            "See [/src/main.rs:42](vscode://file/src/main.rs:42)  for details.",
+            result
+        );
+    }
+
+    #[test]
+    fn citation_followed_by_space_so_they_do_not_run_together() {
+        let markdown = "References on lines 【F:src/foo.rs†L24】【F:src/foo.rs†L42】";
+        let cwd = Path::new("/home/user/project");
+        let result = rewrite_file_citations_with_scheme(markdown, Some("vscode"), cwd);
+
+        assert_eq!(
+            "References on lines [src/foo.rs:24](vscode://file/home/user/project/src/foo.rs:24) [src/foo.rs:42](vscode://file/home/user/project/src/foo.rs:42) ",
+            result
+        );
+    }
+
+    #[test]
+    fn citation_unchanged_without_file_opener() {
+        let markdown = "Look at 【F:file.rs†L1】.";
+        let cwd = Path::new("/");
+        let unchanged = rewrite_file_citations_with_scheme(markdown, Some("vscode"), cwd);
+        // The helper itself always rewrites – this test validates behaviour of
+        // append_markdown when `file_opener` is None.
+        let rendered = render_markdown_text_with_citations(markdown, None, cwd);
+        // Convert lines back to string for comparison.
+        let rendered: String = rendered
+            .lines
+            .iter()
+            .flat_map(|l| l.spans.iter())
+            .map(|s| s.content.clone())
+            .collect::<Vec<_>>()
+            .join("");
+        assert_eq!(markdown, rendered);
+        // Ensure helper rewrites.
+        assert_ne!(markdown, unchanged);
+    }
+}

codex-rs/tui/src/markdown_render_tests.rs

@@ -0,0 +1,995 @@
+use pretty_assertions::assert_eq;
+use ratatui::style::Stylize;
+use ratatui::text::Line;
+use ratatui::text::Span;
+use ratatui::text::Text;
+
+use crate::markdown_render::render_markdown_text;
+use insta::assert_snapshot;
+
+#[test]
+fn empty() {
+    assert_eq!(render_markdown_text(""), Text::default());
+}
+
+#[test]
+fn paragraph_single() {
+    assert_eq!(
+        render_markdown_text("Hello, world!"),
+        Text::from("Hello, world!")
+    );
+}
+
+#[test]
+fn paragraph_soft_break() {
+    assert_eq!(
+        render_markdown_text("Hello\nWorld"),
+        Text::from_iter(["Hello", "World"])
+    );
+}
+
+#[test]
+fn paragraph_multiple() {
+    assert_eq!(
+        render_markdown_text("Paragraph 1\n\nParagraph 2"),
+        Text::from_iter(["Paragraph 1", "", "Paragraph 2"])
+    );
+}
+
+#[test]
+fn headings() {
+    let md = "# Heading 1\n## Heading 2\n### Heading 3\n#### Heading 4\n##### Heading 5\n###### Heading 6\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["# ".bold().underlined(), "Heading 1".bold().underlined()]),
+        Line::default(),
+        Line::from_iter(["## ".bold(), "Heading 2".bold()]),
+        Line::default(),
+        Line::from_iter(["### ".bold().italic(), "Heading 3".bold().italic()]),
+        Line::default(),
+        Line::from_iter(["#### ".italic(), "Heading 4".italic()]),
+        Line::default(),
+        Line::from_iter(["##### ".italic(), "Heading 5".italic()]),
+        Line::default(),
+        Line::from_iter(["###### ".italic(), "Heading 6".italic()]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn blockquote_single() {
+    let text = render_markdown_text("> Blockquote");
+    let expected = Text::from(Line::from_iter(["> ", "Blockquote"]).green());
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn blockquote_soft_break() {
+    // Soft break via lazy continuation should render as a new line in blockquotes.
+    let text = render_markdown_text("> This is a blockquote\nwith a soft break\n");
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(
+        lines,
+        vec![
+            "> This is a blockquote".to_string(),
+            "> with a soft break".to_string()
+        ]
+    );
+}
+
+#[test]
+fn blockquote_multiple_with_break() {
+    let text = render_markdown_text("> Blockquote 1\n\n> Blockquote 2\n");
+    let expected = Text::from_iter([
+        Line::from_iter(["> ", "Blockquote 1"]).green(),
+        Line::default(),
+        Line::from_iter(["> ", "Blockquote 2"]).green(),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn blockquote_three_paragraphs_short_lines() {
+    let md = "> one\n>\n> two\n>\n> three\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["> ", "one"]).green(),
+        Line::from_iter(["> "]).green(),
+        Line::from_iter(["> ", "two"]).green(),
+        Line::from_iter(["> "]).green(),
+        Line::from_iter(["> ", "three"]).green(),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn blockquote_nested_two_levels() {
+    let md = "> Level 1\n>> Level 2\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["> ", "Level 1"]).green(),
+        Line::from_iter(["> "]).green(),
+        Line::from_iter(["> ", "> ", "Level 2"]).green(),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn blockquote_with_list_items() {
+    let md = "> - item 1\n> - item 2\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["> ", "- ", "item 1"]).green(),
+        Line::from_iter(["> ", "- ", "item 2"]).green(),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn blockquote_with_ordered_list() {
+    let md = "> 1. first\n> 2. second\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(vec![
+            Span::from("> "),
+            "1. ".light_blue(),
+            Span::from("first"),
+        ])
+        .green(),
+        Line::from_iter(vec![
+            Span::from("> "),
+            "2. ".light_blue(),
+            Span::from("second"),
+        ])
+        .green(),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn blockquote_list_then_nested_blockquote() {
+    let md = "> - parent\n>   > child\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["> ", "- ", "parent"]).green(),
+        Line::from_iter(["> ", "  ", "> ", "child"]).green(),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn list_item_with_inline_blockquote_on_same_line() {
+    let md = "1. > quoted\n";
+    let text = render_markdown_text(md);
+    let mut lines = text.lines.iter();
+    let first = lines.next().expect("one line");
+    // Expect content to include the ordered marker, a space, "> ", and the text
+    let s: String = first.spans.iter().map(|sp| sp.content.clone()).collect();
+    assert_eq!(s, "1. > quoted");
+}
+
+#[test]
+fn blockquote_surrounded_by_blank_lines() {
+    let md = "foo\n\n> bar\n\nbaz\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(
+        lines,
+        vec![
+            "foo".to_string(),
+            "".to_string(),
+            "> bar".to_string(),
+            "".to_string(),
+            "baz".to_string(),
+        ]
+    );
+}
+
+#[test]
+fn blockquote_in_ordered_list_on_next_line() {
+    // Blockquote begins on a new line within an ordered list item; it should
+    // render inline on the same marker line.
+    let md = "1.\n   > quoted\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["1. > quoted".to_string()]);
+}
+
+#[test]
+fn blockquote_in_unordered_list_on_next_line() {
+    // Blockquote begins on a new line within an unordered list item; it should
+    // render inline on the same marker line.
+    let md = "-\n  > quoted\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["- > quoted".to_string()]);
+}
+
+#[test]
+fn blockquote_two_paragraphs_inside_ordered_list_has_blank_line() {
+    // Two blockquote paragraphs inside a list item should be separated by a blank line.
+    let md = "1.\n   > para 1\n   >\n   > para 2\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(
+        lines,
+        vec![
+            "1. > para 1".to_string(),
+            "   > ".to_string(),
+            "   > para 2".to_string(),
+        ],
+        "expected blockquote content to stay aligned after list marker"
+    );
+}
+
+#[test]
+fn blockquote_inside_nested_list() {
+    let md = "1. A\n    - B\n      > inner\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["1. A", "    - B", "      > inner"]);
+}
+
+#[test]
+fn list_item_text_then_blockquote() {
+    let md = "1. before\n   > quoted\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["1. before", "   > quoted"]);
+}
+
+#[test]
+fn list_item_blockquote_then_text() {
+    let md = "1.\n   > quoted\n   after\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["1. > quoted", "   > after"]);
+}
+
+#[test]
+fn list_item_text_blockquote_text() {
+    let md = "1. before\n   > quoted\n   after\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["1. before", "   > quoted", "   > after"]);
+}
+
+#[test]
+fn blockquote_with_heading_and_paragraph() {
+    let md = "> # Heading\n> paragraph text\n";
+    let text = render_markdown_text(md);
+    // Validate on content shape; styling is handled elsewhere
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(
+        lines,
+        vec![
+            "> # Heading".to_string(),
+            "> ".to_string(),
+            "> paragraph text".to_string(),
+        ]
+    );
+}
+
+#[test]
+fn blockquote_heading_inherits_heading_style() {
+    let text = render_markdown_text("> # test header\n> in blockquote\n");
+    assert_eq!(
+        text.lines,
+        [
+            Line::from_iter([
+                "> ".into(),
+                "# ".bold().underlined(),
+                "test header".bold().underlined(),
+            ])
+            .green(),
+            Line::from_iter(["> "]).green(),
+            Line::from_iter(["> ", "in blockquote"]).green(),
+        ]
+    );
+}
+
+#[test]
+fn blockquote_with_code_block() {
+    let md = "> ```\n> code\n> ```\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["> code".to_string()]);
+}
+
+#[test]
+fn blockquote_with_multiline_code_block() {
+    let md = "> ```\n> first\n> second\n> ```\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["> first", "> second"]);
+}
+
+#[test]
+fn nested_blockquote_with_inline_and_fenced_code() {
+    /*
+    let md = \"> Nested quote with code:\n\
+    > > Inner quote and `inline code`\n\
+    > >\n\
+    > > ```\n\
+    > > # fenced code inside a quote\n\
+    > > echo \"hello from a quote\"\n\
+    > > ```\n";
+    */
+    let md = r#"> Nested quote with code:
+> > Inner quote and `inline code`
+> >
+> > ```
+> > # fenced code inside a quote
+> > echo "hello from a quote"
+> > ```
+"#;
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(
+        lines,
+        vec![
+            "> Nested quote with code:".to_string(),
+            "> ".to_string(),
+            "> > Inner quote and inline code".to_string(),
+            "> > ".to_string(),
+            "> > # fenced code inside a quote".to_string(),
+            "> > echo \"hello from a quote\"".to_string(),
+        ]
+    );
+}
+
+#[test]
+fn list_unordered_single() {
+    let text = render_markdown_text("- List item 1\n");
+    let expected = Text::from_iter([Line::from_iter(["- ", "List item 1"])]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn list_unordered_multiple() {
+    let text = render_markdown_text("- List item 1\n- List item 2\n");
+    let expected = Text::from_iter([
+        Line::from_iter(["- ", "List item 1"]),
+        Line::from_iter(["- ", "List item 2"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn list_ordered() {
+    let text = render_markdown_text("1. List item 1\n2. List item 2\n");
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "List item 1".into()]),
+        Line::from_iter(["2. ".light_blue(), "List item 2".into()]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn list_nested() {
+    let text = render_markdown_text("- List item 1\n  - Nested list item 1\n");
+    let expected = Text::from_iter([
+        Line::from_iter(["- ", "List item 1"]),
+        Line::from_iter(["    - ", "Nested list item 1"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn list_ordered_custom_start() {
+    let text = render_markdown_text("3. First\n4. Second\n");
+    let expected = Text::from_iter([
+        Line::from_iter(["3. ".light_blue(), "First".into()]),
+        Line::from_iter(["4. ".light_blue(), "Second".into()]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn nested_unordered_in_ordered() {
+    let md = "1. Outer\n    - Inner A\n    - Inner B\n2. Next\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "Outer".into()]),
+        Line::from_iter(["    - ", "Inner A"]),
+        Line::from_iter(["    - ", "Inner B"]),
+        Line::from_iter(["2. ".light_blue(), "Next".into()]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn nested_ordered_in_unordered() {
+    let md = "- Outer\n    1. One\n    2. Two\n- Last\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["- ", "Outer"]),
+        Line::from_iter(["    1. ".light_blue(), "One".into()]),
+        Line::from_iter(["    2. ".light_blue(), "Two".into()]),
+        Line::from_iter(["- ", "Last"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn loose_list_item_multiple_paragraphs() {
+    let md = "1. First paragraph\n\n   Second paragraph of same item\n\n2. Next item\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "First paragraph".into()]),
+        Line::default(),
+        Line::from_iter(["   ", "Second paragraph of same item"]),
+        Line::from_iter(["2. ".light_blue(), "Next item".into()]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn tight_item_with_soft_break() {
+    let md = "- item line1\n  item line2\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["- ", "item line1"]),
+        Line::from_iter(["  ", "item line2"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn deeply_nested_mixed_three_levels() {
+    let md = "1. A\n    - B\n        1. C\n2. D\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "A".into()]),
+        Line::from_iter(["    - ", "B"]),
+        Line::from_iter(["        1. ".light_blue(), "C".into()]),
+        Line::from_iter(["2. ".light_blue(), "D".into()]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn loose_items_due_to_blank_line_between_items() {
+    let md = "1. First\n\n2. Second\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "First".into()]),
+        Line::from_iter(["2. ".light_blue(), "Second".into()]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn mixed_tight_then_loose_in_one_list() {
+    let md = "1. Tight\n\n2.\n   Loose\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "Tight".into()]),
+        Line::from_iter(["2. ".light_blue(), "Loose".into()]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn ordered_item_with_indented_continuation_is_tight() {
+    let md = "1. Foo\n   Bar\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "Foo".into()]),
+        Line::from_iter(["   ", "Bar"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn inline_code() {
+    let text = render_markdown_text("Example of `Inline code`");
+    let expected = Line::from_iter(["Example of ".into(), "Inline code".dim()]).into();
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn strong() {
+    assert_eq!(
+        render_markdown_text("**Strong**"),
+        Text::from(Line::from("Strong".bold()))
+    );
+}
+
+#[test]
+fn emphasis() {
+    assert_eq!(
+        render_markdown_text("*Emphasis*"),
+        Text::from(Line::from("Emphasis".italic()))
+    );
+}
+
+#[test]
+fn strikethrough() {
+    assert_eq!(
+        render_markdown_text("~~Strikethrough~~"),
+        Text::from(Line::from("Strikethrough".crossed_out()))
+    );
+}
+
+#[test]
+fn strong_emphasis() {
+    let text = render_markdown_text("**Strong *emphasis***");
+    let expected = Text::from(Line::from_iter([
+        "Strong ".bold(),
+        "emphasis".bold().italic(),
+    ]));
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn link() {
+    let text = render_markdown_text("[Link](https://example.com)");
+    let expected = Text::from(Line::from_iter([
+        "Link".into(),
+        " (".into(),
+        "https://example.com".cyan().underlined(),
+        ")".into(),
+    ]));
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn code_block_unhighlighted() {
+    let text = render_markdown_text("```rust\nfn main() {}\n```\n");
+    let expected = Text::from_iter([Line::from_iter(["", "fn main() {}"])]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn code_block_multiple_lines_root() {
+    let md = "```\nfirst\nsecond\n```\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["", "first"]),
+        Line::from_iter(["", "second"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn code_block_indented() {
+    let md = "    function greet() {\n      console.log(\"Hi\");\n    }\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["    ", "function greet() {"]),
+        Line::from_iter(["    ", "  console.log(\"Hi\");"]),
+        Line::from_iter(["    ", "}"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn horizontal_rule_renders_em_dashes() {
+    let md = "Before\n\n---\n\nAfter\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["Before", "", "———", "", "After"]);
+}
+
+#[test]
+fn code_block_with_inner_triple_backticks_outer_four() {
+    let md = r#"````text
+Here is a code block that shows another fenced block:
+
+```md
+# Inside fence
+- bullet
+- `inline code`
+```
+````
+"#;
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(
+        lines,
+        vec![
+            "Here is a code block that shows another fenced block:".to_string(),
+            String::new(),
+            "```md".to_string(),
+            "# Inside fence".to_string(),
+            "- bullet".to_string(),
+            "- `inline code`".to_string(),
+            "```".to_string(),
+        ]
+    );
+}
+
+#[test]
+fn code_block_inside_unordered_list_item_is_indented() {
+    let md = "- Item\n\n  ```\n  code line\n  ```\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["- Item", "", "  code line"]);
+}
+
+#[test]
+fn code_block_multiple_lines_inside_unordered_list() {
+    let md = "- Item\n\n  ```\n  first\n  second\n  ```\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["- Item", "", "  first", "  second"]);
+}
+
+#[test]
+fn code_block_inside_unordered_list_item_multiple_lines() {
+    let md = "- Item\n\n  ```\n  first\n  second\n  ```\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(lines, vec!["- Item", "", "  first", "  second"]);
+}
+
+#[test]
+fn markdown_render_complex_snapshot() {
+    let md = r#"# H1: Markdown Streaming Test
+Intro paragraph with bold **text**, italic *text*, and inline code `x=1`.
+Combined bold-italic ***both*** and escaped asterisks \*literal\*.
+Auto-link: <https://example.com> and reference link [ref][r1].
+Link with title: [hover me](https://example.com "Example") and mailto <mailto:test@example.com>.
+Image: ![alt text](https://example.com/img.png "Title")
+> Blockquote level 1
+>> Blockquote level 2 with `inline code`
+- Unordered list item 1
+  - Nested bullet with italics _inner_
+- Unordered list item 2 with ~~strikethrough~~
+1. Ordered item one
+2. Ordered item two with sublist:
+   1) Alt-numbered subitem
+- [ ] Task: unchecked
+- [x] Task: checked with link [home](https://example.org)
+---
+Table below (alignment test):
+| Left | Center | Right |
+|:-----|:------:|------:|
+| a    |   b    |     c |
+Inline HTML: <sup>sup</sup> and <sub>sub</sub>.
+HTML block:
+<div style="border:1px solid #ccc;padding:2px">inline block</div>
+Escapes: \_underscores\_, backslash \\, ticks ``code with `backtick` inside``.
+Emoji shortcodes: :sparkles: :tada: (if supported).
+Hard break test (line ends with two spaces)  
+Next line should be close to previous.
+Footnote reference here[^1] and another[^longnote].
+Horizontal rule with asterisks:
+***
+Fenced code block (JSON):
+```json
+{ "a": 1, "b": [true, false] }
+```
+Fenced code with tildes and triple backticks inside:
+~~~markdown
+To close ``` you need tildes.
+~~~
+Indented code block:
+    for i in range(3): print(i)
+Definition-like list:
+Term
+: Definition with `code`.
+Character entities: &amp; &lt; &gt; &quot; &#39;
+[^1]: This is the first footnote.
+[^longnote]: A longer footnote with a link to [Rust](https://www.rust-lang.org/).
+Escaped pipe in text: a \| b \| c.
+URL with parentheses: [link](https://example.com/path_(with)_parens).
+[r1]: https://example.com/ref "Reference link title"
+"#;
+
+    let text = render_markdown_text(md);
+    // Convert to plain text lines for snapshot (ignore styles)
+    let rendered = text
+        .lines
+        .iter()
+        .map(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+        })
+        .collect::<Vec<_>>()
+        .join("\n");
+
+    assert_snapshot!(rendered);
+}
+
+#[test]
+fn ordered_item_with_code_block_and_nested_bullet() {
+    let md = "1. **item 1**\n\n2. **item 2**\n   ```\n   code\n   ```\n   - `PROCESS_START` (a `OnceLock<Instant>`) keeps the start time for the entire process.\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|line| {
+            line.spans
+                .iter()
+                .map(|span| span.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(
+        lines,
+        vec![
+            "1. item 1".to_string(),
+            "2. item 2".to_string(),
+            String::new(),
+            "   code".to_string(),
+            "    - PROCESS_START (a OnceLock<Instant>) keeps the start time for the entire process.".to_string(),
+        ]
+    );
+}
+
+#[test]
+fn nested_five_levels_mixed_lists() {
+    let md = "1. First\n   - Second level\n     1. Third level (ordered)\n        - Fourth level (bullet)\n          - Fifth level to test indent consistency\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "First".into()]),
+        Line::from_iter(["    - ", "Second level"]),
+        Line::from_iter(["        1. ".light_blue(), "Third level (ordered)".into()]),
+        Line::from_iter(["            - ", "Fourth level (bullet)"]),
+        Line::from_iter([
+            "                - ",
+            "Fifth level to test indent consistency",
+        ]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn html_inline_is_verbatim() {
+    let md = "Hello <span>world</span>!";
+    let text = render_markdown_text(md);
+    let expected: Text = Line::from_iter(["Hello ", "<span>", "world", "</span>", "!"]).into();
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn html_block_is_verbatim_multiline() {
+    let md = "<div>\n  <span>hi</span>\n</div>\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["<div>"]),
+        Line::from_iter(["  <span>hi</span>"]),
+        Line::from_iter(["</div>"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn html_in_tight_ordered_item_soft_breaks_with_space() {
+    let md = "1. Foo\n   <i>Bar</i>\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "Foo".into()]),
+        Line::from_iter(["   ", "<i>", "Bar", "</i>"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn html_continuation_paragraph_in_unordered_item_indented() {
+    let md = "- Item\n\n  <em>continued</em>\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["- ", "Item"]),
+        Line::default(),
+        Line::from_iter(["  ", "<em>", "continued", "</em>"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn unordered_item_continuation_paragraph_is_indented() {
+    let md = "- Intro\n\n  Continuation paragraph line 1\n  Continuation paragraph line 2\n";
+    let text = render_markdown_text(md);
+    let lines: Vec<String> = text
+        .lines
+        .iter()
+        .map(|line| {
+            line.spans
+                .iter()
+                .map(|span| span.content.clone())
+                .collect::<String>()
+        })
+        .collect();
+    assert_eq!(
+        lines,
+        vec![
+            "- Intro".to_string(),
+            String::new(),
+            "  Continuation paragraph line 1".to_string(),
+            "  Continuation paragraph line 2".to_string(),
+        ]
+    );
+}
+
+#[test]
+fn ordered_item_continuation_paragraph_is_indented() {
+    let md = "1. Intro\n\n   More details about intro\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "Intro".into()]),
+        Line::default(),
+        Line::from_iter(["   ", "More details about intro"]),
+    ]);
+    assert_eq!(text, expected);
+}
+
+#[test]
+fn nested_item_continuation_paragraph_is_indented() {
+    let md = "1. A\n    - B\n\n      Continuation for B\n2. C\n";
+    let text = render_markdown_text(md);
+    let expected = Text::from_iter([
+        Line::from_iter(["1. ".light_blue(), "A".into()]),
+        Line::from_iter(["    - ", "B"]),
+        Line::default(),
+        Line::from_iter(["      ", "Continuation for B"]),
+        Line::from_iter(["2. ".light_blue(), "C".into()]),
+    ]);
+    assert_eq!(text, expected);
+}

codex-rs/tui/src/markdown_stream.rs

@@ -4,8 +4,6 @@ use codex_core::config::Config;
 use ratatui::text::Line;
 
 use crate::markdown;
-use crate::render::markdown_utils::is_inside_unclosed_fence;
-use crate::render::markdown_utils::strip_empty_fenced_code_blocks;
 
 /// Newline-gated accumulator that renders markdown and commits only fully
 /// completed logical lines.
@@ -42,6 +40,7 @@ impl MarkdownStreamCollector {
     }
 
     pub fn push_delta(&mut self, delta: &str) {
+        tracing::trace!("push_delta: {delta:?}");
         self.buffer.push_str(delta);
     }
 
@@ -49,14 +48,15 @@ impl MarkdownStreamCollector {
     /// since the last commit. When the buffer does not end with a newline, the
     /// final rendered line is considered incomplete and is not emitted.
     pub fn commit_complete_lines(&mut self, config: &Config) -> Vec<Line<'static>> {
-        // In non-test builds, unwrap an outer ```markdown fence during commit as well,
-        // so fence markers never appear in streamed history.
-        let source = unwrap_markdown_language_fence_if_enabled(self.buffer.clone());
-        let source = strip_empty_fenced_code_blocks(&source);
-
+        let source = self.buffer.clone();
+        let last_newline_idx = source.rfind('\n');
+        let source = if let Some(last_newline_idx) = last_newline_idx {
+            source[..=last_newline_idx].to_string()
+        } else {
+            return Vec::new();
+        };
         let mut rendered: Vec<Line<'static>> = Vec::new();
         markdown::append_markdown(&source, &mut rendered, config);
-
         let mut complete_line_count = rendered.len();
         if complete_line_count > 0
             && crate::render::line_utils::is_blank_line_spaces_only(
@@ -65,87 +65,12 @@ impl MarkdownStreamCollector {
         {
             complete_line_count -= 1;
         }
-        // Heuristic: if the buffer ends with a double newline and the last non-blank
-        // rendered line looks like a list bullet with inline content (e.g., "- item"),
-        // defer committing that line. Subsequent context (e.g., another list item)
-        // can cause the renderer to split the bullet marker and text into separate
-        // logical lines ("- " then "item"), which would otherwise duplicate content.
-        if self.buffer.ends_with("\n\n") && complete_line_count > 0 {
-            let last = &rendered[complete_line_count - 1];
-            let mut text = String::new();
-            for s in &last.spans {
-                text.push_str(&s.content);
-            }
-            if text.starts_with("- ") && text.trim() != "-" {
-                complete_line_count = complete_line_count.saturating_sub(1);
-            }
-        }
-        if !self.buffer.ends_with('\n') {
-            complete_line_count = complete_line_count.saturating_sub(1);
-            // If we're inside an unclosed fenced code block, also drop the
-            // last rendered line to avoid committing a partial code line.
-            if is_inside_unclosed_fence(&source) {
-                complete_line_count = complete_line_count.saturating_sub(1);
-            }
-            // If the next (incomplete) line appears to begin a list item,
-            // also defer the previous completed line because the renderer may
-            // retroactively treat it as part of the list (e.g., ordered list item 1).
-            if let Some(last_nl) = source.rfind('\n') {
-                let tail = &source[last_nl + 1..];
-                if starts_with_list_marker(tail) {
-                    complete_line_count = complete_line_count.saturating_sub(1);
-                }
-            }
-        }
-
-        // Conservatively withhold trailing list-like lines (unordered or ordered)
-        // because streaming mid-item can cause the renderer to later split or
-        // restructure them (e.g., duplicating content or separating the marker).
-        // Only defers lines at the end of the out slice so previously committed
-        // lines remain stable.
-        if complete_line_count > self.committed_line_count {
-            let mut safe_count = complete_line_count;
-            while safe_count > self.committed_line_count {
-                let l = &rendered[safe_count - 1];
-                let mut text = String::new();
-                for s in &l.spans {
-                    text.push_str(&s.content);
-                }
-                let listish = is_potentially_volatile_list_line(&text);
-                if listish {
-                    safe_count -= 1;
-                    continue;
-                }
-                break;
-            }
-            complete_line_count = safe_count;
-        }
 
         if self.committed_line_count >= complete_line_count {
             return Vec::new();
         }
 
         let out_slice = &rendered[self.committed_line_count..complete_line_count];
-        // Strong correctness: while a fenced code block is open (no closing fence yet),
-        // do not emit any new lines from inside it. Wait until the fence closes to emit
-        // the entire block together. This avoids stray backticks and misformatted content.
-        if is_inside_unclosed_fence(&source) {
-            return Vec::new();
-        }
-
-        // Additional conservative hold-back: if exactly one short, plain word
-        // line would be emitted, defer it. This avoids committing a lone word
-        // that might become the first ordered-list item once the next delta
-        // arrives (e.g., next line starts with "2 " or "2. ").
-        if out_slice.len() == 1 {
-            let mut s = String::new();
-            for sp in &out_slice[0].spans {
-                s.push_str(&sp.content);
-            }
-            if is_short_plain_word(&s) {
-                return Vec::new();
-            }
-        }
 
         let out = out_slice.to_vec();
         self.committed_line_count = complete_line_count;
@@ -157,12 +82,19 @@ impl MarkdownStreamCollector {
     /// for rendering. Optionally unwraps ```markdown language fences in
     /// non-test builds.
     pub fn finalize_and_drain(&mut self, config: &Config) -> Vec<Line<'static>> {
-        let mut source: String = self.buffer.clone();
+        let raw_buffer = self.buffer.clone();
+        let mut source: String = raw_buffer.clone();
         if !source.ends_with('\n') {
             source.push('\n');
         }
-        let source = unwrap_markdown_language_fence_if_enabled(source);
-        let source = strip_empty_fenced_code_blocks(&source);
+        tracing::debug!(
+            raw_len = raw_buffer.len(),
+            source_len = source.len(),
+            "markdown finalize (raw length: {}, rendered length: {})",
+            raw_buffer.len(),
+            source.len()
+        );
+        tracing::trace!("markdown finalize (raw source):\n---\n{source}\n---");
 
         let mut rendered: Vec<Line<'static>> = Vec::new();
         markdown::append_markdown(&source, &mut rendered, config);
@@ -179,122 +111,6 @@ impl MarkdownStreamCollector {
     }
 }
 
-#[inline]
-fn is_potentially_volatile_list_line(text: &str) -> bool {
-    let t = text.trim_end();
-    if t == "-" || t == "*" || t == "- " || t == "* " {
-        return true;
-    }
-    if t.starts_with("- ") || t.starts_with("* ") {
-        return true;
-    }
-    // ordered list like "1. " or "23. "
-    let mut it = t.chars().peekable();
-    let mut saw_digit = false;
-    while let Some(&ch) = it.peek() {
-        if ch.is_ascii_digit() {
-            saw_digit = true;
-            it.next();
-            continue;
-        }
-        break;
-    }
-    if saw_digit && it.peek() == Some(&'.') {
-        // consume '.'
-        it.next();
-        if it.peek() == Some(&' ') {
-            return true;
-        }
-    }
-    false
-}
-
-#[inline]
-fn starts_with_list_marker(text: &str) -> bool {
-    let t = text.trim_start();
-    if t.starts_with("- ") || t.starts_with("* ") || t.starts_with("-\t") || t.starts_with("*\t") {
-        return true;
-    }
-    // ordered list marker like "1 ", "1. ", "23 ", "23. "
-    let mut it = t.chars().peekable();
-    let mut saw_digit = false;
-    while let Some(&ch) = it.peek() {
-        if ch.is_ascii_digit() {
-            saw_digit = true;
-            it.next();
-        } else {
-            break;
-        }
-    }
-    if !saw_digit {
-        return false;
-    }
-    match it.peek() {
-        Some('.') => {
-            it.next();
-            matches!(it.peek(), Some(' '))
-        }
-        Some(' ') => true,
-        _ => false,
-    }
-}
-
-#[inline]
-fn is_short_plain_word(s: &str) -> bool {
-    let t = s.trim();
-    if t.is_empty() || t.len() > 5 {
-        return false;
-    }
-    t.chars().all(|c| c.is_alphanumeric())
-}
-
-/// fence helpers are provided by `crate::render::markdown_utils`
-#[cfg(test)]
-fn unwrap_markdown_language_fence_if_enabled(s: String) -> String {
-    // In tests, keep content exactly as provided to simplify assertions.
-    s
-}
-
-#[cfg(not(test))]
-fn unwrap_markdown_language_fence_if_enabled(s: String) -> String {
-    // Best-effort unwrap of a single outer fenced markdown block.
-    // Recognizes common forms like ```markdown, ```md (any case), optional
-    // surrounding whitespace, and flexible trailing newlines/CRLF.
-    // If the block is not recognized, return the input unchanged.
-    let lines = s.lines().collect::<Vec<_>>();
-    if lines.len() < 2 {
-        return s;
-    }
-
-    // Identify opening fence and language.
-    let open = lines.first().map(|l| l.trim_start()).unwrap_or("");
-    if !open.starts_with("```") {
-        return s;
-    }
-    let lang = open.trim_start_matches("```").trim();
-    let is_markdown_lang = lang.eq_ignore_ascii_case("markdown") || lang.eq_ignore_ascii_case("md");
-    if !is_markdown_lang {
-        return s;
-    }
-
-    // Find the last non-empty line and ensure it is a closing fence.
-    let mut last_idx = lines.len() - 1;
-    while last_idx > 0 && lines[last_idx].trim().is_empty() {
-        last_idx -= 1;
-    }
-    if lines[last_idx].trim() != "```" {
-        return s;
-    }
-
-    // Reconstruct the inner content between the fences.
-    let mut out = String::new();
-    for l in lines.iter().take(last_idx).skip(1) {
-        out.push_str(l);
-        out.push('\n');
-    }
-    out
-}
-
 pub(crate) struct StepResult {
     pub history: Vec<Line<'static>>, // lines to insert into history this step
 }
@@ -373,6 +189,7 @@ mod tests {
     use super::*;
     use codex_core::config::Config;
     use codex_core::config::ConfigOverrides;
+    use ratatui::style::Color;
 
     fn test_config() -> Config {
         let overrides = ConfigOverrides {
@@ -406,6 +223,125 @@ mod tests {
         assert_eq!(out.len(), 1);
     }
 
+    #[test]
+    fn e2e_stream_blockquote_simple_is_green() {
+        let cfg = test_config();
+        let out = super::simulate_stream_markdown_for_tests(&["> Hello\n"], true, &cfg);
+        assert_eq!(out.len(), 1);
+        let l = &out[0];
+        assert_eq!(
+            l.style.fg,
+            Some(Color::Green),
+            "expected blockquote line fg green, got {:?}",
+            l.style.fg
+        );
+    }
+
+    #[test]
+    fn e2e_stream_blockquote_nested_is_green() {
+        let cfg = test_config();
+        let out =
+            super::simulate_stream_markdown_for_tests(&["> Level 1\n>> Level 2\n"], true, &cfg);
+        // Filter out any blank lines that may be inserted at paragraph starts.
+        let non_blank: Vec<_> = out
+            .into_iter()
+            .filter(|l| {
+                let s = l
+                    .spans
+                    .iter()
+                    .map(|sp| sp.content.clone())
+                    .collect::<Vec<_>>()
+                    .join("");
+                let t = s.trim();
+                // Ignore quote-only blank lines like ">" inserted at paragraph boundaries.
+                !(t.is_empty() || t == ">")
+            })
+            .collect();
+        assert_eq!(non_blank.len(), 2);
+        assert_eq!(non_blank[0].style.fg, Some(Color::Green));
+        assert_eq!(non_blank[1].style.fg, Some(Color::Green));
+    }
+
+    #[test]
+    fn e2e_stream_blockquote_with_list_items_is_green() {
+        let cfg = test_config();
+        let out =
+            super::simulate_stream_markdown_for_tests(&["> - item 1\n> - item 2\n"], true, &cfg);
+        assert_eq!(out.len(), 2);
+        assert_eq!(out[0].style.fg, Some(Color::Green));
+        assert_eq!(out[1].style.fg, Some(Color::Green));
+    }
+
+    #[test]
+    fn e2e_stream_nested_mixed_lists_ordered_marker_is_light_blue() {
+        let cfg = test_config();
+        let md = [
+            "1. First\n",
+            "   - Second level\n",
+            "     1. Third level (ordered)\n",
+            "        - Fourth level (bullet)\n",
+            "          - Fifth level to test indent consistency\n",
+        ];
+        let out = super::simulate_stream_markdown_for_tests(&md, true, &cfg);
+        // Find the line that contains the third-level ordered text
+        let find_idx = out.iter().position(|l| {
+            l.spans
+                .iter()
+                .map(|s| s.content.clone())
+                .collect::<String>()
+                .contains("Third level (ordered)")
+        });
+        let idx = find_idx.expect("expected third-level ordered line");
+        let line = &out[idx];
+        // Expect at least one span on this line to be styled light blue
+        let has_light_blue = line
+            .spans
+            .iter()
+            .any(|s| s.style.fg == Some(ratatui::style::Color::LightBlue));
+        assert!(
+            has_light_blue,
+            "expected an ordered-list marker span with light blue fg on: {line:?}"
+        );
+    }
+
+    #[test]
+    fn e2e_stream_blockquote_wrap_preserves_green_style() {
+        let cfg = test_config();
+        let long = "> This is a very long quoted line that should wrap across multiple columns to verify style preservation.";
+        let out = super::simulate_stream_markdown_for_tests(&[long, "\n"], true, &cfg);
+        // Wrap to a narrow width to force multiple output lines.
+        let wrapped = crate::wrapping::word_wrap_lines(
+            out.iter().collect::<Vec<_>>(),
+            crate::wrapping::RtOptions::new(24),
+        );
+        // Filter out purely blank lines
+        let non_blank: Vec<_> = wrapped
+            .into_iter()
+            .filter(|l| {
+                let s = l
+                    .spans
+                    .iter()
+                    .map(|sp| sp.content.clone())
+                    .collect::<Vec<_>>()
+                    .join("");
+                !s.trim().is_empty()
+            })
+            .collect();
+        assert!(
+            non_blank.len() >= 2,
+            "expected wrapped blockquote to span multiple lines"
+        );
+        for (i, l) in non_blank.iter().enumerate() {
+            assert_eq!(
+                l.style.fg,
+                Some(Color::Green),
+                "wrapped line {} should preserve green style, got {:?}",
+                i,
+                l.style.fg
+            );
+        }
+    }
+
     #[test]
     fn heading_starts_on_new_line_when_following_paragraph() {
         let cfg = test_config();
@@ -490,7 +426,7 @@ mod tests {
             .collect();
         assert_eq!(
             s1,
-            vec!["Sounds good!", ""],
+            vec!["Sounds good!"],
             "expected paragraph followed by blank separator before heading chunk"
         );
 
@@ -509,7 +445,7 @@ mod tests {
             .collect();
         assert_eq!(
             s2,
-            vec!["## Adding Bird subcommand"],
+            vec!["", "## Adding Bird subcommand"],
             "expected the heading line only on the final commit"
         );
 
@@ -531,18 +467,6 @@ mod tests {
             vec!["Hello."],
             "unexpected markdown lines: {rendered_strings:?}"
         );
-
-        let line_to_string = |l: &ratatui::text::Line<'_>| -> String {
-            l.spans
-                .iter()
-                .map(|s| s.content.clone())
-                .collect::<Vec<_>>()
-                .join("")
-        };
-
-        assert_eq!(line_to_string(&out1[0]), "Sounds good!");
-        assert_eq!(line_to_string(&out1[1]), "");
-        assert_eq!(line_to_string(&out2[0]), "## Adding Bird subcommand");
     }
 
     fn lines_to_plain_strings(lines: &[ratatui::text::Line<'_>]) -> Vec<String> {
@@ -560,35 +484,11 @@ mod tests {
 
     #[test]
     fn lists_and_fences_commit_without_duplication() {
-        let cfg = test_config();
-
         // List case
-        let deltas = vec!["- a\n- ", "b\n- c\n"];
-        let streamed = simulate_stream_markdown_for_tests(&deltas, true, &cfg);
-        let streamed_str = lines_to_plain_strings(&streamed);
-
-        let mut rendered_all: Vec<ratatui::text::Line<'static>> = Vec::new();
-        crate::markdown::append_markdown("- a\n- b\n- c\n", &mut rendered_all, &cfg);
-        let rendered_all_str = lines_to_plain_strings(&rendered_all);
-
-        assert_eq!(
-            streamed_str, rendered_all_str,
-            "list streaming should equal full render without duplication"
-        );
+        assert_streamed_equals_full(&["- a\n- ", "b\n- c\n"]);
 
         // Fenced code case: stream in small chunks
-        let deltas2 = vec!["```", "\nco", "de 1\ncode 2\n", "```\n"];
-        let streamed2 = simulate_stream_markdown_for_tests(&deltas2, true, &cfg);
-        let streamed2_str = lines_to_plain_strings(&streamed2);
-
-        let mut rendered_all2: Vec<ratatui::text::Line<'static>> = Vec::new();
-        crate::markdown::append_markdown("```\ncode 1\ncode 2\n```\n", &mut rendered_all2, &cfg);
-        let rendered_all2_str = lines_to_plain_strings(&rendered_all2);
-
-        assert_eq!(
-            streamed2_str, rendered_all2_str,
-            "fence streaming should equal full render without duplication"
-        );
+        assert_streamed_equals_full(&["```", "\nco", "de 1\ncode 2\n", "```\n"]);
     }
 
     #[test]
@@ -622,6 +522,56 @@ mod tests {
         );
     }
 
+    #[test]
+    fn e2e_stream_deep_nested_third_level_marker_is_light_blue() {
+        let cfg = test_config();
+        let md = "1. First\n   - Second level\n     1. Third level (ordered)\n        - Fourth level (bullet)\n          - Fifth level to test indent consistency\n";
+        let streamed = super::simulate_stream_markdown_for_tests(&[md], true, &cfg);
+        let streamed_strs = lines_to_plain_strings(&streamed);
+
+        // Locate the third-level line in the streamed output; avoid relying on exact indent.
+        let target_suffix = "1. Third level (ordered)";
+        let mut found = None;
+        for line in &streamed {
+            let s: String = line.spans.iter().map(|sp| sp.content.clone()).collect();
+            if s.contains(target_suffix) {
+                found = Some(line.clone());
+                break;
+            }
+        }
+        let line = found.unwrap_or_else(|| {
+            panic!("expected to find the third-level ordered list line; got: {streamed_strs:?}")
+        });
+
+        // The marker (including indent and "1.") is expected to be in the first span
+        // and colored LightBlue; following content should be default color.
+        assert!(
+            !line.spans.is_empty(),
+            "expected non-empty spans for the third-level line"
+        );
+        let marker_span = &line.spans[0];
+        assert_eq!(
+            marker_span.style.fg,
+            Some(Color::LightBlue),
+            "expected LightBlue 3rd-level ordered marker, got {:?}",
+            marker_span.style.fg
+        );
+        // Find the first non-empty non-space content span and verify it is default color.
+        let mut content_fg = None;
+        for sp in &line.spans[1..] {
+            let t = sp.content.trim();
+            if !t.is_empty() {
+                content_fg = Some(sp.style.fg);
+                break;
+            }
+        }
+        assert_eq!(
+            content_fg.flatten(),
+            None,
+            "expected default color for 3rd-level content, got {content_fg:?}"
+        );
+    }
+
     #[test]
     fn empty_fenced_block_is_dropped_and_separator_preserved_before_heading() {
         let cfg = test_config();
@@ -768,16 +718,12 @@ mod tests {
         let expected = vec![
             "Loose vs. tight list items:".to_string(),
             "".to_string(),
-            "1. ".to_string(),
-            "Tight item".to_string(),
-            "2. ".to_string(),
-            "Another tight item".to_string(),
-            "3. ".to_string(),
-            "Loose item with its own paragraph.".to_string(),
+            "1. Tight item".to_string(),
+            "2. Another tight item".to_string(),
+            "3. Loose item with its own paragraph.".to_string(),
             "".to_string(),
-            "This paragraph belongs to the same list item.".to_string(),
-            "4. ".to_string(),
-            "Second loose item with a nested list after a blank line.".to_string(),
+            "   This paragraph belongs to the same list item.".to_string(),
+            "4. Second loose item with a nested list after a blank line.".to_string(),
             "    - Nested bullet under a loose item".to_string(),
             "    - Another nested bullet".to_string(),
         ];
@@ -788,63 +734,39 @@ mod tests {
     }
 
     // Targeted tests derived from fuzz findings. Each asserts streamed == full render.
-
-    #[test]
-    fn fuzz_class_bare_dash_then_task_item() {
+    fn assert_streamed_equals_full(deltas: &[&str]) {
         let cfg = test_config();
-        // Case similar to: ["two\n", "- \n* [x] done "]
-        let deltas = vec!["two\n", "- \n* [x] done \n"];
-        let streamed = simulate_stream_markdown_for_tests(&deltas, true, &cfg);
+        let streamed = simulate_stream_markdown_for_tests(deltas, true, &cfg);
         let streamed_strs = lines_to_plain_strings(&streamed);
         let full: String = deltas.iter().copied().collect();
         let mut rendered: Vec<ratatui::text::Line<'static>> = Vec::new();
         crate::markdown::append_markdown(&full, &mut rendered, &cfg);
         let rendered_strs = lines_to_plain_strings(&rendered);
-        assert_eq!(streamed_strs, rendered_strs);
+        assert_eq!(streamed_strs, rendered_strs, "full:\n---\n{full}\n---");
     }
 
     #[test]
     fn fuzz_class_bullet_duplication_variant_1() {
-        let cfg = test_config();
-        // Case similar to: ["aph.\n- let one\n- bull", "et two\n\n  second paragraph "]
-        let deltas = vec!["aph.\n- let one\n- bull", "et two\n\n  second paragraph \n"];
-        let streamed = simulate_stream_markdown_for_tests(&deltas, true, &cfg);
-        let streamed_strs = lines_to_plain_strings(&streamed);
-        let full: String = deltas.iter().copied().collect();
-        let mut rendered: Vec<ratatui::text::Line<'static>> = Vec::new();
-        crate::markdown::append_markdown(&full, &mut rendered, &cfg);
-        let rendered_strs = lines_to_plain_strings(&rendered);
-        assert_eq!(streamed_strs, rendered_strs);
+        assert_streamed_equals_full(&[
+            "aph.\n- let one\n- bull",
+            "et two\n\n  second paragraph \n",
+        ]);
     }
 
     #[test]
     fn fuzz_class_bullet_duplication_variant_2() {
-        let cfg = test_config();
-        // Case similar to: ["- e\n  c", "e\n- bullet two\n\n  second paragraph in bullet two\n"]
-        let deltas = vec![
+        assert_streamed_equals_full(&[
             "- e\n  c",
             "e\n- bullet two\n\n  second paragraph in bullet two\n",
-        ];
-        let streamed = simulate_stream_markdown_for_tests(&deltas, true, &cfg);
-        let streamed_strs = lines_to_plain_strings(&streamed);
-        let full: String = deltas.iter().copied().collect();
-        let mut rendered: Vec<ratatui::text::Line<'static>> = Vec::new();
-        crate::markdown::append_markdown(&full, &mut rendered, &cfg);
-        let rendered_strs = lines_to_plain_strings(&rendered);
-        assert_eq!(streamed_strs, rendered_strs);
+        ]);
     }
 
     #[test]
-    fn fuzz_class_ordered_list_split_weirdness() {
-        let cfg = test_config();
-        // Case similar to: ["one\n2", " two\n- \n* [x] d"]
-        let deltas = vec!["one\n2", " two\n- \n* [x] d\n"];
-        let streamed = simulate_stream_markdown_for_tests(&deltas, true, &cfg);
-        let streamed_strs = lines_to_plain_strings(&streamed);
-        let full: String = deltas.iter().copied().collect();
-        let mut rendered: Vec<ratatui::text::Line<'static>> = Vec::new();
-        crate::markdown::append_markdown(&full, &mut rendered, &cfg);
-        let rendered_strs = lines_to_plain_strings(&rendered);
-        assert_eq!(streamed_strs, rendered_strs);
+    fn streaming_html_block_then_text_matches_full() {
+        assert_streamed_equals_full(&[
+            "HTML block:\n",
+            "<div>inline block</div>\n",
+            "more stuff\n",
+        ]);
     }
 }

codex-rs/tui/src/onboarding/auth.rs

@@ -2,7 +2,6 @@
 
 use codex_core::AuthManager;
 use codex_core::auth::CLIENT_ID;
-use codex_core::config::Config;
 use codex_login::ServerOptions;
 use codex_login::ShutdownHandle;
 use codex_login::run_login_server;
@@ -114,7 +113,6 @@ pub(crate) struct AuthModeWidget {
     pub login_status: LoginStatus,
     pub preferred_auth_method: AuthMode,
     pub auth_manager: Arc<AuthManager>,
-    pub config: Config,
 }
 
 impl AuthModeWidget {
@@ -316,11 +314,7 @@ impl AuthModeWidget {
         }
 
         self.error = None;
-        let opts = ServerOptions::new(
-            self.codex_home.clone(),
-            CLIENT_ID.to_string(),
-            self.config.responses_originator_header.clone(),
-        );
+        let opts = ServerOptions::new(self.codex_home.clone(), CLIENT_ID.to_string());
         match run_login_server(opts) {
             Ok(child) => {
                 let sign_in_state = self.sign_in_state.clone();

codex-rs/tui/src/onboarding/onboarding_screen.rs

@@ -85,7 +85,6 @@ impl OnboardingScreen {
                 login_status,
                 auth_manager,
                 preferred_auth_method,
-                config,
             }))
         }
         let is_git_repo = get_git_repo_root(&cwd).is_some();

codex-rs/tui/src/render/markdown_utils.rs

@@ -1,72 +0,0 @@
-/// Returns true if the provided text contains an unclosed fenced code block
-/// (opened by ``` or ~~~, closed by a matching fence on its own line).
-pub fn is_inside_unclosed_fence(s: &str) -> bool {
-    let mut open = false;
-    for line in s.lines() {
-        let t = line.trim_start();
-        if t.starts_with("```") || t.starts_with("~~~") {
-            if !open {
-                open = true;
-            } else {
-                // closing fence on same pattern toggles off
-                open = false;
-            }
-        }
-    }
-    open
-}
-
-/// Remove fenced code blocks that contain no content (whitespace-only) to avoid
-/// streaming empty code blocks like ```lang\n``` or ```\n```.
-pub fn strip_empty_fenced_code_blocks(s: &str) -> String {
-    // Only remove complete fenced blocks that contain no non-whitespace content.
-    // Leave all other content unchanged to avoid affecting partial streams.
-    let lines: Vec<&str> = s.lines().collect();
-    let mut out = String::with_capacity(s.len());
-    let mut i = 0usize;
-    while i < lines.len() {
-        let line = lines[i];
-        let trimmed_start = line.trim_start();
-        let fence_token = if trimmed_start.starts_with("```") {
-            "```"
-        } else if trimmed_start.starts_with("~~~") {
-            "~~~"
-        } else {
-            ""
-        };
-        if !fence_token.is_empty() {
-            // Find a matching closing fence on its own line.
-            let mut j = i + 1;
-            let mut has_content = false;
-            let mut found_close = false;
-            while j < lines.len() {
-                let l = lines[j];
-                if l.trim() == fence_token {
-                    found_close = true;
-                    break;
-                }
-                if !l.trim().is_empty() {
-                    has_content = true;
-                }
-                j += 1;
-            }
-            if found_close && !has_content {
-                // Drop i..=j and insert at most a single blank separator line.
-                if !out.ends_with('\n') {
-                    out.push('\n');
-                }
-                i = j + 1;
-                continue;
-            }
-            // Not an empty fenced block; emit as-is.
-            out.push_str(line);
-            out.push('\n');
-            i += 1;
-        } else {
-            out.push_str(line);
-            out.push('\n');
-            i += 1;
-        }
-    }
-    out
-}

codex-rs/tui/src/render/mod.rs

@@ -1,3 +1,2 @@
 pub mod highlight;
 pub mod line_utils;
-pub mod markdown_utils;

codex-rs/tui/src/resume_picker.rs

@@ -2,7 +2,6 @@ use std::path::Path;
 use std::path::PathBuf;
 
 use chrono::DateTime;
-use chrono::TimeZone;
 use chrono::Utc;
 use codex_core::ConversationItem;
 use codex_core::ConversationsPage;
@@ -255,19 +254,10 @@ impl PickerState {
 }
 
 fn to_rows(page: ConversationsPage) -> Vec<Row> {
-    use std::cmp::Reverse;
-    let mut rows: Vec<Row> = page
-        .items
-        .into_iter()
-        .filter_map(|it| head_to_row(&it))
-        .collect();
-    // Ensure newest-first ordering within the page by timestamp when available.
-    let epoch = Utc.timestamp_opt(0, 0).single().unwrap_or_else(Utc::now);
-    rows.sort_by_key(|r| Reverse(r.ts.unwrap_or(epoch)));
-    rows
+    page.items.into_iter().map(|it| head_to_row(&it)).collect()
 }
 
-fn head_to_row(item: &ConversationItem) -> Option<Row> {
+fn head_to_row(item: &ConversationItem) -> Row {
     let mut ts: Option<DateTime<Utc>> = None;
     if let Some(first) = item.head.first()
         && let Some(t) = first.get("timestamp").and_then(|v| v.as_str())
@@ -276,16 +266,16 @@ fn head_to_row(item: &ConversationItem) -> Option<Row> {
         ts = Some(parsed.with_timezone(&Utc));
     }
 
-    let preview = preview_from_head(&item.head)?;
-    let preview = preview.trim().to_string();
-    if preview.is_empty() {
-        return None;
-    }
-    Some(Row {
+    let preview = preview_from_head(&item.head)
+        .map(|s| s.trim().to_string())
+        .filter(|s| !s.is_empty())
+        .unwrap_or_else(|| String::from("(no message yet)"));
+
+    Row {
         path: item.path.clone(),
         preview,
         ts,
-    })
+    }
 }
 
 fn preview_from_head(head: &[serde_json::Value]) -> Option<String> {
@@ -483,7 +473,7 @@ mod tests {
     }
 
     #[test]
-    fn to_rows_sorts_descending_by_timestamp() {
+    fn to_rows_preserves_backend_order() {
         // Construct two items with different timestamps and real user text.
         let a = ConversationItem {
             path: PathBuf::from("/tmp/a.jsonl"),
@@ -500,8 +490,8 @@ mod tests {
             reached_scan_cap: false,
         });
         assert_eq!(rows.len(), 2);
-        // Expect the newer timestamp (B) first
-        assert!(rows[0].preview.contains('B'));
-        assert!(rows[1].preview.contains('A'));
+        // Preserve the given order; backend already provides newest-first
+        assert!(rows[0].preview.contains('A'));
+        assert!(rows[1].preview.contains('B'));
     }
 }

codex-rs/tui/src/snapshots/codex_tui__markdown_render__markdown_render_tests__markdown_render_complex_snapshot.snap

@@ -0,0 +1,62 @@
+---
+source: tui/src/markdown_render_tests.rs
+expression: rendered
+---
+# H1: Markdown Streaming Test
+
+Intro paragraph with bold text, italic text, and inline code x=1.
+Combined bold-italic both and escaped asterisks *literal*.
+Auto-link: https://example.com (https://example.com) and reference link [ref][r1].
+Link with title: hover me (https://example.com) and mailto mailto:test@example.com (mailto:test@example.com).
+Image: alt text
+
+> Blockquote level 1
+> 
+> > Blockquote level 2 with inline code
+
+- Unordered list item 1
+    - Nested bullet with italics inner
+- Unordered list item 2 with strikethrough
+
+1. Ordered item one
+2. Ordered item two with sublist:
+    1. Alt-numbered subitem
+
+- [ ] Task: unchecked
+- [x] Task: checked with link home (https://example.org)
+
+———
+
+Table below (alignment test):
+| Left | Center | Right |
+|:-----|:------:|------:|
+| a    |   b    |     c |
+Inline HTML: <sup>sup</sup> and <sub>sub</sub>.
+HTML block:
+<div style="border:1px solid #ccc;padding:2px">inline block</div>
+Escapes: \_underscores\_, backslash \\, ticks ``code with `backtick` inside``.
+Emoji shortcodes: :sparkles: :tada: (if supported).
+Hard break test (line ends with two spaces)  
+Next line should be close to previous.
+Footnote reference here[^1] and another[^longnote].
+Horizontal rule with asterisks:
+***
+Fenced code block (JSON):
+```json
+{ "a": 1, "b": [true, false] }
+```
+Fenced code with tildes and triple backticks inside:
+~~~markdown
+To close ``` you need tildes.
+~~~
+Indented code block:
+    for i in range(3): print(i)
+Definition-like list:
+Term
+: Definition with `code`.
+Character entities: &amp; &lt; &gt; &quot; &#39;
+[^1]: This is the first footnote.
+[^longnote]: A longer footnote with a link to [Rust](https://www.rust-lang.org/).
+Escaped pipe in text: a \| b \| c.
+URL with parentheses: [link](https://example.com/path_(with)_parens).
+[r1]: https://example.com/ref "Reference link title"

codex-rs/tui/src/streaming/controller.rs

@@ -380,16 +380,12 @@ mod tests {
         let expected = vec![
             "Loose vs. tight list items:".to_string(),
             "".to_string(),
-            "1. ".to_string(),
-            "Tight item".to_string(),
-            "2. ".to_string(),
-            "Another tight item".to_string(),
-            "3. ".to_string(),
-            "Loose item with its own paragraph.".to_string(),
+            "1. Tight item".to_string(),
+            "2. Another tight item".to_string(),
+            "3. Loose item with its own paragraph.".to_string(),
             "".to_string(),
-            "This paragraph belongs to the same list item.".to_string(),
-            "4. ".to_string(),
-            "Second loose item with a nested list after a blank line.".to_string(),
+            "   This paragraph belongs to the same list item.".to_string(),
+            "4. Second loose item with a nested list after a blank line.".to_string(),
             "    - Nested bullet under a loose item".to_string(),
             "    - Another nested bullet".to_string(),
         ];