[prompt] apply_patch updates

Previously, `codex exec` was printing `Warning: no file to write last
message to` as a warning to stderr even though `--output-last-message`
was not specified, which is wrong. This fixes the code and changes
`handle_last_message()` so that it is only called when
`last_message_path` is `Some`.

.github/actions/codex/bun.lock

@@ -11,7 +11,7 @@
         "@types/bun": "^1.2.19",
         "@types/node": "^24.1.0",
         "prettier": "^3.6.2",
-        "typescript": "^5.8.3",
+        "typescript": "^5.9.2",
       },
     },
   },
@@ -68,7 +68,7 @@
 
     "tunnel": ["tunnel@0.0.6", "", {}, "sha512-1h/Lnq9yajKY2PEbBadPXj3VxsDDu844OnaAo52UVmIzIvwwtBPIuNvkjuzBlTWpfJyUbG3ez0KSBibQkj4ojg=="],
 
-    "typescript": ["typescript@5.8.3", "", { "bin": { "tsc": "bin/tsc", "tsserver": "bin/tsserver" } }, "sha512-p1diW6TqL9L07nNxvRMM7hMMw4c5XOo/1ibL4aAIGmSAt9slTE1Xgw5KWuof2uTOvCg9BY7ZRi+GaF+7sfgPeQ=="],
+    "typescript": ["typescript@5.9.2", "", { "bin": { "tsc": "bin/tsc", "tsserver": "bin/tsserver" } }, "sha512-CWBzXQrc/qOkhidw1OzBTQuYRbfyxDXJMVJ1XNwUHGROVmuaeiEm3OslpZ1RV96d7SKKjZKrSJu3+t/xlw3R9A=="],
 
     "undici": ["undici@5.29.0", "", { "dependencies": { "@fastify/busboy": "^2.0.0" } }, "sha512-raqeBD6NQK4SkWhQzeYKd1KmIG6dllBOTt55Rmkt4HtI9mwdWtJljnrXjAFUBLTSN67HWrOIZ3EPF4kjUw80Bg=="],
 

.github/actions/codex/package.json

@@ -16,6 +16,6 @@
         "@types/bun": "^1.2.19",
         "@types/node": "^24.1.0",
         "prettier": "^3.6.2",
-        "typescript": "^5.8.3"
+        "typescript": "^5.9.2"
     }
 }

.github/codex/labels/codex-rust-review.md

@@ -0,0 +1,23 @@
+Review this PR and respond with a very concise final message, formatted in Markdown.
+
+There should be a summary of the changes (1-2 sentences) and a few bullet points if necessary.
+
+Then provide the **review** (1-2 sentences plus bullet points, friendly tone).
+
+Things to look out for when doing the review:
+
+- **Make sure the pull request body explains the motivation behind the change.** If the author has failed to do this, call it out, and if you think you can deduce the motivation behind the change, propose copy.
+- Ideally, the PR body also contains a small summary of the change. For small changes, the PR title may be sufficient.
+- Each PR should ideally do one conceptual thing. For example, if a PR does a refactoring as well as introducing a new feature, push back and suggest the refactoring be done in a separate PR. This makes things easier for the reviewer, as refactoring changes can often be far-reaching, yet quick to review.
+- If the nature of the change seems to have a visual component (which is often the case for changes to `codex-rs/tui`), recommend including a screenshot or video to demonstrate the change, if appropriate.
+- Rust files should generally be organized such that the public parts of the API appear near the top of the file and helper functions go below. This is analagous to the "inverted pyramid" structure that is favored in journalism.
+- Encourage the use of small enums or the newtype pattern in Rust if it helps readability without adding significant cognitive load or lines of code.
+- Be wary of large files and offer suggestions for how to break things into more reasonably-sized files.
+- When modifying a `Cargo.toml` file, make sure that dependency lists stay alphabetically sorted. Also consider whether a new dependency is added to the appropriate place (e.g., `[dependencies]` versus `[dev-dependencies]`)
+- If you see opportunities for the changes in a diff to use more idiomatic Rust, please make specific recommendations. For example, favor the use of expressions over `return`.
+- When introducing new code, be on the lookout for code that duplicates existing code. When found, propose a way to refactor the existing code such that it should be reused.
+- Each create in the Cargo workspace in `codex-rs` has a specific purpose: make a note if you believe new code is not introduced in the correct crate.
+- When possible, try to keep the `core` crate as small as possible. Non-core but shared logic is often a good candidate for `codex-rs/common`.
+- References to existing GitHub issues and PRs are encouraged, where appropriate, though you likely do not have network access, so may not be able to help here.
+
+{CODEX_ACTION_GITHUB_EVENT_PATH} contains the JSON that triggered this GitHub workflow. It contains the `base` and `head` refs that define this PR. Both refs are available locally.

.github/workflows/codex.yml

@@ -20,7 +20,7 @@ jobs:
       (github.event_name == 'issues' && (
         (github.event.action == 'labeled' && (github.event.label.name == 'codex-attempt' || github.event.label.name == 'codex-triage'))
       )) ||
-      (github.event_name == 'pull_request' && github.event.action == 'labeled' && github.event.label.name == 'codex-review')
+      (github.event_name == 'pull_request' && github.event.action == 'labeled' && (github.event.label.name == 'codex-review' || github.event.label.name == 'codex-rust-review'))
     runs-on: ubuntu-latest
     permissions:
       contents: write # can push or create branches

.vscode/extensions.json

@@ -0,0 +1,5 @@
+{
+    "recommendations": [
+        "tamasfe.even-better-toml",
+    ]
+}

.vscode/settings.json

@@ -6,5 +6,11 @@
     "[rust]": {
         "editor.defaultFormatter": "rust-lang.rust-analyzer",
         "editor.formatOnSave": true,
-    }
+    },
+    "[toml]": {
+        "editor.defaultFormatter": "tamasfe.even-better-toml",
+        "editor.formatOnSave": true,
+    },
+    "evenBetterToml.formatter.reorderArrays": true,
+    "evenBetterToml.formatter.reorderKeys": true,
 }

AGENTS.md

@@ -2,7 +2,9 @@
 
 In the codex-rs folder where the rust code lives:
 
-- Never add or modify any code related to `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR`. You operate in a sandbox where `CODEX_SANDBOX_NETWORK_DISABLED=1` will be set whenever you use the `shell` tool. Any existing code that uses `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR` was authored with this fact in mind. It is often used to early exit out of tests that the author knew you would not be able to run given your sandbox limitations.
+- Never add or modify any code related to `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR` or `CODEX_SANDBOX_ENV_VAR`.
+  - You operate in a sandbox where `CODEX_SANDBOX_NETWORK_DISABLED=1` will be set whenever you use the `shell` tool. Any existing code that uses `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR` was authored with this fact in mind. It is often used to early exit out of tests that the author knew you would not be able to run given your sandbox limitations.
+  - Similarly, when you spawn a process using Seatbelt (`/usr/bin/sandbox-exec`), `CODEX_SANDBOX=seatbelt` will be set on the child process. Integration tests that want to run Seatbelt themselves cannot be run under Seatbelt, so checks for `CODEX_SANDBOX=seatbelt` are also often used to early exit out of tests, as appropriate.
 
 Before creating a pull request with changes to `codex-rs`, run `just fmt` (in `codex-rs` directory) to format the code and `just fix` (in `codex-rs` directory) to fix any linter issues in the code, ensure the test suite passes by running `cargo test --all-features` in the `codex-rs` directory.
 

NOTICE

@@ -1,2 +1,6 @@
 OpenAI Codex
 Copyright 2025 OpenAI
+
+This project includes code derived from [Ratatui](https://github.com/ratatui/ratatui), licensed under the MIT license.
+Copyright (c) 2016-2022 Florian Dehau
+Copyright (c) 2023-2025 The Ratatui Developers

SUMMARY.md

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

codex-cli/bin/codex.js

@@ -83,6 +83,7 @@ if (wantsNative && process.platform !== 'win32') {
 
   const child = spawn(binaryPath, process.argv.slice(2), {
     stdio: "inherit",
+    env: { ...process.env, CODEX_MANAGED_BY_NPM: "1" },
   });
 
   child.on("error", (err) => {

codex-cli/src/utils/agent/sandbox/macos-seatbelt.ts

@@ -147,4 +147,8 @@ const READ_ONLY_SEATBELT_POLICY = `
   (sysctl-name "kern.version")
   (sysctl-name "sysctl.proc_cputype")
   (sysctl-name-prefix "hw.perflevel")
-)`.trim();
+)
+
+; Added on top of Chrome profile
+; Needed for python multiprocessing on MacOS for the SemLock
+(allow ipc-posix-sem)`.trim();

codex-rs/Cargo.lock

@@ -661,7 +661,7 @@ dependencies = [
  "clap",
  "codex-core",
  "serde",
- "toml 0.9.2",
+ "toml 0.9.4",
 ]
 
 [[package]]
@@ -695,9 +695,11 @@ dependencies = [
  "reqwest",
  "seccompiler",
  "serde",
+ "serde_bytes",
  "serde_json",
  "sha1",
  "shlex",
+ "similar",
  "strum_macros 0.27.2",
  "tempfile",
  "thiserror 2.0.12",
@@ -705,7 +707,7 @@ dependencies = [
  "tokio",
  "tokio-test",
  "tokio-util",
- "toml 0.9.2",
+ "toml 0.9.4",
  "tracing",
  "tree-sitter",
  "tree-sitter-bash",
@@ -793,6 +795,7 @@ dependencies = [
  "reqwest",
  "serde",
  "serde_json",
+ "tempfile",
  "tokio",
 ]
 
@@ -828,7 +831,7 @@ dependencies = [
  "tempfile",
  "tokio",
  "tokio-test",
- "toml 0.9.2",
+ "toml 0.9.4",
  "tracing",
  "tracing-subscriber",
  "uuid",
@@ -841,6 +844,7 @@ version = "0.0.0"
 dependencies = [
  "anyhow",
  "base64 0.22.1",
+ "chrono",
  "clap",
  "codex-ansi-escape",
  "codex-arg0",
@@ -856,20 +860,24 @@ dependencies = [
  "mcp-types",
  "path-clean",
  "pretty_assertions",
+ "rand 0.8.5",
  "ratatui",
  "ratatui-image",
  "regex-lite",
+ "reqwest",
+ "serde",
  "serde_json",
  "shlex",
  "strum 0.27.2",
  "strum_macros 0.27.2",
+ "supports-color",
+ "textwrap 0.16.2",
  "tokio",
  "tracing",
  "tracing-appender",
  "tracing-subscriber",
  "tui-input",
  "tui-markdown",
- "tui-textarea",
  "unicode-segmentation",
  "unicode-width 0.1.14",
  "uuid",
@@ -2009,7 +2017,7 @@ dependencies = [
  "libc",
  "percent-encoding",
  "pin-project-lite",
- "socket2 0.6.0",
+ "socket2",
  "system-configuration",
  "tokio",
  "tower-service",
@@ -2331,6 +2339,12 @@ dependencies = [
  "windows-sys 0.59.0",
 ]
 
+[[package]]
+name = "is_ci"
+version = "1.2.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7655c9839580ee829dfacba1d1278c2b7883e50a277ff7541299489d6bdfdc45"
+
 [[package]]
 name = "is_terminal_polyfill"
 version = "1.70.1"
@@ -2642,6 +2656,7 @@ version = "0.0.0"
 dependencies = [
  "anyhow",
  "assert_cmd",
+ "codex-core",
  "codex-mcp-server",
  "mcp-types",
  "pretty_assertions",
@@ -2649,6 +2664,7 @@ dependencies = [
  "shlex",
  "tempfile",
  "tokio",
+ "uuid",
  "wiremock",
 ]
 
@@ -3948,6 +3964,15 @@ dependencies = [
  "serde_derive",
 ]
 
+[[package]]
+name = "serde_bytes"
+version = "0.11.17"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "8437fd221bde2d4ca316d61b90e337e9e702b3820b87d63caa9ba6c02bd06d96"
+dependencies = [
+ "serde",
+]
+
 [[package]]
 name = "serde_derive"
 version = "1.0.219"
@@ -3972,9 +3997,9 @@ dependencies = [
 
 [[package]]
 name = "serde_json"
-version = "1.0.141"
+version = "1.0.142"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "30b9eff21ebe718216c6ec64e1d9ac57087aad11efc64e32002bce4a0d4c03d3"
+checksum = "030fedb782600dcbd6f02d479bf0d817ac3bb40d644745b769d6a96bc3afc5a7"
 dependencies = [
  "indexmap 2.10.0",
  "itoa",
@@ -4158,14 +4183,10 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "67b1b7a3b5fe4f1376887184045fcf45c69e92af734b7aaddc05fb777b6fbd03"
 
 [[package]]
-name = "socket2"
-version = "0.5.10"
+name = "smawk"
+version = "0.3.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e22376abed350d73dd1cd119b57ffccad95b4e585a7cda43e286245ce23c0678"
-dependencies = [
- "libc",
- "windows-sys 0.52.0",
-]
+checksum = "b7c388c1b5e93756d0c740965c41e8822f866621d41acbdf6336a6a168f8840c"
 
 [[package]]
 name = "socket2"
@@ -4219,7 +4240,7 @@ dependencies = [
  "starlark_syntax",
  "static_assertions",
  "strsim 0.10.0",
- "textwrap",
+ "textwrap 0.11.0",
  "thiserror 1.0.69",
 ]
 
@@ -4355,6 +4376,15 @@ version = "2.6.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "13c2bddecc57b384dee18652358fb23172facb8a2c51ccc10d74c157bdea3292"
 
+[[package]]
+name = "supports-color"
+version = "3.0.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c64fc7232dd8d2e4ac5ce4ef302b1d81e0b80d055b9d77c7c4f51f6aa4c867d6"
+dependencies = [
+ "is_ci",
+]
+
 [[package]]
 name = "syn"
 version = "1.0.109"
@@ -4508,6 +4538,17 @@ dependencies = [
  "unicode-width 0.1.14",
 ]
 
+[[package]]
+name = "textwrap"
+version = "0.16.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c13547615a44dc9c452a8a534638acdf07120d4b6847c8178705da06306a3057"
+dependencies = [
+ "smawk",
+ "unicode-linebreak",
+ "unicode-width 0.2.0",
+]
+
 [[package]]
 name = "thiserror"
 version = "1.0.69"
@@ -4622,9 +4663,9 @@ dependencies = [
 
 [[package]]
 name = "tokio"
-version = "1.46.1"
+version = "1.47.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0cc3a2344dafbe23a245241fe8b09735b521110d30fcefbbd5feb1797ca35d17"
+checksum = "89e49afdadebb872d3145a5638b59eb0691ea23e46ca484037cfab3b76b95038"
 dependencies = [
  "backtrace",
  "bytes",
@@ -4635,9 +4676,9 @@ dependencies = [
  "pin-project-lite",
  "signal-hook-registry",
  "slab",
- "socket2 0.5.10",
+ "socket2",
  "tokio-macros",
- "windows-sys 0.52.0",
+ "windows-sys 0.59.0",
 ]
 
 [[package]]
@@ -4722,9 +4763,9 @@ dependencies = [
 
 [[package]]
 name = "toml"
-version = "0.9.2"
+version = "0.9.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "ed0aee96c12fa71097902e0bb061a5e1ebd766a6636bb605ba401c45c1650eac"
+checksum = "41ae868b5a0f67631c14589f7e250c1ea2c574ee5ba21c6c8dd4b1485705a5a1"
 dependencies = [
  "indexmap 2.10.0",
  "serde",
@@ -4972,17 +5013,6 @@ dependencies = [
  "tracing",
 ]
 
-[[package]]
-name = "tui-textarea"
-version = "0.7.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0a5318dd619ed73c52a9417ad19046724effc1287fb75cdcc4eca1d6ac1acbae"
-dependencies = [
- "crossterm",
- "ratatui",
- "unicode-width 0.2.0",
-]
-
 [[package]]
 name = "typenum"
 version = "1.18.0"
@@ -5001,6 +5031,12 @@ version = "1.0.18"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "5a5f39404a5da50712a4c1eecf25e90dd62b613502b7e925fd4e4d19b5c96512"
 
+[[package]]
+name = "unicode-linebreak"
+version = "0.1.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3b09c83c3c29d37506a3e260c08c03743a6bb66a9cd432c6934ab501a190571f"
+
 [[package]]
 name = "unicode-segmentation"
 version = "1.12.0"

codex-rs/Cargo.toml

@@ -1,5 +1,4 @@
 [workspace]
-resolver = "2"
 members = [
     "ansi-escape",
     "apply-patch",
@@ -17,6 +16,7 @@ members = [
     "mcp-types",
     "tui",
 ]
+resolver = "2"
 
 [workspace.package]
 version = "0.0.0"
@@ -45,4 +45,3 @@ codegen-units = 1
 [patch.crates-io]
 # ratatui = { path = "../../ratatui" }
 ratatui = { git = "https://github.com/nornagon/ratatui", branch = "nornagon-v0.29.0-patch" }
-

codex-rs/ansi-escape/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-ansi-escape"
 version = { workspace = true }
-edition = "2024"
 
 [lib]
 name = "codex_ansi_escape"
@@ -10,7 +10,7 @@ path = "src/lib.rs"
 [dependencies]
 ansi-to-tui = "7.0.0"
 ratatui = { version = "0.29.0", features = [
-    "unstable-widget-ref",
     "unstable-rendered-line-info",
+    "unstable-widget-ref",
 ] }
 tracing = { version = "0.1.41", features = ["log"] }

codex-rs/apply-patch/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-apply-patch"
 version = { workspace = true }
-edition = "2024"
 
 [lib]
 name = "codex_apply_patch"

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

@@ -58,16 +58,24 @@ impl PartialEq for IoError {
 
 #[derive(Debug, PartialEq)]
 pub enum MaybeApplyPatch {
-    Body(Vec<Hunk>),
+    Body(ApplyPatchArgs),
     ShellParseError(ExtractHeredocError),
     PatchParseError(ParseError),
     NotApplyPatch,
 }
 
+/// Both the raw PATCH argument to `apply_patch` as well as the PATCH argument
+/// parsed into hunks.
+#[derive(Debug, PartialEq)]
+pub struct ApplyPatchArgs {
+    pub patch: String,
+    pub hunks: Vec<Hunk>,
+}
+
 pub fn maybe_parse_apply_patch(argv: &[String]) -> MaybeApplyPatch {
     match argv {
         [cmd, body] if cmd == "apply_patch" => match parse_patch(body) {
-            Ok(hunks) => MaybeApplyPatch::Body(hunks),
+            Ok(source) => MaybeApplyPatch::Body(source),
             Err(e) => MaybeApplyPatch::PatchParseError(e),
         },
         [bash, flag, script]
@@ -77,7 +85,7 @@ pub fn maybe_parse_apply_patch(argv: &[String]) -> MaybeApplyPatch {
         {
             match extract_heredoc_body_from_apply_patch_command(script) {
                 Ok(body) => match parse_patch(&body) {
-                    Ok(hunks) => MaybeApplyPatch::Body(hunks),
+                    Ok(source) => MaybeApplyPatch::Body(source),
                     Err(e) => MaybeApplyPatch::PatchParseError(e),
                 },
                 Err(e) => MaybeApplyPatch::ShellParseError(e),
@@ -116,11 +124,19 @@ pub enum MaybeApplyPatchVerified {
     NotApplyPatch,
 }
 
-#[derive(Debug, PartialEq)]
 /// ApplyPatchAction is the result of parsing an `apply_patch` command. By
 /// construction, all paths should be absolute paths.
+#[derive(Debug, PartialEq)]
 pub struct ApplyPatchAction {
     changes: HashMap<PathBuf, ApplyPatchFileChange>,
+
+    /// The raw patch argument that can be used with `apply_patch` as an exec
+    /// call. i.e., if the original arg was parsed in "lenient" mode with a
+    /// heredoc, this should be the value without the heredoc wrapper.
+    pub patch: String,
+
+    /// The working directory that was used to resolve relative paths in the patch.
+    pub cwd: PathBuf,
 }
 
 impl ApplyPatchAction {
@@ -140,8 +156,28 @@ impl ApplyPatchAction {
             panic!("path must be absolute");
         }
 
+        #[allow(clippy::expect_used)]
+        let filename = path
+            .file_name()
+            .expect("path should not be empty")
+            .to_string_lossy();
+        let patch = format!(
+            r#"*** Begin Patch
+*** Update File: {filename}
+@@
++ {content}
+*** End Patch"#,
+        );
         let changes = HashMap::from([(path.to_path_buf(), ApplyPatchFileChange::Add { content })]);
-        Self { changes }
+        #[allow(clippy::expect_used)]
+        Self {
+            changes,
+            cwd: path
+                .parent()
+                .expect("path should have parent")
+                .to_path_buf(),
+            patch,
+        }
     }
 }
 
@@ -149,7 +185,7 @@ impl ApplyPatchAction {
 /// patch.
 pub fn maybe_parse_apply_patch_verified(argv: &[String], cwd: &Path) -> MaybeApplyPatchVerified {
     match maybe_parse_apply_patch(argv) {
-        MaybeApplyPatch::Body(hunks) => {
+        MaybeApplyPatch::Body(ApplyPatchArgs { patch, hunks }) => {
             let mut changes = HashMap::new();
             for hunk in hunks {
                 let path = hunk.resolve_path(cwd);
@@ -183,7 +219,11 @@ pub fn maybe_parse_apply_patch_verified(argv: &[String], cwd: &Path) -> MaybeApp
                     }
                 }
             }
-            MaybeApplyPatchVerified::Body(ApplyPatchAction { changes })
+            MaybeApplyPatchVerified::Body(ApplyPatchAction {
+                changes,
+                patch,
+                cwd: cwd.to_path_buf(),
+            })
         }
         MaybeApplyPatch::ShellParseError(e) => MaybeApplyPatchVerified::ShellParseError(e),
         MaybeApplyPatch::PatchParseError(e) => MaybeApplyPatchVerified::CorrectnessError(e.into()),
@@ -264,7 +304,7 @@ pub fn apply_patch(
     stderr: &mut impl std::io::Write,
 ) -> Result<(), ApplyPatchError> {
     let hunks = match parse_patch(patch) {
-        Ok(hunks) => hunks,
+        Ok(source) => source.hunks,
         Err(e) => {
             match &e {
                 InvalidPatchError(message) => {
@@ -652,7 +692,7 @@ mod tests {
         ]);
 
         match maybe_parse_apply_patch(&args) {
-            MaybeApplyPatch::Body(hunks) => {
+            MaybeApplyPatch::Body(ApplyPatchArgs { hunks, patch: _ }) => {
                 assert_eq!(
                     hunks,
                     vec![Hunk::AddFile {
@@ -679,7 +719,7 @@ PATCH"#,
         ]);
 
         match maybe_parse_apply_patch(&args) {
-            MaybeApplyPatch::Body(hunks) => {
+            MaybeApplyPatch::Body(ApplyPatchArgs { hunks, patch: _ }) => {
                 assert_eq!(
                     hunks,
                     vec![Hunk::AddFile {
@@ -954,7 +994,7 @@ PATCH"#,
         ));
         let patch = parse_patch(&patch).unwrap();
 
-        let update_file_chunks = match patch.as_slice() {
+        let update_file_chunks = match patch.hunks.as_slice() {
             [Hunk::UpdateFile { chunks, .. }] => chunks,
             _ => panic!("Expected a single UpdateFile hunk"),
         };
@@ -992,7 +1032,7 @@ PATCH"#,
         ));
 
         let patch = parse_patch(&patch).unwrap();
-        let chunks = match patch.as_slice() {
+        let chunks = match patch.hunks.as_slice() {
             [Hunk::UpdateFile { chunks, .. }] => chunks,
             _ => panic!("Expected a single UpdateFile hunk"),
         };
@@ -1029,7 +1069,7 @@ PATCH"#,
         ));
 
         let patch = parse_patch(&patch).unwrap();
-        let chunks = match patch.as_slice() {
+        let chunks = match patch.hunks.as_slice() {
             [Hunk::UpdateFile { chunks, .. }] => chunks,
             _ => panic!("Expected a single UpdateFile hunk"),
         };
@@ -1064,7 +1104,7 @@ PATCH"#,
         ));
 
         let patch = parse_patch(&patch).unwrap();
-        let chunks = match patch.as_slice() {
+        let chunks = match patch.hunks.as_slice() {
             [Hunk::UpdateFile { chunks, .. }] => chunks,
             _ => panic!("Expected a single UpdateFile hunk"),
         };
@@ -1110,7 +1150,7 @@ PATCH"#,
 
         // Extract chunks then build the unified diff.
         let parsed = parse_patch(&patch).unwrap();
-        let chunks = match parsed.as_slice() {
+        let chunks = match parsed.hunks.as_slice() {
             [Hunk::UpdateFile { chunks, .. }] => chunks,
             _ => panic!("Expected a single UpdateFile hunk"),
         };
@@ -1193,6 +1233,8 @@ g
                         new_content: "updated session directory content\n".to_string(),
                     },
                 )]),
+                patch: argv[1].clone(),
+                cwd: session_dir.path().to_path_buf(),
             })
         );
     }

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

@@ -22,6 +22,7 @@
 //!
 //! The parser below is a little more lenient than the explicit spec and allows for
 //! leading/trailing whitespace around patch markers.
+use crate::ApplyPatchArgs;
 use std::path::Path;
 use std::path::PathBuf;
 
@@ -102,7 +103,7 @@ pub struct UpdateFileChunk {
     pub is_end_of_file: bool,
 }
 
-pub fn parse_patch(patch: &str) -> Result<Vec<Hunk>, ParseError> {
+pub fn parse_patch(patch: &str) -> Result<ApplyPatchArgs, ParseError> {
     let mode = if PARSE_IN_STRICT_MODE {
         ParseMode::Strict
     } else {
@@ -150,7 +151,7 @@ enum ParseMode {
     Lenient,
 }
 
-fn parse_patch_text(patch: &str, mode: ParseMode) -> Result<Vec<Hunk>, ParseError> {
+fn parse_patch_text(patch: &str, mode: ParseMode) -> Result<ApplyPatchArgs, ParseError> {
     let lines: Vec<&str> = patch.trim().lines().collect();
     let lines: &[&str] = match check_patch_boundaries_strict(&lines) {
         Ok(()) => &lines,
@@ -173,7 +174,8 @@ fn parse_patch_text(patch: &str, mode: ParseMode) -> Result<Vec<Hunk>, ParseErro
         line_number += hunk_lines;
         remaining_lines = &remaining_lines[hunk_lines..]
     }
-    Ok(hunks)
+    let patch = lines.join("\n");
+    Ok(ApplyPatchArgs { hunks, patch })
 }
 
 /// Checks the start and end lines of the patch text for `apply_patch`,
@@ -425,6 +427,7 @@ fn parse_update_file_chunk(
 }
 
 #[test]
+#[allow(clippy::unwrap_used)]
 fn test_parse_patch() {
     assert_eq!(
         parse_patch_text("bad", ParseMode::Strict),
@@ -455,8 +458,10 @@ fn test_parse_patch() {
             "*** Begin Patch\n\
              *** End Patch",
             ParseMode::Strict
-        ),
-        Ok(Vec::new())
+        )
+        .unwrap()
+        .hunks,
+        Vec::new()
     );
     assert_eq!(
         parse_patch_text(
@@ -472,8 +477,10 @@ fn test_parse_patch() {
              +    return 123\n\
              *** End Patch",
             ParseMode::Strict
-        ),
-        Ok(vec![
+        )
+        .unwrap()
+        .hunks,
+        vec![
             AddFile {
                 path: PathBuf::from("path/add.py"),
                 contents: "abc\ndef\n".to_string()
@@ -491,7 +498,7 @@ fn test_parse_patch() {
                     is_end_of_file: false
                 }]
             }
-        ])
+        ]
     );
     // Update hunk followed by another hunk (Add File).
     assert_eq!(
@@ -504,8 +511,10 @@ fn test_parse_patch() {
              +content\n\
              *** End Patch",
             ParseMode::Strict
-        ),
-        Ok(vec![
+        )
+        .unwrap()
+        .hunks,
+        vec![
             UpdateFile {
                 path: PathBuf::from("file.py"),
                 move_path: None,
@@ -520,7 +529,7 @@ fn test_parse_patch() {
                 path: PathBuf::from("other.py"),
                 contents: "content\n".to_string()
             }
-        ])
+        ]
     );
 
     // Update hunk without an explicit @@ header for the first chunk should parse.
@@ -533,8 +542,10 @@ fn test_parse_patch() {
 +bar
 *** End Patch"#,
             ParseMode::Strict
-        ),
-        Ok(vec![UpdateFile {
+        )
+        .unwrap()
+        .hunks,
+        vec![UpdateFile {
             path: PathBuf::from("file2.py"),
             move_path: None,
             chunks: vec![UpdateFileChunk {
@@ -543,7 +554,7 @@ fn test_parse_patch() {
                 new_lines: vec!["import foo".to_string(), "bar".to_string()],
                 is_end_of_file: false,
             }],
-        }])
+        }]
     );
 }
 
@@ -574,7 +585,10 @@ fn test_parse_patch_lenient() {
     );
     assert_eq!(
         parse_patch_text(&patch_text_in_heredoc, ParseMode::Lenient),
-        Ok(expected_patch.clone())
+        Ok(ApplyPatchArgs {
+            hunks: expected_patch.clone(),
+            patch: patch_text.to_string()
+        })
     );
 
     let patch_text_in_single_quoted_heredoc = format!("<<'EOF'\n{patch_text}\nEOF\n");
@@ -584,7 +598,10 @@ fn test_parse_patch_lenient() {
     );
     assert_eq!(
         parse_patch_text(&patch_text_in_single_quoted_heredoc, ParseMode::Lenient),
-        Ok(expected_patch.clone())
+        Ok(ApplyPatchArgs {
+            hunks: expected_patch.clone(),
+            patch: patch_text.to_string()
+        })
     );
 
     let patch_text_in_double_quoted_heredoc = format!("<<\"EOF\"\n{patch_text}\nEOF\n");
@@ -594,7 +611,10 @@ fn test_parse_patch_lenient() {
     );
     assert_eq!(
         parse_patch_text(&patch_text_in_double_quoted_heredoc, ParseMode::Lenient),
-        Ok(expected_patch.clone())
+        Ok(ApplyPatchArgs {
+            hunks: expected_patch.clone(),
+            patch: patch_text.to_string()
+        })
     );
 
     let patch_text_in_mismatched_quotes_heredoc = format!("<<\"EOF'\n{patch_text}\nEOF\n");

codex-rs/arg0/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-arg0"
 version = { workspace = true }
-edition = "2024"
 
 [lib]
 name = "codex_arg0"

codex-rs/arg0/src/lib.rs

@@ -2,6 +2,8 @@ use std::future::Future;
 use std::path::Path;
 use std::path::PathBuf;
 
+use codex_core::CODEX_APPLY_PATCH_ARG1;
+
 /// While we want to deploy the Codex CLI as a single executable for simplicity,
 /// we also want to expose some of its functionality as distinct CLIs, so we use
 /// the "arg0 trick" to determine which CLI to dispatch. This effectively allows
@@ -43,7 +45,7 @@ where
     }
 
     let argv1 = args.next().unwrap_or_default();
-    if argv1 == "--codex-run-as-apply-patch" {
+    if argv1 == CODEX_APPLY_PATCH_ARG1 {
         let patch_arg = args.next().and_then(|s| s.to_str().map(|s| s.to_owned()));
         let exit_code = match patch_arg {
             Some(patch_arg) => {
@@ -55,7 +57,7 @@ where
                 }
             }
             None => {
-                eprintln!("Error: --codex-run-as-apply-patch requires a UTF-8 PATCH argument.");
+                eprintln!("Error: {CODEX_APPLY_PATCH_ARG1} requires a UTF-8 PATCH argument.");
                 1
             }
         };

codex-rs/chatgpt/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-chatgpt"
 version = { workspace = true }
-edition = "2024"
 
 [lints]
 workspace = true
@@ -9,12 +9,12 @@ workspace = true
 [dependencies]
 anyhow = "1"
 clap = { version = "4", features = ["derive"] }
-serde = { version = "1", features = ["derive"] }
-serde_json = "1"
 codex-common = { path = "../common", features = ["cli"] }
 codex-core = { path = "../core" }
 codex-login = { path = "../login" }
 reqwest = { version = "0.12", features = ["json", "stream"] }
+serde = { version = "1", features = ["derive"] }
+serde_json = "1"
 tokio = { version = "1", features = ["full"] }
 
 [dev-dependencies]

codex-rs/chatgpt/src/chatgpt_token.rs

@@ -18,7 +18,7 @@ pub fn set_chatgpt_token_data(value: TokenData) {
 
 /// Initialize the ChatGPT token from auth.json file
 pub async fn init_chatgpt_token_from_auth(codex_home: &Path) -> std::io::Result<()> {
-    let auth = codex_login::load_auth(codex_home)?;
+    let auth = codex_login::load_auth(codex_home, true)?;
     if let Some(auth) = auth {
         let token_data = auth.get_token_data().await?;
         set_chatgpt_token_data(token_data);

codex-rs/chatgpt/tests/apply_command_e2e.rs

@@ -10,8 +10,13 @@ use tokio::process::Command;
 async fn create_temp_git_repo() -> anyhow::Result<TempDir> {
     let temp_dir = TempDir::new()?;
     let repo_path = temp_dir.path();
+    let envs = vec![
+        ("GIT_CONFIG_GLOBAL", "/dev/null"),
+        ("GIT_CONFIG_NOSYSTEM", "1"),
+    ];
 
     let output = Command::new("git")
+        .envs(envs.clone())
         .args(["init"])
         .current_dir(repo_path)
         .output()
@@ -25,12 +30,14 @@ async fn create_temp_git_repo() -> anyhow::Result<TempDir> {
     }
 
     Command::new("git")
+        .envs(envs.clone())
         .args(["config", "user.email", "test@example.com"])
         .current_dir(repo_path)
         .output()
         .await?;
 
     Command::new("git")
+        .envs(envs.clone())
         .args(["config", "user.name", "Test User"])
         .current_dir(repo_path)
         .output()
@@ -39,12 +46,14 @@ async fn create_temp_git_repo() -> anyhow::Result<TempDir> {
     std::fs::write(repo_path.join("README.md"), "# Test Repo\n")?;
 
     Command::new("git")
+        .envs(envs.clone())
         .args(["add", "README.md"])
         .current_dir(repo_path)
         .output()
         .await?;
 
     let output = Command::new("git")
+        .envs(envs.clone())
         .args(["commit", "-m", "Initial commit"])
         .current_dir(repo_path)
         .output()

codex-rs/cli/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-cli"
 version = { workspace = true }
-edition = "2024"
 
 [[bin]]
 name = "codex"
@@ -20,8 +20,8 @@ clap = { version = "4", features = ["derive"] }
 clap_complete = "4"
 codex-arg0 = { path = "../arg0" }
 codex-chatgpt = { path = "../chatgpt" }
-codex-core = { path = "../core" }
 codex-common = { path = "../common", features = ["cli"] }
+codex-core = { path = "../core" }
 codex-exec = { path = "../exec" }
 codex-login = { path = "../login" }
 codex-mcp-server = { path = "../mcp-server" }

codex-rs/cli/src/debug_sandbox.rs

@@ -4,10 +4,10 @@ use codex_common::CliConfigOverrides;
 use codex_core::config::Config;
 use codex_core::config::ConfigOverrides;
 use codex_core::config_types::SandboxMode;
-use codex_core::exec::StdioPolicy;
 use codex_core::exec::spawn_command_under_linux_sandbox;
-use codex_core::exec::spawn_command_under_seatbelt;
 use codex_core::exec_env::create_env;
+use codex_core::seatbelt::spawn_command_under_seatbelt;
+use codex_core::spawn::StdioPolicy;
 
 use crate::LandlockCommand;
 use crate::SeatbeltCommand;

codex-rs/cli/src/login.rs

@@ -1,8 +1,12 @@
+use std::env;
+
 use codex_common::CliConfigOverrides;
 use codex_core::config::Config;
 use codex_core::config::ConfigOverrides;
 use codex_login::AuthMode;
+use codex_login::OPENAI_API_KEY_ENV_VAR;
 use codex_login::load_auth;
+use codex_login::login_with_api_key;
 use codex_login::login_with_chatgpt;
 
 pub async fn run_login_with_chatgpt(cli_config_overrides: CliConfigOverrides) -> ! {
@@ -21,14 +25,40 @@ pub async fn run_login_with_chatgpt(cli_config_overrides: CliConfigOverrides) ->
     }
 }
 
+pub async fn run_login_with_api_key(
+    cli_config_overrides: CliConfigOverrides,
+    api_key: String,
+) -> ! {
+    let config = load_config_or_exit(cli_config_overrides);
+
+    match login_with_api_key(&config.codex_home, &api_key) {
+        Ok(_) => {
+            eprintln!("Successfully logged in");
+            std::process::exit(0);
+        }
+        Err(e) => {
+            eprintln!("Error logging in: {e}");
+            std::process::exit(1);
+        }
+    }
+}
+
 pub async fn run_login_status(cli_config_overrides: CliConfigOverrides) -> ! {
     let config = load_config_or_exit(cli_config_overrides);
 
-    match load_auth(&config.codex_home) {
+    match load_auth(&config.codex_home, true) {
         Ok(Some(auth)) => match auth.mode {
             AuthMode::ApiKey => {
                 if let Some(api_key) = auth.api_key.as_deref() {
                     eprintln!("Logged in using an API key - {}", safe_format_key(api_key));
+
+                    if let Ok(env_api_key) = env::var(OPENAI_API_KEY_ENV_VAR) {
+                        if env_api_key == api_key {
+                            eprintln!(
+                                "   API loaded from OPENAI_API_KEY environment variable or .env file"
+                            );
+                        }
+                    }
                 } else {
                     eprintln!("Logged in using an API key");
                 }

codex-rs/cli/src/main.rs

@@ -8,6 +8,7 @@ use codex_chatgpt::apply_command::run_apply_command;
 use codex_cli::LandlockCommand;
 use codex_cli::SeatbeltCommand;
 use codex_cli::login::run_login_status;
+use codex_cli::login::run_login_with_api_key;
 use codex_cli::login::run_login_with_chatgpt;
 use codex_cli::proto;
 use codex_common::CliConfigOverrides;
@@ -92,6 +93,9 @@ struct LoginCommand {
     #[clap(skip)]
     config_overrides: CliConfigOverrides,
 
+    #[arg(long = "api-key", value_name = "API_KEY")]
+    api_key: Option<String>,
+
     #[command(subcommand)]
     action: Option<LoginSubcommand>,
 }
@@ -133,7 +137,11 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
                     run_login_status(login_cli.config_overrides).await;
                 }
                 None => {
-                    run_login_with_chatgpt(login_cli.config_overrides).await;
+                    if let Some(api_key) = login_cli.api_key {
+                        run_login_with_api_key(login_cli.config_overrides, api_key).await;
+                    } else {
+                        run_login_with_chatgpt(login_cli.config_overrides).await;
+                    }
                 }
             }
         }

codex-rs/cli/src/proto.rs

@@ -36,7 +36,7 @@ pub async fn run_main(opts: ProtoCli) -> anyhow::Result<()> {
         .map_err(anyhow::Error::msg)?;
 
     let config = Config::load_with_cli_overrides(overrides_vec, ConfigOverrides::default())?;
-    let auth = load_auth(&config.codex_home)?;
+    let auth = load_auth(&config.codex_home, true)?;
     let ctrl_c = notify_on_sigint();
     let CodexSpawnOk { codex, .. } = Codex::spawn(config, auth, ctrl_c.clone()).await?;
     let codex = Arc::new(codex);

codex-rs/common/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-common"
 version = { workspace = true }
-edition = "2024"
 
 [lints]
 workspace = true
@@ -9,11 +9,11 @@ workspace = true
 [dependencies]
 clap = { version = "4", features = ["derive", "wrap_help"], optional = true }
 codex-core = { path = "../core" }
-toml = { version = "0.9", optional = true }
 serde = { version = "1", optional = true }
+toml = { version = "0.9", optional = true }
 
 [features]
 # Separate feature so that `clap` is not a mandatory dependency.
-cli = ["clap", "toml", "serde"]
+cli = ["clap", "serde", "toml"]
 elapsed = []
 sandbox_summary = []

codex-rs/common/src/sandbox_summary.rs

@@ -7,6 +7,7 @@ pub fn summarize_sandbox_policy(sandbox_policy: &SandboxPolicy) -> String {
         SandboxPolicy::WorkspaceWrite {
             writable_roots,
             network_access,
+            include_default_writable_roots,
         } => {
             let mut summary = "workspace-write".to_string();
             if !writable_roots.is_empty() {
@@ -19,6 +20,9 @@ pub fn summarize_sandbox_policy(sandbox_policy: &SandboxPolicy) -> String {
                         .join(", ")
                 ));
             }
+            if !*include_default_writable_roots {
+                summary.push_str(" (exact writable roots)");
+            }
             if *network_access {
                 summary.push_str(" (network access enabled)");
             }

codex-rs/config.md

@@ -259,6 +259,8 @@ disk, but attempts to write a file or access the network will be blocked.
 
 A more relaxed policy is `workspace-write`. When specified, the current working directory for the Codex task will be writable (as well as `$TMPDIR` on macOS). Note that the CLI defaults to using the directory where it was spawned as `cwd`, though this can be overridden using `--cwd/-C`.
 
+On macOS (and soon Linux), all writable roots (including `cwd`) that contain a `.git/` folder _as an immediate child_ will configure the `.git/` folder to be read-only while the rest of the Git repository will be writable. This means that commands like `git commit` will fail, by default (as it entails writing to `.git/`), and will require Codex to ask for permission.
+
 ```toml
 # same as `--sandbox workspace-write`
 sandbox_mode = "workspace-write"

codex-rs/core/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-core"
 version = { workspace = true }
-edition = "2024"
 
 [lib]
 name = "codex_core"
@@ -15,10 +15,10 @@ anyhow = "1"
 async-channel = "2.3.1"
 base64 = "0.22"
 bytes = "1.10.1"
-codex-apply-patch = { path = "../apply-patch" }
-codex-mcp-client = { path = "../mcp-client" }
 chrono = { version = "0.4", features = ["serde"] }
+codex-apply-patch = { path = "../apply-patch" }
 codex-login = { path = "../login" }
+codex-mcp-client = { path = "../mcp-client" }
 dirs = "6"
 env-flags = "0.1.1"
 eventsource-stream = "0.2.3"
@@ -31,8 +31,10 @@ rand = "0.9"
 reqwest = { version = "0.12", features = ["json", "stream"] }
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
+serde_bytes = "0.11"
 sha1 = "0.10.6"
 shlex = "1.3.0"
+similar = "2.7.0"
 strum_macros = "0.27.2"
 thiserror = "2.0.12"
 time = { version = "0.3", features = ["formatting", "local-offset", "macros"] }
@@ -44,13 +46,13 @@ tokio = { version = "1", features = [
     "signal",
 ] }
 tokio-util = "0.7.14"
-toml = "0.9.2"
+toml = "0.9.4"
 tracing = { version = "0.1.41", features = ["log"] }
 tree-sitter = "0.25.8"
 tree-sitter-bash = "0.25.0"
 uuid = { version = "1", features = ["serde", "v4"] }
-wildmatch = "2.4.0"
 whoami = "1.6.0"
+wildmatch = "2.4.0"
 
 
 [target.'cfg(target_os = "linux")'.dependencies]

codex-rs/core/prompt.md

@@ -10,7 +10,7 @@ You MUST adhere to the following criteria when executing the task:
 - Showing user code and tool call details is allowed.
 - User instructions may overwrite the _CODING GUIDELINES_ section in this developer message.
 - Do not use \`ls -R\`, \`find\`, or \`grep\` - these are slow in large repos. Use \`rg\` and \`rg --files\`.
-- Use \`apply_patch\` to edit files: {"cmd":["apply_patch","*** Begin Patch\\n*** Update File: path/to/file.py\\n@@ def example():\\n- pass\\n+ return 123\\n*** End Patch"]}
+- Use the \`apply_patch\` shell command to edit files: {"command":["apply_patch","*** Begin Patch\\n*** Update File: path/to/file.py\\n@@ def example():\\n- pass\\n+ return 123\\n*** End Patch"]}
 - If completing the user's task requires writing or modifying files:
   - Your code and final answer should follow these _CODING GUIDELINES_:
     - Fix the problem at the root cause rather than applying surface-level patches, when possible.
@@ -40,16 +40,16 @@ You MUST adhere to the following criteria when executing the task:
 
 Your patch language is a stripped‑down, file‑oriented diff format designed to be easy to parse and safe to apply. You can think of it as a high‑level envelope:
 
-**_ Begin Patch
+*** Begin Patch
 [ one or more file sections ]
-_** End Patch
+*** End Patch
 
 Within that envelope, you get a sequence of file operations.
 You MUST include a header to specify the action you are taking.
 Each operation starts with one of three headers:
 
-**_ Add File: <path> - create a new file. Every following line is a + line (the initial contents).
-_** Delete File: <path> - remove an existing file. Nothing follows.
+*** Add File: <path> - create a new file. Every following line is a + line (the initial contents).
+*** Delete File: <path> - remove an existing file. Nothing follows.
 \*\*\* Update File: <path> - patch an existing file in place (optionally with a rename).
 
 May be immediately followed by \*\*\* Move to: <new path> if you want to rename the file.
@@ -63,28 +63,28 @@ Within a hunk each line starts with:
   At the end of a truncated hunk you can emit \*\*\* End of File.
 
 Patch := Begin { FileOp } End
-Begin := "**_ Begin Patch" NEWLINE
-End := "_** End Patch" NEWLINE
+Begin := "*** Begin Patch" NEWLINE
+End := "*** End Patch" NEWLINE
 FileOp := AddFile | DeleteFile | UpdateFile
-AddFile := "**_ Add File: " path NEWLINE { "+" line NEWLINE }
-DeleteFile := "_** Delete File: " path NEWLINE
-UpdateFile := "**_ Update File: " path NEWLINE [ MoveTo ] { Hunk }
-MoveTo := "_** Move to: " newPath NEWLINE
+AddFile := "*** Add File: " path NEWLINE { "+" line NEWLINE }
+DeleteFile := "*** Delete File: " path NEWLINE
+UpdateFile := "*** Update File: " path NEWLINE [ MoveTo ] { Hunk }
+MoveTo := "*** Move to: " newPath NEWLINE
 Hunk := "@@" [ header ] NEWLINE { HunkLine } [ "*** End of File" NEWLINE ]
 HunkLine := (" " | "-" | "+") text NEWLINE
 
 A full patch can combine several operations:
 
-**_ Begin Patch
-_** Add File: hello.txt
+*** Begin Patch
+*** Add File: hello.txt
 +Hello world
-**_ Update File: src/app.py
-_** Move to: src/main.py
+*** Update File: src/app.py
+*** Move to: src/main.py
 @@ def greet():
 -print("Hi")
 +print("Hello, world!")
-**_ Delete File: obsolete.txt
-_** End Patch
+*** Delete File: obsolete.txt
+*** End Patch
 
 It is important to remember:
 
@@ -96,3 +96,12 @@ You can invoke apply_patch like:
 ```
 shell {"command":["apply_patch","*** Begin Patch\n*** Add File: hello.txt\n+Hello, world!\n*** End Patch\n"]}
 ```
+
+Plan updates
+
+A tool named `update_plan` is available. Use it to keep an up‑to‑date, step‑by‑step plan for the task so you can follow your progress. When making your plans, keep in mind that you are a deployed coding agent - `update_plan` calls should not involve doing anything that you aren't capable of doing. For example, `update_plan` calls should NEVER contain tasks to merge your own pull requests. Only stop to ask the user if you genuinely need their feedback on a change.
+
+- At the start of any nontrivial task, call `update_plan` with an initial plan: a short list of 1‑sentence steps with a `status` for each step (`pending`, `in_progress`, or `completed`). There should always be exactly one `in_progress` step until everything is done.
+- Whenever you finish a step, call `update_plan` again, marking the finished step as `completed` and the next step as `in_progress`.
+- If your plan needs to change, call `update_plan` with the revised steps and include an `explanation` describing the change.
+- When all steps are complete, make a final `update_plan` call with all steps marked `completed`.

codex-rs/core/src/apply_patch.rs

@@ -1,277 +1,107 @@
 use crate::codex::Session;
 use crate::models::FunctionCallOutputPayload;
 use crate::models::ResponseInputItem;
-use crate::protocol::Event;
-use crate::protocol::EventMsg;
 use crate::protocol::FileChange;
-use crate::protocol::PatchApplyBeginEvent;
-use crate::protocol::PatchApplyEndEvent;
 use crate::protocol::ReviewDecision;
 use crate::safety::SafetyCheck;
 use crate::safety::assess_patch_safety;
-use anyhow::Context;
-use codex_apply_patch::AffectedPaths;
 use codex_apply_patch::ApplyPatchAction;
 use codex_apply_patch::ApplyPatchFileChange;
-use codex_apply_patch::print_summary;
 use std::collections::HashMap;
 use std::path::Path;
 use std::path::PathBuf;
 
+pub const CODEX_APPLY_PATCH_ARG1: &str = "--codex-run-as-apply-patch";
+
+pub(crate) enum InternalApplyPatchInvocation {
+    /// The `apply_patch` call was handled programmatically, without any sort
+    /// of sandbox, because the user explicitly approved it. This is the
+    /// result to use with the `shell` function call that contained `apply_patch`.
+    Output(ResponseInputItem),
+
+    /// The `apply_patch` call was approved, either automatically because it
+    /// appears that it should be allowed based on the user's sandbox policy
+    /// *or* because the user explicitly approved it. In either case, we use
+    /// exec with [`CODEX_APPLY_PATCH_ARG1`] to realize the `apply_patch` call,
+    /// but [`ApplyPatchExec::auto_approved`] is used to determine the sandbox
+    /// used with the `exec()`.
+    DelegateToExec(ApplyPatchExec),
+}
+
+pub(crate) struct ApplyPatchExec {
+    pub(crate) action: ApplyPatchAction,
+    pub(crate) user_explicitly_approved_this_action: bool,
+}
+
+impl From<ResponseInputItem> for InternalApplyPatchInvocation {
+    fn from(item: ResponseInputItem) -> Self {
+        InternalApplyPatchInvocation::Output(item)
+    }
+}
+
 pub(crate) async fn apply_patch(
     sess: &Session,
-    sub_id: String,
-    call_id: String,
+    sub_id: &str,
+    call_id: &str,
     action: ApplyPatchAction,
-) -> ResponseInputItem {
+) -> InternalApplyPatchInvocation {
     let writable_roots_snapshot = {
         #[allow(clippy::unwrap_used)]
         let guard = sess.writable_roots.lock().unwrap();
         guard.clone()
     };
 
-    let auto_approved = match assess_patch_safety(
+    match assess_patch_safety(
         &action,
         sess.approval_policy,
         &writable_roots_snapshot,
         &sess.cwd,
     ) {
-        SafetyCheck::AutoApprove { .. } => true,
+        SafetyCheck::AutoApprove { .. } => {
+            InternalApplyPatchInvocation::DelegateToExec(ApplyPatchExec {
+                action,
+                user_explicitly_approved_this_action: false,
+            })
+        }
         SafetyCheck::AskUser => {
             // Compute a readable summary of path changes to include in the
             // approval request so the user can make an informed decision.
+            //
+            // Note that it might be worth expanding this approval request to
+            // give the user the option to expand the set of writable roots so
+            // that similar patches can be auto-approved in the future during
+            // this session.
             let rx_approve = sess
-                .request_patch_approval(sub_id.clone(), call_id.clone(), &action, None, None)
+                .request_patch_approval(sub_id.to_owned(), call_id.to_owned(), &action, None, None)
                 .await;
             match rx_approve.await.unwrap_or_default() {
-                ReviewDecision::Approved | ReviewDecision::ApprovedForSession => false,
+                ReviewDecision::Approved | ReviewDecision::ApprovedForSession => {
+                    InternalApplyPatchInvocation::DelegateToExec(ApplyPatchExec {
+                        action,
+                        user_explicitly_approved_this_action: true,
+                    })
+                }
                 ReviewDecision::Denied | ReviewDecision::Abort => {
-                    return ResponseInputItem::FunctionCallOutput {
-                        call_id,
+                    ResponseInputItem::FunctionCallOutput {
+                        call_id: call_id.to_owned(),
                         output: FunctionCallOutputPayload {
                             content: "patch rejected by user".to_string(),
                             success: Some(false),
                         },
-                    };
+                    }
+                    .into()
                 }
             }
         }
-        SafetyCheck::Reject { reason } => {
-            return ResponseInputItem::FunctionCallOutput {
-                call_id,
-                output: FunctionCallOutputPayload {
-                    content: format!("patch rejected: {reason}"),
-                    success: Some(false),
-                },
-            };
-        }
-    };
-
-    // Verify write permissions before touching the filesystem.
-    let writable_snapshot = {
-        #[allow(clippy::unwrap_used)]
-        sess.writable_roots.lock().unwrap().clone()
-    };
-
-    if let Some(offending) = first_offending_path(&action, &writable_snapshot, &sess.cwd) {
-        let root = offending.parent().unwrap_or(&offending).to_path_buf();
-
-        let reason = Some(format!(
-            "grant write access to {} for this session",
-            root.display()
-        ));
-
-        let rx = sess
-            .request_patch_approval(
-                sub_id.clone(),
-                call_id.clone(),
-                &action,
-                reason.clone(),
-                Some(root.clone()),
-            )
-            .await;
-
-        if !matches!(
-            rx.await.unwrap_or_default(),
-            ReviewDecision::Approved | ReviewDecision::ApprovedForSession
-        ) {
-            return ResponseInputItem::FunctionCallOutput {
-                call_id,
-                output: FunctionCallOutputPayload {
-                    content: "patch rejected by user".to_string(),
-                    success: Some(false),
-                },
-            };
-        }
-
-        // user approved, extend writable roots for this session
-        #[allow(clippy::unwrap_used)]
-        sess.writable_roots.lock().unwrap().push(root);
-    }
-
-    let _ = sess
-        .tx_event
-        .send(Event {
-            id: sub_id.clone(),
-            msg: EventMsg::PatchApplyBegin(PatchApplyBeginEvent {
-                call_id: call_id.clone(),
-                auto_approved,
-                changes: convert_apply_patch_to_protocol(&action),
-            }),
-        })
-        .await;
-
-    let mut stdout = Vec::new();
-    let mut stderr = Vec::new();
-    // Enforce writable roots. If a write is blocked, collect offending root
-    // and prompt the user to extend permissions.
-    let mut result = apply_changes_from_apply_patch_and_report(&action, &mut stdout, &mut stderr);
-
-    if let Err(err) = &result {
-        if err.kind() == std::io::ErrorKind::PermissionDenied {
-            // Determine first offending path.
-            let offending_opt = action
-                .changes()
-                .iter()
-                .flat_map(|(path, change)| match change {
-                    ApplyPatchFileChange::Add { .. } => vec![path.as_ref()],
-                    ApplyPatchFileChange::Delete => vec![path.as_ref()],
-                    ApplyPatchFileChange::Update {
-                        move_path: Some(move_path),
-                        ..
-                    } => {
-                        vec![path.as_ref(), move_path.as_ref()]
-                    }
-                    ApplyPatchFileChange::Update {
-                        move_path: None, ..
-                    } => vec![path.as_ref()],
-                })
-                .find_map(|path: &Path| {
-                    // ApplyPatchAction promises to guarantee absolute paths.
-                    if !path.is_absolute() {
-                        panic!("apply_patch invariant failed: path is not absolute: {path:?}");
-                    }
-
-                    let writable = {
-                        #[allow(clippy::unwrap_used)]
-                        let roots = sess.writable_roots.lock().unwrap();
-                        roots.iter().any(|root| path.starts_with(root))
-                    };
-                    if writable {
-                        None
-                    } else {
-                        Some(path.to_path_buf())
-                    }
-                });
-
-            if let Some(offending) = offending_opt {
-                let root = offending.parent().unwrap_or(&offending).to_path_buf();
-
-                let reason = Some(format!(
-                    "grant write access to {} for this session",
-                    root.display()
-                ));
-                let rx = sess
-                    .request_patch_approval(
-                        sub_id.clone(),
-                        call_id.clone(),
-                        &action,
-                        reason.clone(),
-                        Some(root.clone()),
-                    )
-                    .await;
-                if matches!(
-                    rx.await.unwrap_or_default(),
-                    ReviewDecision::Approved | ReviewDecision::ApprovedForSession
-                ) {
-                    // Extend writable roots.
-                    #[allow(clippy::unwrap_used)]
-                    sess.writable_roots.lock().unwrap().push(root);
-                    stdout.clear();
-                    stderr.clear();
-                    result = apply_changes_from_apply_patch_and_report(
-                        &action,
-                        &mut stdout,
-                        &mut stderr,
-                    );
-                }
-            }
-        }
-    }
-
-    // Emit PatchApplyEnd event.
-    let success_flag = result.is_ok();
-    let _ = sess
-        .tx_event
-        .send(Event {
-            id: sub_id.clone(),
-            msg: EventMsg::PatchApplyEnd(PatchApplyEndEvent {
-                call_id: call_id.clone(),
-                stdout: String::from_utf8_lossy(&stdout).to_string(),
-                stderr: String::from_utf8_lossy(&stderr).to_string(),
-                success: success_flag,
-            }),
-        })
-        .await;
-
-    match result {
-        Ok(_) => ResponseInputItem::FunctionCallOutput {
-            call_id,
+        SafetyCheck::Reject { reason } => ResponseInputItem::FunctionCallOutput {
+            call_id: call_id.to_owned(),
             output: FunctionCallOutputPayload {
-                content: String::from_utf8_lossy(&stdout).to_string(),
-                success: None,
-            },
-        },
-        Err(e) => ResponseInputItem::FunctionCallOutput {
-            call_id,
-            output: FunctionCallOutputPayload {
-                content: format!("error: {e:#}, stderr: {}", String::from_utf8_lossy(&stderr)),
+                content: format!("patch rejected: {reason}"),
                 success: Some(false),
             },
-        },
-    }
-}
-
-/// Return the first path in `hunks` that is NOT under any of the
-/// `writable_roots` (after normalising). If all paths are acceptable,
-/// returns None.
-fn first_offending_path(
-    action: &ApplyPatchAction,
-    writable_roots: &[PathBuf],
-    cwd: &Path,
-) -> Option<PathBuf> {
-    let changes = action.changes();
-    for (path, change) in changes {
-        let candidate = match change {
-            ApplyPatchFileChange::Add { .. } => path,
-            ApplyPatchFileChange::Delete => path,
-            ApplyPatchFileChange::Update { move_path, .. } => move_path.as_ref().unwrap_or(path),
-        };
-
-        let abs = if candidate.is_absolute() {
-            candidate.clone()
-        } else {
-            cwd.join(candidate)
-        };
-
-        let mut allowed = false;
-        for root in writable_roots {
-            let root_abs = if root.is_absolute() {
-                root.clone()
-            } else {
-                cwd.join(root)
-            };
-            if abs.starts_with(&root_abs) {
-                allowed = true;
-                break;
-            }
-        }
-
-        if !allowed {
-            return Some(candidate.clone());
         }
+        .into(),
     }
-    None
 }
 
 pub(crate) fn convert_apply_patch_to_protocol(
@@ -299,85 +129,6 @@ pub(crate) fn convert_apply_patch_to_protocol(
     result
 }
 
-fn apply_changes_from_apply_patch_and_report(
-    action: &ApplyPatchAction,
-    stdout: &mut impl std::io::Write,
-    stderr: &mut impl std::io::Write,
-) -> std::io::Result<()> {
-    match apply_changes_from_apply_patch(action) {
-        Ok(affected_paths) => {
-            print_summary(&affected_paths, stdout)?;
-        }
-        Err(err) => {
-            writeln!(stderr, "{err:?}")?;
-        }
-    }
-
-    Ok(())
-}
-
-fn apply_changes_from_apply_patch(action: &ApplyPatchAction) -> anyhow::Result<AffectedPaths> {
-    let mut added: Vec<PathBuf> = Vec::new();
-    let mut modified: Vec<PathBuf> = Vec::new();
-    let mut deleted: Vec<PathBuf> = Vec::new();
-
-    let changes = action.changes();
-    for (path, change) in changes {
-        match change {
-            ApplyPatchFileChange::Add { content } => {
-                if let Some(parent) = path.parent() {
-                    if !parent.as_os_str().is_empty() {
-                        std::fs::create_dir_all(parent).with_context(|| {
-                            format!("Failed to create parent directories for {}", path.display())
-                        })?;
-                    }
-                }
-                std::fs::write(path, content)
-                    .with_context(|| format!("Failed to write file {}", path.display()))?;
-                added.push(path.clone());
-            }
-            ApplyPatchFileChange::Delete => {
-                std::fs::remove_file(path)
-                    .with_context(|| format!("Failed to delete file {}", path.display()))?;
-                deleted.push(path.clone());
-            }
-            ApplyPatchFileChange::Update {
-                unified_diff: _unified_diff,
-                move_path,
-                new_content,
-            } => {
-                if let Some(move_path) = move_path {
-                    if let Some(parent) = move_path.parent() {
-                        if !parent.as_os_str().is_empty() {
-                            std::fs::create_dir_all(parent).with_context(|| {
-                                format!(
-                                    "Failed to create parent directories for {}",
-                                    move_path.display()
-                                )
-                            })?;
-                        }
-                    }
-
-                    std::fs::rename(path, move_path)
-                        .with_context(|| format!("Failed to rename file {}", path.display()))?;
-                    std::fs::write(move_path, new_content)?;
-                    modified.push(move_path.clone());
-                    deleted.push(path.clone());
-                } else {
-                    std::fs::write(path, new_content)?;
-                    modified.push(path.clone());
-                }
-            }
-        }
-    }
-
-    Ok(AffectedPaths {
-        added,
-        modified,
-        deleted,
-    })
-}
-
 pub(crate) fn get_writable_roots(cwd: &Path) -> Vec<PathBuf> {
     let mut writable_roots = Vec::new();
     if cfg!(target_os = "macos") {

codex-rs/core/src/client.rs

@@ -192,7 +192,7 @@ impl ModelClient {
         loop {
             attempt += 1;
 
-            let req_builder = self
+            let mut req_builder = self
                 .client
                 .post(format!("{base_url}/responses"))
                 .header("OpenAI-Beta", "responses=experimental")
@@ -201,7 +201,20 @@ impl ModelClient {
                 .header(reqwest::header::ACCEPT, "text/event-stream")
                 .json(&payload);
 
-            let req_builder = self.provider.apply_http_headers(req_builder);
+            if auth.mode == AuthMode::ChatGPT {
+                if let Some(account_id) = auth.get_account_id().await {
+                    req_builder = req_builder.header("chatgpt-account-id", account_id);
+                }
+            }
+
+            req_builder = self.provider.apply_http_headers(req_builder);
+
+            let originator = self
+                .config
+                .internal_originator
+                .as_deref()
+                .unwrap_or("codex_cli_rs");
+            req_builder = req_builder.header("originator", originator);
 
             let res = req_builder.send().await;
             if let Ok(resp) = &res {

codex-rs/core/src/client_common.rs

@@ -44,7 +44,7 @@ impl Prompt {
             .as_deref()
             .unwrap_or(BASE_INSTRUCTIONS);
         let mut sections: Vec<&str> = vec![base];
-        if model.starts_with("gpt-4.1") {
+        if model.starts_with("gpt-4.1") || model.starts_with("gpt-oss") {
             sections.push(APPLY_PATCH_TOOL_INSTRUCTIONS);
         }
         Cow::Owned(sections.join("\n"))

codex-rs/core/src/codex.rs

@@ -30,6 +30,9 @@ use tracing::trace;
 use tracing::warn;
 use uuid::Uuid;
 
+use crate::apply_patch::ApplyPatchExec;
+use crate::apply_patch::CODEX_APPLY_PATCH_ARG1;
+use crate::apply_patch::InternalApplyPatchInvocation;
 use crate::apply_patch::convert_apply_patch_to_protocol;
 use crate::apply_patch::get_writable_roots;
 use crate::apply_patch::{self};
@@ -45,6 +48,7 @@ use crate::error::SandboxErr;
 use crate::exec::ExecParams;
 use crate::exec::ExecToolCallOutput;
 use crate::exec::SandboxType;
+use crate::exec::StdoutStream;
 use crate::exec::process_exec_tool_call;
 use crate::exec_env::create_env;
 use crate::mcp_connection_manager::McpConnectionManager;
@@ -71,17 +75,23 @@ use crate::protocol::EventMsg;
 use crate::protocol::ExecApprovalRequestEvent;
 use crate::protocol::ExecCommandBeginEvent;
 use crate::protocol::ExecCommandEndEvent;
+use crate::protocol::FileChange;
 use crate::protocol::InputItem;
 use crate::protocol::Op;
+use crate::protocol::PatchApplyBeginEvent;
+use crate::protocol::PatchApplyEndEvent;
 use crate::protocol::ReviewDecision;
 use crate::protocol::SandboxPolicy;
 use crate::protocol::SessionConfiguredEvent;
 use crate::protocol::Submission;
 use crate::protocol::TaskCompleteEvent;
+use crate::protocol::TurnDiffEvent;
 use crate::rollout::RolloutRecorder;
 use crate::safety::SafetyCheck;
 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::user_notification::UserNotification;
 use crate::util::backoff;
 
@@ -354,39 +364,101 @@ impl Session {
         }
     }
 
-    async fn notify_exec_command_begin(&self, sub_id: &str, call_id: &str, params: &ExecParams) {
+    async fn on_exec_command_begin(
+        &self,
+        turn_diff_tracker: &mut TurnDiffTracker,
+        exec_command_context: ExecCommandContext,
+    ) {
+        let ExecCommandContext {
+            sub_id,
+            call_id,
+            command_for_display,
+            cwd,
+            apply_patch,
+        } = exec_command_context;
+        let msg = match apply_patch {
+            Some(ApplyPatchCommandContext {
+                user_explicitly_approved_this_action,
+                changes,
+            }) => {
+                turn_diff_tracker.on_patch_begin(&changes);
+
+                EventMsg::PatchApplyBegin(PatchApplyBeginEvent {
+                    call_id,
+                    auto_approved: !user_explicitly_approved_this_action,
+                    changes,
+                })
+            }
+            None => EventMsg::ExecCommandBegin(ExecCommandBeginEvent {
+                call_id,
+                command: command_for_display.clone(),
+                cwd,
+            }),
+        };
         let event = Event {
             id: sub_id.to_string(),
-            msg: EventMsg::ExecCommandBegin(ExecCommandBeginEvent {
-                call_id: call_id.to_string(),
-                command: params.command.clone(),
-                cwd: params.cwd.clone(),
-            }),
+            msg,
         };
         let _ = self.tx_event.send(event).await;
     }
 
-    async fn notify_exec_command_end(
+    #[allow(clippy::too_many_arguments)]
+    async fn on_exec_command_end(
         &self,
+        turn_diff_tracker: &mut TurnDiffTracker,
         sub_id: &str,
         call_id: &str,
-        stdout: &str,
-        stderr: &str,
-        exit_code: i32,
+        output: &ExecToolCallOutput,
+        is_apply_patch: bool,
     ) {
+        let ExecToolCallOutput {
+            stdout,
+            stderr,
+            duration,
+            exit_code,
+        } = output;
+        // Because stdout and stderr could each be up to 100 KiB, we send
+        // truncated versions.
         const MAX_STREAM_OUTPUT: usize = 5 * 1024; // 5KiB
+        let stdout = stdout.chars().take(MAX_STREAM_OUTPUT).collect();
+        let stderr = stderr.chars().take(MAX_STREAM_OUTPUT).collect();
+
+        let msg = if is_apply_patch {
+            EventMsg::PatchApplyEnd(PatchApplyEndEvent {
+                call_id: call_id.to_string(),
+                stdout,
+                stderr,
+                success: *exit_code == 0,
+            })
+        } else {
+            EventMsg::ExecCommandEnd(ExecCommandEndEvent {
+                call_id: call_id.to_string(),
+                stdout,
+                stderr,
+                duration: *duration,
+                exit_code: *exit_code,
+            })
+        };
+
         let event = Event {
             id: sub_id.to_string(),
-            // Because stdout and stderr could each be up to 100 KiB, we send
-            // truncated versions.
-            msg: EventMsg::ExecCommandEnd(ExecCommandEndEvent {
-                call_id: call_id.to_string(),
-                stdout: stdout.chars().take(MAX_STREAM_OUTPUT).collect(),
-                stderr: stderr.chars().take(MAX_STREAM_OUTPUT).collect(),
-                exit_code,
-            }),
+            msg,
         };
         let _ = self.tx_event.send(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.
+        if is_apply_patch {
+            let unified_diff = turn_diff_tracker.get_unified_diff();
+            if let Ok(Some(unified_diff)) = unified_diff {
+                let msg = EventMsg::TurnDiff(TurnDiffEvent { unified_diff });
+                let event = Event {
+                    id: sub_id.into(),
+                    msg,
+                };
+                let _ = self.tx_event.send(event).await;
+            }
+        }
     }
 
     /// Helper that emits a BackgroundEvent with the given message. This keeps
@@ -402,6 +474,12 @@ impl Session {
         let _ = self.tx_event.send(event).await;
     }
 
+    /// Build the full turn input by concatenating the current conversation
+    /// history with additional items for this turn.
+    pub fn turn_input_with_history(&self, extra: Vec<ResponseItem>) -> Vec<ResponseItem> {
+        [self.state.lock().unwrap().history.contents(), extra].concat()
+    }
+
     /// Returns the input if there was no task running to inject into
     pub fn inject_input(&self, input: Vec<InputItem>) -> Result<(), Vec<InputItem>> {
         let mut state = self.state.lock().unwrap();
@@ -492,6 +570,21 @@ impl State {
     }
 }
 
+#[derive(Clone, Debug)]
+pub(crate) struct ExecCommandContext {
+    pub(crate) sub_id: String,
+    pub(crate) call_id: String,
+    pub(crate) command_for_display: Vec<String>,
+    pub(crate) cwd: PathBuf,
+    pub(crate) apply_patch: Option<ApplyPatchCommandContext>,
+}
+
+#[derive(Clone, Debug)]
+pub(crate) struct ApplyPatchCommandContext {
+    pub(crate) user_explicitly_approved_this_action: bool,
+    pub(crate) changes: HashMap<PathBuf, FileChange>,
+}
+
 /// A series of Turns in response to user input.
 pub(crate) struct AgentTask {
     sess: Arc<Session>,
@@ -509,6 +602,25 @@ impl AgentTask {
             handle,
         }
     }
+    fn compact(
+        sess: Arc<Session>,
+        sub_id: String,
+        input: Vec<InputItem>,
+        compact_instructions: String,
+    ) -> Self {
+        let handle = tokio::spawn(run_compact_task(
+            Arc::clone(&sess),
+            sub_id.clone(),
+            input,
+            compact_instructions,
+        ))
+        .abort_handle();
+        Self {
+            sess,
+            sub_id,
+            handle,
+        }
+    }
 
     fn abort(self) {
         if !self.handle.is_finished() {
@@ -829,6 +941,31 @@ async fn submission_loop(
                     }
                 });
             }
+            Op::Compact => {
+                let sess = match sess.as_ref() {
+                    Some(sess) => sess,
+                    None => {
+                        send_no_session_event(sub.id).await;
+                        continue;
+                    }
+                };
+
+                // Create a summarization request as user input
+                const SUMMARIZATION_PROMPT: &str = include_str!("../../../SUMMARY.md");
+
+                // Attempt to inject input into current task
+                if let Err(items) = sess.inject_input(vec![InputItem::Text {
+                    text: "Start Summarization".to_string(),
+                }]) {
+                    let task = AgentTask::compact(
+                        sess.clone(),
+                        sub.id,
+                        items,
+                        SUMMARIZATION_PROMPT.to_string(),
+                    );
+                    sess.set_task(task);
+                }
+            }
             Op::Shutdown => {
                 info!("Shutting down Codex instance");
 
@@ -890,11 +1027,15 @@ async fn run_task(sess: Arc<Session>, sub_id: String, input: Vec<InputItem>) {
         return;
     }
 
-    let initial_input_for_turn = ResponseInputItem::from(input);
+    let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
     sess.record_conversation_items(&[initial_input_for_turn.clone().into()])
         .await;
 
     let last_agent_message: Option<String>;
+    // Although from the perspective of codex.rs, TurnDiffTracker has the lifecycle of a Task which contains
+    // many turns, from the perspective of the user, it is a single turn.
+    let mut turn_diff_tracker = TurnDiffTracker::new();
+
     loop {
         // Note that pending_input would be something like a message the user
         // submitted through the UI while the model was running. Though the UI
@@ -911,8 +1052,7 @@ async fn run_task(sess: Arc<Session>, sub_id: String, input: Vec<InputItem>) {
         // conversation history on each turn. The rollout file, however, should
         // only record the new items that originated in this turn so that it
         // represents an append-only log without duplicates.
-        let turn_input: Vec<ResponseItem> =
-            [sess.state.lock().unwrap().history.contents(), pending_input].concat();
+        let turn_input: Vec<ResponseItem> = sess.turn_input_with_history(pending_input);
 
         let turn_input_messages: Vec<String> = turn_input
             .iter()
@@ -927,7 +1067,7 @@ async fn run_task(sess: Arc<Session>, sub_id: String, input: Vec<InputItem>) {
                 })
             })
             .collect();
-        match run_turn(&sess, sub_id.clone(), turn_input).await {
+        match run_turn(&sess, &mut turn_diff_tracker, sub_id.clone(), turn_input).await {
             Ok(turn_output) => {
                 let mut items_to_record_in_conversation_history = Vec::<ResponseItem>::new();
                 let mut responses = Vec::<ResponseInputItem>::new();
@@ -1053,6 +1193,7 @@ async fn run_task(sess: Arc<Session>, sub_id: String, input: Vec<InputItem>) {
 
 async fn run_turn(
     sess: &Session,
+    turn_diff_tracker: &mut TurnDiffTracker,
     sub_id: String,
     input: Vec<ResponseItem>,
 ) -> CodexResult<Vec<ProcessedResponseItem>> {
@@ -1067,7 +1208,7 @@ async fn run_turn(
 
     let mut retries = 0;
     loop {
-        match try_run_turn(sess, &sub_id, &prompt).await {
+        match try_run_turn(sess, turn_diff_tracker, &sub_id, &prompt).await {
             Ok(output) => return Ok(output),
             Err(CodexErr::Interrupted) => return Err(CodexErr::Interrupted),
             Err(CodexErr::EnvVar(var)) => return Err(CodexErr::EnvVar(var)),
@@ -1113,6 +1254,7 @@ struct ProcessedResponseItem {
 
 async fn try_run_turn(
     sess: &Session,
+    turn_diff_tracker: &mut TurnDiffTracker,
     sub_id: &str,
     prompt: &Prompt,
 ) -> CodexResult<Vec<ProcessedResponseItem>> {
@@ -1200,7 +1342,8 @@ async fn try_run_turn(
         match event {
             ResponseEvent::Created => {}
             ResponseEvent::OutputItemDone(item) => {
-                let response = handle_response_item(sess, sub_id, item.clone()).await?;
+                let response =
+                    handle_response_item(sess, turn_diff_tracker, sub_id, item.clone()).await?;
 
                 output.push(ProcessedResponseItem { item, response });
             }
@@ -1218,6 +1361,16 @@ async fn try_run_turn(
                         .ok();
                 }
 
+                let unified_diff = turn_diff_tracker.get_unified_diff();
+                if let Ok(Some(unified_diff)) = unified_diff {
+                    let msg = EventMsg::TurnDiff(TurnDiffEvent { unified_diff });
+                    let event = Event {
+                        id: sub_id.to_string(),
+                        msg,
+                    };
+                    let _ = sess.tx_event.send(event).await;
+                }
+
                 return Ok(output);
             }
             ResponseEvent::OutputTextDelta(delta) => {
@@ -1238,8 +1391,91 @@ async fn try_run_turn(
     }
 }
 
+async fn run_compact_task(
+    sess: Arc<Session>,
+    sub_id: String,
+    input: Vec<InputItem>,
+    compact_instructions: String,
+) {
+    let start_event = Event {
+        id: sub_id.clone(),
+        msg: EventMsg::TaskStarted,
+    };
+    if sess.tx_event.send(start_event).await.is_err() {
+        return;
+    }
+
+    let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
+    let turn_input: Vec<ResponseItem> =
+        sess.turn_input_with_history(vec![initial_input_for_turn.clone().into()]);
+
+    let prompt = Prompt {
+        input: turn_input,
+        user_instructions: None,
+        store: !sess.disable_response_storage,
+        extra_tools: HashMap::new(),
+        base_instructions_override: Some(compact_instructions.clone()),
+    };
+
+    let max_retries = sess.client.get_provider().stream_max_retries();
+    let mut retries = 0;
+
+    loop {
+        let attempt_result = drain_to_completed(&sess, &sub_id, &prompt).await;
+
+        match attempt_result {
+            Ok(()) => break,
+            Err(CodexErr::Interrupted) => return,
+            Err(e) => {
+                if retries < max_retries {
+                    retries += 1;
+                    let delay = backoff(retries);
+                    sess.notify_background_event(
+                        &sub_id,
+                        format!(
+                            "stream error: {e}; retrying {retries}/{max_retries} in {delay:?}…"
+                        ),
+                    )
+                    .await;
+                    tokio::time::sleep(delay).await;
+                    continue;
+                } else {
+                    let event = Event {
+                        id: sub_id.clone(),
+                        msg: EventMsg::Error(ErrorEvent {
+                            message: e.to_string(),
+                        }),
+                    };
+                    sess.send_event(event).await;
+                    return;
+                }
+            }
+        }
+    }
+
+    sess.remove_task(&sub_id);
+    let event = Event {
+        id: sub_id.clone(),
+        msg: EventMsg::AgentMessage(AgentMessageEvent {
+            message: "Compact task completed".to_string(),
+        }),
+    };
+    sess.send_event(event).await;
+    let event = Event {
+        id: sub_id.clone(),
+        msg: EventMsg::TaskComplete(TaskCompleteEvent {
+            last_agent_message: None,
+        }),
+    };
+    sess.send_event(event).await;
+
+    let mut state = sess.state.lock().unwrap();
+    state.history.keep_last_messages(1);
+}
+
 async fn handle_response_item(
     sess: &Session,
+    turn_diff_tracker: &mut TurnDiffTracker,
     sub_id: &str,
     item: ResponseItem,
 ) -> CodexResult<Option<ResponseInputItem>> {
@@ -1277,7 +1513,17 @@ async fn handle_response_item(
             ..
         } => {
             info!("FunctionCall: {arguments}");
-            Some(handle_function_call(sess, sub_id.to_string(), name, arguments, call_id).await)
+            Some(
+                handle_function_call(
+                    sess,
+                    turn_diff_tracker,
+                    sub_id.to_string(),
+                    name,
+                    arguments,
+                    call_id,
+                )
+                .await,
+            )
         }
         ResponseItem::LocalShellCall {
             id,
@@ -1312,6 +1558,7 @@ async fn handle_response_item(
                 handle_container_exec_with_params(
                     exec_params,
                     sess,
+                    turn_diff_tracker,
                     sub_id.to_string(),
                     effective_call_id,
                 )
@@ -1329,6 +1576,7 @@ async fn handle_response_item(
 
 async fn handle_function_call(
     sess: &Session,
+    turn_diff_tracker: &mut TurnDiffTracker,
     sub_id: String,
     name: String,
     arguments: String,
@@ -1342,7 +1590,8 @@ async fn handle_function_call(
                     return *output;
                 }
             };
-            handle_container_exec_with_params(params, sess, sub_id, call_id).await
+            handle_container_exec_with_params(params, sess, turn_diff_tracker, sub_id, call_id)
+                .await
         }
         "update_plan" => handle_update_plan(sess, arguments, sub_id, call_id).await,
         _ => {
@@ -1416,13 +1665,19 @@ fn maybe_run_with_user_profile(params: ExecParams, sess: &Session) -> ExecParams
 async fn handle_container_exec_with_params(
     params: ExecParams,
     sess: &Session,
+    turn_diff_tracker: &mut TurnDiffTracker,
     sub_id: String,
     call_id: String,
 ) -> ResponseInputItem {
     // check if this was a patch, and apply it if so
-    match maybe_parse_apply_patch_verified(&params.command, &params.cwd) {
+    let apply_patch_exec = match maybe_parse_apply_patch_verified(&params.command, &params.cwd) {
         MaybeApplyPatchVerified::Body(changes) => {
-            return apply_patch::apply_patch(sess, sub_id, call_id, changes).await;
+            match apply_patch::apply_patch(sess, &sub_id, &call_id, changes).await {
+                InternalApplyPatchInvocation::Output(item) => return item,
+                InternalApplyPatchInvocation::DelegateToExec(apply_patch_exec) => {
+                    Some(apply_patch_exec)
+                }
+            }
         }
         MaybeApplyPatchVerified::CorrectnessError(parse_error) => {
             // It looks like an invocation of `apply_patch`, but we
@@ -1438,20 +1693,67 @@ async fn handle_container_exec_with_params(
         }
         MaybeApplyPatchVerified::ShellParseError(error) => {
             trace!("Failed to parse shell command, {error:?}");
+            None
         }
-        MaybeApplyPatchVerified::NotApplyPatch => (),
-    }
-
-    // safety checks
-    let safety = {
-        let state = sess.state.lock().unwrap();
-        assess_command_safety(
-            &params.command,
-            sess.approval_policy,
-            &sess.sandbox_policy,
-            &state.approved_commands,
-        )
+        MaybeApplyPatchVerified::NotApplyPatch => None,
     };
+
+    let (params, safety, command_for_display) = match &apply_patch_exec {
+        Some(ApplyPatchExec {
+            action: ApplyPatchAction { patch, cwd, .. },
+            user_explicitly_approved_this_action,
+        }) => {
+            let path_to_codex = std::env::current_exe()
+                .ok()
+                .map(|p| p.to_string_lossy().to_string());
+            let Some(path_to_codex) = path_to_codex else {
+                return ResponseInputItem::FunctionCallOutput {
+                    call_id,
+                    output: FunctionCallOutputPayload {
+                        content: "failed to determine path to codex executable".to_string(),
+                        success: None,
+                    },
+                };
+            };
+
+            let params = ExecParams {
+                command: vec![
+                    path_to_codex,
+                    CODEX_APPLY_PATCH_ARG1.to_string(),
+                    patch.clone(),
+                ],
+                cwd: cwd.clone(),
+                timeout_ms: params.timeout_ms,
+                env: HashMap::new(),
+            };
+            let safety = if *user_explicitly_approved_this_action {
+                SafetyCheck::AutoApprove {
+                    sandbox_type: SandboxType::None,
+                }
+            } else {
+                assess_safety_for_untrusted_command(sess.approval_policy, &sess.sandbox_policy)
+            };
+            (
+                params,
+                safety,
+                vec!["apply_patch".to_string(), patch.clone()],
+            )
+        }
+        None => {
+            let safety = {
+                let state = sess.state.lock().unwrap();
+                assess_command_safety(
+                    &params.command,
+                    sess.approval_policy,
+                    &sess.sandbox_policy,
+                    &state.approved_commands,
+                )
+            };
+            let command_for_display = params.command.clone();
+            (params, safety, command_for_display)
+        }
+    };
+
     let sandbox_type = match safety {
         SafetyCheck::AutoApprove { sandbox_type } => sandbox_type,
         SafetyCheck::AskUser => {
@@ -1496,7 +1798,22 @@ async fn handle_container_exec_with_params(
         }
     };
 
-    sess.notify_exec_command_begin(&sub_id, &call_id, &params)
+    let exec_command_context = ExecCommandContext {
+        sub_id: sub_id.clone(),
+        call_id: call_id.clone(),
+        command_for_display: command_for_display.clone(),
+        cwd: params.cwd.clone(),
+        apply_patch: apply_patch_exec.map(
+            |ApplyPatchExec {
+                 action,
+                 user_explicitly_approved_this_action,
+             }| ApplyPatchCommandContext {
+                user_explicitly_approved_this_action,
+                changes: convert_apply_patch_to_protocol(&action),
+            },
+        ),
+    };
+    sess.on_exec_command_begin(turn_diff_tracker, exec_command_context.clone())
         .await;
 
     let params = maybe_run_with_user_profile(params, sess);
@@ -1506,6 +1823,11 @@ async fn handle_container_exec_with_params(
         sess.ctrl_c.clone(),
         &sess.sandbox_policy,
         &sess.codex_linux_sandbox_exe,
+        Some(StdoutStream {
+            sub_id: sub_id.clone(),
+            call_id: call_id.clone(),
+            tx_event: sess.tx_event.clone(),
+        }),
     )
     .await;
 
@@ -1516,16 +1838,22 @@ async fn handle_container_exec_with_params(
                 stdout,
                 stderr,
                 duration,
-            } = output;
+            } = &output;
 
-            sess.notify_exec_command_end(&sub_id, &call_id, &stdout, &stderr, exit_code)
-                .await;
+            sess.on_exec_command_end(
+                turn_diff_tracker,
+                &sub_id,
+                &call_id,
+                &output,
+                exec_command_context.apply_patch.is_some(),
+            )
+            .await;
 
-            let is_success = exit_code == 0;
+            let is_success = *exit_code == 0;
             let content = format_exec_output(
-                if is_success { &stdout } else { &stderr },
-                exit_code,
-                duration,
+                if is_success { stdout } else { stderr },
+                *exit_code,
+                *duration,
             );
 
             ResponseInputItem::FunctionCallOutput {
@@ -1537,7 +1865,15 @@ async fn handle_container_exec_with_params(
             }
         }
         Err(CodexErr::Sandbox(error)) => {
-            handle_sandbox_error(error, sandbox_type, params, sess, sub_id, call_id).await
+            handle_sandbox_error(
+                turn_diff_tracker,
+                params,
+                exec_command_context,
+                error,
+                sandbox_type,
+                sess,
+            )
+            .await
         }
         Err(e) => {
             // Handle non-sandbox errors
@@ -1553,13 +1889,18 @@ async fn handle_container_exec_with_params(
 }
 
 async fn handle_sandbox_error(
+    turn_diff_tracker: &mut TurnDiffTracker,
+    params: ExecParams,
+    exec_command_context: ExecCommandContext,
     error: SandboxErr,
     sandbox_type: SandboxType,
-    params: ExecParams,
     sess: &Session,
-    sub_id: String,
-    call_id: String,
 ) -> ResponseInputItem {
+    let call_id = exec_command_context.call_id.clone();
+    let sub_id = exec_command_context.sub_id.clone();
+    let cwd = exec_command_context.cwd.clone();
+    let is_apply_patch = exec_command_context.apply_patch.is_some();
+
     // Early out if the user never wants to be asked for approval; just return to the model immediately
     if sess.approval_policy == AskForApproval::Never {
         return ResponseInputItem::FunctionCallOutput {
@@ -1589,7 +1930,7 @@ async fn handle_sandbox_error(
             sub_id.clone(),
             call_id.clone(),
             params.command.clone(),
-            params.cwd.clone(),
+            cwd.clone(),
             Some("command failed; retry without sandbox?".to_string()),
         )
         .await;
@@ -1605,7 +1946,7 @@ async fn handle_sandbox_error(
             sess.notify_background_event(&sub_id, "retrying command without sandbox")
                 .await;
 
-            sess.notify_exec_command_begin(&sub_id, &call_id, &params)
+            sess.on_exec_command_begin(turn_diff_tracker, exec_command_context)
                 .await;
 
             // This is an escalated retry; the policy will not be
@@ -1616,6 +1957,11 @@ async fn handle_sandbox_error(
                 sess.ctrl_c.clone(),
                 &sess.sandbox_policy,
                 &sess.codex_linux_sandbox_exe,
+                Some(StdoutStream {
+                    sub_id: sub_id.clone(),
+                    call_id: call_id.clone(),
+                    tx_event: sess.tx_event.clone(),
+                }),
             )
             .await;
 
@@ -1626,16 +1972,22 @@ async fn handle_sandbox_error(
                         stdout,
                         stderr,
                         duration,
-                    } = retry_output;
+                    } = &retry_output;
 
-                    sess.notify_exec_command_end(&sub_id, &call_id, &stdout, &stderr, exit_code)
-                        .await;
+                    sess.on_exec_command_end(
+                        turn_diff_tracker,
+                        &sub_id,
+                        &call_id,
+                        &retry_output,
+                        is_apply_patch,
+                    )
+                    .await;
 
-                    let is_success = exit_code == 0;
+                    let is_success = *exit_code == 0;
                     let content = format_exec_output(
-                        if is_success { &stdout } else { &stderr },
-                        exit_code,
-                        duration,
+                        if is_success { stdout } else { stderr },
+                        *exit_code,
+                        *duration,
                     );
 
                     ResponseInputItem::FunctionCallOutput {
@@ -1719,3 +2071,45 @@ fn get_last_assistant_message_from_turn(responses: &[ResponseItem]) -> Option<St
         }
     })
 }
+
+async fn drain_to_completed(sess: &Session, sub_id: &str, prompt: &Prompt) -> CodexResult<()> {
+    let mut stream = sess.client.clone().stream(prompt).await?;
+    loop {
+        let maybe_event = stream.next().await;
+        let Some(event) = maybe_event else {
+            return Err(CodexErr::Stream(
+                "stream closed before response.completed".into(),
+            ));
+        };
+        match event {
+            Ok(ResponseEvent::OutputItemDone(item)) => {
+                // Record only to in-memory conversation history; avoid state snapshot.
+                let mut state = sess.state.lock().unwrap();
+                state.history.record_items(std::slice::from_ref(&item));
+            }
+            Ok(ResponseEvent::Completed {
+                response_id: _,
+                token_usage,
+            }) => {
+                let token_usage = match token_usage {
+                    Some(usage) => usage,
+                    None => {
+                        return Err(CodexErr::Stream(
+                            "token_usage was None in ResponseEvent::Completed".into(),
+                        ));
+                    }
+                };
+                sess.tx_event
+                    .send(Event {
+                        id: sub_id.to_string(),
+                        msg: EventMsg::TokenCount(token_usage),
+                    })
+                    .await
+                    .ok();
+                return Ok(());
+            }
+            Ok(_) => continue,
+            Err(e) => return Err(e),
+        }
+    }
+}

codex-rs/core/src/codex_wrapper.rs

@@ -26,7 +26,7 @@ pub struct CodexConversation {
 /// that callers can surface the information to the UI.
 pub async fn init_codex(config: Config) -> anyhow::Result<CodexConversation> {
     let ctrl_c = notify_on_sigint();
-    let auth = load_auth(&config.codex_home)?;
+    let auth = load_auth(&config.codex_home, true)?;
     let CodexSpawnOk {
         codex,
         init_id,

codex-rs/core/src/config.rs

@@ -146,6 +146,9 @@ pub struct Config {
 
     /// Include an experimental plan tool that the model can use to update its current plan and status of each step.
     pub include_plan_tool: bool,
+
+    /// The value for the `originator` header included with Responses API requests.
+    pub internal_originator: Option<String>,
 }
 
 impl Config {
@@ -336,6 +339,9 @@ pub struct ConfigToml {
 
     /// Experimental path to a file whose contents replace the built-in BASE_INSTRUCTIONS.
     pub experimental_instructions_file: Option<PathBuf>,
+
+    /// The value for the `originator` header included with Responses API requests.
+    pub internal_originator: Option<String>,
 }
 
 impl ConfigToml {
@@ -350,6 +356,7 @@ impl ConfigToml {
                 Some(s) => SandboxPolicy::WorkspaceWrite {
                     writable_roots: s.writable_roots.clone(),
                     network_access: s.network_access,
+                    include_default_writable_roots: true,
                 },
                 None => SandboxPolicy::new_workspace_write_policy(),
             },
@@ -473,10 +480,12 @@ impl Config {
         // Load base instructions override from a file if specified. If the
         // path is relative, resolve it against the effective cwd so the
         // behaviour matches other path-like config values.
-        let file_base_instructions = Self::get_base_instructions(
-            cfg.experimental_instructions_file.as_ref(),
-            &resolved_cwd,
-        )?;
+        let experimental_instructions_path = config_profile
+            .experimental_instructions_file
+            .as_ref()
+            .or(cfg.experimental_instructions_file.as_ref());
+        let file_base_instructions =
+            Self::get_base_instructions(experimental_instructions_path, &resolved_cwd)?;
         let base_instructions = base_instructions.or(file_base_instructions);
 
         let config = Self {
@@ -529,6 +538,7 @@ impl Config {
 
             experimental_resume,
             include_plan_tool: include_plan_tool.unwrap_or(false),
+            internal_originator: cfg.internal_originator,
         };
         Ok(config)
     }
@@ -720,6 +730,7 @@ writable_roots = [
             SandboxPolicy::WorkspaceWrite {
                 writable_roots: vec![PathBuf::from("/tmp")],
                 network_access: false,
+                include_default_writable_roots: true,
             },
             sandbox_workspace_write_cfg.derive_sandbox_policy(sandbox_mode_override)
         );
@@ -887,6 +898,7 @@ disable_response_storage = true
                 experimental_resume: None,
                 base_instructions: None,
                 include_plan_tool: false,
+                internal_originator: None,
             },
             o3_profile_config
         );
@@ -936,6 +948,7 @@ disable_response_storage = true
             experimental_resume: None,
             base_instructions: None,
             include_plan_tool: false,
+            internal_originator: None,
         };
 
         assert_eq!(expected_gpt3_profile_config, gpt3_profile_config);
@@ -1000,6 +1013,7 @@ disable_response_storage = true
             experimental_resume: None,
             base_instructions: None,
             include_plan_tool: false,
+            internal_originator: None,
         };
 
         assert_eq!(expected_zdr_profile_config, zdr_profile_config);

codex-rs/core/src/config_profile.rs

@@ -1,4 +1,5 @@
 use serde::Deserialize;
+use std::path::PathBuf;
 
 use crate::config_types::ReasoningEffort;
 use crate::config_types::ReasoningSummary;
@@ -17,4 +18,5 @@ pub struct ConfigProfile {
     pub model_reasoning_effort: Option<ReasoningEffort>,
     pub model_reasoning_summary: Option<ReasoningSummary>,
     pub chatgpt_base_url: Option<String>,
+    pub experimental_instructions_file: Option<PathBuf>,
 }

codex-rs/core/src/conversation_history.rs

@@ -30,6 +30,34 @@ impl ConversationHistory {
             }
         }
     }
+
+    pub(crate) fn keep_last_messages(&mut self, n: usize) {
+        if n == 0 {
+            self.items.clear();
+            return;
+        }
+
+        // Collect the last N message items (assistant/user), newest to oldest.
+        let mut kept: Vec<ResponseItem> = Vec::with_capacity(n);
+        for item in self.items.iter().rev() {
+            if let ResponseItem::Message { role, content, .. } = item {
+                kept.push(ResponseItem::Message {
+                    // we need to remove the id or the model will complain that messages are sent without
+                    // their reasonings
+                    id: None,
+                    role: role.clone(),
+                    content: content.clone(),
+                });
+                if kept.len() == n {
+                    break;
+                }
+            }
+        }
+
+        // Preserve chronological order (oldest to newest) within the kept slice.
+        kept.reverse();
+        self.items = kept;
+    }
 }
 
 /// Anything that is not a system message or "reasoning" message is considered

codex-rs/core/src/exec.rs

@@ -6,23 +6,29 @@ use std::io;
 use std::path::Path;
 use std::path::PathBuf;
 use std::process::ExitStatus;
-use std::process::Stdio;
 use std::sync::Arc;
 use std::time::Duration;
 use std::time::Instant;
 
+use async_channel::Sender;
 use tokio::io::AsyncRead;
 use tokio::io::AsyncReadExt;
 use tokio::io::BufReader;
 use tokio::process::Child;
-use tokio::process::Command;
 use tokio::sync::Notify;
-use tracing::trace;
 
 use crate::error::CodexErr;
 use crate::error::Result;
 use crate::error::SandboxErr;
+use crate::protocol::Event;
+use crate::protocol::EventMsg;
+use crate::protocol::ExecCommandOutputDeltaEvent;
+use crate::protocol::ExecOutputStream;
 use crate::protocol::SandboxPolicy;
+use crate::seatbelt::spawn_command_under_seatbelt;
+use crate::spawn::StdioPolicy;
+use crate::spawn::spawn_child_async;
+use serde_bytes::ByteBuf;
 
 // Maximum we send for each stream, which is either:
 // - 10KiB OR
@@ -37,24 +43,6 @@ const DEFAULT_TIMEOUT_MS: u64 = 10_000;
 const SIGKILL_CODE: i32 = 9;
 const TIMEOUT_CODE: i32 = 64;
 
-const MACOS_SEATBELT_BASE_POLICY: &str = include_str!("seatbelt_base_policy.sbpl");
-
-/// When working with `sandbox-exec`, only consider `sandbox-exec` in `/usr/bin`
-/// to defend against an attacker trying to inject a malicious version on the
-/// PATH. If /usr/bin/sandbox-exec has been tampered with, then the attacker
-/// already has root access.
-const MACOS_PATH_TO_SEATBELT_EXECUTABLE: &str = "/usr/bin/sandbox-exec";
-
-/// Experimental environment variable that will be set to some non-empty value
-/// if both of the following are true:
-///
-/// 1. The process was spawned by Codex as part of a shell tool call.
-/// 2. SandboxPolicy.has_full_network_access() was false for the tool call.
-///
-/// We may try to have just one environment variable for all sandboxing
-/// attributes, so this may change in the future.
-pub const CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR: &str = "CODEX_SANDBOX_NETWORK_DISABLED";
-
 #[derive(Debug, Clone)]
 pub struct ExecParams {
     pub command: Vec<String>,
@@ -74,18 +62,26 @@ pub enum SandboxType {
     LinuxSeccomp,
 }
 
+#[derive(Clone)]
+pub struct StdoutStream {
+    pub sub_id: String,
+    pub call_id: String,
+    pub tx_event: Sender<Event>,
+}
+
 pub async fn process_exec_tool_call(
     params: ExecParams,
     sandbox_type: SandboxType,
     ctrl_c: Arc<Notify>,
     sandbox_policy: &SandboxPolicy,
     codex_linux_sandbox_exe: &Option<PathBuf>,
+    stdout_stream: Option<StdoutStream>,
 ) -> Result<ExecToolCallOutput> {
     let start = Instant::now();
 
     let raw_output_result: std::result::Result<RawExecToolCallOutput, CodexErr> = match sandbox_type
     {
-        SandboxType::None => exec(params, sandbox_policy, ctrl_c).await,
+        SandboxType::None => exec(params, sandbox_policy, ctrl_c, stdout_stream.clone()).await,
         SandboxType::MacosSeatbelt => {
             let ExecParams {
                 command,
@@ -101,7 +97,7 @@ pub async fn process_exec_tool_call(
                 env,
             )
             .await?;
-            consume_truncated_output(child, ctrl_c, timeout_ms).await
+            consume_truncated_output(child, ctrl_c, timeout_ms, stdout_stream.clone()).await
         }
         SandboxType::LinuxSeccomp => {
             let ExecParams {
@@ -124,7 +120,7 @@ pub async fn process_exec_tool_call(
             )
             .await?;
 
-            consume_truncated_output(child, ctrl_c, timeout_ms).await
+            consume_truncated_output(child, ctrl_c, timeout_ms, stdout_stream).await
         }
     };
     let duration = start.elapsed();
@@ -144,11 +140,7 @@ pub async fn process_exec_tool_call(
 
             let exit_code = raw_output.exit_status.code().unwrap_or(-1);
 
-            // NOTE(ragona): This is much less restrictive than the previous check. If we exec
-            // a command, and it returns anything other than success, we assume that it may have
-            // been a sandboxing error and allow the user to retry. (The user of course may choose
-            // not to retry, or in a non-interactive mode, would automatically reject the approval.)
-            if exit_code != 0 && sandbox_type != SandboxType::None {
+            if exit_code != 0 && is_likely_sandbox_denied(sandbox_type, exit_code) {
                 return Err(CodexErr::Sandbox(SandboxErr::Denied(
                     exit_code, stdout, stderr,
                 )));
@@ -168,27 +160,6 @@ pub async fn process_exec_tool_call(
     }
 }
 
-pub async fn spawn_command_under_seatbelt(
-    command: Vec<String>,
-    sandbox_policy: &SandboxPolicy,
-    cwd: PathBuf,
-    stdio_policy: StdioPolicy,
-    env: HashMap<String, String>,
-) -> std::io::Result<Child> {
-    let args = create_seatbelt_command_args(command, sandbox_policy, &cwd);
-    let arg0 = None;
-    spawn_child_async(
-        PathBuf::from(MACOS_PATH_TO_SEATBELT_EXECUTABLE),
-        args,
-        arg0,
-        cwd,
-        sandbox_policy,
-        stdio_policy,
-        env,
-    )
-    .await
-}
-
 /// Spawn a shell tool command under the Linux Landlock+seccomp sandbox helper
 /// (codex-linux-sandbox).
 ///
@@ -248,63 +219,24 @@ fn create_linux_sandbox_command_args(
     linux_cmd
 }
 
-fn create_seatbelt_command_args(
-    command: Vec<String>,
-    sandbox_policy: &SandboxPolicy,
-    cwd: &Path,
-) -> Vec<String> {
-    let (file_write_policy, extra_cli_args) = {
-        if sandbox_policy.has_full_disk_write_access() {
-            // Allegedly, this is more permissive than `(allow file-write*)`.
-            (
-                r#"(allow file-write* (regex #"^/"))"#.to_string(),
-                Vec::<String>::new(),
-            )
-        } else {
-            let writable_roots = sandbox_policy.get_writable_roots_with_cwd(cwd);
-            let (writable_folder_policies, cli_args): (Vec<String>, Vec<String>) = writable_roots
-                .iter()
-                .enumerate()
-                .map(|(index, root)| {
-                    let param_name = format!("WRITABLE_ROOT_{index}");
-                    let policy: String = format!("(subpath (param \"{param_name}\"))");
-                    let cli_arg = format!("-D{param_name}={}", root.to_string_lossy());
-                    (policy, cli_arg)
-                })
-                .unzip();
-            if writable_folder_policies.is_empty() {
-                ("".to_string(), Vec::<String>::new())
-            } else {
-                let file_write_policy = format!(
-                    "(allow file-write*\n{}\n)",
-                    writable_folder_policies.join(" ")
-                );
-                (file_write_policy, cli_args)
-            }
-        }
-    };
+/// We don't have a fully deterministic way to tell if our command failed
+/// because of the sandbox - a command in the user's zshrc file might hit an
+/// error, but the command itself might fail or succeed for other reasons.
+/// For now, we conservatively check for 'command not found' (exit code 127),
+/// and can add additional cases as necessary.
+fn is_likely_sandbox_denied(sandbox_type: SandboxType, exit_code: i32) -> bool {
+    if sandbox_type == SandboxType::None {
+        return false;
+    }
 
-    let file_read_policy = if sandbox_policy.has_full_disk_read_access() {
-        "; allow read-only file operations\n(allow file-read*)"
-    } else {
-        ""
-    };
+    // Quick rejects: well-known non-sandbox shell exit codes
+    // 127: command not found, 2: misuse of shell builtins
+    if exit_code == 127 {
+        return false;
+    }
 
-    // TODO(mbolin): apply_patch calls must also honor the SandboxPolicy.
-    let network_policy = if sandbox_policy.has_full_network_access() {
-        "(allow network-outbound)\n(allow network-inbound)\n(allow system-socket)"
-    } else {
-        ""
-    };
-
-    let full_policy = format!(
-        "{MACOS_SEATBELT_BASE_POLICY}\n{file_read_policy}\n{file_write_policy}\n{network_policy}"
-    );
-    let mut seatbelt_args: Vec<String> = vec!["-p".to_string(), full_policy];
-    seatbelt_args.extend(extra_cli_args);
-    seatbelt_args.push("--".to_string());
-    seatbelt_args.extend(command);
-    seatbelt_args
+    // For all other cases, we assume the sandbox is the cause
+    true
 }
 
 #[derive(Debug)]
@@ -331,6 +263,7 @@ async fn exec(
     }: ExecParams,
     sandbox_policy: &SandboxPolicy,
     ctrl_c: Arc<Notify>,
+    stdout_stream: Option<StdoutStream>,
 ) -> Result<RawExecToolCallOutput> {
     let (program, args) = command.split_first().ok_or_else(|| {
         CodexErr::Io(io::Error::new(
@@ -349,91 +282,7 @@ async fn exec(
         env,
     )
     .await?;
-    consume_truncated_output(child, ctrl_c, timeout_ms).await
-}
-
-#[derive(Debug, Clone, Copy)]
-pub enum StdioPolicy {
-    RedirectForShellTool,
-    Inherit,
-}
-
-/// Spawns the appropriate child process for the ExecParams and SandboxPolicy,
-/// ensuring the args and environment variables used to create the `Command`
-/// (and `Child`) honor the configuration.
-///
-/// For now, we take `SandboxPolicy` as a parameter to spawn_child() because
-/// we need to determine whether to set the
-/// `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR` environment variable.
-async fn spawn_child_async(
-    program: PathBuf,
-    args: Vec<String>,
-    #[cfg_attr(not(unix), allow(unused_variables))] arg0: Option<&str>,
-    cwd: PathBuf,
-    sandbox_policy: &SandboxPolicy,
-    stdio_policy: StdioPolicy,
-    env: HashMap<String, String>,
-) -> std::io::Result<Child> {
-    trace!(
-        "spawn_child_async: {program:?} {args:?} {arg0:?} {cwd:?} {sandbox_policy:?} {stdio_policy:?} {env:?}"
-    );
-
-    let mut cmd = Command::new(&program);
-    #[cfg(unix)]
-    cmd.arg0(arg0.map_or_else(|| program.to_string_lossy().to_string(), String::from));
-    cmd.args(args);
-    cmd.current_dir(cwd);
-    cmd.env_clear();
-    cmd.envs(env);
-
-    if !sandbox_policy.has_full_network_access() {
-        cmd.env(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR, "1");
-    }
-
-    // If this Codex process dies (including being killed via SIGKILL), we want
-    // any child processes that were spawned as part of a `"shell"` tool call
-    // to also be terminated.
-
-    // This relies on prctl(2), so it only works on Linux.
-    #[cfg(target_os = "linux")]
-    unsafe {
-        cmd.pre_exec(|| {
-            // This prctl call effectively requests, "deliver SIGTERM when my
-            // current parent dies."
-            if libc::prctl(libc::PR_SET_PDEATHSIG, libc::SIGTERM) == -1 {
-                return Err(io::Error::last_os_error());
-            }
-
-            // Though if there was a race condition and this pre_exec() block is
-            // run _after_ the parent (i.e., the Codex process) has already
-            // exited, then the parent is the _init_ process (which will never
-            // die), so we should just terminate the child process now.
-            if libc::getppid() == 1 {
-                libc::raise(libc::SIGTERM);
-            }
-            Ok(())
-        });
-    }
-
-    match stdio_policy {
-        StdioPolicy::RedirectForShellTool => {
-            // Do not create a file descriptor for stdin because otherwise some
-            // commands may hang forever waiting for input. For example, ripgrep has
-            // a heuristic where it may try to read from stdin as explained here:
-            // https://github.com/BurntSushi/ripgrep/blob/e2362d4d5185d02fa857bf381e7bd52e66fafc73/crates/core/flags/hiargs.rs#L1101-L1103
-            cmd.stdin(Stdio::null());
-
-            cmd.stdout(Stdio::piped()).stderr(Stdio::piped());
-        }
-        StdioPolicy::Inherit => {
-            // Inherit stdin, stdout, and stderr from the parent process.
-            cmd.stdin(Stdio::inherit())
-                .stdout(Stdio::inherit())
-                .stderr(Stdio::inherit());
-        }
-    }
-
-    cmd.kill_on_drop(true).spawn()
+    consume_truncated_output(child, ctrl_c, timeout_ms, stdout_stream).await
 }
 
 /// Consumes the output of a child process, truncating it so it is suitable for
@@ -442,6 +291,7 @@ pub(crate) async fn consume_truncated_output(
     mut child: Child,
     ctrl_c: Arc<Notify>,
     timeout_ms: Option<u64>,
+    stdout_stream: Option<StdoutStream>,
 ) -> Result<RawExecToolCallOutput> {
     // Both stdout and stderr were configured with `Stdio::piped()`
     // above, therefore `take()` should normally return `Some`.  If it doesn't
@@ -462,11 +312,15 @@ pub(crate) async fn consume_truncated_output(
         BufReader::new(stdout_reader),
         MAX_STREAM_OUTPUT,
         MAX_STREAM_OUTPUT_LINES,
+        stdout_stream.clone(),
+        false,
     ));
     let stderr_handle = tokio::spawn(read_capped(
         BufReader::new(stderr_reader),
         MAX_STREAM_OUTPUT,
         MAX_STREAM_OUTPUT_LINES,
+        stdout_stream.clone(),
+        true,
     ));
 
     let interrupted = ctrl_c.notified();
@@ -500,10 +354,12 @@ pub(crate) async fn consume_truncated_output(
     })
 }
 
-async fn read_capped<R: AsyncRead + Unpin>(
+async fn read_capped<R: AsyncRead + Unpin + Send + 'static>(
     mut reader: R,
     max_output: usize,
     max_lines: usize,
+    stream: Option<StdoutStream>,
+    is_stderr: bool,
 ) -> io::Result<Vec<u8>> {
     let mut buf = Vec::with_capacity(max_output.min(8 * 1024));
     let mut tmp = [0u8; 8192];
@@ -517,6 +373,25 @@ async fn read_capped<R: AsyncRead + Unpin>(
             break;
         }
 
+        if let Some(stream) = &stream {
+            let chunk = tmp[..n].to_vec();
+            let msg = EventMsg::ExecCommandOutputDelta(ExecCommandOutputDeltaEvent {
+                call_id: stream.call_id.clone(),
+                stream: if is_stderr {
+                    ExecOutputStream::Stderr
+                } else {
+                    ExecOutputStream::Stdout
+                },
+                chunk: ByteBuf::from(chunk),
+            });
+            let event = Event {
+                id: stream.sub_id.clone(),
+                msg,
+            };
+            #[allow(clippy::let_unit_value)]
+            let _ = stream.tx_event.send(event).await;
+        }
+
         // Copy into the buffer only while we still have byte and line budget.
         if remaining_bytes > 0 && remaining_lines > 0 {
             let mut copy_len = 0;

codex-rs/core/src/git_info.rs

@@ -111,9 +111,14 @@ mod tests {
     // Helper function to create a test git repository
     async fn create_test_git_repo(temp_dir: &TempDir) -> PathBuf {
         let repo_path = temp_dir.path().to_path_buf();
+        let envs = vec![
+            ("GIT_CONFIG_GLOBAL", "/dev/null"),
+            ("GIT_CONFIG_NOSYSTEM", "1"),
+        ];
 
         // Initialize git repo
         Command::new("git")
+            .envs(envs.clone())
             .args(["init"])
             .current_dir(&repo_path)
             .output()
@@ -122,6 +127,7 @@ mod tests {
 
         // Configure git user (required for commits)
         Command::new("git")
+            .envs(envs.clone())
             .args(["config", "user.name", "Test User"])
             .current_dir(&repo_path)
             .output()
@@ -129,6 +135,7 @@ mod tests {
             .expect("Failed to set git user name");
 
         Command::new("git")
+            .envs(envs.clone())
             .args(["config", "user.email", "test@example.com"])
             .current_dir(&repo_path)
             .output()
@@ -140,6 +147,7 @@ mod tests {
         fs::write(&test_file, "test content").expect("Failed to write test file");
 
         Command::new("git")
+            .envs(envs.clone())
             .args(["add", "."])
             .current_dir(&repo_path)
             .output()
@@ -147,6 +155,7 @@ mod tests {
             .expect("Failed to add files");
 
         Command::new("git")
+            .envs(envs.clone())
             .args(["commit", "-m", "Initial commit"])
             .current_dir(&repo_path)
             .output()

codex-rs/core/src/lib.rs

@@ -38,9 +38,14 @@ pub mod plan_tool;
 mod project_doc;
 pub mod protocol;
 mod rollout;
-mod safety;
+pub(crate) mod safety;
+pub mod seatbelt;
 pub mod shell;
+pub mod spawn;
+pub mod turn_diff_tracker;
 mod user_notification;
 pub mod util;
 
+pub use apply_patch::CODEX_APPLY_PATCH_ARG1;
 pub use client_common::model_supports_reasoning_summaries;
+pub use safety::get_platform_sandbox;

codex-rs/core/src/model_provider_info.rs

@@ -12,10 +12,6 @@ use std::env::VarError;
 use std::time::Duration;
 
 use crate::error::EnvVarError;
-
-/// Value for the `OpenAI-Originator` header that is sent with requests to
-/// OpenAI.
-const OPENAI_ORIGINATOR_HEADER: &str = "codex_cli_rs";
 const DEFAULT_STREAM_IDLE_TIMEOUT_MS: u64 = 300_000;
 const DEFAULT_STREAM_MAX_RETRIES: u64 = 10;
 const DEFAULT_REQUEST_MAX_RETRIES: u64 = 4;
@@ -229,15 +225,9 @@ pub fn built_in_model_providers() -> HashMap<String, ModelProviderInfo> {
             wire_api: WireApi::Responses,
             query_params: None,
             http_headers: Some(
-                [
-                    (
-                        "originator".to_string(),
-                        OPENAI_ORIGINATOR_HEADER.to_string(),
-                    ),
-                    ("version".to_string(), env!("CARGO_PKG_VERSION").to_string()),
-                ]
-                .into_iter()
-                .collect(),
+                [("version".to_string(), env!("CARGO_PKG_VERSION").to_string())]
+                    .into_iter()
+                    .collect(),
             ),
             env_http_headers: Some(
                 [

codex-rs/core/src/protocol.rs

@@ -13,6 +13,7 @@ use std::time::Duration;
 use mcp_types::CallToolResult;
 use serde::Deserialize;
 use serde::Serialize;
+use serde_bytes::ByteBuf;
 use strum_macros::Display;
 use uuid::Uuid;
 
@@ -121,6 +122,10 @@ pub enum Op {
     /// Request a single history entry identified by `log_id` + `offset`.
     GetHistoryEntryRequest { offset: usize, log_id: u64 },
 
+    /// Request the agent to summarize the current conversation context.
+    /// The agent will use its existing context (either conversation history or previous response id)
+    /// to generate a summary which will be returned as an AgentMessage event.
+    Compact,
     /// Request to shut down codex instance.
     Shutdown,
 }
@@ -175,9 +180,29 @@ pub enum SandboxPolicy {
         /// default.
         #[serde(default)]
         network_access: bool,
+
+        /// When set to `true`, will include defaults like the current working
+        /// directory and TMPDIR (on macOS). When `false`, only `writable_roots`
+        /// are used. (Mainly used for testing.)
+        #[serde(default = "default_true")]
+        include_default_writable_roots: bool,
     },
 }
 
+/// A writable root path accompanied by a list of subpaths that should remain
+/// read‑only even when the root is writable. This is primarily used to ensure
+/// top‑level VCS metadata directories (e.g. `.git`) under a writable root are
+/// not modified by the agent.
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub struct WritableRoot {
+    pub root: PathBuf,
+    pub read_only_subpaths: Vec<PathBuf>,
+}
+
+fn default_true() -> bool {
+    true
+}
+
 impl FromStr for SandboxPolicy {
     type Err = serde_json::Error;
 
@@ -199,6 +224,7 @@ impl SandboxPolicy {
         SandboxPolicy::WorkspaceWrite {
             writable_roots: vec![],
             network_access: false,
+            include_default_writable_roots: true,
         }
     }
 
@@ -224,27 +250,51 @@ impl SandboxPolicy {
         }
     }
 
-    /// Returns the list of writable roots that should be passed down to the
-    /// Landlock rules installer, tailored to the current working directory.
-    pub fn get_writable_roots_with_cwd(&self, cwd: &Path) -> Vec<PathBuf> {
+    /// Returns the list of writable roots (tailored to the current working
+    /// directory) together with subpaths that should remain read‑only under
+    /// each writable root.
+    pub fn get_writable_roots_with_cwd(&self, cwd: &Path) -> Vec<WritableRoot> {
         match self {
             SandboxPolicy::DangerFullAccess => Vec::new(),
             SandboxPolicy::ReadOnly => Vec::new(),
-            SandboxPolicy::WorkspaceWrite { writable_roots, .. } => {
-                let mut roots = writable_roots.clone();
-                roots.push(cwd.to_path_buf());
+            SandboxPolicy::WorkspaceWrite {
+                writable_roots,
+                include_default_writable_roots,
+                ..
+            } => {
+                // Start from explicitly configured writable roots.
+                let mut roots: Vec<PathBuf> = writable_roots.clone();
 
-                // Also include the per-user tmp dir on macOS.
-                // Note this is added dynamically rather than storing it in
-                // writable_roots because writable_roots contains only static
-                // values deserialized from the config file.
-                if cfg!(target_os = "macos") {
-                    if let Some(tmpdir) = std::env::var_os("TMPDIR") {
-                        roots.push(PathBuf::from(tmpdir));
+                // Optionally include defaults (cwd and TMPDIR on macOS).
+                if *include_default_writable_roots {
+                    roots.push(cwd.to_path_buf());
+
+                    // Also include the per-user tmp dir on macOS.
+                    // Note this is added dynamically rather than storing it in
+                    // `writable_roots` because `writable_roots` contains only static
+                    // values deserialized from the config file.
+                    if cfg!(target_os = "macos") {
+                        if let Some(tmpdir) = std::env::var_os("TMPDIR") {
+                            roots.push(PathBuf::from(tmpdir));
+                        }
                     }
                 }
 
+                // For each root, compute subpaths that should remain read-only.
                 roots
+                    .into_iter()
+                    .map(|writable_root| {
+                        let mut subpaths = Vec::new();
+                        let top_level_git = writable_root.join(".git");
+                        if top_level_git.is_dir() {
+                            subpaths.push(top_level_git);
+                        }
+                        WritableRoot {
+                            root: writable_root,
+                            read_only_subpaths: subpaths,
+                        }
+                    })
+                    .collect()
             }
         }
     }
@@ -319,6 +369,9 @@ pub enum EventMsg {
     /// Notification that the server is about to execute a command.
     ExecCommandBegin(ExecCommandBeginEvent),
 
+    /// Incremental chunk of output from a running command.
+    ExecCommandOutputDelta(ExecCommandOutputDeltaEvent),
+
     ExecCommandEnd(ExecCommandEndEvent),
 
     ExecApprovalRequest(ExecApprovalRequestEvent),
@@ -334,6 +387,8 @@ pub enum EventMsg {
     /// Notification that a patch application has finished.
     PatchApplyEnd(PatchApplyEndEvent),
 
+    TurnDiff(TurnDiffEvent),
+
     /// Response to GetHistoryEntryRequest.
     GetHistoryEntryResponse(GetHistoryEntryResponseEvent),
 
@@ -470,6 +525,26 @@ pub struct ExecCommandEndEvent {
     pub stderr: String,
     /// The command's exit code.
     pub exit_code: i32,
+    /// The duration of the command execution.
+    pub duration: Duration,
+}
+
+#[derive(Debug, Clone, Deserialize, Serialize)]
+#[serde(rename_all = "snake_case")]
+pub enum ExecOutputStream {
+    Stdout,
+    Stderr,
+}
+
+#[derive(Debug, Clone, Deserialize, Serialize)]
+pub struct ExecCommandOutputDeltaEvent {
+    /// Identifier for the ExecCommandBegin that produced this chunk.
+    pub call_id: String,
+    /// Which stream produced this chunk.
+    pub stream: ExecOutputStream,
+    /// Raw bytes from the stream (may not be valid UTF-8).
+    #[serde(with = "serde_bytes")]
+    pub chunk: ByteBuf,
 }
 
 #[derive(Debug, Clone, Deserialize, Serialize)]
@@ -525,6 +600,11 @@ pub struct PatchApplyEndEvent {
     pub success: bool,
 }
 
+#[derive(Debug, Clone, Deserialize, Serialize)]
+pub struct TurnDiffEvent {
+    pub unified_diff: String,
+}
+
 #[derive(Debug, Clone, Deserialize, Serialize)]
 pub struct GetHistoryEntryResponseEvent {
     pub offset: usize,

codex-rs/core/src/safety.rs

@@ -41,11 +41,13 @@ pub fn assess_patch_safety(
         }
     }
 
-    if is_write_patch_constrained_to_writable_paths(action, writable_roots, cwd) {
-        SafetyCheck::AutoApprove {
-            sandbox_type: SandboxType::None,
-        }
-    } else if policy == AskForApproval::OnFailure {
+    // Even though the patch *appears* to be constrained to writable paths, it
+    // is possible that paths in the patch are hard links to files outside the
+    // writable roots, so we should still run `apply_patch` in a sandbox in that
+    // case.
+    if is_write_patch_constrained_to_writable_paths(action, writable_roots, cwd)
+        || policy == AskForApproval::OnFailure
+    {
         // Only auto‑approve when we can actually enforce a sandbox. Otherwise
         // fall back to asking the user because the patch may touch arbitrary
         // paths outside the project.
@@ -75,9 +77,6 @@ pub fn assess_command_safety(
     sandbox_policy: &SandboxPolicy,
     approved: &HashSet<Vec<String>>,
 ) -> SafetyCheck {
-    use AskForApproval::*;
-    use SandboxPolicy::*;
-
     // A command is "trusted" because either:
     // - it belongs to a set of commands we consider "safe" by default, or
     // - the user has explicitly approved the command for this session
@@ -97,6 +96,16 @@ pub fn assess_command_safety(
         };
     }
 
+    assess_safety_for_untrusted_command(approval_policy, sandbox_policy)
+}
+
+pub(crate) fn assess_safety_for_untrusted_command(
+    approval_policy: AskForApproval,
+    sandbox_policy: &SandboxPolicy,
+) -> SafetyCheck {
+    use AskForApproval::*;
+    use SandboxPolicy::*;
+
     match (approval_policy, sandbox_policy) {
         (UnlessTrusted, _) => {
             // Even though the user may have opted into DangerFullAccess,

codex-rs/core/src/seatbelt.rs

@@ -0,0 +1,312 @@
+use std::collections::HashMap;
+use std::path::Path;
+use std::path::PathBuf;
+use tokio::process::Child;
+
+use crate::protocol::SandboxPolicy;
+use crate::spawn::CODEX_SANDBOX_ENV_VAR;
+use crate::spawn::StdioPolicy;
+use crate::spawn::spawn_child_async;
+
+const MACOS_SEATBELT_BASE_POLICY: &str = include_str!("seatbelt_base_policy.sbpl");
+
+/// When working with `sandbox-exec`, only consider `sandbox-exec` in `/usr/bin`
+/// to defend against an attacker trying to inject a malicious version on the
+/// PATH. If /usr/bin/sandbox-exec has been tampered with, then the attacker
+/// already has root access.
+const MACOS_PATH_TO_SEATBELT_EXECUTABLE: &str = "/usr/bin/sandbox-exec";
+
+pub async fn spawn_command_under_seatbelt(
+    command: Vec<String>,
+    sandbox_policy: &SandboxPolicy,
+    cwd: PathBuf,
+    stdio_policy: StdioPolicy,
+    mut env: HashMap<String, String>,
+) -> std::io::Result<Child> {
+    let args = create_seatbelt_command_args(command, sandbox_policy, &cwd);
+    let arg0 = None;
+    env.insert(CODEX_SANDBOX_ENV_VAR.to_string(), "seatbelt".to_string());
+    spawn_child_async(
+        PathBuf::from(MACOS_PATH_TO_SEATBELT_EXECUTABLE),
+        args,
+        arg0,
+        cwd,
+        sandbox_policy,
+        stdio_policy,
+        env,
+    )
+    .await
+}
+
+fn create_seatbelt_command_args(
+    command: Vec<String>,
+    sandbox_policy: &SandboxPolicy,
+    cwd: &Path,
+) -> Vec<String> {
+    let (file_write_policy, extra_cli_args) = {
+        if sandbox_policy.has_full_disk_write_access() {
+            // Allegedly, this is more permissive than `(allow file-write*)`.
+            (
+                r#"(allow file-write* (regex #"^/"))"#.to_string(),
+                Vec::<String>::new(),
+            )
+        } else {
+            let writable_roots = sandbox_policy.get_writable_roots_with_cwd(cwd);
+
+            let mut writable_folder_policies: Vec<String> = Vec::new();
+            let mut cli_args: Vec<String> = Vec::new();
+
+            for (index, wr) in writable_roots.iter().enumerate() {
+                // Canonicalize to avoid mismatches like /var vs /private/var on macOS.
+                let canonical_root = wr.root.canonicalize().unwrap_or_else(|_| wr.root.clone());
+                let root_param = format!("WRITABLE_ROOT_{index}");
+                cli_args.push(format!(
+                    "-D{root_param}={}",
+                    canonical_root.to_string_lossy()
+                ));
+
+                if wr.read_only_subpaths.is_empty() {
+                    writable_folder_policies.push(format!("(subpath (param \"{root_param}\"))"));
+                } else {
+                    // Add parameters for each read-only subpath and generate
+                    // the `(require-not ...)` clauses.
+                    let mut require_parts: Vec<String> = Vec::new();
+                    require_parts.push(format!("(subpath (param \"{root_param}\"))"));
+                    for (subpath_index, ro) in wr.read_only_subpaths.iter().enumerate() {
+                        let canonical_ro = ro.canonicalize().unwrap_or_else(|_| ro.clone());
+                        let ro_param = format!("WRITABLE_ROOT_{index}_RO_{subpath_index}");
+                        cli_args.push(format!("-D{ro_param}={}", canonical_ro.to_string_lossy()));
+                        require_parts
+                            .push(format!("(require-not (subpath (param \"{ro_param}\")))"));
+                    }
+                    let policy_component = format!("(require-all {} )", require_parts.join(" "));
+                    writable_folder_policies.push(policy_component);
+                }
+            }
+
+            if writable_folder_policies.is_empty() {
+                ("".to_string(), Vec::<String>::new())
+            } else {
+                let file_write_policy = format!(
+                    "(allow file-write*\n{}\n)",
+                    writable_folder_policies.join(" ")
+                );
+                (file_write_policy, cli_args)
+            }
+        }
+    };
+
+    let file_read_policy = if sandbox_policy.has_full_disk_read_access() {
+        "; allow read-only file operations\n(allow file-read*)"
+    } else {
+        ""
+    };
+
+    // TODO(mbolin): apply_patch calls must also honor the SandboxPolicy.
+    let network_policy = if sandbox_policy.has_full_network_access() {
+        "(allow network-outbound)\n(allow network-inbound)\n(allow system-socket)"
+    } else {
+        ""
+    };
+
+    let full_policy = format!(
+        "{MACOS_SEATBELT_BASE_POLICY}\n{file_read_policy}\n{file_write_policy}\n{network_policy}"
+    );
+
+    let mut seatbelt_args: Vec<String> = vec!["-p".to_string(), full_policy];
+    seatbelt_args.extend(extra_cli_args);
+    seatbelt_args.push("--".to_string());
+    seatbelt_args.extend(command);
+    seatbelt_args
+}
+
+#[cfg(test)]
+mod tests {
+    #![expect(clippy::expect_used)]
+    use super::MACOS_SEATBELT_BASE_POLICY;
+    use super::create_seatbelt_command_args;
+    use crate::protocol::SandboxPolicy;
+    use pretty_assertions::assert_eq;
+    use std::fs;
+    use std::path::Path;
+    use std::path::PathBuf;
+    use tempfile::TempDir;
+
+    #[test]
+    fn create_seatbelt_args_with_read_only_git_subpath() {
+        // Create a temporary workspace with two writable roots: one containing
+        // a top-level .git directory and one without it.
+        let tmp = TempDir::new().expect("tempdir");
+        let PopulatedTmp {
+            root_with_git,
+            root_without_git,
+            root_with_git_canon,
+            root_with_git_git_canon,
+            root_without_git_canon,
+        } = populate_tmpdir(tmp.path());
+
+        // Build a policy that only includes the two test roots as writable and
+        // does not automatically include defaults like cwd or TMPDIR.
+        let policy = SandboxPolicy::WorkspaceWrite {
+            writable_roots: vec![root_with_git.clone(), root_without_git.clone()],
+            network_access: false,
+            include_default_writable_roots: false,
+        };
+
+        let args = create_seatbelt_command_args(
+            vec!["/bin/echo".to_string(), "hello".to_string()],
+            &policy,
+            tmp.path(),
+        );
+
+        // Build the expected policy text using a raw string for readability.
+        // Note that the policy includes:
+        // - the base policy,
+        // - read-only access to the filesystem,
+        // - write access to WRITABLE_ROOT_0 (but not its .git) and WRITABLE_ROOT_1.
+        let expected_policy = format!(
+            r#"{MACOS_SEATBELT_BASE_POLICY}
+; allow read-only file operations
+(allow file-read*)
+(allow file-write*
+(require-all (subpath (param "WRITABLE_ROOT_0")) (require-not (subpath (param "WRITABLE_ROOT_0_RO_0"))) ) (subpath (param "WRITABLE_ROOT_1"))
+)
+"#,
+        );
+
+        let expected_args = vec![
+            "-p".to_string(),
+            expected_policy,
+            format!(
+                "-DWRITABLE_ROOT_0={}",
+                root_with_git_canon.to_string_lossy()
+            ),
+            format!(
+                "-DWRITABLE_ROOT_0_RO_0={}",
+                root_with_git_git_canon.to_string_lossy()
+            ),
+            format!(
+                "-DWRITABLE_ROOT_1={}",
+                root_without_git_canon.to_string_lossy()
+            ),
+            "--".to_string(),
+            "/bin/echo".to_string(),
+            "hello".to_string(),
+        ];
+
+        assert_eq!(args, expected_args);
+    }
+
+    #[test]
+    fn create_seatbelt_args_for_cwd_as_git_repo() {
+        // Create a temporary workspace with two writable roots: one containing
+        // a top-level .git directory and one without it.
+        let tmp = TempDir::new().expect("tempdir");
+        let PopulatedTmp {
+            root_with_git,
+            root_with_git_canon,
+            root_with_git_git_canon,
+            ..
+        } = populate_tmpdir(tmp.path());
+
+        // Build a policy that does not specify any writable_roots, but does
+        // use the default ones (cwd and TMPDIR) and verifies the `.git` check
+        // is done properly for cwd.
+        let policy = SandboxPolicy::WorkspaceWrite {
+            writable_roots: vec![],
+            network_access: false,
+            include_default_writable_roots: true,
+        };
+
+        let args = create_seatbelt_command_args(
+            vec!["/bin/echo".to_string(), "hello".to_string()],
+            &policy,
+            root_with_git.as_path(),
+        );
+
+        let tmpdir_env_var = if cfg!(target_os = "macos") {
+            std::env::var("TMPDIR")
+                .ok()
+                .map(PathBuf::from)
+                .and_then(|p| p.canonicalize().ok())
+                .map(|p| p.to_string_lossy().to_string())
+        } else {
+            None
+        };
+        let tempdir_policy_entry = if tmpdir_env_var.is_some() {
+            " (subpath (param \"WRITABLE_ROOT_1\"))"
+        } else {
+            ""
+        };
+
+        // Build the expected policy text using a raw string for readability.
+        // Note that the policy includes:
+        // - the base policy,
+        // - read-only access to the filesystem,
+        // - write access to WRITABLE_ROOT_0 (but not its .git) and WRITABLE_ROOT_1.
+        let expected_policy = format!(
+            r#"{MACOS_SEATBELT_BASE_POLICY}
+; allow read-only file operations
+(allow file-read*)
+(allow file-write*
+(require-all (subpath (param "WRITABLE_ROOT_0")) (require-not (subpath (param "WRITABLE_ROOT_0_RO_0"))) ){tempdir_policy_entry}
+)
+"#,
+        );
+
+        let mut expected_args = vec![
+            "-p".to_string(),
+            expected_policy,
+            format!(
+                "-DWRITABLE_ROOT_0={}",
+                root_with_git_canon.to_string_lossy()
+            ),
+            format!(
+                "-DWRITABLE_ROOT_0_RO_0={}",
+                root_with_git_git_canon.to_string_lossy()
+            ),
+        ];
+
+        if let Some(p) = tmpdir_env_var {
+            expected_args.push(format!("-DWRITABLE_ROOT_1={p}"));
+        }
+
+        expected_args.extend(vec![
+            "--".to_string(),
+            "/bin/echo".to_string(),
+            "hello".to_string(),
+        ]);
+
+        assert_eq!(args, expected_args);
+    }
+
+    struct PopulatedTmp {
+        root_with_git: PathBuf,
+        root_without_git: PathBuf,
+        root_with_git_canon: PathBuf,
+        root_with_git_git_canon: PathBuf,
+        root_without_git_canon: PathBuf,
+    }
+
+    fn populate_tmpdir(tmp: &Path) -> PopulatedTmp {
+        let root_with_git = tmp.join("with_git");
+        let root_without_git = tmp.join("no_git");
+        fs::create_dir_all(&root_with_git).expect("create with_git");
+        fs::create_dir_all(&root_without_git).expect("create no_git");
+        fs::create_dir_all(root_with_git.join(".git")).expect("create .git");
+
+        // Ensure we have canonical paths for -D parameter matching.
+        let root_with_git_canon = root_with_git.canonicalize().expect("canonicalize with_git");
+        let root_with_git_git_canon = root_with_git_canon.join(".git");
+        let root_without_git_canon = root_without_git
+            .canonicalize()
+            .expect("canonicalize no_git");
+        PopulatedTmp {
+            root_with_git,
+            root_without_git,
+            root_with_git_canon,
+            root_with_git_git_canon,
+            root_without_git_canon,
+        }
+    }
+}

codex-rs/core/src/seatbelt_base_policy.sbpl

@@ -65,3 +65,7 @@
   (sysctl-name "sysctl.proc_cputype")
   (sysctl-name-prefix "hw.perflevel")
 )
+
+; Added on top of Chrome profile
+; Needed for python multiprocessing on MacOS for the SemLock
+(allow ipc-posix-sem)

codex-rs/core/src/shell.rs

@@ -220,6 +220,7 @@ mod tests {
                 Arc::new(Notify::new()),
                 &SandboxPolicy::DangerFullAccess,
                 &None,
+                None,
             )
             .await
             .unwrap();

codex-rs/core/src/spawn.rs

@@ -0,0 +1,107 @@
+use std::collections::HashMap;
+use std::path::PathBuf;
+use std::process::Stdio;
+use tokio::process::Child;
+use tokio::process::Command;
+use tracing::trace;
+
+use crate::protocol::SandboxPolicy;
+
+/// Experimental environment variable that will be set to some non-empty value
+/// if both of the following are true:
+///
+/// 1. The process was spawned by Codex as part of a shell tool call.
+/// 2. SandboxPolicy.has_full_network_access() was false for the tool call.
+///
+/// We may try to have just one environment variable for all sandboxing
+/// attributes, so this may change in the future.
+pub const CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR: &str = "CODEX_SANDBOX_NETWORK_DISABLED";
+
+/// Should be set when the process is spawned under a sandbox. Currently, the
+/// value is "seatbelt" for macOS, but it may change in the future to
+/// accommodate sandboxing configuration and other sandboxing mechanisms.
+pub const CODEX_SANDBOX_ENV_VAR: &str = "CODEX_SANDBOX";
+
+#[derive(Debug, Clone, Copy)]
+pub enum StdioPolicy {
+    RedirectForShellTool,
+    Inherit,
+}
+
+/// Spawns the appropriate child process for the ExecParams and SandboxPolicy,
+/// ensuring the args and environment variables used to create the `Command`
+/// (and `Child`) honor the configuration.
+///
+/// For now, we take `SandboxPolicy` as a parameter to spawn_child() because
+/// we need to determine whether to set the
+/// `CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR` environment variable.
+pub(crate) async fn spawn_child_async(
+    program: PathBuf,
+    args: Vec<String>,
+    #[cfg_attr(not(unix), allow(unused_variables))] arg0: Option<&str>,
+    cwd: PathBuf,
+    sandbox_policy: &SandboxPolicy,
+    stdio_policy: StdioPolicy,
+    env: HashMap<String, String>,
+) -> std::io::Result<Child> {
+    trace!(
+        "spawn_child_async: {program:?} {args:?} {arg0:?} {cwd:?} {sandbox_policy:?} {stdio_policy:?} {env:?}"
+    );
+
+    let mut cmd = Command::new(&program);
+    #[cfg(unix)]
+    cmd.arg0(arg0.map_or_else(|| program.to_string_lossy().to_string(), String::from));
+    cmd.args(args);
+    cmd.current_dir(cwd);
+    cmd.env_clear();
+    cmd.envs(env);
+
+    if !sandbox_policy.has_full_network_access() {
+        cmd.env(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR, "1");
+    }
+
+    // If this Codex process dies (including being killed via SIGKILL), we want
+    // any child processes that were spawned as part of a `"shell"` tool call
+    // to also be terminated.
+
+    // This relies on prctl(2), so it only works on Linux.
+    #[cfg(target_os = "linux")]
+    unsafe {
+        cmd.pre_exec(|| {
+            // This prctl call effectively requests, "deliver SIGTERM when my
+            // current parent dies."
+            if libc::prctl(libc::PR_SET_PDEATHSIG, libc::SIGTERM) == -1 {
+                return Err(std::io::Error::last_os_error());
+            }
+
+            // Though if there was a race condition and this pre_exec() block is
+            // run _after_ the parent (i.e., the Codex process) has already
+            // exited, then the parent is the _init_ process (which will never
+            // die), so we should just terminate the child process now.
+            if libc::getppid() == 1 {
+                libc::raise(libc::SIGTERM);
+            }
+            Ok(())
+        });
+    }
+
+    match stdio_policy {
+        StdioPolicy::RedirectForShellTool => {
+            // Do not create a file descriptor for stdin because otherwise some
+            // commands may hang forever waiting for input. For example, ripgrep has
+            // a heuristic where it may try to read from stdin as explained here:
+            // https://github.com/BurntSushi/ripgrep/blob/e2362d4d5185d02fa857bf381e7bd52e66fafc73/crates/core/flags/hiargs.rs#L1101-L1103
+            cmd.stdin(Stdio::null());
+
+            cmd.stdout(Stdio::piped()).stderr(Stdio::piped());
+        }
+        StdioPolicy::Inherit => {
+            // Inherit stdin, stdout, and stderr from the parent process.
+            cmd.stdin(Stdio::inherit())
+                .stdout(Stdio::inherit())
+                .stderr(Stdio::inherit());
+        }
+    }
+
+    cmd.kill_on_drop(true).spawn()
+}

codex-rs/core/src/turn_diff_tracker.rs

@@ -0,0 +1,887 @@
+use std::collections::HashMap;
+use std::fs;
+use std::path::Path;
+use std::path::PathBuf;
+use std::process::Command;
+
+use anyhow::Context;
+use anyhow::Result;
+use anyhow::anyhow;
+use sha1::digest::Output;
+use uuid::Uuid;
+
+use crate::protocol::FileChange;
+
+const ZERO_OID: &str = "0000000000000000000000000000000000000000";
+const DEV_NULL: &str = "/dev/null";
+
+struct BaselineFileInfo {
+    path: PathBuf,
+    content: Vec<u8>,
+    mode: FileMode,
+    oid: String,
+}
+
+/// Tracks sets of changes to files and exposes the overall unified diff.
+/// Internally, the way this works is now:
+/// 1. Maintain an in-memory baseline snapshot of files when they are first seen.
+///    For new additions, do not create a baseline so that diffs are shown as proper additions (using /dev/null).
+/// 2. Keep a stable internal filename (uuid) per external path for rename tracking.
+/// 3. To compute the aggregated unified diff, compare each baseline snapshot to the current file on disk entirely in-memory
+///    using the `similar` crate and emit unified diffs with rewritten external paths.
+#[derive(Default)]
+pub struct TurnDiffTracker {
+    /// Map external path -> internal filename (uuid).
+    external_to_temp_name: HashMap<PathBuf, String>,
+    /// Internal filename -> baseline file info.
+    baseline_file_info: HashMap<String, BaselineFileInfo>,
+    /// Internal filename -> external path as of current accumulated state (after applying all changes).
+    /// This is where renames are tracked.
+    temp_name_to_current_path: HashMap<String, PathBuf>,
+    /// Cache of known git worktree roots to avoid repeated filesystem walks.
+    git_root_cache: Vec<PathBuf>,
+}
+
+impl TurnDiffTracker {
+    pub fn new() -> Self {
+        Self::default()
+    }
+
+    /// Front-run apply patch calls to track the starting contents of any modified files.
+    /// - Creates an in-memory baseline snapshot for files that already exist on disk when first seen.
+    /// - For additions, we intentionally do not create a baseline snapshot so that diffs are proper additions.
+    /// - Also updates internal mappings for move/rename events.
+    pub fn on_patch_begin(&mut self, changes: &HashMap<PathBuf, FileChange>) {
+        for (path, change) in changes.iter() {
+            // Ensure a stable internal filename exists for this external path.
+            if !self.external_to_temp_name.contains_key(path) {
+                let internal = Uuid::new_v4().to_string();
+                self.external_to_temp_name
+                    .insert(path.clone(), internal.clone());
+                self.temp_name_to_current_path
+                    .insert(internal.clone(), path.clone());
+
+                // If the file exists on disk now, snapshot as baseline; else leave missing to represent /dev/null.
+                let baseline_file_info = if path.exists() {
+                    let mode = file_mode_for_path(path);
+                    let mode_val = mode.unwrap_or(FileMode::Regular);
+                    let content = blob_bytes(path, &mode_val).unwrap_or_default();
+                    let oid = if mode == Some(FileMode::Symlink) {
+                        format!("{:x}", git_blob_sha1_hex_bytes(&content))
+                    } else {
+                        self.git_blob_oid_for_path(path)
+                            .unwrap_or_else(|| format!("{:x}", git_blob_sha1_hex_bytes(&content)))
+                    };
+                    Some(BaselineFileInfo {
+                        path: path.clone(),
+                        content,
+                        mode: mode_val,
+                        oid,
+                    })
+                } else {
+                    Some(BaselineFileInfo {
+                        path: path.clone(),
+                        content: vec![],
+                        mode: FileMode::Regular,
+                        oid: ZERO_OID.to_string(),
+                    })
+                };
+
+                if let Some(baseline_file_info) = baseline_file_info {
+                    self.baseline_file_info
+                        .insert(internal.clone(), baseline_file_info);
+                }
+            }
+
+            // Track rename/move in current mapping if provided in an Update.
+            if let FileChange::Update {
+                move_path: Some(dest),
+                ..
+            } = change
+            {
+                let uuid_filename = match self.external_to_temp_name.get(path) {
+                    Some(i) => i.clone(),
+                    None => {
+                        // This should be rare, but if we haven't mapped the source, create it with no baseline.
+                        let i = Uuid::new_v4().to_string();
+                        self.baseline_file_info.insert(
+                            i.clone(),
+                            BaselineFileInfo {
+                                path: path.clone(),
+                                content: vec![],
+                                mode: FileMode::Regular,
+                                oid: ZERO_OID.to_string(),
+                            },
+                        );
+                        i
+                    }
+                };
+                // Update current external mapping for temp file name.
+                self.temp_name_to_current_path
+                    .insert(uuid_filename.clone(), dest.clone());
+                // Update forward file_mapping: external current -> internal name.
+                self.external_to_temp_name.remove(path);
+                self.external_to_temp_name
+                    .insert(dest.clone(), uuid_filename);
+            };
+        }
+    }
+
+    fn get_path_for_internal(&self, internal: &str) -> Option<PathBuf> {
+        self.temp_name_to_current_path
+            .get(internal)
+            .cloned()
+            .or_else(|| {
+                self.baseline_file_info
+                    .get(internal)
+                    .map(|info| info.path.clone())
+            })
+    }
+
+    /// Find the git worktree root for a file/directory by walking up to the first ancestor containing a `.git` entry.
+    /// Uses a simple cache of known roots and avoids negative-result caching for simplicity.
+    fn find_git_root_cached(&mut self, start: &Path) -> Option<PathBuf> {
+        let dir = if start.is_dir() {
+            start
+        } else {
+            start.parent()?
+        };
+
+        // Fast path: if any cached root is an ancestor of this path, use it.
+        if let Some(root) = self
+            .git_root_cache
+            .iter()
+            .find(|r| dir.starts_with(r))
+            .cloned()
+        {
+            return Some(root);
+        }
+
+        // Walk up to find a `.git` marker.
+        let mut cur = dir.to_path_buf();
+        loop {
+            let git_marker = cur.join(".git");
+            if git_marker.is_dir() || git_marker.is_file() {
+                if !self.git_root_cache.iter().any(|r| r == &cur) {
+                    self.git_root_cache.push(cur.clone());
+                }
+                return Some(cur);
+            }
+
+            // On Windows, avoid walking above the drive or UNC share root.
+            #[cfg(windows)]
+            {
+                if is_windows_drive_or_unc_root(&cur) {
+                    return None;
+                }
+            }
+
+            if let Some(parent) = cur.parent() {
+                cur = parent.to_path_buf();
+            } else {
+                return None;
+            }
+        }
+    }
+
+    /// Return a display string for `path` relative to its git root if found, else absolute.
+    fn relative_to_git_root_str(&mut self, path: &Path) -> String {
+        let s = if let Some(root) = self.find_git_root_cached(path) {
+            if let Ok(rel) = path.strip_prefix(&root) {
+                rel.display().to_string()
+            } else {
+                path.display().to_string()
+            }
+        } else {
+            path.display().to_string()
+        };
+        s.replace('\\', "/")
+    }
+
+    /// Ask git to compute the blob SHA-1 for the file at `path` within its repository.
+    /// Returns None if no repository is found or git invocation fails.
+    fn git_blob_oid_for_path(&mut self, path: &Path) -> Option<String> {
+        let root = self.find_git_root_cached(path)?;
+        // Compute a path relative to the repo root for better portability across platforms.
+        let rel = path.strip_prefix(&root).unwrap_or(path);
+        let output = Command::new("git")
+            .arg("-C")
+            .arg(&root)
+            .arg("hash-object")
+            .arg("--")
+            .arg(rel)
+            .output()
+            .ok()?;
+        if !output.status.success() {
+            return None;
+        }
+        let s = String::from_utf8_lossy(&output.stdout).trim().to_string();
+        if s.len() == 40 { Some(s) } else { None }
+    }
+
+    /// Recompute the aggregated unified diff by comparing all of the in-memory snapshots that were
+    /// collected before the first time they were touched by apply_patch during this turn with
+    /// the current repo state.
+    pub fn get_unified_diff(&mut self) -> Result<Option<String>> {
+        let mut aggregated = String::new();
+
+        // Compute diffs per tracked internal file in a stable order by external path.
+        let mut baseline_file_names: Vec<String> =
+            self.baseline_file_info.keys().cloned().collect();
+        // Sort lexicographically by full repo-relative path to match git behavior.
+        baseline_file_names.sort_by_key(|internal| {
+            self.get_path_for_internal(internal)
+                .map(|p| self.relative_to_git_root_str(&p))
+                .unwrap_or_default()
+        });
+
+        for internal in baseline_file_names {
+            aggregated.push_str(self.get_file_diff(&internal).as_str());
+            if !aggregated.ends_with('\n') {
+                aggregated.push('\n');
+            }
+        }
+
+        if aggregated.trim().is_empty() {
+            Ok(None)
+        } else {
+            Ok(Some(aggregated))
+        }
+    }
+
+    fn get_file_diff(&mut self, internal_file_name: &str) -> String {
+        let mut aggregated = String::new();
+
+        // Snapshot lightweight fields only.
+        let (baseline_external_path, baseline_mode, left_oid) = {
+            if let Some(info) = self.baseline_file_info.get(internal_file_name) {
+                (info.path.clone(), info.mode, info.oid.clone())
+            } else {
+                (PathBuf::new(), FileMode::Regular, ZERO_OID.to_string())
+            }
+        };
+        let current_external_path = match self.get_path_for_internal(internal_file_name) {
+            Some(p) => p,
+            None => return aggregated,
+        };
+
+        let current_mode = file_mode_for_path(&current_external_path).unwrap_or(FileMode::Regular);
+        let right_bytes = blob_bytes(&current_external_path, &current_mode);
+
+        // Compute displays with &mut self before borrowing any baseline content.
+        let left_display = self.relative_to_git_root_str(&baseline_external_path);
+        let right_display = self.relative_to_git_root_str(&current_external_path);
+
+        // Compute right oid before borrowing baseline content.
+        let right_oid = if let Some(b) = right_bytes.as_ref() {
+            if current_mode == FileMode::Symlink {
+                format!("{:x}", git_blob_sha1_hex_bytes(b))
+            } else {
+                self.git_blob_oid_for_path(&current_external_path)
+                    .unwrap_or_else(|| format!("{:x}", git_blob_sha1_hex_bytes(b)))
+            }
+        } else {
+            ZERO_OID.to_string()
+        };
+
+        // Borrow baseline content only after all &mut self uses are done.
+        let left_present = left_oid.as_str() != ZERO_OID;
+        let left_bytes: Option<&[u8]> = if left_present {
+            self.baseline_file_info
+                .get(internal_file_name)
+                .map(|i| i.content.as_slice())
+        } else {
+            None
+        };
+
+        // Fast path: identical bytes or both missing.
+        if left_bytes == right_bytes.as_deref() {
+            return aggregated;
+        }
+
+        aggregated.push_str(&format!("diff --git a/{left_display} b/{right_display}\n"));
+
+        let is_add = !left_present && right_bytes.is_some();
+        let is_delete = left_present && right_bytes.is_none();
+
+        if is_add {
+            aggregated.push_str(&format!("new file mode {current_mode}\n"));
+        } else if is_delete {
+            aggregated.push_str(&format!("deleted file mode {baseline_mode}\n"));
+        } else if baseline_mode != current_mode {
+            aggregated.push_str(&format!("old mode {baseline_mode}\n"));
+            aggregated.push_str(&format!("new mode {current_mode}\n"));
+        }
+
+        let left_text = left_bytes.and_then(|b| std::str::from_utf8(b).ok());
+        let right_text = right_bytes
+            .as_deref()
+            .and_then(|b| std::str::from_utf8(b).ok());
+
+        let can_text_diff = matches!(
+            (left_text, right_text, is_add, is_delete),
+            (Some(_), Some(_), _, _) | (_, Some(_), true, _) | (Some(_), _, _, true)
+        );
+
+        if can_text_diff {
+            let l = left_text.unwrap_or("");
+            let r = right_text.unwrap_or("");
+
+            aggregated.push_str(&format!("index {left_oid}..{right_oid}\n"));
+
+            let old_header = if left_present {
+                format!("a/{left_display}")
+            } else {
+                DEV_NULL.to_string()
+            };
+            let new_header = if right_bytes.is_some() {
+                format!("b/{right_display}")
+            } else {
+                DEV_NULL.to_string()
+            };
+
+            let diff = similar::TextDiff::from_lines(l, r);
+            let unified = diff
+                .unified_diff()
+                .context_radius(3)
+                .header(&old_header, &new_header)
+                .to_string();
+
+            aggregated.push_str(&unified);
+        } else {
+            aggregated.push_str(&format!("index {left_oid}..{right_oid}\n"));
+            let old_header = if left_present {
+                format!("a/{left_display}")
+            } else {
+                DEV_NULL.to_string()
+            };
+            let new_header = if right_bytes.is_some() {
+                format!("b/{right_display}")
+            } else {
+                DEV_NULL.to_string()
+            };
+            aggregated.push_str(&format!("--- {old_header}\n"));
+            aggregated.push_str(&format!("+++ {new_header}\n"));
+            aggregated.push_str("Binary files differ\n");
+        }
+        aggregated
+    }
+}
+
+/// Compute the Git SHA-1 blob object ID for the given content (bytes).
+fn git_blob_sha1_hex_bytes(data: &[u8]) -> Output<sha1::Sha1> {
+    // Git blob hash is sha1 of: "blob <len>\0<data>"
+    let header = format!("blob {}\0", data.len());
+    use sha1::Digest;
+    let mut hasher = sha1::Sha1::new();
+    hasher.update(header.as_bytes());
+    hasher.update(data);
+    hasher.finalize()
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+enum FileMode {
+    Regular,
+    #[cfg(unix)]
+    Executable,
+    Symlink,
+}
+
+impl FileMode {
+    fn as_str(&self) -> &'static str {
+        match self {
+            FileMode::Regular => "100644",
+            #[cfg(unix)]
+            FileMode::Executable => "100755",
+            FileMode::Symlink => "120000",
+        }
+    }
+}
+
+impl std::fmt::Display for FileMode {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        f.write_str(self.as_str())
+    }
+}
+
+#[cfg(unix)]
+fn file_mode_for_path(path: &Path) -> Option<FileMode> {
+    use std::os::unix::fs::PermissionsExt;
+    let meta = fs::symlink_metadata(path).ok()?;
+    let ft = meta.file_type();
+    if ft.is_symlink() {
+        return Some(FileMode::Symlink);
+    }
+    let mode = meta.permissions().mode();
+    let is_exec = (mode & 0o111) != 0;
+    Some(if is_exec {
+        FileMode::Executable
+    } else {
+        FileMode::Regular
+    })
+}
+
+#[cfg(not(unix))]
+fn file_mode_for_path(_path: &Path) -> Option<FileMode> {
+    // Default to non-executable on non-unix.
+    Some(FileMode::Regular)
+}
+
+fn blob_bytes(path: &Path, mode: &FileMode) -> Option<Vec<u8>> {
+    if path.exists() {
+        let contents = if *mode == FileMode::Symlink {
+            symlink_blob_bytes(path)
+                .ok_or_else(|| anyhow!("failed to read symlink target for {}", path.display()))
+        } else {
+            fs::read(path)
+                .with_context(|| format!("failed to read current file for diff {}", path.display()))
+        };
+        contents.ok()
+    } else {
+        None
+    }
+}
+
+#[cfg(unix)]
+fn symlink_blob_bytes(path: &Path) -> Option<Vec<u8>> {
+    use std::os::unix::ffi::OsStrExt;
+    let target = std::fs::read_link(path).ok()?;
+    Some(target.as_os_str().as_bytes().to_vec())
+}
+
+#[cfg(not(unix))]
+fn symlink_blob_bytes(_path: &Path) -> Option<Vec<u8>> {
+    None
+}
+
+#[cfg(windows)]
+fn is_windows_drive_or_unc_root(p: &std::path::Path) -> bool {
+    use std::path::Component;
+    let mut comps = p.components();
+    matches!(
+        (comps.next(), comps.next(), comps.next()),
+        (Some(Component::Prefix(_)), Some(Component::RootDir), None)
+    )
+}
+
+#[cfg(test)]
+mod tests {
+    #![allow(clippy::unwrap_used)]
+    use super::*;
+    use pretty_assertions::assert_eq;
+    use tempfile::tempdir;
+
+    /// Compute the Git SHA-1 blob object ID for the given content (string).
+    /// This delegates to the bytes version to avoid UTF-8 lossy conversions here.
+    fn git_blob_sha1_hex(data: &str) -> String {
+        format!("{:x}", git_blob_sha1_hex_bytes(data.as_bytes()))
+    }
+
+    fn normalize_diff_for_test(input: &str, root: &Path) -> String {
+        let root_str = root.display().to_string().replace('\\', "/");
+        let replaced = input.replace(&root_str, "<TMP>");
+        // Split into blocks on lines starting with "diff --git ", sort blocks for determinism, and rejoin
+        let mut blocks: Vec<String> = Vec::new();
+        let mut current = String::new();
+        for line in replaced.lines() {
+            if line.starts_with("diff --git ") && !current.is_empty() {
+                blocks.push(current);
+                current = String::new();
+            }
+            if !current.is_empty() {
+                current.push('\n');
+            }
+            current.push_str(line);
+        }
+        if !current.is_empty() {
+            blocks.push(current);
+        }
+        blocks.sort();
+        let mut out = blocks.join("\n");
+        if !out.ends_with('\n') {
+            out.push('\n');
+        }
+        out
+    }
+
+    #[test]
+    fn accumulates_add_and_update() {
+        let mut acc = TurnDiffTracker::new();
+
+        let dir = tempdir().unwrap();
+        let file = dir.path().join("a.txt");
+
+        // First patch: add file (baseline should be /dev/null).
+        let add_changes = HashMap::from([(
+            file.clone(),
+            FileChange::Add {
+                content: "foo\n".to_string(),
+            },
+        )]);
+        acc.on_patch_begin(&add_changes);
+
+        // Simulate apply: create the file on disk.
+        fs::write(&file, "foo\n").unwrap();
+        let first = acc.get_unified_diff().unwrap().unwrap();
+        let first = normalize_diff_for_test(&first, dir.path());
+        let expected_first = {
+            let mode = file_mode_for_path(&file).unwrap_or(FileMode::Regular);
+            let right_oid = git_blob_sha1_hex("foo\n");
+            format!(
+                r#"diff --git a/<TMP>/a.txt b/<TMP>/a.txt
+new file mode {mode}
+index {ZERO_OID}..{right_oid}
+--- {DEV_NULL}
++++ b/<TMP>/a.txt
+@@ -0,0 +1 @@
++foo
+"#,
+            )
+        };
+        assert_eq!(first, expected_first);
+
+        // Second patch: update the file on disk.
+        let update_changes = HashMap::from([(
+            file.clone(),
+            FileChange::Update {
+                unified_diff: "".to_owned(),
+                move_path: None,
+            },
+        )]);
+        acc.on_patch_begin(&update_changes);
+
+        // Simulate apply: append a new line.
+        fs::write(&file, "foo\nbar\n").unwrap();
+        let combined = acc.get_unified_diff().unwrap().unwrap();
+        let combined = normalize_diff_for_test(&combined, dir.path());
+        let expected_combined = {
+            let mode = file_mode_for_path(&file).unwrap_or(FileMode::Regular);
+            let right_oid = git_blob_sha1_hex("foo\nbar\n");
+            format!(
+                r#"diff --git a/<TMP>/a.txt b/<TMP>/a.txt
+new file mode {mode}
+index {ZERO_OID}..{right_oid}
+--- {DEV_NULL}
++++ b/<TMP>/a.txt
+@@ -0,0 +1,2 @@
++foo
++bar
+"#,
+            )
+        };
+        assert_eq!(combined, expected_combined);
+    }
+
+    #[test]
+    fn accumulates_delete() {
+        let dir = tempdir().unwrap();
+        let file = dir.path().join("b.txt");
+        fs::write(&file, "x\n").unwrap();
+
+        let mut acc = TurnDiffTracker::new();
+        let del_changes = HashMap::from([(file.clone(), FileChange::Delete)]);
+        acc.on_patch_begin(&del_changes);
+
+        // Simulate apply: delete the file from disk.
+        let baseline_mode = file_mode_for_path(&file).unwrap_or(FileMode::Regular);
+        fs::remove_file(&file).unwrap();
+        let diff = acc.get_unified_diff().unwrap().unwrap();
+        let diff = normalize_diff_for_test(&diff, dir.path());
+        let expected = {
+            let left_oid = git_blob_sha1_hex("x\n");
+            format!(
+                r#"diff --git a/<TMP>/b.txt b/<TMP>/b.txt
+deleted file mode {baseline_mode}
+index {left_oid}..{ZERO_OID}
+--- a/<TMP>/b.txt
++++ {DEV_NULL}
+@@ -1 +0,0 @@
+-x
+"#,
+            )
+        };
+        assert_eq!(diff, expected);
+    }
+
+    #[test]
+    fn accumulates_move_and_update() {
+        let dir = tempdir().unwrap();
+        let src = dir.path().join("src.txt");
+        let dest = dir.path().join("dst.txt");
+        fs::write(&src, "line\n").unwrap();
+
+        let mut acc = TurnDiffTracker::new();
+        let mv_changes = HashMap::from([(
+            src.clone(),
+            FileChange::Update {
+                unified_diff: "".to_owned(),
+                move_path: Some(dest.clone()),
+            },
+        )]);
+        acc.on_patch_begin(&mv_changes);
+
+        // Simulate apply: move and update content.
+        fs::rename(&src, &dest).unwrap();
+        fs::write(&dest, "line2\n").unwrap();
+
+        let out = acc.get_unified_diff().unwrap().unwrap();
+        let out = normalize_diff_for_test(&out, dir.path());
+        let expected = {
+            let left_oid = git_blob_sha1_hex("line\n");
+            let right_oid = git_blob_sha1_hex("line2\n");
+            format!(
+                r#"diff --git a/<TMP>/src.txt b/<TMP>/dst.txt
+index {left_oid}..{right_oid}
+--- a/<TMP>/src.txt
++++ b/<TMP>/dst.txt
+@@ -1 +1 @@
+-line
++line2
+"#
+            )
+        };
+        assert_eq!(out, expected);
+    }
+
+    #[test]
+    fn move_without_1change_yields_no_diff() {
+        let dir = tempdir().unwrap();
+        let src = dir.path().join("moved.txt");
+        let dest = dir.path().join("renamed.txt");
+        fs::write(&src, "same\n").unwrap();
+
+        let mut acc = TurnDiffTracker::new();
+        let mv_changes = HashMap::from([(
+            src.clone(),
+            FileChange::Update {
+                unified_diff: "".to_owned(),
+                move_path: Some(dest.clone()),
+            },
+        )]);
+        acc.on_patch_begin(&mv_changes);
+
+        // Simulate apply: move only, no content change.
+        fs::rename(&src, &dest).unwrap();
+
+        let diff = acc.get_unified_diff().unwrap();
+        assert_eq!(diff, None);
+    }
+
+    #[test]
+    fn move_declared_but_file_only_appears_at_dest_is_add() {
+        let dir = tempdir().unwrap();
+        let src = dir.path().join("src.txt");
+        let dest = dir.path().join("dest.txt");
+        let mut acc = TurnDiffTracker::new();
+        let mv = HashMap::from([(
+            src.clone(),
+            FileChange::Update {
+                unified_diff: "".into(),
+                move_path: Some(dest.clone()),
+            },
+        )]);
+        acc.on_patch_begin(&mv);
+        // No file existed initially; create only dest
+        fs::write(&dest, "hello\n").unwrap();
+        let diff = acc.get_unified_diff().unwrap().unwrap();
+        let diff = normalize_diff_for_test(&diff, dir.path());
+        let expected = {
+            let mode = file_mode_for_path(&dest).unwrap_or(FileMode::Regular);
+            let right_oid = git_blob_sha1_hex("hello\n");
+            format!(
+                r#"diff --git a/<TMP>/src.txt b/<TMP>/dest.txt
+new file mode {mode}
+index {ZERO_OID}..{right_oid}
+--- {DEV_NULL}
++++ b/<TMP>/dest.txt
+@@ -0,0 +1 @@
++hello
+"#,
+            )
+        };
+        assert_eq!(diff, expected);
+    }
+
+    #[test]
+    fn update_persists_across_new_baseline_for_new_file() {
+        let dir = tempdir().unwrap();
+        let a = dir.path().join("a.txt");
+        let b = dir.path().join("b.txt");
+        fs::write(&a, "foo\n").unwrap();
+        fs::write(&b, "z\n").unwrap();
+
+        let mut acc = TurnDiffTracker::new();
+
+        // First: update existing a.txt (baseline snapshot is created for a).
+        let update_a = HashMap::from([(
+            a.clone(),
+            FileChange::Update {
+                unified_diff: "".to_owned(),
+                move_path: None,
+            },
+        )]);
+        acc.on_patch_begin(&update_a);
+        // Simulate apply: modify a.txt on disk.
+        fs::write(&a, "foo\nbar\n").unwrap();
+        let first = acc.get_unified_diff().unwrap().unwrap();
+        let first = normalize_diff_for_test(&first, dir.path());
+        let expected_first = {
+            let left_oid = git_blob_sha1_hex("foo\n");
+            let right_oid = git_blob_sha1_hex("foo\nbar\n");
+            format!(
+                r#"diff --git a/<TMP>/a.txt b/<TMP>/a.txt
+index {left_oid}..{right_oid}
+--- a/<TMP>/a.txt
++++ b/<TMP>/a.txt
+@@ -1 +1,2 @@
+ foo
++bar
+"#
+            )
+        };
+        assert_eq!(first, expected_first);
+
+        // Next: introduce a brand-new path b.txt into baseline snapshots via a delete change.
+        let del_b = HashMap::from([(b.clone(), FileChange::Delete)]);
+        acc.on_patch_begin(&del_b);
+        // Simulate apply: delete b.txt.
+        let baseline_mode = file_mode_for_path(&b).unwrap_or(FileMode::Regular);
+        fs::remove_file(&b).unwrap();
+
+        let combined = acc.get_unified_diff().unwrap().unwrap();
+        let combined = normalize_diff_for_test(&combined, dir.path());
+        let expected = {
+            let left_oid_a = git_blob_sha1_hex("foo\n");
+            let right_oid_a = git_blob_sha1_hex("foo\nbar\n");
+            let left_oid_b = git_blob_sha1_hex("z\n");
+            format!(
+                r#"diff --git a/<TMP>/a.txt b/<TMP>/a.txt
+index {left_oid_a}..{right_oid_a}
+--- a/<TMP>/a.txt
++++ b/<TMP>/a.txt
+@@ -1 +1,2 @@
+ foo
++bar
+diff --git a/<TMP>/b.txt b/<TMP>/b.txt
+deleted file mode {baseline_mode}
+index {left_oid_b}..{ZERO_OID}
+--- a/<TMP>/b.txt
++++ {DEV_NULL}
+@@ -1 +0,0 @@
+-z
+"#,
+            )
+        };
+        assert_eq!(combined, expected);
+    }
+
+    #[test]
+    fn binary_files_differ_update() {
+        let dir = tempdir().unwrap();
+        let file = dir.path().join("bin.dat");
+
+        // Initial non-UTF8 bytes
+        let left_bytes: Vec<u8> = vec![0xff, 0xfe, 0xfd, 0x00];
+        // Updated non-UTF8 bytes
+        let right_bytes: Vec<u8> = vec![0x01, 0x02, 0x03, 0x00];
+
+        fs::write(&file, &left_bytes).unwrap();
+
+        let mut acc = TurnDiffTracker::new();
+        let update_changes = HashMap::from([(
+            file.clone(),
+            FileChange::Update {
+                unified_diff: "".to_owned(),
+                move_path: None,
+            },
+        )]);
+        acc.on_patch_begin(&update_changes);
+
+        // Apply update on disk
+        fs::write(&file, &right_bytes).unwrap();
+
+        let diff = acc.get_unified_diff().unwrap().unwrap();
+        let diff = normalize_diff_for_test(&diff, dir.path());
+        let expected = {
+            let left_oid = format!("{:x}", git_blob_sha1_hex_bytes(&left_bytes));
+            let right_oid = format!("{:x}", git_blob_sha1_hex_bytes(&right_bytes));
+            format!(
+                r#"diff --git a/<TMP>/bin.dat b/<TMP>/bin.dat
+index {left_oid}..{right_oid}
+--- a/<TMP>/bin.dat
++++ b/<TMP>/bin.dat
+Binary files differ
+"#
+            )
+        };
+        assert_eq!(diff, expected);
+    }
+
+    #[test]
+    fn filenames_with_spaces_add_and_update() {
+        let mut acc = TurnDiffTracker::new();
+
+        let dir = tempdir().unwrap();
+        let file = dir.path().join("name with spaces.txt");
+
+        // First patch: add file (baseline should be /dev/null).
+        let add_changes = HashMap::from([(
+            file.clone(),
+            FileChange::Add {
+                content: "foo\n".to_string(),
+            },
+        )]);
+        acc.on_patch_begin(&add_changes);
+
+        // Simulate apply: create the file on disk.
+        fs::write(&file, "foo\n").unwrap();
+        let first = acc.get_unified_diff().unwrap().unwrap();
+        let first = normalize_diff_for_test(&first, dir.path());
+        let expected_first = {
+            let mode = file_mode_for_path(&file).unwrap_or(FileMode::Regular);
+            let right_oid = git_blob_sha1_hex("foo\n");
+            format!(
+                r#"diff --git a/<TMP>/name with spaces.txt b/<TMP>/name with spaces.txt
+new file mode {mode}
+index {ZERO_OID}..{right_oid}
+--- {DEV_NULL}
++++ b/<TMP>/name with spaces.txt
+@@ -0,0 +1 @@
++foo
+"#,
+            )
+        };
+        assert_eq!(first, expected_first);
+
+        // Second patch: update the file on disk.
+        let update_changes = HashMap::from([(
+            file.clone(),
+            FileChange::Update {
+                unified_diff: "".to_owned(),
+                move_path: None,
+            },
+        )]);
+        acc.on_patch_begin(&update_changes);
+
+        // Simulate apply: append a new line with a space.
+        fs::write(&file, "foo\nbar baz\n").unwrap();
+        let combined = acc.get_unified_diff().unwrap().unwrap();
+        let combined = normalize_diff_for_test(&combined, dir.path());
+        let expected_combined = {
+            let mode = file_mode_for_path(&file).unwrap_or(FileMode::Regular);
+            let right_oid = git_blob_sha1_hex("foo\nbar baz\n");
+            format!(
+                r#"diff --git a/<TMP>/name with spaces.txt b/<TMP>/name with spaces.txt
+new file mode {mode}
+index {ZERO_OID}..{right_oid}
+--- {DEV_NULL}
++++ b/<TMP>/name with spaces.txt
+@@ -0,0 +1,2 @@
++foo
++bar baz
+"#,
+            )
+        };
+        assert_eq!(combined, expected_combined);
+    }
+}

codex-rs/core/tests/cli_stream.rs

@@ -1,7 +1,7 @@
 #![expect(clippy::unwrap_used)]
 
 use assert_cmd::Command as AssertCommand;
-use codex_core::exec::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
+use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use std::time::Duration;
 use std::time::Instant;
 use tempfile::TempDir;
@@ -460,9 +460,14 @@ async fn integration_git_info_unit_test() {
     // 1. Create temp directory for git repo
     let temp_dir = TempDir::new().unwrap();
     let git_repo = temp_dir.path().to_path_buf();
+    let envs = vec![
+        ("GIT_CONFIG_GLOBAL", "/dev/null"),
+        ("GIT_CONFIG_NOSYSTEM", "1"),
+    ];
 
     // 2. Initialize a git repository with some content
     let init_output = std::process::Command::new("git")
+        .envs(envs.clone())
         .args(["init"])
         .current_dir(&git_repo)
         .output()
@@ -471,12 +476,14 @@ async fn integration_git_info_unit_test() {
 
     // Configure git user (required for commits)
     std::process::Command::new("git")
+        .envs(envs.clone())
         .args(["config", "user.name", "Integration Test"])
         .current_dir(&git_repo)
         .output()
         .unwrap();
 
     std::process::Command::new("git")
+        .envs(envs.clone())
         .args(["config", "user.email", "test@example.com"])
         .current_dir(&git_repo)
         .output()
@@ -487,12 +494,14 @@ async fn integration_git_info_unit_test() {
     std::fs::write(&test_file, "integration test content").unwrap();
 
     std::process::Command::new("git")
+        .envs(envs.clone())
         .args(["add", "."])
         .current_dir(&git_repo)
         .output()
         .unwrap();
 
     let commit_output = std::process::Command::new("git")
+        .envs(envs.clone())
         .args(["commit", "-m", "Integration test commit"])
         .current_dir(&git_repo)
         .output()
@@ -501,6 +510,7 @@ async fn integration_git_info_unit_test() {
 
     // Create a branch to test branch detection
     std::process::Command::new("git")
+        .envs(envs.clone())
         .args(["checkout", "-b", "integration-test-branch"])
         .current_dir(&git_repo)
         .output()
@@ -508,6 +518,7 @@ async fn integration_git_info_unit_test() {
 
     // Add a remote to test repository URL detection
     std::process::Command::new("git")
+        .envs(envs.clone())
         .args([
             "remote",
             "add",

codex-rs/core/tests/client.rs

@@ -5,11 +5,11 @@ use codex_core::Codex;
 use codex_core::CodexSpawnOk;
 use codex_core::ModelProviderInfo;
 use codex_core::built_in_model_providers;
-use codex_core::exec::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::InputItem;
 use codex_core::protocol::Op;
 use codex_core::protocol::SessionConfiguredEvent;
+use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_login::AuthDotJson;
 use codex_login::AuthMode;
 use codex_login::CodexAuth;
@@ -95,8 +95,8 @@ async fn includes_session_id_and_model_headers_in_request() {
     // get request from the server
     let request = &server.received_requests().await.unwrap()[0];
     let request_session_id = request.headers.get("session_id").unwrap();
-    let request_originator = request.headers.get("originator").unwrap();
     let request_authorization = request.headers.get("authorization").unwrap();
+    let request_originator = request.headers.get("originator").unwrap();
 
     assert!(current_session_id.is_some());
     assert_eq!(
@@ -170,6 +170,59 @@ async fn includes_base_instructions_override_in_request() {
     );
 }
 
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn originator_config_override_is_used() {
+    #![allow(clippy::unwrap_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.internal_originator = Some("my_override".to_string());
+
+    let ctrl_c = std::sync::Arc::new(tokio::sync::Notify::new());
+    let CodexSpawnOk { codex, .. } = Codex::spawn(
+        config,
+        Some(CodexAuth::from_api_key("Test API Key".to_string())),
+        ctrl_c.clone(),
+    )
+    .await
+    .unwrap();
+
+    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() {
     #![allow(clippy::unwrap_used)]
@@ -235,8 +288,9 @@ async fn chatgpt_auth_sends_correct_request() {
     // get request from the server
     let request = &server.received_requests().await.unwrap()[0];
     let request_session_id = request.headers.get("session_id").unwrap();
-    let request_originator = request.headers.get("originator").unwrap();
     let request_authorization = request.headers.get("authorization").unwrap();
+    let request_originator = request.headers.get("originator").unwrap();
+    let request_chatgpt_account_id = request.headers.get("chatgpt-account-id").unwrap();
     let request_body = request.body_json::<serde_json::Value>().unwrap();
 
     assert!(current_session_id.is_some());
@@ -249,6 +303,7 @@ async fn chatgpt_auth_sends_correct_request() {
         request_authorization.to_str().unwrap(),
         "Bearer Access Token"
     );
+    assert_eq!(request_chatgpt_account_id.to_str().unwrap(), "account_id");
     assert!(!request_body["store"].as_bool().unwrap());
     assert!(request_body["stream"].as_bool().unwrap());
     assert_eq!(
@@ -327,14 +382,14 @@ fn auth_from_token(id_token: String) -> CodexAuth {
         AuthMode::ChatGPT,
         PathBuf::new(),
         Some(AuthDotJson {
-            tokens: TokenData {
+            openai_api_key: None,
+            tokens: Some(TokenData {
                 id_token,
                 access_token: "Access Token".to_string(),
                 refresh_token: "test".to_string(),
-                account_id: None,
-            },
-            last_refresh: Utc::now(),
-            openai_api_key: None,
+                account_id: Some("account_id".to_string()),
+            }),
+            last_refresh: Some(Utc::now()),
         }),
     )
 }

codex-rs/core/tests/compact.rs

@@ -0,0 +1,254 @@
+#![expect(clippy::unwrap_used)]
+
+use codex_core::Codex;
+use codex_core::CodexSpawnOk;
+use codex_core::ModelProviderInfo;
+use codex_core::built_in_model_providers;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::InputItem;
+use codex_core::protocol::Op;
+use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
+use codex_login::CodexAuth;
+use core_test_support::load_default_config_for_test;
+use core_test_support::wait_for_event;
+use serde_json::Value;
+use tempfile::TempDir;
+use wiremock::Mock;
+use wiremock::MockServer;
+use wiremock::ResponseTemplate;
+use wiremock::matchers::method;
+use wiremock::matchers::path;
+
+use pretty_assertions::assert_eq;
+
+// --- Test helpers -----------------------------------------------------------
+
+/// Build an SSE stream body from a list of JSON events.
+fn sse(events: Vec<Value>) -> String {
+    use std::fmt::Write as _;
+    let mut out = String::new();
+    for ev in events {
+        let kind = ev.get("type").and_then(|v| v.as_str()).unwrap();
+        writeln!(&mut out, "event: {kind}").unwrap();
+        if !ev.as_object().map(|o| o.len() == 1).unwrap_or(false) {
+            write!(&mut out, "data: {ev}\n\n").unwrap();
+        } else {
+            out.push('\n');
+        }
+    }
+    out
+}
+
+/// Convenience: SSE event for a completed response with a specific id.
+fn ev_completed(id: &str) -> Value {
+    serde_json::json!({
+        "type": "response.completed",
+        "response": {
+            "id": id,
+            "usage": {"input_tokens":0,"input_tokens_details":null,"output_tokens":0,"output_tokens_details":null,"total_tokens":0}
+        }
+    })
+}
+
+/// Convenience: SSE event for a single assistant message output item.
+fn ev_assistant_message(id: &str, text: &str) -> Value {
+    serde_json::json!({
+        "type": "response.output_item.done",
+        "item": {
+            "type": "message",
+            "role": "assistant",
+            "id": id,
+            "content": [{"type": "output_text", "text": text}]
+        }
+    })
+}
+
+fn sse_response(body: String) -> ResponseTemplate {
+    ResponseTemplate::new(200)
+        .insert_header("content-type", "text/event-stream")
+        .set_body_raw(body, "text/event-stream")
+}
+
+async fn mount_sse_once<M>(server: &MockServer, matcher: M, body: String)
+where
+    M: wiremock::Match + Send + Sync + 'static,
+{
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(matcher)
+        .respond_with(sse_response(body))
+        .expect(1)
+        .mount(server)
+        .await;
+}
+
+const FIRST_REPLY: &str = "FIRST_REPLY";
+const SUMMARY_TEXT: &str = "SUMMARY_ONLY_CONTEXT";
+const SUMMARIZE_TRIGGER: &str = "Start Summarization";
+const THIRD_USER_MSG: &str = "next turn";
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn summarize_context_three_requests_and_instructions() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    // Set up a mock server that we can inspect after the run.
+    let server = MockServer::start().await;
+
+    // SSE 1: assistant replies normally so it is recorded in history.
+    let sse1 = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed("r1"),
+    ]);
+
+    // SSE 2: summarizer returns a summary message.
+    let sse2 = sse(vec![
+        ev_assistant_message("m2", SUMMARY_TEXT),
+        ev_completed("r2"),
+    ]);
+
+    // SSE 3: minimal completed; we only need to capture the request body.
+    let sse3 = sse(vec![ev_completed("r3")]);
+
+    // Mount three expectations, one per request, matched by body content.
+    let first_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("\"text\":\"hello world\"")
+            && !body.contains(&format!("\"text\":\"{SUMMARIZE_TRIGGER}\""))
+    };
+    mount_sse_once(&server, first_matcher, sse1).await;
+
+    let second_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(&format!("\"text\":\"{SUMMARIZE_TRIGGER}\""))
+    };
+    mount_sse_once(&server, second_matcher, sse2).await;
+
+    let third_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(&format!("\"text\":\"{THIRD_USER_MSG}\""))
+    };
+    mount_sse_once(&server, third_matcher, sse3).await;
+
+    // Build config pointing to the mock server and spawn Codex.
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+    let home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&home);
+    config.model_provider = model_provider;
+    let ctrl_c = std::sync::Arc::new(tokio::sync::Notify::new());
+    let CodexSpawnOk { codex, .. } = Codex::spawn(
+        config,
+        Some(CodexAuth::from_api_key("dummy".to_string())),
+        ctrl_c.clone(),
+    )
+    .await
+    .unwrap();
+
+    // 1) Normal user input – should hit server once.
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: "hello world".into(),
+            }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // 2) Summarize – second hit with summarization instructions.
+    codex.submit(Op::Compact).await.unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // 3) Next user input – third hit; history should include only the summary.
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: THIRD_USER_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Inspect the three captured requests.
+    let requests = server.received_requests().await.unwrap();
+    assert_eq!(requests.len(), 3, "expected exactly three requests");
+
+    let req1 = &requests[0];
+    let req2 = &requests[1];
+    let req3 = &requests[2];
+
+    let body1 = req1.body_json::<serde_json::Value>().unwrap();
+    let body2 = req2.body_json::<serde_json::Value>().unwrap();
+    let body3 = req3.body_json::<serde_json::Value>().unwrap();
+
+    // System instructions should change for the summarization turn.
+    let instr1 = body1.get("instructions").and_then(|v| v.as_str()).unwrap();
+    let instr2 = body2.get("instructions").and_then(|v| v.as_str()).unwrap();
+    assert_ne!(
+        instr1, instr2,
+        "summarization should override base instructions"
+    );
+    assert!(
+        instr2.contains("You are a summarization assistant"),
+        "summarization instructions not applied"
+    );
+
+    // The summarization request should include the injected user input marker.
+    let input2 = body2.get("input").and_then(|v| v.as_array()).unwrap();
+    // The last item is the user message created from the injected input.
+    let last2 = input2.last().unwrap();
+    assert_eq!(last2.get("type").unwrap().as_str().unwrap(), "message");
+    assert_eq!(last2.get("role").unwrap().as_str().unwrap(), "user");
+    let text2 = last2["content"][0]["text"].as_str().unwrap();
+    assert!(text2.contains(SUMMARIZE_TRIGGER));
+
+    // Third request must contain only the summary from step 2 as prior history plus new user msg.
+    let input3 = body3.get("input").and_then(|v| v.as_array()).unwrap();
+    println!("third request body: {body3}");
+    assert!(
+        input3.len() >= 2,
+        "expected summary + new user message in third request"
+    );
+
+    // Collect all (role, text) message tuples.
+    let mut messages: Vec<(String, String)> = Vec::new();
+    for item in input3 {
+        if item["type"].as_str() == Some("message") {
+            let role = item["role"].as_str().unwrap_or_default().to_string();
+            let text = item["content"][0]["text"]
+                .as_str()
+                .unwrap_or_default()
+                .to_string();
+            messages.push((role, text));
+        }
+    }
+
+    // Exactly one assistant message should remain after compaction and the new user message is present.
+    let assistant_count = messages.iter().filter(|(r, _)| r == "assistant").count();
+    assert_eq!(
+        assistant_count, 1,
+        "exactly one assistant message should remain after compaction"
+    );
+    assert!(
+        messages
+            .iter()
+            .any(|(r, t)| r == "user" && t == THIRD_USER_MSG),
+        "third request should include the new user message"
+    );
+    assert!(
+        !messages.iter().any(|(_, t)| t.contains("hello world")),
+        "third request should not include the original user input"
+    );
+    assert!(
+        !messages.iter().any(|(_, t)| t.contains(SUMMARIZE_TRIGGER)),
+        "third request should not include the summarize trigger"
+    );
+}

codex-rs/core/tests/exec.rs

@@ -0,0 +1,69 @@
+#![cfg(target_os = "macos")]
+#![expect(clippy::expect_used)]
+
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use codex_core::exec::ExecParams;
+use codex_core::exec::SandboxType;
+use codex_core::exec::process_exec_tool_call;
+use codex_core::protocol::SandboxPolicy;
+use codex_core::spawn::CODEX_SANDBOX_ENV_VAR;
+use tempfile::TempDir;
+use tokio::sync::Notify;
+
+use codex_core::get_platform_sandbox;
+
+async fn run_test_cmd(tmp: TempDir, cmd: Vec<&str>, should_be_ok: bool) {
+    if std::env::var(CODEX_SANDBOX_ENV_VAR) == Ok("seatbelt".to_string()) {
+        eprintln!("{CODEX_SANDBOX_ENV_VAR} is set to 'seatbelt', skipping test.");
+        return;
+    }
+
+    let sandbox_type = get_platform_sandbox().expect("should be able to get sandbox type");
+    assert_eq!(sandbox_type, SandboxType::MacosSeatbelt);
+
+    let params = ExecParams {
+        command: cmd.iter().map(|s| s.to_string()).collect(),
+        cwd: tmp.path().to_path_buf(),
+        timeout_ms: Some(1000),
+        env: HashMap::new(),
+    };
+
+    let ctrl_c = Arc::new(Notify::new());
+    let policy = SandboxPolicy::new_read_only_policy();
+
+    let result = process_exec_tool_call(params, sandbox_type, ctrl_c, &policy, &None, None).await;
+
+    assert!(result.is_ok() == should_be_ok);
+}
+
+/// Command succeeds with exit code 0 normally
+#[tokio::test]
+async fn exit_code_0_succeeds() {
+    let tmp = TempDir::new().expect("should be able to create temp dir");
+    let cmd = vec!["echo", "hello"];
+
+    run_test_cmd(tmp, cmd, true).await
+}
+
+/// Command not found returns exit code 127, this is not considered a sandbox error
+#[tokio::test]
+async fn exit_command_not_found_is_ok() {
+    let tmp = TempDir::new().expect("should be able to create temp dir");
+    let cmd = vec!["/bin/bash", "-c", "nonexistent_command_12345"];
+    run_test_cmd(tmp, cmd, true).await
+}
+
+/// Writing a file fails and should be considered a sandbox error
+#[tokio::test]
+async fn write_file_fails_as_sandbox_error() {
+    let tmp = TempDir::new().expect("should be able to create temp dir");
+    let path = tmp.path().join("test.txt");
+    let cmd = vec![
+        "/user/bin/touch",
+        path.to_str().expect("should be able to get path"),
+    ];
+
+    run_test_cmd(tmp, cmd, false).await;
+}

codex-rs/core/tests/exec_stream_events.rs

@@ -0,0 +1,143 @@
+#![cfg(unix)]
+
+use std::collections::HashMap;
+use std::path::PathBuf;
+use std::sync::Arc;
+
+use async_channel::Receiver;
+use codex_core::exec::ExecParams;
+use codex_core::exec::SandboxType;
+use codex_core::exec::StdoutStream;
+use codex_core::exec::process_exec_tool_call;
+use codex_core::protocol::Event;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::ExecCommandOutputDeltaEvent;
+use codex_core::protocol::ExecOutputStream;
+use codex_core::protocol::SandboxPolicy;
+use tokio::sync::Notify;
+
+fn collect_stdout_events(rx: Receiver<Event>) -> Vec<u8> {
+    let mut out = Vec::new();
+    while let Ok(ev) = rx.try_recv() {
+        if let EventMsg::ExecCommandOutputDelta(ExecCommandOutputDeltaEvent {
+            stream: ExecOutputStream::Stdout,
+            chunk,
+            ..
+        }) = ev.msg
+        {
+            out.extend_from_slice(&chunk);
+        }
+    }
+    out
+}
+
+#[tokio::test]
+async fn test_exec_stdout_stream_events_echo() {
+    let (tx, rx) = async_channel::unbounded::<Event>();
+
+    let stdout_stream = StdoutStream {
+        sub_id: "test-sub".to_string(),
+        call_id: "call-1".to_string(),
+        tx_event: tx,
+    };
+
+    let cmd = vec![
+        "/bin/sh".to_string(),
+        "-c".to_string(),
+        // Use printf for predictable behavior across shells
+        "printf 'hello-world\n'".to_string(),
+    ];
+
+    let params = ExecParams {
+        command: cmd,
+        cwd: std::env::current_dir().unwrap_or_else(|_| PathBuf::from(".")),
+        timeout_ms: Some(5_000),
+        env: HashMap::new(),
+    };
+
+    let ctrl_c = Arc::new(Notify::new());
+    let policy = SandboxPolicy::new_read_only_policy();
+
+    let result = process_exec_tool_call(
+        params,
+        SandboxType::None,
+        ctrl_c,
+        &policy,
+        &None,
+        Some(stdout_stream),
+    )
+    .await;
+
+    let result = match result {
+        Ok(r) => r,
+        Err(e) => panic!("process_exec_tool_call failed: {e}"),
+    };
+
+    assert_eq!(result.exit_code, 0);
+    assert_eq!(result.stdout, "hello-world\n");
+
+    let streamed = collect_stdout_events(rx);
+    // We should have received at least the same contents (possibly in one chunk)
+    assert_eq!(String::from_utf8_lossy(&streamed), "hello-world\n");
+}
+
+#[tokio::test]
+async fn test_exec_stderr_stream_events_echo() {
+    let (tx, rx) = async_channel::unbounded::<Event>();
+
+    let stdout_stream = StdoutStream {
+        sub_id: "test-sub".to_string(),
+        call_id: "call-2".to_string(),
+        tx_event: tx,
+    };
+
+    let cmd = vec![
+        "/bin/sh".to_string(),
+        "-c".to_string(),
+        // Write to stderr explicitly
+        "printf 'oops\n' 1>&2".to_string(),
+    ];
+
+    let params = ExecParams {
+        command: cmd,
+        cwd: std::env::current_dir().unwrap_or_else(|_| PathBuf::from(".")),
+        timeout_ms: Some(5_000),
+        env: HashMap::new(),
+    };
+
+    let ctrl_c = Arc::new(Notify::new());
+    let policy = SandboxPolicy::new_read_only_policy();
+
+    let result = process_exec_tool_call(
+        params,
+        SandboxType::None,
+        ctrl_c,
+        &policy,
+        &None,
+        Some(stdout_stream),
+    )
+    .await;
+
+    let result = match result {
+        Ok(r) => r,
+        Err(e) => panic!("process_exec_tool_call failed: {e}"),
+    };
+
+    assert_eq!(result.exit_code, 0);
+    assert_eq!(result.stdout, "");
+    assert_eq!(result.stderr, "oops\n");
+
+    // Collect only stderr delta events
+    let mut err = Vec::new();
+    while let Ok(ev) = rx.try_recv() {
+        if let EventMsg::ExecCommandOutputDelta(ExecCommandOutputDeltaEvent {
+            stream: ExecOutputStream::Stderr,
+            chunk,
+            ..
+        }) = ev.msg
+        {
+            err.extend_from_slice(&chunk);
+        }
+    }
+    assert_eq!(String::from_utf8_lossy(&err), "oops\n");
+}

codex-rs/core/tests/live_agent.rs

@@ -177,8 +177,7 @@ async fn live_shell_function_call() {
         match ev.msg {
             EventMsg::ExecCommandBegin(codex_core::protocol::ExecCommandBeginEvent {
                 command,
-                call_id: _,
-                cwd: _,
+                ..
             }) => {
                 assert_eq!(command, vec!["echo", MARKER]);
                 saw_begin = true;
@@ -186,8 +185,7 @@ async fn live_shell_function_call() {
             EventMsg::ExecCommandEnd(codex_core::protocol::ExecCommandEndEvent {
                 stdout,
                 exit_code,
-                call_id: _,
-                stderr: _,
+                ..
             }) => {
                 assert_eq!(exit_code, 0, "echo returned non‑zero exit code");
                 assert!(stdout.contains(MARKER));

codex-rs/core/tests/sandbox.rs

@@ -0,0 +1,195 @@
+#![cfg(target_os = "macos")]
+#![expect(clippy::expect_used)]
+
+use std::collections::HashMap;
+use std::path::Path;
+use std::path::PathBuf;
+
+use codex_core::protocol::SandboxPolicy;
+use codex_core::seatbelt::spawn_command_under_seatbelt;
+use codex_core::spawn::CODEX_SANDBOX_ENV_VAR;
+use codex_core::spawn::StdioPolicy;
+use tempfile::TempDir;
+
+struct TestScenario {
+    repo_parent: PathBuf,
+    file_outside_repo: PathBuf,
+    repo_root: PathBuf,
+    file_in_repo_root: PathBuf,
+    file_in_dot_git_dir: PathBuf,
+}
+
+struct TestExpectations {
+    file_outside_repo_is_writable: bool,
+    file_in_repo_root_is_writable: bool,
+    file_in_dot_git_dir_is_writable: bool,
+}
+
+impl TestScenario {
+    async fn run_test(&self, policy: &SandboxPolicy, expectations: TestExpectations) {
+        if std::env::var(CODEX_SANDBOX_ENV_VAR) == Ok("seatbelt".to_string()) {
+            eprintln!("{CODEX_SANDBOX_ENV_VAR} is set to 'seatbelt', skipping test.");
+            return;
+        }
+
+        assert_eq!(
+            touch(&self.file_outside_repo, policy).await,
+            expectations.file_outside_repo_is_writable
+        );
+        assert_eq!(
+            self.file_outside_repo.exists(),
+            expectations.file_outside_repo_is_writable
+        );
+
+        assert_eq!(
+            touch(&self.file_in_repo_root, policy).await,
+            expectations.file_in_repo_root_is_writable
+        );
+        assert_eq!(
+            self.file_in_repo_root.exists(),
+            expectations.file_in_repo_root_is_writable
+        );
+
+        assert_eq!(
+            touch(&self.file_in_dot_git_dir, policy).await,
+            expectations.file_in_dot_git_dir_is_writable
+        );
+        assert_eq!(
+            self.file_in_dot_git_dir.exists(),
+            expectations.file_in_dot_git_dir_is_writable
+        );
+    }
+}
+
+/// If the user has added a workspace root that is not a Git repo root, then
+/// the user has to specify `--skip-git-repo-check` or go through some
+/// interstitial that indicates they are taking on some risk because Git
+/// cannot be used to backup their work before the agent begins.
+///
+/// Because the user has agreed to this risk, we do not try find all .git
+/// folders in the workspace and block them (though we could change our
+/// position on this in the future).
+#[tokio::test]
+async fn if_parent_of_repo_is_writable_then_dot_git_folder_is_writable() {
+    let tmp = TempDir::new().expect("should be able to create temp dir");
+    let test_scenario = create_test_scenario(&tmp);
+    let policy = SandboxPolicy::WorkspaceWrite {
+        writable_roots: vec![test_scenario.repo_parent.clone()],
+        network_access: false,
+        include_default_writable_roots: false,
+    };
+
+    test_scenario
+        .run_test(
+            &policy,
+            TestExpectations {
+                file_outside_repo_is_writable: true,
+                file_in_repo_root_is_writable: true,
+                file_in_dot_git_dir_is_writable: true,
+            },
+        )
+        .await;
+}
+
+/// When the writable root is the root of a Git repository (as evidenced by the
+/// presence of a .git folder), then the .git folder should be read-only if
+/// the policy is `WorkspaceWrite`.
+#[tokio::test]
+async fn if_git_repo_is_writable_root_then_dot_git_folder_is_read_only() {
+    let tmp = TempDir::new().expect("should be able to create temp dir");
+    let test_scenario = create_test_scenario(&tmp);
+    let policy = SandboxPolicy::WorkspaceWrite {
+        writable_roots: vec![test_scenario.repo_root.clone()],
+        network_access: false,
+        include_default_writable_roots: false,
+    };
+
+    test_scenario
+        .run_test(
+            &policy,
+            TestExpectations {
+                file_outside_repo_is_writable: false,
+                file_in_repo_root_is_writable: true,
+                file_in_dot_git_dir_is_writable: false,
+            },
+        )
+        .await;
+}
+
+/// Under DangerFullAccess, all writes should be permitted anywhere on disk,
+/// including inside the .git folder.
+#[tokio::test]
+async fn danger_full_access_allows_all_writes() {
+    let tmp = TempDir::new().expect("should be able to create temp dir");
+    let test_scenario = create_test_scenario(&tmp);
+    let policy = SandboxPolicy::DangerFullAccess;
+
+    test_scenario
+        .run_test(
+            &policy,
+            TestExpectations {
+                file_outside_repo_is_writable: true,
+                file_in_repo_root_is_writable: true,
+                file_in_dot_git_dir_is_writable: true,
+            },
+        )
+        .await;
+}
+
+/// Under ReadOnly, writes should not be permitted anywhere on disk.
+#[tokio::test]
+async fn read_only_forbids_all_writes() {
+    let tmp = TempDir::new().expect("should be able to create temp dir");
+    let test_scenario = create_test_scenario(&tmp);
+    let policy = SandboxPolicy::ReadOnly;
+
+    test_scenario
+        .run_test(
+            &policy,
+            TestExpectations {
+                file_outside_repo_is_writable: false,
+                file_in_repo_root_is_writable: false,
+                file_in_dot_git_dir_is_writable: false,
+            },
+        )
+        .await;
+}
+
+fn create_test_scenario(tmp: &TempDir) -> TestScenario {
+    let repo_parent = tmp.path().to_path_buf();
+    let repo_root = repo_parent.join("repo");
+    let dot_git_dir = repo_root.join(".git");
+
+    std::fs::create_dir(&repo_root).expect("should be able to create repo root");
+    std::fs::create_dir(&dot_git_dir).expect("should be able to create .git dir");
+
+    TestScenario {
+        file_outside_repo: repo_parent.join("outside.txt"),
+        repo_parent,
+        file_in_repo_root: repo_root.join("repo_file.txt"),
+        repo_root,
+        file_in_dot_git_dir: dot_git_dir.join("dot_git_file.txt"),
+    }
+}
+
+/// Note that `path` must be absolute.
+async fn touch(path: &Path, policy: &SandboxPolicy) -> bool {
+    assert!(path.is_absolute(), "Path must be absolute: {path:?}");
+    let mut child = spawn_command_under_seatbelt(
+        vec![
+            "/usr/bin/touch".to_string(),
+            path.to_string_lossy().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 command under seatbelt");
+    child
+        .wait()
+        .await
+        .expect("should be able to wait for child process")
+        .success()
+}

codex-rs/core/tests/stream_no_completed.rs

@@ -6,10 +6,10 @@ use std::time::Duration;
 use codex_core::Codex;
 use codex_core::CodexSpawnOk;
 use codex_core::ModelProviderInfo;
-use codex_core::exec::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::InputItem;
 use codex_core::protocol::Op;
+use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_login::CodexAuth;
 use core_test_support::load_default_config_for_test;
 use core_test_support::load_sse_fixture;

codex-rs/exec/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-exec"
 version = { workspace = true }
-edition = "2024"
 
 [[bin]]
 name = "codex-exec"
@@ -19,12 +19,12 @@ anyhow = "1"
 chrono = "0.4.40"
 clap = { version = "4", features = ["derive"] }
 codex-arg0 = { path = "../arg0" }
-codex-core = { path = "../core" }
 codex-common = { path = "../common", features = [
     "cli",
     "elapsed",
     "sandbox_summary",
 ] }
+codex-core = { path = "../core" }
 owo-colors = "4.2.0"
 serde_json = "1"
 shlex = "1.3.0"

codex-rs/exec/src/event_processor.rs

@@ -44,20 +44,14 @@ pub(crate) fn create_config_summary_entries(config: &Config) -> Vec<(&'static st
     entries
 }
 
-pub(crate) fn handle_last_message(
-    last_agent_message: Option<&str>,
-    last_message_path: Option<&Path>,
-) {
-    match (last_message_path, last_agent_message) {
-        (Some(path), Some(msg)) => write_last_message_file(msg, Some(path)),
-        (Some(path), None) => {
-            write_last_message_file("", Some(path));
-            eprintln!(
-                "Warning: no last agent message; wrote empty content to {}",
-                path.display()
-            );
-        }
-        (None, _) => eprintln!("Warning: no file to write last message to."),
+pub(crate) fn handle_last_message(last_agent_message: Option<&str>, output_file: &Path) {
+    let message = last_agent_message.unwrap_or_default();
+    write_last_message_file(message, Some(output_file));
+    if last_agent_message.is_none() {
+        eprintln!(
+            "Warning: no last agent message; wrote empty content to {}",
+            output_file.display()
+        );
     }
 }
 

codex-rs/exec/src/event_processor_with_human_output.rs

@@ -20,6 +20,7 @@ use codex_core::protocol::PatchApplyEndEvent;
 use codex_core::protocol::SessionConfiguredEvent;
 use codex_core::protocol::TaskCompleteEvent;
 use codex_core::protocol::TokenUsage;
+use codex_core::protocol::TurnDiffEvent;
 use owo_colors::OwoColorize;
 use owo_colors::Style;
 use shlex::try_join;
@@ -106,7 +107,6 @@ impl EventProcessorWithHumanOutput {
 
 struct ExecCommandBegin {
     command: Vec<String>,
-    start_time: Instant,
 }
 
 struct PatchApplyBegin {
@@ -170,10 +170,9 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                 // Ignore.
             }
             EventMsg::TaskComplete(TaskCompleteEvent { last_agent_message }) => {
-                handle_last_message(
-                    last_agent_message.as_deref(),
-                    self.last_message_path.as_deref(),
-                );
+                if let Some(output_file) = self.last_message_path.as_deref() {
+                    handle_last_message(last_agent_message.as_deref(), output_file);
+                }
                 return CodexStatus::InitiateShutdown;
             }
             EventMsg::TokenCount(TokenUsage { total_tokens, .. }) => {
@@ -228,7 +227,6 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                     call_id.clone(),
                     ExecCommandBegin {
                         command: command.clone(),
-                        start_time: Instant::now(),
                     },
                 );
                 ts_println!(
@@ -239,20 +237,19 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                     cwd.to_string_lossy(),
                 );
             }
+            EventMsg::ExecCommandOutputDelta(_) => {}
             EventMsg::ExecCommandEnd(ExecCommandEndEvent {
                 call_id,
                 stdout,
                 stderr,
+                duration,
                 exit_code,
             }) => {
                 let exec_command = self.call_id_to_command.remove(&call_id);
-                let (duration, call) = if let Some(ExecCommandBegin {
-                    command,
-                    start_time,
-                }) = exec_command
+                let (duration, call) = if let Some(ExecCommandBegin { command, .. }) = exec_command
                 {
                     (
-                        format!(" in {}", format_elapsed(start_time)),
+                        format!(" in {}", format_duration(duration)),
                         format!("{}", escape_command(&command).style(self.bold)),
                     )
                 } else {
@@ -402,6 +399,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                 stdout,
                 stderr,
                 success,
+                ..
             }) => {
                 let patch_begin = self.call_id_to_patch.remove(&call_id);
 
@@ -431,6 +429,10 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                     println!("{}", line.style(self.dimmed));
                 }
             }
+            EventMsg::TurnDiff(TurnDiffEvent { unified_diff }) => {
+                ts_println!(self, "{}", "turn diff:".style(self.magenta));
+                println!("{unified_diff}");
+            }
             EventMsg::ExecApprovalRequest(_) => {
                 // Should we exit?
             }

codex-rs/exec/src/event_processor_with_json_output.rs

@@ -46,10 +46,9 @@ impl EventProcessor for EventProcessorWithJsonOutput {
                 CodexStatus::Running
             }
             EventMsg::TaskComplete(TaskCompleteEvent { last_agent_message }) => {
-                handle_last_message(
-                    last_agent_message.as_deref(),
-                    self.last_message_path.as_deref(),
-                );
+                if let Some(output_file) = self.last_message_path.as_deref() {
+                    handle_last_message(last_agent_message.as_deref(), output_file);
+                }
                 CodexStatus::InitiateShutdown
             }
             EventMsg::ShutdownComplete => CodexStatus::Shutdown,

codex-rs/exec/tests/apply_patch.rs

@@ -1,5 +1,6 @@
 use anyhow::Context;
 use assert_cmd::prelude::*;
+use codex_core::CODEX_APPLY_PATCH_ARG1;
 use std::fs;
 use std::process::Command;
 use tempfile::tempdir;
@@ -16,7 +17,7 @@ fn test_standalone_exec_cli_can_use_apply_patch() -> anyhow::Result<()> {
 
     Command::cargo_bin("codex-exec")
         .context("should find binary for codex-exec")?
-        .arg("--codex-run-as-apply-patch")
+        .arg(CODEX_APPLY_PATCH_ARG1)
         .arg(
             r#"*** Begin Patch
 *** Update File: source.txt

codex-rs/execpolicy/Cargo.toml

@@ -26,7 +26,7 @@ multimap = "0.10.0"
 path-absolutize = "3.1.1"
 regex-lite = "0.1"
 serde = { version = "1.0.194", features = ["derive"] }
-serde_json = "1.0.110"
+serde_json = "1.0.142"
 serde_with = { version = "3", features = ["macros"] }
 
 [dev-dependencies]

codex-rs/file-search/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-file-search"
 version = { workspace = true }
-edition = "2024"
 
 [[bin]]
 name = "codex-file-search"
@@ -17,5 +17,5 @@ clap = { version = "4", features = ["derive"] }
 ignore = "0.4.23"
 nucleo-matcher = "0.3.1"
 serde = { version = "1", features = ["derive"] }
-serde_json = "1.0.110"
+serde_json = "1.0.142"
 tokio = { version = "1", features = ["full"] }

codex-rs/linux-sandbox/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-linux-sandbox"
 version = { workspace = true }
-edition = "2024"
 
 [[bin]]
 name = "codex-linux-sandbox"
@@ -19,8 +19,8 @@ anyhow = "1"
 clap = { version = "4", features = ["derive"] }
 codex-common = { path = "../common", features = ["cli"] }
 codex-core = { path = "../core" }
-libc = "0.2.172"
 landlock = "0.4.1"
+libc = "0.2.172"
 seccompiler = "0.5.0"
 
 [target.'cfg(target_os = "linux")'.dev-dependencies]

codex-rs/linux-sandbox/src/landlock.rs

@@ -36,7 +36,11 @@ pub(crate) fn apply_sandbox_policy_to_current_thread(
     }
 
     if !sandbox_policy.has_full_disk_write_access() {
-        let writable_roots = sandbox_policy.get_writable_roots_with_cwd(cwd);
+        let writable_roots = sandbox_policy
+            .get_writable_roots_with_cwd(cwd)
+            .into_iter()
+            .map(|writable_root| writable_root.root)
+            .collect();
         install_filesystem_landlock_rules_on_current_thread(writable_roots)?;
     }
 

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

@@ -49,6 +49,7 @@ async fn run_cmd(cmd: &[&str], writable_roots: &[PathBuf], timeout_ms: u64) {
     let sandbox_policy = SandboxPolicy::WorkspaceWrite {
         writable_roots: writable_roots.to_vec(),
         network_access: false,
+        include_default_writable_roots: true,
     };
     let sandbox_program = env!("CARGO_BIN_EXE_codex-linux-sandbox");
     let codex_linux_sandbox_exe = Some(PathBuf::from(sandbox_program));
@@ -59,6 +60,7 @@ async fn run_cmd(cmd: &[&str], writable_roots: &[PathBuf], timeout_ms: u64) {
         ctrl_c,
         &sandbox_policy,
         &codex_linux_sandbox_exe,
+        None,
     )
     .await
     .unwrap();
@@ -149,6 +151,7 @@ async fn assert_network_blocked(cmd: &[&str]) {
         ctrl_c,
         &sandbox_policy,
         &codex_linux_sandbox_exe,
+        None,
     )
     .await;
 

codex-rs/login/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-login"
 version = { workspace = true }
-edition = "2024"
 
 [lints]
 workspace = true
@@ -18,3 +18,6 @@ tokio = { version = "1", features = [
     "rt-multi-thread",
     "signal",
 ] }
+
+[dev-dependencies]
+tempfile = "3"

codex-rs/login/src/lib.rs

@@ -20,7 +20,7 @@ use tokio::process::Command;
 const SOURCE_FOR_PYTHON_SERVER: &str = include_str!("./login_with_chatgpt.py");
 
 const CLIENT_ID: &str = "app_EMoamEEZ73f0CkXaXp7hrann";
-const OPENAI_API_KEY_ENV_VAR: &str = "OPENAI_API_KEY";
+pub const OPENAI_API_KEY_ENV_VAR: &str = "OPENAI_API_KEY";
 
 #[derive(Clone, Debug, PartialEq)]
 pub enum AuthMode {
@@ -70,13 +70,16 @@ impl CodexAuth {
     pub async fn get_token_data(&self) -> Result<TokenData, std::io::Error> {
         #[expect(clippy::unwrap_used)]
         let auth_dot_json = self.auth_dot_json.lock().unwrap().clone();
-
         match auth_dot_json {
-            Some(auth_dot_json) => {
-                if auth_dot_json.last_refresh < Utc::now() - chrono::Duration::days(28) {
+            Some(AuthDotJson {
+                tokens: Some(mut tokens),
+                last_refresh: Some(last_refresh),
+                ..
+            }) => {
+                if last_refresh < Utc::now() - chrono::Duration::days(28) {
                     let refresh_response = tokio::time::timeout(
                         Duration::from_secs(60),
-                        try_refresh_token(auth_dot_json.tokens.refresh_token.clone()),
+                        try_refresh_token(tokens.refresh_token.clone()),
                     )
                     .await
                     .map_err(|_| {
@@ -92,13 +95,21 @@ impl CodexAuth {
                     )
                     .await?;
 
+                    tokens = updated_auth_dot_json
+                        .tokens
+                        .clone()
+                        .ok_or(std::io::Error::other(
+                            "Token data is not available after refresh.",
+                        ))?;
+
                     #[expect(clippy::unwrap_used)]
-                    let mut auth_dot_json = self.auth_dot_json.lock().unwrap();
-                    *auth_dot_json = Some(updated_auth_dot_json);
+                    let mut auth_lock = self.auth_dot_json.lock().unwrap();
+                    *auth_lock = Some(updated_auth_dot_json);
                 }
-                Ok(auth_dot_json.tokens.clone())
+
+                Ok(tokens)
             }
-            None => Err(std::io::Error::other("Token data is not available.")),
+            _ => Err(std::io::Error::other("Token data is not available.")),
         }
     }
 
@@ -112,11 +123,22 @@ impl CodexAuth {
             }
         }
     }
+
+    pub async fn get_account_id(&self) -> Option<String> {
+        match self.mode {
+            AuthMode::ApiKey => None,
+            AuthMode::ChatGPT => {
+                let token_data = self.get_token_data().await.ok()?;
+
+                token_data.account_id.clone()
+            }
+        }
+    }
 }
 
 // Loads the available auth information from the auth.json or OPENAI_API_KEY environment variable.
-pub fn load_auth(codex_home: &Path) -> std::io::Result<Option<CodexAuth>> {
-    let auth_file = codex_home.join("auth.json");
+pub fn load_auth(codex_home: &Path, include_env_var: bool) -> std::io::Result<Option<CodexAuth>> {
+    let auth_file = get_auth_file(codex_home);
 
     let auth_dot_json = try_read_auth_json(&auth_file).ok();
 
@@ -125,12 +147,21 @@ pub fn load_auth(codex_home: &Path) -> std::io::Result<Option<CodexAuth>> {
         .and_then(|a| a.openai_api_key.clone())
         .filter(|s| !s.is_empty());
 
-    let openai_api_key = env::var(OPENAI_API_KEY_ENV_VAR)
-        .ok()
-        .filter(|s| !s.is_empty())
-        .or(auth_json_api_key);
+    let openai_api_key = if include_env_var {
+        env::var(OPENAI_API_KEY_ENV_VAR)
+            .ok()
+            .filter(|s| !s.is_empty())
+            .or(auth_json_api_key)
+    } else {
+        auth_json_api_key
+    };
 
-    if openai_api_key.is_none() && auth_dot_json.is_none() {
+    let has_tokens = auth_dot_json
+        .as_ref()
+        .and_then(|a| a.tokens.as_ref())
+        .is_some();
+
+    if openai_api_key.is_none() && !has_tokens {
         return Ok(None);
     }
 
@@ -148,6 +179,10 @@ pub fn load_auth(codex_home: &Path) -> std::io::Result<Option<CodexAuth>> {
     }))
 }
 
+fn get_auth_file(codex_home: &Path) -> PathBuf {
+    codex_home.join("auth.json")
+}
+
 /// Run `python3 -c {{SOURCE_FOR_PYTHON_SERVER}}` with the CODEX_HOME
 /// environment variable set to the provided `codex_home` path. If the
 /// subprocess exits 0, read the OPENAI_API_KEY property out of
@@ -187,6 +222,15 @@ pub async fn login_with_chatgpt(codex_home: &Path, capture_output: bool) -> std:
     }
 }
 
+pub fn login_with_api_key(codex_home: &Path, api_key: &str) -> std::io::Result<()> {
+    let auth_dot_json = AuthDotJson {
+        openai_api_key: Some(api_key.to_string()),
+        tokens: None,
+        last_refresh: None,
+    };
+    write_auth_json(&get_auth_file(codex_home), &auth_dot_json)
+}
+
 /// Attempt to read and refresh the `auth.json` file in the given `CODEX_HOME` directory.
 /// Returns the full AuthDotJson structure after refreshing if necessary.
 pub fn try_read_auth_json(auth_file: &Path) -> std::io::Result<AuthDotJson> {
@@ -198,35 +242,38 @@ pub fn try_read_auth_json(auth_file: &Path) -> std::io::Result<AuthDotJson> {
     Ok(auth_dot_json)
 }
 
-async fn update_tokens(
-    auth_file: &Path,
-    id_token: String,
-    access_token: Option<String>,
-    refresh_token: Option<String>,
-) -> std::io::Result<AuthDotJson> {
+fn write_auth_json(auth_file: &Path, auth_dot_json: &AuthDotJson) -> std::io::Result<()> {
+    let json_data = serde_json::to_string_pretty(auth_dot_json)?;
     let mut options = OpenOptions::new();
     options.truncate(true).write(true).create(true);
     #[cfg(unix)]
     {
         options.mode(0o600);
     }
+    let mut file = options.open(auth_file)?;
+    file.write_all(json_data.as_bytes())?;
+    file.flush()?;
+    Ok(())
+}
+
+async fn update_tokens(
+    auth_file: &Path,
+    id_token: String,
+    access_token: Option<String>,
+    refresh_token: Option<String>,
+) -> std::io::Result<AuthDotJson> {
     let mut auth_dot_json = try_read_auth_json(auth_file)?;
 
-    auth_dot_json.tokens.id_token = id_token.to_string();
+    let tokens = auth_dot_json.tokens.get_or_insert_with(TokenData::default);
+    tokens.id_token = id_token.to_string();
     if let Some(access_token) = access_token {
-        auth_dot_json.tokens.access_token = access_token.to_string();
+        tokens.access_token = access_token.to_string();
     }
     if let Some(refresh_token) = refresh_token {
-        auth_dot_json.tokens.refresh_token = refresh_token.to_string();
-    }
-    auth_dot_json.last_refresh = Utc::now();
-
-    let json_data = serde_json::to_string_pretty(&auth_dot_json)?;
-    {
-        let mut file = options.open(auth_file)?;
-        file.write_all(json_data.as_bytes())?;
-        file.flush()?;
+        tokens.refresh_token = refresh_token.to_string();
     }
+    auth_dot_json.last_refresh = Some(Utc::now());
+    write_auth_json(auth_file, &auth_dot_json)?;
     Ok(auth_dot_json)
 }
 
@@ -282,12 +329,14 @@ pub struct AuthDotJson {
     #[serde(rename = "OPENAI_API_KEY")]
     pub openai_api_key: Option<String>,
 
-    pub tokens: TokenData,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub tokens: Option<TokenData>,
 
-    pub last_refresh: DateTime<Utc>,
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub last_refresh: Option<DateTime<Utc>>,
 }
 
-#[derive(Deserialize, Serialize, Clone, Debug, PartialEq)]
+#[derive(Deserialize, Serialize, Clone, Debug, PartialEq, Default)]
 pub struct TokenData {
     /// This is a JWT.
     pub id_token: String,
@@ -299,3 +348,95 @@ pub struct TokenData {
 
     pub account_id: Option<String>,
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use tempfile::tempdir;
+
+    #[test]
+    #[expect(clippy::unwrap_used)]
+    fn writes_api_key_and_loads_auth() {
+        let dir = tempdir().unwrap();
+        login_with_api_key(dir.path(), "sk-test-key").unwrap();
+        let auth = load_auth(dir.path(), false).unwrap().unwrap();
+        assert_eq!(auth.mode, AuthMode::ApiKey);
+        assert_eq!(auth.api_key.as_deref(), Some("sk-test-key"));
+    }
+
+    #[test]
+    #[expect(clippy::unwrap_used)]
+    fn loads_from_env_var_if_env_var_exists() {
+        let dir = tempdir().unwrap();
+
+        let env_var = std::env::var(OPENAI_API_KEY_ENV_VAR);
+
+        if let Ok(env_var) = env_var {
+            let auth = load_auth(dir.path(), true).unwrap().unwrap();
+            assert_eq!(auth.mode, AuthMode::ApiKey);
+            assert_eq!(auth.api_key, Some(env_var));
+        }
+    }
+
+    #[tokio::test]
+    #[expect(clippy::unwrap_used)]
+    async fn loads_token_data_from_auth_json() {
+        let dir = tempdir().unwrap();
+        let auth_file = dir.path().join("auth.json");
+        std::fs::write(
+            auth_file,
+            format!(
+                r#"
+        {{
+            "OPENAI_API_KEY": null,
+            "tokens": {{
+                "id_token": "test-id-token",
+                "access_token": "test-access-token",
+                "refresh_token": "test-refresh-token"
+            }},
+            "last_refresh": "{}"
+        }}
+        "#,
+                Utc::now().to_rfc3339()
+            ),
+        )
+        .unwrap();
+
+        let auth = load_auth(dir.path(), false).unwrap().unwrap();
+        assert_eq!(auth.mode, AuthMode::ChatGPT);
+        assert_eq!(auth.api_key, None);
+        assert_eq!(
+            auth.get_token_data().await.unwrap(),
+            TokenData {
+                id_token: "test-id-token".to_string(),
+                access_token: "test-access-token".to_string(),
+                refresh_token: "test-refresh-token".to_string(),
+                account_id: None,
+            }
+        );
+    }
+
+    #[tokio::test]
+    #[expect(clippy::unwrap_used)]
+    async fn loads_api_key_from_auth_json() {
+        let dir = tempdir().unwrap();
+        let auth_file = dir.path().join("auth.json");
+        std::fs::write(
+            auth_file,
+            r#"
+        {
+            "OPENAI_API_KEY": "sk-test-key",
+            "tokens": null,
+            "last_refresh": null
+        }
+        "#,
+        )
+        .unwrap();
+
+        let auth = load_auth(dir.path(), false).unwrap().unwrap();
+        assert_eq!(auth.mode, AuthMode::ApiKey);
+        assert_eq!(auth.api_key, Some("sk-test-key".to_string()));
+
+        assert!(auth.get_token_data().await.is_err());
+    }
+}

codex-rs/login/src/login_with_chatgpt.py

@@ -686,6 +686,7 @@ LOGIN_SUCCESS_HTML = """<!DOCTYPE html>
         justify-content: center;
         position: relative;
         background: white;
+
         font-family: system-ui, -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, Oxygen, Ubuntu, Cantarell, 'Open Sans', 'Helvetica Neue', sans-serif;
       }
       .inner-container {
@@ -703,6 +704,7 @@ LOGIN_SUCCESS_HTML = """<!DOCTYPE html>
         align-items: center;
         gap: 20px;
         display: flex;
+        margin-top: 15vh;
       }
       .svg-wrapper {
         position: relative;
@@ -710,9 +712,9 @@ LOGIN_SUCCESS_HTML = """<!DOCTYPE html>
       .title {
         text-align: center;
         color: var(--text-primary, #0D0D0D);
-        font-size: 28px;
+        font-size: 32px;
         font-weight: 400;
-        line-height: 36.40px;
+        line-height: 40px;
         word-wrap: break-word;
       }
       .setup-box {
@@ -785,16 +787,26 @@ LOGIN_SUCCESS_HTML = """<!DOCTYPE html>
         word-wrap: break-word;
         text-decoration: none;
       }
+      .logo {
+        display: flex;
+        align-items: center;
+        justify-content: center;
+        width: 4rem;
+        height: 4rem;
+        border-radius: 16px;
+        border: .5px solid rgba(0, 0, 0, 0.1);
+        box-shadow: rgba(0, 0, 0, 0.1) 0px 4px 16px 0px;
+        box-sizing: border-box;
+        background-color: rgb(255, 255, 255);
+      }
     </style>
   </head>
   <body>
     <div class="container">
       <div class="inner-container">
         <div class="content">
-          <div data-svg-wrapper class="svg-wrapper">
-            <svg width="56" height="56" viewBox="0 0 56 56" fill="none" xmlns="http://www.w3.org/2000/svg">
-              <path d="M4.6665 28.0003C4.6665 15.1137 15.1132 4.66699 27.9998 4.66699C40.8865 4.66699 51.3332 15.1137 51.3332 28.0003C51.3332 40.887 40.8865 51.3337 27.9998 51.3337C15.1132 51.3337 4.6665 40.887 4.6665 28.0003ZM37.5093 18.5088C36.4554 17.7672 34.9999 18.0203 34.2583 19.0742L24.8508 32.4427L20.9764 28.1808C20.1095 27.2272 18.6338 27.1569 17.6803 28.0238C16.7267 28.8906 16.6565 30.3664 17.5233 31.3199L23.3566 37.7366C23.833 38.2606 24.5216 38.5399 25.2284 38.4958C25.9353 38.4517 26.5838 38.089 26.9914 37.5098L38.0747 21.7598C38.8163 20.7059 38.5632 19.2504 37.5093 18.5088Z" fill="var(--green-400, #04B84C)"/>
-            </svg>
+          <div class="logo">
+            <svg xmlns="http://www.w3.org/2000/svg" width="32" height="32" fill="none" viewBox="0 0 32 32"><path stroke="#000" stroke-linecap="round" stroke-width="2.484" d="M22.356 19.797H17.17M9.662 12.29l1.979 3.576a.511.511 0 0 1-.005.504l-1.974 3.409M30.758 16c0 8.15-6.607 14.758-14.758 14.758-8.15 0-14.758-6.607-14.758-14.758C1.242 7.85 7.85 1.242 16 1.242c8.15 0 14.758 6.608 14.758 14.758Z"></path></svg>
           </div>
           <div class="title">Signed in to Codex CLI</div>
         </div>

codex-rs/mcp-server/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-mcp-server"
 version = { workspace = true }
-edition = "2024"
 
 [[bin]]
 name = "codex-mcp-server"
@@ -23,9 +23,7 @@ schemars = "0.8.22"
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 shlex = "1.3.0"
-toml = "0.9"
-tracing = { version = "0.1.41", features = ["log"] }
-tracing-subscriber = { version = "0.3", features = ["fmt", "env-filter"] }
+strum_macros = "0.27.2"
 tokio = { version = "1", features = [
     "io-std",
     "macros",
@@ -33,8 +31,10 @@ tokio = { version = "1", features = [
     "rt-multi-thread",
     "signal",
 ] }
+toml = "0.9"
+tracing = { version = "0.1.41", features = ["log"] }
+tracing-subscriber = { version = "0.3", features = ["env-filter", "fmt"] }
 uuid = { version = "1", features = ["serde", "v4"] }
-strum_macros = "0.27.2"
 
 [dev-dependencies]
 assert_cmd = "2"

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

@@ -258,10 +258,12 @@ async fn run_codex_tool_session_inner(
                     | EventMsg::McpToolCallBegin(_)
                     | EventMsg::McpToolCallEnd(_)
                     | EventMsg::ExecCommandBegin(_)
+                    | EventMsg::ExecCommandOutputDelta(_)
                     | EventMsg::ExecCommandEnd(_)
                     | EventMsg::BackgroundEvent(_)
                     | EventMsg::PatchApplyBegin(_)
                     | EventMsg::PatchApplyEnd(_)
+                    | EventMsg::TurnDiff(_)
                     | EventMsg::GetHistoryEntryResponse(_)
                     | EventMsg::PlanUpdate(_)
                     | EventMsg::ShutdownComplete => {

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

@@ -0,0 +1,122 @@
+use std::sync::Arc;
+
+use crate::exec_approval::handle_exec_approval_request;
+use crate::outgoing_message::OutgoingMessageSender;
+use crate::outgoing_message::OutgoingNotificationMeta;
+use crate::patch_approval::handle_patch_approval_request;
+use codex_core::Codex;
+use codex_core::protocol::AgentMessageEvent;
+use codex_core::protocol::ApplyPatchApprovalRequestEvent;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::ExecApprovalRequestEvent;
+use mcp_types::RequestId;
+use tracing::error;
+
+pub async fn run_conversation_loop(
+    codex: Arc<Codex>,
+    outgoing: Arc<OutgoingMessageSender>,
+    request_id: RequestId,
+) {
+    let request_id_str = match &request_id {
+        RequestId::String(s) => s.clone(),
+        RequestId::Integer(n) => n.to_string(),
+    };
+
+    // Stream events until the task needs to pause for user interaction or
+    // completes.
+    loop {
+        match codex.next_event().await {
+            Ok(event) => {
+                outgoing
+                    .send_event_as_notification(
+                        &event,
+                        Some(OutgoingNotificationMeta::new(Some(request_id.clone()))),
+                    )
+                    .await;
+
+                match event.msg {
+                    EventMsg::ExecApprovalRequest(ExecApprovalRequestEvent {
+                        command,
+                        cwd,
+                        call_id,
+                        reason: _,
+                    }) => {
+                        handle_exec_approval_request(
+                            command,
+                            cwd,
+                            outgoing.clone(),
+                            codex.clone(),
+                            request_id.clone(),
+                            request_id_str.clone(),
+                            event.id.clone(),
+                            call_id,
+                        )
+                        .await;
+                        continue;
+                    }
+                    EventMsg::Error(_) => {
+                        error!("Codex runtime error");
+                    }
+                    EventMsg::ApplyPatchApprovalRequest(ApplyPatchApprovalRequestEvent {
+                        call_id,
+                        reason,
+                        grant_root,
+                        changes,
+                    }) => {
+                        handle_patch_approval_request(
+                            call_id,
+                            reason,
+                            grant_root,
+                            changes,
+                            outgoing.clone(),
+                            codex.clone(),
+                            request_id.clone(),
+                            request_id_str.clone(),
+                            event.id.clone(),
+                        )
+                        .await;
+                        continue;
+                    }
+                    EventMsg::TaskComplete(_) => {}
+                    EventMsg::SessionConfigured(_) => {
+                        tracing::error!("unexpected SessionConfigured event");
+                    }
+                    EventMsg::AgentMessageDelta(_) => {
+                        // TODO: think how we want to support this in the MCP
+                    }
+                    EventMsg::AgentReasoningDelta(_) => {
+                        // TODO: think how we want to support this in the MCP
+                    }
+                    EventMsg::AgentMessage(AgentMessageEvent { .. }) => {
+                        // TODO: think how we want to support this in the MCP
+                    }
+                    EventMsg::TaskStarted
+                    | EventMsg::TokenCount(_)
+                    | EventMsg::AgentReasoning(_)
+                    | EventMsg::McpToolCallBegin(_)
+                    | EventMsg::McpToolCallEnd(_)
+                    | EventMsg::ExecCommandBegin(_)
+                    | EventMsg::ExecCommandEnd(_)
+                    | EventMsg::TurnDiff(_)
+                    | EventMsg::BackgroundEvent(_)
+                    | EventMsg::ExecCommandOutputDelta(_)
+                    | EventMsg::PatchApplyBegin(_)
+                    | EventMsg::PatchApplyEnd(_)
+                    | EventMsg::GetHistoryEntryResponse(_)
+                    | EventMsg::PlanUpdate(_)
+                    | EventMsg::ShutdownComplete => {
+                        // For now, we do not do anything extra for these
+                        // events. Note that
+                        // send(codex_event_to_notification(&event)) above has
+                        // already dispatched these events as notifications,
+                        // though we may want to do give different treatment to
+                        // individual events in the future.
+                    }
+                }
+            }
+            Err(e) => {
+                error!("Codex runtime error: {e}");
+            }
+        }
+    }
+}

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

@@ -18,7 +18,7 @@ use crate::codex_tool_runner::INVALID_PARAMS_ERROR_CODE;
 
 /// Conforms to [`mcp_types::ElicitRequestParams`] so that it can be used as the
 /// `params` field of an [`ElicitRequest`].
-#[derive(Debug, Serialize)]
+#[derive(Debug, Deserialize, Serialize)]
 pub struct ExecApprovalElicitRequestParams {
     // These fields are required so that `params`
     // conforms to ElicitRequestParams.

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

@@ -17,12 +17,14 @@ use tracing_subscriber::EnvFilter;
 
 mod codex_tool_config;
 mod codex_tool_runner;
+mod conversation_loop;
 mod exec_approval;
 mod json_to_toml;
-mod mcp_protocol;
-mod message_processor;
+pub mod mcp_protocol;
+pub(crate) mod message_processor;
 mod outgoing_message;
 mod patch_approval;
+pub(crate) mod tool_handlers;
 
 use crate::message_processor::MessageProcessor;
 use crate::outgoing_message::OutgoingMessage;

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

@@ -7,7 +7,10 @@ use serde::Serialize;
 use strum_macros::Display;
 use uuid::Uuid;
 
+use mcp_types::CallToolResult;
+use mcp_types::ContentBlock;
 use mcp_types::RequestId;
+use mcp_types::TextContent;
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
 #[serde(transparent)]
@@ -118,10 +121,47 @@ pub struct ToolCallResponse {
     pub request_id: RequestId,
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub is_error: Option<bool>,
-    #[serde(default, skip_serializing_if = "Option::is_none")]
+    #[serde(default, skip_serializing_if = "Option::is_none", flatten)]
     pub result: Option<ToolCallResponseResult>,
 }
 
+impl From<ToolCallResponse> for CallToolResult {
+    fn from(val: ToolCallResponse) -> Self {
+        let ToolCallResponse {
+            request_id: _request_id,
+            is_error,
+            result,
+        } = val;
+        match result {
+            Some(res) => match serde_json::to_value(&res) {
+                Ok(v) => CallToolResult {
+                    content: vec![ContentBlock::TextContent(TextContent {
+                        r#type: "text".to_string(),
+                        text: v.to_string(),
+                        annotations: None,
+                    })],
+                    is_error,
+                    structured_content: Some(v),
+                },
+                Err(e) => CallToolResult {
+                    content: vec![ContentBlock::TextContent(TextContent {
+                        r#type: "text".to_string(),
+                        text: format!("Failed to serialize tool result: {e}"),
+                        annotations: None,
+                    })],
+                    is_error: Some(true),
+                    structured_content: None,
+                },
+            },
+            None => CallToolResult {
+                content: vec![],
+                is_error,
+                structured_content: None,
+            },
+        }
+    }
+}
+
 #[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
 #[serde(untagged)]
 pub enum ToolCallResponseResult {
@@ -132,17 +172,26 @@ pub enum ToolCallResponseResult {
 }
 
 #[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
-pub struct ConversationCreateResult {
-    pub conversation_id: ConversationId,
-    pub model: String,
+#[serde(untagged)]
+pub enum ConversationCreateResult {
+    Ok {
+        conversation_id: ConversationId,
+        model: String,
+    },
+    Error {
+        message: String,
+    },
 }
 
 #[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
 pub struct ConversationStreamResult {}
 
 #[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
-pub struct ConversationSendMessageResult {
-    pub success: bool,
+// TODO: remove this status because we have is_error field in the response.
+#[serde(tag = "status", rename_all = "camelCase")]
+pub enum ConversationSendMessageResult {
+    Ok,
+    Error { message: String },
 }
 
 #[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
@@ -449,16 +498,19 @@ mod tests {
             request_id: RequestId::Integer(1),
             is_error: None,
             result: Some(ToolCallResponseResult::ConversationCreate(
-                ConversationCreateResult {
+                ConversationCreateResult::Ok {
                     conversation_id: ConversationId(uuid!("d0f6ecbe-84a2-41c1-b23d-b20473b25eab")),
                     model: "o3".into(),
                 },
             )),
         };
-        let observed = to_val(&env);
+        let req_id = env.request_id.clone();
+        let observed = to_val(&CallToolResult::from(env));
         let expected = json!({
-            "requestId": 1,
-            "result": {
+            "content": [
+                { "type": "text", "text": "{\"conversation_id\":\"d0f6ecbe-84a2-41c1-b23d-b20473b25eab\",\"model\":\"o3\"}" }
+            ],
+            "structuredContent": {
                 "conversation_id": "d0f6ecbe-84a2-41c1-b23d-b20473b25eab",
                 "model": "o3"
             }
@@ -467,6 +519,36 @@ mod tests {
             observed, expected,
             "response (ConversationCreate) must match"
         );
+        assert_eq!(req_id, RequestId::Integer(1));
+    }
+
+    #[test]
+    fn response_error_conversation_create_full_schema() {
+        let env = ToolCallResponse {
+            request_id: RequestId::Integer(2),
+            is_error: Some(true),
+            result: Some(ToolCallResponseResult::ConversationCreate(
+                ConversationCreateResult::Error {
+                    message: "Failed to initialize session".into(),
+                },
+            )),
+        };
+        let req_id = env.request_id.clone();
+        let observed = to_val(&CallToolResult::from(env));
+        let expected = json!({
+            "content": [
+                { "type": "text", "text": "{\"message\":\"Failed to initialize session\"}" }
+            ],
+            "isError": true,
+            "structuredContent": {
+                "message": "Failed to initialize session"
+            }
+        });
+        assert_eq!(
+            observed, expected,
+            "error response (ConversationCreate) must match"
+        );
+        assert_eq!(req_id, RequestId::Integer(2));
     }
 
     #[test]
@@ -478,15 +560,17 @@ mod tests {
                 ConversationStreamResult {},
             )),
         };
-        let observed = to_val(&env);
+        let req_id = env.request_id.clone();
+        let observed = to_val(&CallToolResult::from(env));
         let expected = json!({
-            "requestId": 2,
-            "result": {}
+            "content": [ { "type": "text", "text": "{}" } ],
+            "structuredContent": {}
         });
         assert_eq!(
             observed, expected,
             "response (ConversationStream) must have empty object result"
         );
+        assert_eq!(req_id, RequestId::Integer(2));
     }
 
     #[test]
@@ -495,18 +579,20 @@ mod tests {
             request_id: RequestId::Integer(3),
             is_error: None,
             result: Some(ToolCallResponseResult::ConversationSendMessage(
-                ConversationSendMessageResult { success: true },
+                ConversationSendMessageResult::Ok,
             )),
         };
-        let observed = to_val(&env);
+        let req_id = env.request_id.clone();
+        let observed = to_val(&CallToolResult::from(env));
         let expected = json!({
-            "requestId": 3,
-            "result": { "success": true }
+            "content": [ { "type": "text", "text": "{\"status\":\"ok\"}" } ],
+            "structuredContent": { "status": "ok" }
         });
         assert_eq!(
             observed, expected,
             "response (ConversationSendMessageAccepted) must match"
         );
+        assert_eq!(req_id, RequestId::Integer(3));
     }
 
     #[test]
@@ -526,10 +612,13 @@ mod tests {
                 },
             )),
         };
-        let observed = to_val(&env);
+        let req_id = env.request_id.clone();
+        let observed = to_val(&CallToolResult::from(env));
         let expected = json!({
-            "requestId": 4,
-            "result": {
+            "content": [
+                { "type": "text", "text": "{\"conversations\":[{\"conversation_id\":\"67e55044-10b1-426f-9247-bb680e5fe0c8\",\"title\":\"Refactor config loader\"}],\"next_cursor\":\"next123\"}" }
+            ],
+            "structuredContent": {
                 "conversations": [
                     {
                         "conversation_id": "67e55044-10b1-426f-9247-bb680e5fe0c8",
@@ -543,6 +632,7 @@ mod tests {
             observed, expected,
             "response (ConversationsList with cursor) must match"
         );
+        assert_eq!(req_id, RequestId::Integer(4));
     }
 
     #[test]
@@ -552,15 +642,17 @@ mod tests {
             is_error: Some(true),
             result: None,
         };
-        let observed = to_val(&env);
+        let req_id = env.request_id.clone();
+        let observed = to_val(&CallToolResult::from(env));
         let expected = json!({
-            "requestId": 4,
+            "content": [],
             "isError": true
         });
         assert_eq!(
             observed, expected,
             "error response must omit `result` and include `isError`"
         );
+        assert_eq!(req_id, RequestId::Integer(4));
     }
 
     // ----- Notifications -----

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

@@ -1,4 +1,5 @@
 use std::collections::HashMap;
+use std::collections::HashSet;
 use std::path::PathBuf;
 use std::sync::Arc;
 
@@ -6,11 +7,17 @@ use crate::codex_tool_config::CodexToolCallParam;
 use crate::codex_tool_config::CodexToolCallReplyParam;
 use crate::codex_tool_config::create_tool_for_codex_tool_call_param;
 use crate::codex_tool_config::create_tool_for_codex_tool_call_reply_param;
+use crate::mcp_protocol::ToolCallRequestParams;
+use crate::mcp_protocol::ToolCallResponse;
+use crate::mcp_protocol::ToolCallResponseResult;
 use crate::outgoing_message::OutgoingMessageSender;
+use crate::tool_handlers::create_conversation::handle_create_conversation;
+use crate::tool_handlers::send_message::handle_send_message;
 
 use codex_core::Codex;
 use codex_core::config::Config as CodexConfig;
 use codex_core::protocol::Submission;
+use mcp_types::CallToolRequest;
 use mcp_types::CallToolRequestParams;
 use mcp_types::CallToolResult;
 use mcp_types::ClientRequest;
@@ -37,6 +44,7 @@ pub(crate) struct MessageProcessor {
     codex_linux_sandbox_exe: Option<PathBuf>,
     session_map: Arc<Mutex<HashMap<Uuid, Arc<Codex>>>>,
     running_requests_id_to_codex_uuid: Arc<Mutex<HashMap<RequestId, Uuid>>>,
+    running_session_ids: Arc<Mutex<HashSet<Uuid>>>,
 }
 
 impl MessageProcessor {
@@ -52,9 +60,22 @@ impl MessageProcessor {
             codex_linux_sandbox_exe,
             session_map: Arc::new(Mutex::new(HashMap::new())),
             running_requests_id_to_codex_uuid: Arc::new(Mutex::new(HashMap::new())),
+            running_session_ids: Arc::new(Mutex::new(HashSet::new())),
         }
     }
 
+    pub(crate) fn session_map(&self) -> Arc<Mutex<HashMap<Uuid, Arc<Codex>>>> {
+        self.session_map.clone()
+    }
+
+    pub(crate) fn outgoing(&self) -> Arc<OutgoingMessageSender> {
+        self.outgoing.clone()
+    }
+
+    pub(crate) fn running_session_ids(&self) -> Arc<Mutex<HashSet<Uuid>>> {
+        self.running_session_ids.clone()
+    }
+
     pub(crate) async fn process_request(&mut self, request: JSONRPCRequest) {
         // Hold on to the ID so we can respond.
         let request_id = request.id.clone();
@@ -300,6 +321,14 @@ impl MessageProcessor {
         params: <mcp_types::CallToolRequest as mcp_types::ModelContextProtocolRequest>::Params,
     ) {
         tracing::info!("tools/call -> params: {:?}", params);
+        // Serialize params into JSON and try to parse as new type
+        if let Ok(new_params) =
+            serde_json::to_value(&params).and_then(serde_json::from_value::<ToolCallRequestParams>)
+        {
+            // New tool call matched → forward
+            self.handle_new_tool_calls(id, new_params).await;
+            return;
+        }
         let CallToolRequestParams { name, arguments } = params;
 
         match name.as_str() {
@@ -323,6 +352,29 @@ impl MessageProcessor {
             }
         }
     }
+    async fn handle_new_tool_calls(&self, request_id: RequestId, params: ToolCallRequestParams) {
+        match params {
+            ToolCallRequestParams::ConversationCreate(args) => {
+                handle_create_conversation(self, request_id, args).await;
+            }
+            ToolCallRequestParams::ConversationSendMessage(args) => {
+                handle_send_message(self, request_id, args).await;
+            }
+            _ => {
+                let result = CallToolResult {
+                    content: vec![ContentBlock::TextContent(TextContent {
+                        r#type: "text".to_string(),
+                        text: "Unknown tool".to_string(),
+                        annotations: None,
+                    })],
+                    is_error: Some(true),
+                    structured_content: None,
+                };
+                self.send_response::<CallToolRequest>(request_id, result)
+                    .await;
+            }
+        }
+    }
 
     async fn handle_tool_call_codex(&self, id: RequestId, arguments: Option<serde_json::Value>) {
         let (initial_prompt, config): (String, CodexConfig) = match arguments {
@@ -631,4 +683,20 @@ impl MessageProcessor {
     ) {
         tracing::info!("notifications/message -> params: {:?}", params);
     }
+
+    pub(crate) async fn send_response_with_optional_error(
+        &self,
+        id: RequestId,
+        message: Option<ToolCallResponseResult>,
+        error: Option<bool>,
+    ) {
+        let response = ToolCallResponse {
+            request_id: id.clone(),
+            is_error: error,
+            result: message,
+        };
+        let result: CallToolResult = response.into();
+        self.send_response::<mcp_types::CallToolRequest>(id.clone(), result)
+            .await;
+    }
 }

codex-rs/mcp-server/src/tool_handlers/create_conversation.rs

@@ -0,0 +1,160 @@
+use std::collections::HashMap;
+use std::path::PathBuf;
+use std::sync::Arc;
+
+use codex_core::Codex;
+use codex_core::codex_wrapper::init_codex;
+use codex_core::config::Config as CodexConfig;
+use codex_core::config::ConfigOverrides;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::SessionConfiguredEvent;
+use mcp_types::RequestId;
+use tokio::sync::Mutex;
+use uuid::Uuid;
+
+use crate::conversation_loop::run_conversation_loop;
+use crate::json_to_toml::json_to_toml;
+use crate::mcp_protocol::ConversationCreateArgs;
+use crate::mcp_protocol::ConversationCreateResult;
+use crate::mcp_protocol::ConversationId;
+use crate::mcp_protocol::ToolCallResponseResult;
+use crate::message_processor::MessageProcessor;
+
+pub(crate) async fn handle_create_conversation(
+    message_processor: &MessageProcessor,
+    id: RequestId,
+    args: ConversationCreateArgs,
+) {
+    // Build ConfigOverrides from args
+    let ConversationCreateArgs {
+        prompt: _, // not used here; creation only establishes the session
+        model,
+        cwd,
+        approval_policy,
+        sandbox,
+        config,
+        profile,
+        base_instructions,
+    } = args;
+
+    // Convert config overrides JSON into CLI-style TOML overrides
+    let cli_overrides: Vec<(String, toml::Value)> = match config {
+        Some(v) => match v.as_object() {
+            Some(map) => map
+                .into_iter()
+                .map(|(k, v)| (k.clone(), json_to_toml(v.clone())))
+                .collect(),
+            None => Vec::new(),
+        },
+        None => Vec::new(),
+    };
+
+    let overrides = ConfigOverrides {
+        model: Some(model.clone()),
+        cwd: Some(PathBuf::from(cwd)),
+        approval_policy,
+        sandbox_mode: sandbox,
+        model_provider: None,
+        config_profile: profile,
+        codex_linux_sandbox_exe: None,
+        base_instructions,
+        include_plan_tool: None,
+    };
+
+    let cfg: CodexConfig = match CodexConfig::load_with_cli_overrides(cli_overrides, overrides) {
+        Ok(cfg) => cfg,
+        Err(e) => {
+            message_processor
+                .send_response_with_optional_error(
+                    id,
+                    Some(ToolCallResponseResult::ConversationCreate(
+                        ConversationCreateResult::Error {
+                            message: format!("Failed to load config: {e}"),
+                        },
+                    )),
+                    Some(true),
+                )
+                .await;
+            return;
+        }
+    };
+
+    // Initialize Codex session
+    let codex_conversation = match init_codex(cfg).await {
+        Ok(conv) => conv,
+        Err(e) => {
+            message_processor
+                .send_response_with_optional_error(
+                    id,
+                    Some(ToolCallResponseResult::ConversationCreate(
+                        ConversationCreateResult::Error {
+                            message: format!("Failed to initialize session: {e}"),
+                        },
+                    )),
+                    Some(true),
+                )
+                .await;
+            return;
+        }
+    };
+
+    // Expect SessionConfigured; if not, return error.
+    let EventMsg::SessionConfigured(SessionConfiguredEvent { model, .. }) =
+        &codex_conversation.session_configured.msg
+    else {
+        message_processor
+            .send_response_with_optional_error(
+                id,
+                Some(ToolCallResponseResult::ConversationCreate(
+                    ConversationCreateResult::Error {
+                        message: "Expected SessionConfigured event".to_string(),
+                    },
+                )),
+                Some(true),
+            )
+            .await;
+        return;
+    };
+
+    let effective_model = model.clone();
+
+    let session_id = codex_conversation.session_id;
+    let codex_arc = Arc::new(codex_conversation.codex);
+
+    // Store session for future calls
+    insert_session(
+        session_id,
+        codex_arc.clone(),
+        message_processor.session_map(),
+    )
+    .await;
+    // Run the conversation loop in the background so this request can return immediately.
+    let outgoing = message_processor.outgoing();
+    let spawn_id = id.clone();
+    tokio::spawn(async move {
+        run_conversation_loop(codex_arc.clone(), outgoing, spawn_id).await;
+    });
+
+    // Reply with the new conversation id and effective model
+    message_processor
+        .send_response_with_optional_error(
+            id,
+            Some(ToolCallResponseResult::ConversationCreate(
+                ConversationCreateResult::Ok {
+                    conversation_id: ConversationId(session_id),
+                    model: effective_model,
+                },
+            )),
+            Some(false),
+        )
+        .await;
+}
+
+async fn insert_session(
+    session_id: Uuid,
+    codex: Arc<Codex>,
+    session_map: Arc<Mutex<HashMap<Uuid, Arc<Codex>>>>,
+) {
+    let mut guard = session_map.lock().await;
+    guard.insert(session_id, codex);
+}

codex-rs/mcp-server/src/tool_handlers/mod.rs

@@ -0,0 +1,2 @@
+pub(crate) mod create_conversation;
+pub(crate) mod send_message;

codex-rs/mcp-server/src/tool_handlers/send_message.rs

@@ -0,0 +1,124 @@
+use std::collections::HashMap;
+use std::sync::Arc;
+
+use codex_core::Codex;
+use codex_core::protocol::Op;
+use codex_core::protocol::Submission;
+use mcp_types::RequestId;
+use tokio::sync::Mutex;
+use uuid::Uuid;
+
+use crate::mcp_protocol::ConversationSendMessageArgs;
+use crate::mcp_protocol::ConversationSendMessageResult;
+use crate::mcp_protocol::ToolCallResponseResult;
+use crate::message_processor::MessageProcessor;
+
+pub(crate) async fn handle_send_message(
+    message_processor: &MessageProcessor,
+    id: RequestId,
+    arguments: ConversationSendMessageArgs,
+) {
+    let ConversationSendMessageArgs {
+        conversation_id,
+        content: items,
+        parent_message_id: _,
+        conversation_overrides: _,
+    } = arguments;
+
+    if items.is_empty() {
+        message_processor
+            .send_response_with_optional_error(
+                id,
+                Some(ToolCallResponseResult::ConversationSendMessage(
+                    ConversationSendMessageResult::Error {
+                        message: "No content items provided".to_string(),
+                    },
+                )),
+                Some(true),
+            )
+            .await;
+        return;
+    }
+
+    let session_id = conversation_id.0;
+    let Some(codex) = get_session(session_id, message_processor.session_map()).await else {
+        message_processor
+            .send_response_with_optional_error(
+                id,
+                Some(ToolCallResponseResult::ConversationSendMessage(
+                    ConversationSendMessageResult::Error {
+                        message: "Session does not exist".to_string(),
+                    },
+                )),
+                Some(true),
+            )
+            .await;
+        return;
+    };
+
+    let running = {
+        let running_sessions = message_processor.running_session_ids();
+        let mut running_sessions = running_sessions.lock().await;
+        !running_sessions.insert(session_id)
+    };
+
+    if running {
+        message_processor
+            .send_response_with_optional_error(
+                id,
+                Some(ToolCallResponseResult::ConversationSendMessage(
+                    ConversationSendMessageResult::Error {
+                        message: "Session is already running".to_string(),
+                    },
+                )),
+                Some(true),
+            )
+            .await;
+        return;
+    }
+
+    let request_id_string = match &id {
+        RequestId::String(s) => s.clone(),
+        RequestId::Integer(i) => i.to_string(),
+    };
+
+    let submit_res = codex
+        .submit_with_id(Submission {
+            id: request_id_string,
+            op: Op::UserInput { items },
+        })
+        .await;
+
+    if let Err(e) = submit_res {
+        message_processor
+            .send_response_with_optional_error(
+                id,
+                Some(ToolCallResponseResult::ConversationSendMessage(
+                    ConversationSendMessageResult::Error {
+                        message: format!("Failed to submit user input: {e}"),
+                    },
+                )),
+                Some(true),
+            )
+            .await;
+        return;
+    }
+
+    message_processor
+        .send_response_with_optional_error(
+            id,
+            Some(ToolCallResponseResult::ConversationSendMessage(
+                ConversationSendMessageResult::Ok,
+            )),
+            Some(false),
+        )
+        .await;
+}
+
+pub(crate) async fn get_session(
+    session_id: Uuid,
+    session_map: Arc<Mutex<HashMap<Uuid, Arc<Codex>>>>,
+) -> Option<Arc<Codex>> {
+    let guard = session_map.lock().await;
+    guard.get(&session_id).cloned()
+}

codex-rs/mcp-server/tests/codex_tool.rs

@@ -3,9 +3,9 @@ use std::env;
 use std::path::Path;
 use std::path::PathBuf;
 
-use codex_core::exec::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_core::protocol::FileChange;
 use codex_core::protocol::ReviewDecision;
+use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_mcp_server::CodexToolCallParam;
 use codex_mcp_server::ExecApprovalElicitRequestParams;
 use codex_mcp_server::ExecApprovalResponse;
@@ -89,14 +89,18 @@ async fn shell_command_approval_triggers_elicitation() -> anyhow::Result<()> {
     // This is the first request from the server, so the id should be 0 given
     // how things are currently implemented.
     let elicitation_request_id = RequestId::Integer(0);
+    let params = serde_json::from_value::<ExecApprovalElicitRequestParams>(
+        elicitation_request
+            .params
+            .clone()
+            .ok_or_else(|| anyhow::anyhow!("elicitation_request.params must be set"))?,
+    )?;
     let expected_elicitation_request = create_expected_elicitation_request(
         elicitation_request_id.clone(),
         shell_command.clone(),
         workdir_for_shell_function_call.path(),
         codex_request_id.to_string(),
-        // Internal Codex id: empirically it is 1, but this is
-        // admittedly an internal detail that could change.
-        "1".to_string(),
+        params.codex_event_id.clone(),
     )?;
     assert_eq!(expected_elicitation_request, elicitation_request);
 

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

@@ -10,6 +10,7 @@ path = "lib.rs"
 anyhow = "1"
 assert_cmd = "2"
 codex-mcp-server = { path = "../.." }
+codex-core = { path = "../../../core" }
 mcp-types = { path = "../../../mcp-types" }
 pretty_assertions = "1.4.1"
 serde_json = "1"
@@ -22,3 +23,4 @@ tokio = { version = "1", features = [
     "rt-multi-thread",
 ] }
 wiremock = "0.6"
+uuid = { version = "1", features = ["serde", "v4"] }

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

@@ -11,8 +11,14 @@ use tokio::process::ChildStdout;
 
 use anyhow::Context;
 use assert_cmd::prelude::*;
+use codex_core::protocol::InputItem;
 use codex_mcp_server::CodexToolCallParam;
 use codex_mcp_server::CodexToolCallReplyParam;
+use codex_mcp_server::mcp_protocol::ConversationCreateArgs;
+use codex_mcp_server::mcp_protocol::ConversationId;
+use codex_mcp_server::mcp_protocol::ConversationSendMessageArgs;
+use codex_mcp_server::mcp_protocol::ToolCallRequestParams;
+
 use mcp_types::CallToolRequestParams;
 use mcp_types::ClientCapabilities;
 use mcp_types::Implementation;
@@ -29,6 +35,7 @@ use pretty_assertions::assert_eq;
 use serde_json::json;
 use std::process::Command as StdCommand;
 use tokio::process::Command;
+use uuid::Uuid;
 
 pub struct McpProcess {
     next_request_id: AtomicI64,
@@ -174,6 +181,61 @@ impl McpProcess {
         .await
     }
 
+    pub async fn send_user_message_tool_call(
+        &mut self,
+        message: &str,
+        session_id: &str,
+    ) -> anyhow::Result<i64> {
+        let params = ToolCallRequestParams::ConversationSendMessage(ConversationSendMessageArgs {
+            conversation_id: ConversationId(Uuid::parse_str(session_id)?),
+            content: vec![InputItem::Text {
+                text: message.to_string(),
+            }],
+            parent_message_id: None,
+            conversation_overrides: None,
+        });
+        self.send_request(
+            mcp_types::CallToolRequest::METHOD,
+            Some(serde_json::to_value(params)?),
+        )
+        .await
+    }
+
+    pub async fn send_conversation_create_tool_call(
+        &mut self,
+        prompt: &str,
+        model: &str,
+        cwd: &str,
+    ) -> anyhow::Result<i64> {
+        let params = ToolCallRequestParams::ConversationCreate(ConversationCreateArgs {
+            prompt: prompt.to_string(),
+            model: model.to_string(),
+            cwd: cwd.to_string(),
+            approval_policy: None,
+            sandbox: None,
+            config: None,
+            profile: None,
+            base_instructions: None,
+        });
+        self.send_request(
+            mcp_types::CallToolRequest::METHOD,
+            Some(serde_json::to_value(params)?),
+        )
+        .await
+    }
+
+    pub async fn send_conversation_create_with_args(
+        &mut self,
+        args: ConversationCreateArgs,
+    ) -> anyhow::Result<i64> {
+        let params = ToolCallRequestParams::ConversationCreate(args);
+        self.send_request(
+            mcp_types::CallToolRequest::METHOD,
+            Some(serde_json::to_value(params)?),
+        )
+        .await
+    }
+
     async fn send_request(
         &mut self,
         method: &str,

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

@@ -0,0 +1,128 @@
+#![allow(clippy::expect_used, clippy::unwrap_used)]
+
+use std::path::Path;
+
+use mcp_test_support::McpProcess;
+use mcp_test_support::create_final_assistant_message_sse_response;
+use mcp_test_support::create_mock_chat_completions_server;
+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: std::time::Duration = std::time::Duration::from_secs(10);
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn test_conversation_create_and_send_message_ok() {
+    // Mock server – we won't strictly rely on it, but provide one to satisfy any model wiring.
+    let responses = vec![
+        create_final_assistant_message_sse_response("Done").expect("build mock assistant message"),
+    ];
+    let server = create_mock_chat_completions_server(responses).await;
+
+    // Temporary Codex home with config pointing at the mock server.
+    let codex_home = TempDir::new().expect("create temp dir");
+    create_config_toml(codex_home.path(), &server.uri()).expect("write config.toml");
+
+    // Start MCP server process and initialize.
+    let mut mcp = McpProcess::new(codex_home.path())
+        .await
+        .expect("spawn mcp process");
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
+        .await
+        .expect("init timeout")
+        .expect("init failed");
+
+    // Create a conversation via the new tool.
+    let req_id = mcp
+        .send_conversation_create_tool_call("", "o3", "/repo")
+        .await
+        .expect("send conversationCreate");
+
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(req_id)),
+    )
+    .await
+    .expect("create response timeout")
+    .expect("create response error");
+
+    // Structured content must include status=ok, a UUID conversation_id and the model we passed.
+    let sc = &resp.result["structuredContent"];
+    let conv_id = sc["conversation_id"].as_str().expect("uuid string");
+    assert!(!conv_id.is_empty());
+    assert_eq!(sc["model"], json!("o3"));
+
+    // Now send a message to the created conversation and expect an OK result.
+    let send_id = mcp
+        .send_user_message_tool_call("Hello", conv_id)
+        .await
+        .expect("send message");
+
+    let send_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(send_id)),
+    )
+    .await
+    .expect("send response timeout")
+    .expect("send response error");
+    assert_eq!(
+        send_resp.result["structuredContent"],
+        json!({ "status": "ok" })
+    );
+
+    // avoid race condition by waiting for the mock server to receive the chat.completions request
+    let deadline = std::time::Instant::now() + DEFAULT_READ_TIMEOUT;
+    loop {
+        let requests = server.received_requests().await.unwrap_or_default();
+        if !requests.is_empty() {
+            break;
+        }
+        if std::time::Instant::now() >= deadline {
+            panic!("mock server did not receive the chat.completions request in time");
+        }
+        tokio::time::sleep(std::time::Duration::from_millis(10)).await;
+    }
+
+    // Verify the outbound request body matches expectations for Chat Completions.
+    let request = &server.received_requests().await.unwrap()[0];
+    let body = request
+        .body_json::<serde_json::Value>()
+        .expect("parse request body as JSON");
+    assert_eq!(body["model"], json!("o3"));
+    assert!(body["stream"].as_bool().unwrap_or(false));
+    let messages = body["messages"]
+        .as_array()
+        .expect("messages should be array");
+    let last = messages.last().expect("at least one message");
+    assert_eq!(last["role"], json!("user"));
+    assert_eq!(last["content"], json!("Hello"));
+
+    drop(server);
+}
+
+// Helper to create a config.toml pointing at the mock model server.
+fn create_config_toml(codex_home: &Path, server_uri: &str) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        format!(
+            r#"
+model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "danger-full-access"
+
+model_provider = "mock_provider"
+
+[model_providers.mock_provider]
+name = "Mock provider for test"
+base_url = "{server_uri}/v1"
+wire_api = "chat"
+request_max_retries = 0
+stream_max_retries = 0
+"#
+        ),
+    )
+}

codex-rs/mcp-server/tests/interrupt.rs

@@ -3,7 +3,7 @@
 
 use std::path::Path;
 
-use codex_core::exec::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
+use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_mcp_server::CodexToolCallParam;
 use mcp_types::JSONRPCResponse;
 use mcp_types::RequestId;

codex-rs/mcp-server/tests/send_message.rs

@@ -0,0 +1,163 @@
+#![allow(clippy::expect_used)]
+
+use std::path::Path;
+use std::thread::sleep;
+use std::time::Duration;
+
+use codex_mcp_server::CodexToolCallParam;
+use mcp_test_support::McpProcess;
+use mcp_test_support::create_final_assistant_message_sse_response;
+use mcp_test_support::create_mock_chat_completions_server;
+use mcp_types::JSONRPC_VERSION;
+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: std::time::Duration = std::time::Duration::from_secs(10);
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn test_send_message_success() {
+    // Spin up a mock completions server that immediately ends the Codex turn.
+    // Two Codex turns hit the mock model (session start + send-user-message). Provide two SSE responses.
+    let responses = vec![
+        create_final_assistant_message_sse_response("Done").expect("build mock assistant message"),
+        create_final_assistant_message_sse_response("Done").expect("build mock assistant message"),
+    ];
+    let server = create_mock_chat_completions_server(responses).await;
+
+    // Create a temporary Codex home with config pointing at the mock server.
+    let codex_home = TempDir::new().expect("create temp dir");
+    create_config_toml(codex_home.path(), &server.uri()).expect("write config.toml");
+
+    // Start MCP server process and initialize.
+    let mut mcp_process = McpProcess::new(codex_home.path())
+        .await
+        .expect("spawn mcp process");
+    timeout(DEFAULT_READ_TIMEOUT, mcp_process.initialize())
+        .await
+        .expect("init timed out")
+        .expect("init failed");
+
+    // Kick off a Codex session so we have a valid session_id.
+    let codex_request_id = mcp_process
+        .send_codex_tool_call(CodexToolCallParam {
+            prompt: "Start a session".to_string(),
+            ..Default::default()
+        })
+        .await
+        .expect("send codex tool call");
+
+    // Wait for the session_configured event to get the session_id.
+    let session_id = mcp_process
+        .read_stream_until_configured_response_message()
+        .await
+        .expect("read session_configured");
+
+    // The original codex call will finish quickly given our mock; consume its response.
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp_process.read_stream_until_response_message(RequestId::Integer(codex_request_id)),
+    )
+    .await
+    .expect("codex response timeout")
+    .expect("codex response error");
+
+    // Now exercise the send-user-message tool.
+    let send_msg_request_id = mcp_process
+        .send_user_message_tool_call("Hello again", &session_id)
+        .await
+        .expect("send send-message tool call");
+
+    let response: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp_process.read_stream_until_response_message(RequestId::Integer(send_msg_request_id)),
+    )
+    .await
+    .expect("send-user-message response timeout")
+    .expect("send-user-message response error");
+
+    assert_eq!(
+        JSONRPCResponse {
+            jsonrpc: JSONRPC_VERSION.into(),
+            id: RequestId::Integer(send_msg_request_id),
+            result: json!({
+                "content": [
+                    {
+                        "text": "{\"status\":\"ok\"}",
+                        "type": "text",
+                    }
+                ],
+                "isError": false,
+                "structuredContent": {
+                    "status": "ok"
+                }
+            }),
+        },
+        response
+    );
+    // wait for the server to hear the user message
+    sleep(Duration::from_secs(1));
+
+    // Ensure the server and tempdir live until end of test
+    drop(server);
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn test_send_message_session_not_found() {
+    // Start MCP without creating a Codex session
+    let codex_home = TempDir::new().expect("tempdir");
+    let mut mcp = McpProcess::new(codex_home.path()).await.expect("spawn");
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize())
+        .await
+        .expect("timeout")
+        .expect("init");
+
+    let unknown = uuid::Uuid::new_v4().to_string();
+    let req_id = mcp
+        .send_user_message_tool_call("ping", &unknown)
+        .await
+        .expect("send tool");
+
+    let resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(req_id)),
+    )
+    .await
+    .expect("timeout")
+    .expect("resp");
+
+    let result = resp.result.clone();
+    let content = result["content"][0]["text"].as_str().unwrap_or("");
+    assert!(content.contains("Session does not exist"));
+    assert_eq!(result["isError"], json!(true));
+}
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+fn create_config_toml(codex_home: &Path, server_uri: &str) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        format!(
+            r#"
+model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "danger-full-access"
+
+model_provider = "mock_provider"
+
+[model_providers.mock_provider]
+name = "Mock provider for test"
+base_url = "{server_uri}/v1"
+wire_api = "chat"
+request_max_retries = 0
+stream_max_retries = 0
+"#
+        ),
+    )
+}

codex-rs/mcp-types/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "mcp-types"
 version = { workspace = true }
-edition = "2024"
 
 [lints]
 workspace = true

codex-rs/tui/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
+edition = "2024"
 name = "codex-tui"
 version = { workspace = true }
-edition = "2024"
 
 [[bin]]
 name = "codex-tui"
@@ -17,15 +17,16 @@ workspace = true
 [dependencies]
 anyhow = "1"
 base64 = "0.22.1"
+chrono = { version = "0.4", features = ["serde"] }
 clap = { version = "4", features = ["derive"] }
 codex-ansi-escape = { path = "../ansi-escape" }
 codex-arg0 = { path = "../arg0" }
-codex-core = { path = "../core" }
 codex-common = { path = "../common", features = [
     "cli",
     "elapsed",
     "sandbox_summary",
 ] }
+codex-core = { path = "../core" }
 codex-file-search = { path = "../file-search" }
 codex-login = { path = "../login" }
 color-eyre = "0.6.3"
@@ -41,10 +42,14 @@ ratatui = { version = "0.29.0", features = [
 ] }
 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"
 strum = "0.27.2"
 strum_macros = "0.27.2"
+supports-color = "3.0.2"
+textwrap = "0.16.2"
 tokio = { version = "1", features = [
     "io-std",
     "macros",
@@ -57,11 +62,14 @@ tracing-appender = "0.2.3"
 tracing-subscriber = { version = "0.3.19", features = ["env-filter"] }
 tui-input = "0.14.0"
 tui-markdown = "0.3.3"
-tui-textarea = "0.7.0"
 unicode-segmentation = "1.12.0"
 unicode-width = "0.1"
 uuid = "1"
 
+
+
 [dev-dependencies]
 insta = "1.43.1"
 pretty_assertions = "1"
+rand = "0.8"
+chrono = { version = "0.4", features = ["serde"] }

codex-rs/tui/src/app.rs

@@ -9,9 +9,17 @@ use crate::slash_command::SlashCommand;
 use crate::tui;
 use codex_core::config::Config;
 use codex_core::protocol::Event;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::Op;
 use color_eyre::eyre::Result;
+use crossterm::SynchronizedUpdate;
 use crossterm::event::KeyCode;
 use crossterm::event::KeyEvent;
+use crossterm::event::KeyEventKind;
+use crossterm::terminal::supports_keyboard_enhancement;
+use ratatui::layout::Offset;
+use ratatui::prelude::Backend;
+use ratatui::text::Line;
 use std::path::PathBuf;
 use std::sync::Arc;
 use std::sync::atomic::AtomicBool;
@@ -50,9 +58,13 @@ pub(crate) struct App<'a> {
     /// True when a redraw has been scheduled but not yet executed.
     pending_redraw: Arc<AtomicBool>,
 
+    pending_history_lines: Vec<Line<'static>>,
+
     /// Stored parameters needed to instantiate the ChatWidget later, e.g.,
     /// after dismissing the Git-repo warning.
     chat_args: Option<ChatWidgetArgs>,
+
+    enhanced_keys_supported: bool,
 }
 
 /// Aggregate parameters needed to create a `ChatWidget`, as creation may be
@@ -62,6 +74,7 @@ struct ChatWidgetArgs {
     config: Config,
     initial_prompt: Option<String>,
     initial_images: Vec<PathBuf>,
+    enhanced_keys_supported: bool,
 }
 
 impl App<'_> {
@@ -75,6 +88,8 @@ impl App<'_> {
         let app_event_tx = AppEventSender::new(app_event_tx);
         let pending_redraw = Arc::new(AtomicBool::new(false));
 
+        let enhanced_keys_supported = supports_keyboard_enhancement().unwrap_or(false);
+
         // Spawn a dedicated thread for reading the crossterm event loop and
         // re-publishing the events as AppEvents, as appropriate.
         {
@@ -127,6 +142,7 @@ impl App<'_> {
                     config: config.clone(),
                     initial_prompt,
                     initial_images,
+                    enhanced_keys_supported,
                 }),
             )
         } else {
@@ -135,6 +151,7 @@ impl App<'_> {
                 app_event_tx.clone(),
                 initial_prompt,
                 initial_images,
+                enhanced_keys_supported,
             );
             (
                 AppState::Chat {
@@ -147,12 +164,14 @@ impl App<'_> {
         let file_search = FileSearchManager::new(config.cwd.clone(), app_event_tx.clone());
         Self {
             app_event_tx,
+            pending_history_lines: Vec::new(),
             app_event_rx,
             app_state,
             config,
             file_search,
             pending_redraw,
             chat_args,
+            enhanced_keys_supported,
         }
     }
 
@@ -192,20 +211,21 @@ impl App<'_> {
         while let Ok(event) = self.app_event_rx.recv() {
             match event {
                 AppEvent::InsertHistory(lines) => {
-                    crate::insert_history::insert_history_lines(terminal, lines);
+                    self.pending_history_lines.extend(lines);
                     self.app_event_tx.send(AppEvent::RequestRedraw);
                 }
                 AppEvent::RequestRedraw => {
                     self.schedule_redraw();
                 }
                 AppEvent::Redraw => {
-                    self.draw_next_frame(terminal)?;
+                    std::io::stdout().sync_update(|_| self.draw_next_frame(terminal))??;
                 }
                 AppEvent::KeyEvent(key_event) => {
                     match key_event {
                         KeyEvent {
                             code: KeyCode::Char('c'),
                             modifiers: crossterm::event::KeyModifiers::CONTROL,
+                            kind: KeyEventKind::Press,
                             ..
                         } => {
                             match &mut self.app_state {
@@ -220,6 +240,7 @@ impl App<'_> {
                         KeyEvent {
                             code: KeyCode::Char('d'),
                             modifiers: crossterm::event::KeyModifiers::CONTROL,
+                            kind: KeyEventKind::Press,
                             ..
                         } => {
                             match &mut self.app_state {
@@ -238,9 +259,15 @@ impl App<'_> {
                                 }
                             }
                         }
-                        _ => {
+                        KeyEvent {
+                            kind: KeyEventKind::Press | KeyEventKind::Repeat,
+                            ..
+                        } => {
                             self.dispatch_key_event(key_event);
                         }
+                        _ => {
+                            // Ignore Release key events for now.
+                        }
                     };
                 }
                 AppEvent::Paste(text) => {
@@ -267,10 +294,17 @@ impl App<'_> {
                             self.app_event_tx.clone(),
                             None,
                             Vec::new(),
+                            self.enhanced_keys_supported,
                         ));
                         self.app_state = AppState::Chat { widget: new_widget };
                         self.app_event_tx.send(AppEvent::RequestRedraw);
                     }
+                    SlashCommand::Compact => {
+                        if let AppState::Chat { widget } = &mut self.app_state {
+                            widget.clear_token_usage();
+                            self.app_event_tx.send(AppEvent::CodexOp(Op::Compact));
+                        }
+                    }
                     SlashCommand::Quit => {
                         break;
                     }
@@ -295,6 +329,45 @@ impl App<'_> {
                             widget.add_diff_output(text);
                         }
                     }
+                    #[cfg(debug_assertions)]
+                    SlashCommand::TestApproval => {
+                        use std::collections::HashMap;
+
+                        use codex_core::protocol::ApplyPatchApprovalRequestEvent;
+                        use codex_core::protocol::FileChange;
+
+                        self.app_event_tx.send(AppEvent::CodexEvent(Event {
+                            id: "1".to_string(),
+                            // msg: EventMsg::ExecApprovalRequest(ExecApprovalRequestEvent {
+                            //     call_id: "1".to_string(),
+                            //     command: vec!["git".into(), "apply".into()],
+                            //     cwd: self.config.cwd.clone(),
+                            //     reason: Some("test".to_string()),
+                            // }),
+                            msg: EventMsg::ApplyPatchApprovalRequest(
+                                ApplyPatchApprovalRequestEvent {
+                                    call_id: "1".to_string(),
+                                    changes: HashMap::from([
+                                        (
+                                            PathBuf::from("/tmp/test.txt"),
+                                            FileChange::Add {
+                                                content: "test".to_string(),
+                                            },
+                                        ),
+                                        (
+                                            PathBuf::from("/tmp/test2.txt"),
+                                            FileChange::Update {
+                                                unified_diff: "+test\n-test2".to_string(),
+                                                move_path: None,
+                                            },
+                                        ),
+                                    ]),
+                                    reason: None,
+                                    grant_root: Some(PathBuf::from("/tmp")),
+                                },
+                            ),
+                        }));
+                    }
                 },
                 AppEvent::StartFileSearch(query) => {
                     self.file_search.on_user_query(query);
@@ -319,16 +392,61 @@ impl App<'_> {
     }
 
     fn draw_next_frame(&mut self, terminal: &mut tui::Tui) -> Result<()> {
-        // TODO: add a throttle to avoid redrawing too often
+        let screen_size = terminal.size()?;
+        let last_known_screen_size = terminal.last_known_screen_size;
+        if screen_size != last_known_screen_size {
+            let cursor_pos = terminal.get_cursor_position()?;
+            let last_known_cursor_pos = terminal.last_known_cursor_pos;
+            if cursor_pos.y != last_known_cursor_pos.y {
+                // The terminal was resized. The only point of reference we have for where our viewport
+                // was moved is the cursor position.
+                // NB this assumes that the cursor was not wrapped as part of the resize.
+                let cursor_delta = cursor_pos.y as i32 - last_known_cursor_pos.y as i32;
 
-        match &mut self.app_state {
-            AppState::Chat { widget } => {
-                terminal.draw(|frame| frame.render_widget_ref(&**widget, frame.area()))?;
-            }
-            AppState::GitWarning { screen } => {
-                terminal.draw(|frame| frame.render_widget_ref(&*screen, frame.area()))?;
+                let new_viewport_area = terminal.viewport_area.offset(Offset {
+                    x: 0,
+                    y: cursor_delta,
+                });
+                terminal.set_viewport_area(new_viewport_area);
+                terminal.clear()?;
             }
         }
+
+        let size = terminal.size()?;
+        let desired_height = match &self.app_state {
+            AppState::Chat { widget } => widget.desired_height(size.width),
+            AppState::GitWarning { .. } => 10,
+        };
+
+        let mut area = terminal.viewport_area;
+        area.height = desired_height.min(size.height);
+        area.width = size.width;
+        if area.bottom() > size.height {
+            terminal
+                .backend_mut()
+                .scroll_region_up(0..area.top(), area.bottom() - size.height)?;
+            area.y = size.height - area.height;
+        }
+        if area != terminal.viewport_area {
+            terminal.clear()?;
+            terminal.set_viewport_area(area);
+        }
+        if !self.pending_history_lines.is_empty() {
+            crate::insert_history::insert_history_lines(
+                terminal,
+                self.pending_history_lines.clone(),
+            );
+            self.pending_history_lines.clear();
+        }
+        terminal.draw(|frame| match &mut self.app_state {
+            AppState::Chat { widget } => {
+                if let Some((x, y)) = widget.cursor_pos(frame.area()) {
+                    frame.set_cursor_position((x, y));
+                }
+                frame.render_widget_ref(&**widget, frame.area())
+            }
+            AppState::GitWarning { screen } => frame.render_widget_ref(&*screen, frame.area()),
+        })?;
         Ok(())
     }
 
@@ -352,6 +470,7 @@ impl App<'_> {
                         self.app_event_tx.clone(),
                         args.initial_prompt,
                         args.initial_images,
+                        args.enhanced_keys_supported,
                     ));
                     self.app_state = AppState::Chat { widget };
                     self.app_event_tx.send(AppEvent::RequestRedraw);

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

@@ -57,6 +57,10 @@ impl<'a> BottomPaneView<'a> for ApprovalModalView<'a> {
         self.current.is_complete() && self.queue.is_empty()
     }
 
+    fn desired_height(&self, width: u16) -> u16 {
+        self.current.desired_height(width)
+    }
+
     fn render(&self, area: Rect, buf: &mut Buffer) {
         (&self.current).render_ref(area, buf);
     }
@@ -95,6 +99,7 @@ mod tests {
         let mut pane = BottomPane::new(super::super::BottomPaneParams {
             app_event_tx: AppEventSender::new(tx_raw2),
             has_input_focus: true,
+            enhanced_keys_supported: false,
         });
         assert_eq!(CancellationEvent::Handled, view.on_ctrl_c(&mut pane));
         assert!(view.queue.is_empty());

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

@@ -28,6 +28,9 @@ pub(crate) trait BottomPaneView<'a> {
         CancellationEvent::Ignored
     }
 
+    /// Return the desired height of the view.
+    fn desired_height(&self, width: u16) -> u16;
+
     /// Render the view: this will be displayed in place of the composer.
     fn render(&self, area: Rect, buf: &mut Buffer);
 

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

@@ -1,8 +1,5 @@
 use std::collections::HashMap;
 
-use tui_textarea::CursorMove;
-use tui_textarea::TextArea;
-
 use crate::app_event::AppEvent;
 use crate::app_event_sender::AppEventSender;
 use codex_core::protocol::Op;
@@ -67,59 +64,52 @@ impl ChatComposerHistory {
 
     /// Should Up/Down key presses be interpreted as history navigation given
     /// the current content and cursor position of `textarea`?
-    pub fn should_handle_navigation(&self, textarea: &TextArea) -> bool {
+    pub fn should_handle_navigation(&self, text: &str, cursor: usize) -> bool {
         if self.history_entry_count == 0 && self.local_history.is_empty() {
             return false;
         }
 
-        if textarea.is_empty() {
+        if text.is_empty() {
             return true;
         }
 
         // Textarea is not empty – only navigate when cursor is at start and
         // text matches last recalled history entry so regular editing is not
         // hijacked.
-        let (row, col) = textarea.cursor();
-        if row != 0 || col != 0 {
+        if cursor != 0 {
             return false;
         }
 
-        let lines = textarea.lines();
-        matches!(&self.last_history_text, Some(prev) if prev == &lines.join("\n"))
+        matches!(&self.last_history_text, Some(prev) if prev == text)
     }
 
     /// Handle <Up>. Returns true when the key was consumed and the caller
     /// should request a redraw.
-    pub fn navigate_up(&mut self, textarea: &mut TextArea, app_event_tx: &AppEventSender) -> bool {
+    pub fn navigate_up(&mut self, app_event_tx: &AppEventSender) -> Option<String> {
         let total_entries = self.history_entry_count + self.local_history.len();
         if total_entries == 0 {
-            return false;
+            return None;
         }
 
         let next_idx = match self.history_cursor {
             None => (total_entries as isize) - 1,
-            Some(0) => return true, // already at oldest
+            Some(0) => return None, // already at oldest
             Some(idx) => idx - 1,
         };
 
         self.history_cursor = Some(next_idx);
-        self.populate_history_at_index(next_idx as usize, textarea, app_event_tx);
-        true
+        self.populate_history_at_index(next_idx as usize, app_event_tx)
     }
 
     /// Handle <Down>.
-    pub fn navigate_down(
-        &mut self,
-        textarea: &mut TextArea,
-        app_event_tx: &AppEventSender,
-    ) -> bool {
+    pub fn navigate_down(&mut self, app_event_tx: &AppEventSender) -> Option<String> {
         let total_entries = self.history_entry_count + self.local_history.len();
         if total_entries == 0 {
-            return false;
+            return None;
         }
 
         let next_idx_opt = match self.history_cursor {
-            None => return false, // not browsing
+            None => return None, // not browsing
             Some(idx) if (idx as usize) + 1 >= total_entries => None,
             Some(idx) => Some(idx + 1),
         };
@@ -127,16 +117,15 @@ impl ChatComposerHistory {
         match next_idx_opt {
             Some(idx) => {
                 self.history_cursor = Some(idx);
-                self.populate_history_at_index(idx as usize, textarea, app_event_tx);
+                self.populate_history_at_index(idx as usize, app_event_tx)
             }
             None => {
                 // Past newest – clear and exit browsing mode.
                 self.history_cursor = None;
                 self.last_history_text = None;
-                self.replace_textarea_content(textarea, "");
+                Some(String::new())
             }
         }
-        true
     }
 
     /// Integrate a GetHistoryEntryResponse event.
@@ -145,19 +134,18 @@ impl ChatComposerHistory {
         log_id: u64,
         offset: usize,
         entry: Option<String>,
-        textarea: &mut TextArea,
-    ) -> bool {
+    ) -> Option<String> {
         if self.history_log_id != Some(log_id) {
-            return false;
+            return None;
         }
-        let Some(text) = entry else { return false };
+        let text = entry?;
         self.fetched_history.insert(offset, text.clone());
 
         if self.history_cursor == Some(offset as isize) {
-            self.replace_textarea_content(textarea, &text);
-            return true;
+            self.last_history_text = Some(text.clone());
+            return Some(text);
         }
-        false
+        None
     }
 
     // ---------------------------------------------------------------------
@@ -167,21 +155,20 @@ impl ChatComposerHistory {
     fn populate_history_at_index(
         &mut self,
         global_idx: usize,
-        textarea: &mut TextArea,
         app_event_tx: &AppEventSender,
-    ) {
+    ) -> Option<String> {
         if global_idx >= self.history_entry_count {
             // Local entry.
             if let Some(text) = self
                 .local_history
                 .get(global_idx - self.history_entry_count)
             {
-                let t = text.clone();
-                self.replace_textarea_content(textarea, &t);
+                self.last_history_text = Some(text.clone());
+                return Some(text.clone());
             }
         } else if let Some(text) = self.fetched_history.get(&global_idx) {
-            let t = text.clone();
-            self.replace_textarea_content(textarea, &t);
+            self.last_history_text = Some(text.clone());
+            return Some(text.clone());
         } else if let Some(log_id) = self.history_log_id {
             let op = Op::GetHistoryEntryRequest {
                 offset: global_idx,
@@ -189,14 +176,7 @@ impl ChatComposerHistory {
             };
             app_event_tx.send(AppEvent::CodexOp(op));
         }
-    }
-
-    fn replace_textarea_content(&mut self, textarea: &mut TextArea, text: &str) {
-        textarea.select_all();
-        textarea.cut();
-        let _ = textarea.insert_str(text);
-        textarea.move_cursor(CursorMove::Jump(0, 0));
-        self.last_history_text = Some(text.to_string());
+        None
     }
 }
 
@@ -217,11 +197,9 @@ mod tests {
         // Pretend there are 3 persistent entries.
         history.set_metadata(1, 3);
 
-        let mut textarea = TextArea::default();
-
         // First Up should request offset 2 (latest) and await async data.
-        assert!(history.should_handle_navigation(&textarea));
-        assert!(history.navigate_up(&mut textarea, &tx));
+        assert!(history.should_handle_navigation("", 0));
+        assert!(history.navigate_up(&tx).is_none()); // don't replace the text yet
 
         // Verify that an AppEvent::CodexOp with the correct GetHistoryEntryRequest was sent.
         let event = rx.try_recv().expect("expected AppEvent to be sent");
@@ -235,14 +213,15 @@ mod tests {
             },
             history_request1
         );
-        assert_eq!(textarea.lines().join("\n"), ""); // still empty
 
         // Inject the async response.
-        assert!(history.on_entry_response(1, 2, Some("latest".into()), &mut textarea));
-        assert_eq!(textarea.lines().join("\n"), "latest");
+        assert_eq!(
+            Some("latest".into()),
+            history.on_entry_response(1, 2, Some("latest".into()))
+        );
 
         // Next Up should move to offset 1.
-        assert!(history.navigate_up(&mut textarea, &tx));
+        assert!(history.navigate_up(&tx).is_none()); // don't replace the text yet
 
         // Verify second CodexOp event for offset 1.
         let event2 = rx.try_recv().expect("expected second event");
@@ -257,7 +236,9 @@ mod tests {
             history_request_2
         );
 
-        history.on_entry_response(1, 1, Some("older".into()), &mut textarea);
-        assert_eq!(textarea.lines().join("\n"), "older");
+        assert_eq!(
+            Some("older".into()),
+            history.on_entry_response(1, 1, Some("older".into()))
+        );
     }
 }

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

@@ -3,9 +3,9 @@ use ratatui::layout::Rect;
 use ratatui::style::Color;
 use ratatui::style::Style;
 use ratatui::style::Stylize;
-use ratatui::widgets::Block;
-use ratatui::widgets::BorderType;
-use ratatui::widgets::Borders;
+use ratatui::symbols::border::QUADRANT_LEFT_HALF;
+use ratatui::text::Line;
+use ratatui::text::Span;
 use ratatui::widgets::Cell;
 use ratatui::widgets::Row;
 use ratatui::widgets::Table;
@@ -71,12 +71,8 @@ impl CommandPopup {
     /// Determine the preferred height of the popup. This is the number of
     /// rows required to show **at most** `MAX_POPUP_ROWS` commands plus the
     /// table/border overhead (one line at the top and one at the bottom).
-    pub(crate) fn calculate_required_height(&self, _area: &Rect) -> u16 {
-        let matches = self.filtered_commands();
-        let row_count = matches.len().clamp(1, MAX_POPUP_ROWS) as u16;
-        // Account for the border added by the Block that wraps the table.
-        // 2 = one line at the top, one at the bottom.
-        row_count + 2
+    pub(crate) fn calculate_required_height(&self) -> u16 {
+        self.filtered_commands().len().clamp(1, MAX_POPUP_ROWS) as u16
     }
 
     /// Return the list of commands that match the current filter. Matching is
@@ -158,18 +154,19 @@ impl WidgetRef for CommandPopup {
             let default_style = Style::default();
             let command_style = Style::default().fg(Color::LightBlue);
             for (idx, cmd) in visible_matches.iter().enumerate() {
-                let (cmd_style, desc_style) = if Some(idx) == self.selected_idx {
-                    (
-                        command_style.bg(Color::DarkGray),
-                        default_style.bg(Color::DarkGray),
-                    )
-                } else {
-                    (command_style, default_style)
-                };
-
                 rows.push(Row::new(vec![
-                    Cell::from(format!("/{}", cmd.command())).style(cmd_style),
-                    Cell::from(cmd.description().to_string()).style(desc_style),
+                    Cell::from(Line::from(vec![
+                        if Some(idx) == self.selected_idx {
+                            Span::styled(
+                                "›",
+                                Style::default().bg(Color::DarkGray).fg(Color::LightCyan),
+                            )
+                        } else {
+                            Span::styled(QUADRANT_LEFT_HALF, Style::default().fg(Color::DarkGray))
+                        },
+                        Span::styled(format!("/{}", cmd.command()), command_style),
+                    ])),
+                    Cell::from(cmd.description().to_string()).style(default_style),
                 ]));
             }
         }
@@ -180,12 +177,13 @@ impl WidgetRef for CommandPopup {
             rows,
             [Constraint::Length(FIRST_COLUMN_WIDTH), Constraint::Min(10)],
         )
-        .column_spacing(0)
-        .block(
-            Block::default()
-                .borders(Borders::ALL)
-                .border_type(BorderType::Rounded),
-        );
+        .column_spacing(0);
+        // .block(
+        //     Block::default()
+        //         .borders(Borders::LEFT)
+        //         .border_type(BorderType::QuadrantOutside)
+        //         .border_style(Style::default().fg(Color::DarkGray)),
+        // );
 
         table.render(area, buf);
     }

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

@@ -109,18 +109,14 @@ impl FileSearchPopup {
     }
 
     /// Preferred height (rows) including border.
-    pub(crate) fn calculate_required_height(&self, _area: &Rect) -> u16 {
+    pub(crate) fn calculate_required_height(&self) -> u16 {
         // Row count depends on whether we already have matches. If no matches
         // yet (e.g. initial search or query with no results) reserve a single
         // row so the popup is still visible. When matches are present we show
         // up to MAX_RESULTS regardless of the waiting flag so the list
         // remains stable while a newer search is in-flight.
-        let rows = if self.matches.is_empty() {
-            1
-        } else {
-            self.matches.len().clamp(1, MAX_RESULTS)
-        } as u16;
-        rows + 2 // border
+
+        self.matches.len().clamp(1, MAX_RESULTS) as u16
     }
 }
 
@@ -128,7 +124,14 @@ impl WidgetRef for &FileSearchPopup {
     fn render_ref(&self, area: Rect, buf: &mut Buffer) {
         // Prepare rows.
         let rows: Vec<Row> = if self.matches.is_empty() {
-            vec![Row::new(vec![Cell::from(" no matches ")])]
+            vec![Row::new(vec![
+                Cell::from(if self.waiting {
+                    "(searching …)"
+                } else {
+                    "no matches"
+                })
+                .style(Style::new().add_modifier(Modifier::ITALIC | Modifier::DIM)),
+            ])]
         } else {
             self.matches
                 .iter()
@@ -169,17 +172,12 @@ impl WidgetRef for &FileSearchPopup {
                 .collect()
         };
 
-        let mut title = format!(" @{} ", self.pending_query);
-        if self.waiting {
-            title.push_str(" (searching …)");
-        }
-
         let table = Table::new(rows, vec![Constraint::Percentage(100)])
             .block(
                 Block::default()
-                    .borders(Borders::ALL)
-                    .border_type(BorderType::Rounded)
-                    .title(title),
+                    .borders(Borders::LEFT)
+                    .border_type(BorderType::QuadrantOutside)
+                    .border_style(Style::default().fg(Color::DarkGray)),
             )
             .widths([Constraint::Percentage(100)]);
 

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

@@ -19,6 +19,7 @@ mod chat_composer_history;
 mod command_popup;
 mod file_search_popup;
 mod status_indicator_view;
+mod textarea;
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub(crate) enum CancellationEvent {
@@ -36,7 +37,7 @@ use status_indicator_view::StatusIndicatorView;
 pub(crate) struct BottomPane<'a> {
     /// Composer is retained even when a BottomPaneView is displayed so the
     /// input state is retained when the view is closed.
-    composer: ChatComposer<'a>,
+    composer: ChatComposer,
 
     /// If present, this is displayed instead of the `composer`.
     active_view: Option<Box<dyn BottomPaneView<'a> + 'a>>,
@@ -50,12 +51,18 @@ pub(crate) struct BottomPane<'a> {
 pub(crate) struct BottomPaneParams {
     pub(crate) app_event_tx: AppEventSender,
     pub(crate) has_input_focus: bool,
+    pub(crate) enhanced_keys_supported: bool,
 }
 
 impl BottomPane<'_> {
     pub fn new(params: BottomPaneParams) -> Self {
+        let enhanced_keys_supported = params.enhanced_keys_supported;
         Self {
-            composer: ChatComposer::new(params.has_input_focus, params.app_event_tx.clone()),
+            composer: ChatComposer::new(
+                params.has_input_focus,
+                params.app_event_tx.clone(),
+                enhanced_keys_supported,
+            ),
             active_view: None,
             app_event_tx: params.app_event_tx,
             has_input_focus: params.has_input_focus,
@@ -64,6 +71,25 @@ impl BottomPane<'_> {
         }
     }
 
+    pub fn desired_height(&self, width: u16) -> u16 {
+        self.active_view
+            .as_ref()
+            .map(|v| v.desired_height(width))
+            .unwrap_or(self.composer.desired_height(width))
+    }
+
+    pub fn cursor_pos(&self, area: Rect) -> Option<(u16, u16)> {
+        // Hide the cursor whenever an overlay view is active (e.g. the
+        // status indicator shown while a task is running, or approval modal).
+        // In these states the textarea is not interactable, so we should not
+        // show its caret.
+        if self.active_view.is_some() {
+            None
+        } else {
+            self.composer.cursor_pos(area)
+        }
+    }
+
     /// Forward a key event to the active view or the composer.
     pub fn handle_key_event(&mut self, key_event: KeyEvent) -> InputResult {
         if let Some(mut view) = self.active_view.take() {
@@ -291,6 +317,7 @@ mod tests {
         let mut pane = BottomPane::new(BottomPaneParams {
             app_event_tx: tx,
             has_input_focus: true,
+            enhanced_keys_supported: false,
         });
         pane.push_approval_request(exec_request());
         assert_eq!(CancellationEvent::Handled, pane.on_ctrl_c());

codex-rs/tui/src/bottom_pane/snapshots/codex_tui__bottom_pane__chat_composer__tests__backspace_after_pastes.snap

@@ -2,13 +2,13 @@
 source: tui/src/bottom_pane/chat_composer.rs
 expression: terminal.backend()
 ---
-"╭──────────────────────────────────────────────────────────────────────────────────────────────────╮"
-"│[Pasted Content 1002 chars][Pasted Content 1004 chars]                                            │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"╰───────────────────────────────────────────────Enter to send | Ctrl+D to quit | Ctrl+J for newline╯"
+"▌[Pasted Content 1002 chars][Pasted Content 1004 chars]                                             "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+" ⏎ send   Ctrl+J newline   Ctrl+C quit                                                              "

codex-rs/tui/src/bottom_pane/snapshots/codex_tui__bottom_pane__chat_composer__tests__empty.snap

@@ -2,13 +2,13 @@
 source: tui/src/bottom_pane/chat_composer.rs
 expression: terminal.backend()
 ---
-"╭──────────────────────────────────────────────────────────────────────────────────────────────────╮"
-"│ send a message                                                                                   │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"│                                                                                                  │"
-"╰───────────────────────────────────────────────Enter to send | Ctrl+D to quit | Ctrl+J for newline╯"
+"▌ ...                                                                                               "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+"▌                                                                                                   "
+" ⏎ send   Ctrl+J newline   Ctrl+C quit                                                              "