initial tests

This update replaces the previous ratatui history widget with an
append-only log so that the terminal can handle text selection and
scrolling. It also disables streaming responses, which we'll do our best
to bring back in a later PR. It also adds a small summary of token use
after the TUI exits.

.devcontainer/devcontainer.json

@@ -21,7 +21,7 @@
       "settings": {
         "terminal.integrated.defaultProfile.linux": "bash"
       },
-      "extensions": ["rust-lang.rust-analyzer"]
+      "extensions": ["rust-lang.rust-analyzer", "tamasfe.even-better-toml"]
     }
   }
 }

.github/actions/codex/bun.lock

@@ -8,8 +8,8 @@
         "@actions/github": "^6.0.1",
       },
       "devDependencies": {
-        "@types/bun": "^1.2.18",
-        "@types/node": "^24.0.13",
+        "@types/bun": "^1.2.19",
+        "@types/node": "^24.1.0",
         "prettier": "^3.6.2",
         "typescript": "^5.8.3",
       },
@@ -48,15 +48,15 @@
 
     "@octokit/types": ["@octokit/types@13.10.0", "", { "dependencies": { "@octokit/openapi-types": "^24.2.0" } }, "sha512-ifLaO34EbbPj0Xgro4G5lP5asESjwHracYJvVaPIyXMuiuXLlhic3S47cBdTb+jfODkTE5YtGCLt3Ay3+J97sA=="],
 
-    "@types/bun": ["@types/bun@1.2.18", "", { "dependencies": { "bun-types": "1.2.18" } }, "sha512-Xf6RaWVheyemaThV0kUfaAUvCNokFr+bH8Jxp+tTZfx7dAPA8z9ePnP9S9+Vspzuxxx9JRAXhnyccRj3GyCMdQ=="],
+    "@types/bun": ["@types/bun@1.2.19", "", { "dependencies": { "bun-types": "1.2.19" } }, "sha512-d9ZCmrH3CJ2uYKXQIUuZ/pUnTqIvLDS0SK7pFmbx8ma+ziH/FRMoAq5bYpRG7y+w1gl+HgyNZbtqgMq4W4e2Lg=="],
 
-    "@types/node": ["@types/node@24.0.13", "", { "dependencies": { "undici-types": "~7.8.0" } }, "sha512-Qm9OYVOFHFYg3wJoTSrz80hoec5Lia/dPp84do3X7dZvLikQvM1YpmvTBEdIr/e+U8HTkFjLHLnl78K/qjf+jQ=="],
+    "@types/node": ["@types/node@24.1.0", "", { "dependencies": { "undici-types": "~7.8.0" } }, "sha512-ut5FthK5moxFKH2T1CUOC6ctR67rQRvvHdFLCD2Ql6KXmMuCrjsSsRI9UsLCm9M18BMwClv4pn327UvB7eeO1w=="],
 
     "@types/react": ["@types/react@19.1.8", "", { "dependencies": { "csstype": "^3.0.2" } }, "sha512-AwAfQ2Wa5bCx9WP8nZL2uMZWod7J7/JSplxbTmBQ5ms6QpqNYm672H0Vu9ZVKVngQ+ii4R/byguVEUZQyeg44g=="],
 
     "before-after-hook": ["before-after-hook@2.2.3", "", {}, "sha512-NzUnlZexiaH/46WDhANlyR2bXRopNg4F/zuSA3OpZnllCUgRaOF2znDioDWrmbNVsuZk6l9pMquQB38cfBZwkQ=="],
 
-    "bun-types": ["bun-types@1.2.18", "", { "dependencies": { "@types/node": "*" }, "peerDependencies": { "@types/react": "^19" } }, "sha512-04+Eha5NP7Z0A9YgDAzMk5PHR16ZuLVa83b26kH5+cp1qZW4F6FmAURngE7INf4tKOvCE69vYvDEwoNl1tGiWw=="],
+    "bun-types": ["bun-types@1.2.19", "", { "dependencies": { "@types/node": "*" }, "peerDependencies": { "@types/react": "^19" } }, "sha512-uAOTaZSPuYsWIXRpj7o56Let0g/wjihKCkeRqUBhlLVM/Bt+Fj9xTo+LhC1OV1XDaGkz4hNC80et5xgy+9KTHQ=="],
 
     "csstype": ["csstype@3.1.3", "", {}, "sha512-M1uQkMl8rQK/szD0LNhtqxIPLpimGm8sOBwU7lLnCpSbTyY3yeU1Vc7l4KT5zT4s/yOxHH5O7tIuuLOCnLADRw=="],
 
@@ -82,6 +82,8 @@
 
     "@octokit/plugin-rest-endpoint-methods/@octokit/types": ["@octokit/types@12.6.0", "", { "dependencies": { "@octokit/openapi-types": "^20.0.0" } }, "sha512-1rhSOfRa6H9w4YwK0yrf5faDaDTb+yLyBUKOCV4xtCDB5VmIPqd/v9yr9o6SAzOAlRxMiRiCic6JVM1/kunVkw=="],
 
+    "bun-types/@types/node": ["@types/node@24.0.13", "", { "dependencies": { "undici-types": "~7.8.0" } }, "sha512-Qm9OYVOFHFYg3wJoTSrz80hoec5Lia/dPp84do3X7dZvLikQvM1YpmvTBEdIr/e+U8HTkFjLHLnl78K/qjf+jQ=="],
+
     "@octokit/plugin-paginate-rest/@octokit/types/@octokit/openapi-types": ["@octokit/openapi-types@20.0.0", "", {}, "sha512-EtqRBEjp1dL/15V7WiX5LJMIxxkdiGJnabzYx5Apx4FkQIFgAfKumXeYAqqJCj1s+BMX4cPFIFC4OLCR6stlnA=="],
 
     "@octokit/plugin-rest-endpoint-methods/@octokit/types/@octokit/openapi-types": ["@octokit/openapi-types@20.0.0", "", {}, "sha512-EtqRBEjp1dL/15V7WiX5LJMIxxkdiGJnabzYx5Apx4FkQIFgAfKumXeYAqqJCj1s+BMX4cPFIFC4OLCR6stlnA=="],

.github/actions/codex/package.json

@@ -13,8 +13,8 @@
         "@actions/github": "^6.0.1"
     },
     "devDependencies": {
-        "@types/bun": "^1.2.18",
-        "@types/node": "^24.0.13",
+        "@types/bun": "^1.2.19",
+        "@types/node": "^24.1.0",
         "prettier": "^3.6.2",
         "typescript": "^5.8.3"
     }

.github/workflows/rust-release.yml

@@ -93,7 +93,7 @@ jobs:
           sudo apt install -y musl-tools pkg-config
 
       - name: Cargo build
-        run: cargo build --target ${{ matrix.target }} --release --all-targets --all-features
+        run: cargo build --target ${{ matrix.target }} --release --bin codex --bin codex-exec --bin codex-linux-sandbox
 
       - name: Stage artifacts
         shell: bash

codex-rs/Cargo.lock

@@ -649,7 +649,7 @@ dependencies = [
  "clap",
  "codex-core",
  "serde",
- "toml 0.9.1",
+ "toml 0.9.2",
 ]
 
 [[package]]
@@ -663,6 +663,7 @@ dependencies = [
  "bytes",
  "codex-apply-patch",
  "codex-mcp-client",
+ "core_test_support",
  "dirs",
  "env-flags",
  "eventsource-stream",
@@ -676,20 +677,20 @@ dependencies = [
  "openssl-sys",
  "predicates",
  "pretty_assertions",
- "rand 0.9.1",
+ "rand 0.9.2",
  "reqwest",
  "seccompiler",
  "serde",
  "serde_json",
  "sha1",
- "strum_macros 0.27.1",
+ "strum_macros 0.27.2",
  "tempfile",
  "thiserror 2.0.12",
  "time",
  "tokio",
  "tokio-test",
  "tokio-util",
- "toml 0.9.1",
+ "toml 0.9.2",
  "tracing",
  "tree-sitter",
  "tree-sitter-bash",
@@ -756,7 +757,9 @@ version = "0.0.0"
 dependencies = [
  "anyhow",
  "clap",
+ "codex-common",
  "codex-core",
+ "dotenvy",
  "landlock",
  "libc",
  "seccompiler",
@@ -797,6 +800,7 @@ dependencies = [
  "codex-core",
  "codex-linux-sandbox",
  "mcp-types",
+ "mcp_test_support",
  "pretty_assertions",
  "schemars 0.8.22",
  "serde",
@@ -805,7 +809,7 @@ dependencies = [
  "tempfile",
  "tokio",
  "tokio-test",
- "toml 0.9.1",
+ "toml 0.9.2",
  "tracing",
  "tracing-subscriber",
  "uuid",
@@ -838,8 +842,8 @@ dependencies = [
  "regex-lite",
  "serde_json",
  "shlex",
- "strum 0.27.1",
- "strum_macros 0.27.1",
+ "strum 0.27.2",
+ "strum_macros 0.27.2",
  "tokio",
  "tracing",
  "tracing-appender",
@@ -848,6 +852,7 @@ dependencies = [
  "tui-markdown",
  "tui-textarea",
  "unicode-segmentation",
+ "unicode-width 0.1.14",
  "uuid",
 ]
 
@@ -950,6 +955,16 @@ version = "0.8.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "773648b94d0e5d620f64f280777445740e61fe701025087ec8b57f45c791888b"
 
+[[package]]
+name = "core_test_support"
+version = "0.0.0"
+dependencies = [
+ "codex-core",
+ "serde_json",
+ "tempfile",
+ "tokio",
+]
+
 [[package]]
 name = "cpufeatures"
 version = "0.2.17"
@@ -1272,6 +1287,12 @@ version = "0.3.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "fea41bba32d969b513997752735605054bc0dfa92b4c56bf1189f2e174be7a10"
 
+[[package]]
+name = "dotenvy"
+version = "0.15.7"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1aaf95b3e5c8f23aa320147307562d361db0ae0d51242340f558153b4eb2439b"
+
 [[package]]
 name = "dupe"
 version = "0.9.1"
@@ -2596,6 +2617,22 @@ dependencies = [
  "serde_json",
 ]
 
+[[package]]
+name = "mcp_test_support"
+version = "0.0.0"
+dependencies = [
+ "anyhow",
+ "assert_cmd",
+ "codex-mcp-server",
+ "mcp-types",
+ "pretty_assertions",
+ "serde_json",
+ "shlex",
+ "tempfile",
+ "tokio",
+ "wiremock",
+]
+
 [[package]]
 name = "memchr"
 version = "2.7.5"
@@ -3271,9 +3308,9 @@ dependencies = [
 
 [[package]]
 name = "rand"
-version = "0.9.1"
+version = "0.9.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "9fbfd9d094a40bf3ae768db9361049ace4c0e04a4fd6b359518bd7b73a73dd97"
+checksum = "6db2770f06117d490610c7488547d543617b21bfa07796d7a12f6f1bd53850d1"
 dependencies = [
  "rand_chacha 0.9.0",
  "rand_core 0.9.3",
@@ -4255,9 +4292,9 @@ dependencies = [
 
 [[package]]
 name = "strum"
-version = "0.27.1"
+version = "0.27.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f64def088c51c9510a8579e3c5d67c65349dcf755e5479ad3d010aa6454e2c32"
+checksum = "af23d6f6c1a224baef9d3f61e287d2761385a5b88fdab4eb4c6f11aeb54c4bcf"
 
 [[package]]
 name = "strum_macros"
@@ -4274,14 +4311,13 @@ dependencies = [
 
 [[package]]
 name = "strum_macros"
-version = "0.27.1"
+version = "0.27.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "c77a8c5abcaf0f9ce05d62342b7d298c346515365c36b673df4ebe3ced01fde8"
+checksum = "7695ce3845ea4b33927c055a39dc438a45b059f7c1b3d91d38d10355fb8cbca7"
 dependencies = [
  "heck",
  "proc-macro2",
  "quote",
- "rustversion",
  "syn 2.0.104",
 ]
 
@@ -4658,9 +4694,9 @@ dependencies = [
 
 [[package]]
 name = "toml"
-version = "0.9.1"
+version = "0.9.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0207d6ed1852c2a124c1fbec61621acb8330d2bf969a5d0643131e9affd985a5"
+checksum = "ed0aee96c12fa71097902e0bb061a5e1ebd766a6636bb605ba401c45c1650eac"
 dependencies = [
  "indexmap 2.10.0",
  "serde",
@@ -4704,9 +4740,9 @@ dependencies = [
 
 [[package]]
 name = "toml_parser"
-version = "1.0.0"
+version = "1.0.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "b5c1c469eda89749d2230d8156a5969a69ffe0d6d01200581cdc6110674d293e"
+checksum = "97200572db069e74c512a14117b296ba0a80a30123fbbb5aa1f4a348f639ca30"
 dependencies = [
  "winnow",
 ]
@@ -4848,9 +4884,9 @@ dependencies = [
 
 [[package]]
 name = "tree-sitter"
-version = "0.25.6"
+version = "0.25.8"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a7cf18d43cbf0bfca51f657132cc616a5097edc4424d538bae6fa60142eaf9f0"
+checksum = "6d7b8994f367f16e6fa14b5aebbcb350de5d7cbea82dc5b00ae997dd71680dd2"
 dependencies = [
  "cc",
  "regex",

codex-rs/apply-patch/Cargo.toml

@@ -14,7 +14,7 @@ workspace = true
 anyhow = "1"
 similar = "2.7.0"
 thiserror = "2.0.12"
-tree-sitter = "0.25.3"
+tree-sitter = "0.25.8"
 tree-sitter-bash = "0.25.0"
 
 [dev-dependencies]

codex-rs/chatgpt/src/apply_command.rs

@@ -1,3 +1,5 @@
+use std::path::PathBuf;
+
 use clap::Parser;
 use codex_common::CliConfigOverrides;
 use codex_core::config::Config;
@@ -17,7 +19,10 @@ pub struct ApplyCommand {
     #[clap(flatten)]
     pub config_overrides: CliConfigOverrides,
 }
-pub async fn run_apply_command(apply_cli: ApplyCommand) -> anyhow::Result<()> {
+pub async fn run_apply_command(
+    apply_cli: ApplyCommand,
+    cwd: Option<PathBuf>,
+) -> anyhow::Result<()> {
     let config = Config::load_with_cli_overrides(
         apply_cli
             .config_overrides
@@ -29,10 +34,13 @@ pub async fn run_apply_command(apply_cli: ApplyCommand) -> anyhow::Result<()> {
     init_chatgpt_token_from_auth(&config.codex_home).await?;
 
     let task_response = get_task(&config, apply_cli.task_id).await?;
-    apply_diff_from_task(task_response).await
+    apply_diff_from_task(task_response, cwd).await
 }
 
-pub async fn apply_diff_from_task(task_response: GetTaskResponse) -> anyhow::Result<()> {
+pub async fn apply_diff_from_task(
+    task_response: GetTaskResponse,
+    cwd: Option<PathBuf>,
+) -> anyhow::Result<()> {
     let diff_turn = match task_response.current_diff_task_turn {
         Some(turn) => turn,
         None => anyhow::bail!("No diff turn found"),
@@ -42,13 +50,17 @@ pub async fn apply_diff_from_task(task_response: GetTaskResponse) -> anyhow::Res
         _ => None,
     });
     match output_diff {
-        Some(output_diff) => apply_diff(&output_diff.diff).await,
+        Some(output_diff) => apply_diff(&output_diff.diff, cwd).await,
         None => anyhow::bail!("No PR output item found"),
     }
 }
 
-async fn apply_diff(diff: &str) -> anyhow::Result<()> {
-    let toplevel_output = tokio::process::Command::new("git")
+async fn apply_diff(diff: &str, cwd: Option<PathBuf>) -> anyhow::Result<()> {
+    let mut cmd = tokio::process::Command::new("git");
+    if let Some(cwd) = cwd {
+        cmd.current_dir(cwd);
+    }
+    let toplevel_output = cmd
         .args(vec!["rev-parse", "--show-toplevel"])
         .output()
         .await?;

codex-rs/chatgpt/tests/apply_command_e2e.rs

@@ -78,17 +78,7 @@ async fn test_apply_command_creates_fibonacci_file() {
         .await
         .expect("Failed to load fixture");
 
-    let original_dir = std::env::current_dir().expect("Failed to get current dir");
-    std::env::set_current_dir(repo_path).expect("Failed to change directory");
-    struct DirGuard(std::path::PathBuf);
-    impl Drop for DirGuard {
-        fn drop(&mut self) {
-            let _ = std::env::set_current_dir(&self.0);
-        }
-    }
-    let _guard = DirGuard(original_dir);
-
-    apply_diff_from_task(task_response)
+    apply_diff_from_task(task_response, Some(repo_path.to_path_buf()))
         .await
         .expect("Failed to apply diff from task");
 
@@ -173,7 +163,7 @@ console.log(fib(10));
         .await
         .expect("Failed to load fixture");
 
-    let apply_result = apply_diff_from_task(task_response).await;
+    let apply_result = apply_diff_from_task(task_response, Some(repo_path.to_path_buf())).await;
 
     assert!(
         apply_result.is_err(),

codex-rs/cli/src/main.rs

@@ -105,7 +105,8 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
         None => {
             let mut tui_cli = cli.interactive;
             prepend_config_flags(&mut tui_cli.config_overrides, cli.config_overrides);
-            codex_tui::run_main(tui_cli, codex_linux_sandbox_exe)?;
+            let usage = codex_tui::run_main(tui_cli, codex_linux_sandbox_exe)?;
+            println!("{}", codex_core::protocol::FinalOutput::from(usage));
         }
         Some(Subcommand::Exec(mut exec_cli)) => {
             prepend_config_flags(&mut exec_cli.config_overrides, cli.config_overrides);
@@ -145,7 +146,7 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
         },
         Some(Subcommand::Apply(mut apply_cli)) => {
             prepend_config_flags(&mut apply_cli.config_overrides, cli.config_overrides);
-            run_apply_command(apply_cli).await?;
+            run_apply_command(apply_cli, None).await?;
         }
     }
 

codex-rs/config.md

@@ -498,14 +498,5 @@ Options that are specific to the TUI.
 
 ```toml
 [tui]
-# This will make it so that Codex does not try to process mouse events, which
-# means your Terminal's native drag-to-text to text selection and copy/paste
-# should work. The tradeoff is that Codex will not receive any mouse events, so
-# it will not be possible to use the mouse to scroll conversation history.
-#
-# Note that most terminals support holding down a modifier key when using the
-# mouse to support text selection. For example, even if Codex mouse capture is
-# enabled (i.e., this is set to `false`), you can still hold down alt while
-# dragging the mouse to select text.
-disable_mouse_capture = true  # defaults to `false`
+# More to come here
 ```

codex-rs/core/Cargo.toml

@@ -30,7 +30,7 @@ reqwest = { version = "0.12", features = ["json", "stream"] }
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 sha1 = "0.10.6"
-strum_macros = "0.27.1"
+strum_macros = "0.27.2"
 thiserror = "2.0.12"
 time = { version = "0.3", features = ["formatting", "local-offset", "macros"] }
 tokio = { version = "1", features = [
@@ -41,9 +41,9 @@ tokio = { version = "1", features = [
     "signal",
 ] }
 tokio-util = "0.7.14"
-toml = "0.9.1"
+toml = "0.9.2"
 tracing = { version = "0.1.41", features = ["log"] }
-tree-sitter = "0.25.3"
+tree-sitter = "0.25.8"
 tree-sitter-bash = "0.25.0"
 uuid = { version = "1", features = ["serde", "v4"] }
 wildmatch = "2.4.0"
@@ -62,6 +62,7 @@ openssl-sys = { version = "*", features = ["vendored"] }
 
 [dev-dependencies]
 assert_cmd = "2"
+core_test_support = { path = "tests/common" }
 maplit = "1.0.2"
 predicates = "3"
 pretty_assertions = "1.4.1"

codex-rs/core/src/bash.rs

@@ -0,0 +1,219 @@
+use tree_sitter::Parser;
+use tree_sitter::Tree;
+use tree_sitter_bash::LANGUAGE as BASH;
+
+/// Parse the provided bash source using tree-sitter-bash, returning a Tree on
+/// success or None if parsing failed.
+pub fn try_parse_bash(bash_lc_arg: &str) -> Option<Tree> {
+    let lang = BASH.into();
+    let mut parser = Parser::new();
+    #[expect(clippy::expect_used)]
+    parser.set_language(&lang).expect("load bash grammar");
+    let old_tree: Option<&Tree> = None;
+    parser.parse(bash_lc_arg, old_tree)
+}
+
+/// Parse a script which may contain multiple simple commands joined only by
+/// the safe logical/pipe/sequencing operators: `&&`, `||`, `;`, `|`.
+///
+/// Returns `Some(Vec<command_words>)` if every command is a plain word‑only
+/// command and the parse tree does not contain disallowed constructs
+/// (parentheses, redirections, substitutions, control flow, etc.). Otherwise
+/// returns `None`.
+pub fn try_parse_word_only_commands_sequence(tree: &Tree, src: &str) -> Option<Vec<Vec<String>>> {
+    if tree.root_node().has_error() {
+        return None;
+    }
+
+    // List of allowed (named) node kinds for a "word only commands sequence".
+    // If we encounter a named node that is not in this list we reject.
+    const ALLOWED_KINDS: &[&str] = &[
+        // top level containers
+        "program",
+        "list",
+        "pipeline",
+        // commands & words
+        "command",
+        "command_name",
+        "word",
+        "string",
+        "string_content",
+        "raw_string",
+        "number",
+    ];
+    // Allow only safe punctuation / operator tokens; anything else causes reject.
+    const ALLOWED_PUNCT_TOKENS: &[&str] = &["&&", "||", ";", "|", "\"", "'"];
+
+    let root = tree.root_node();
+    let mut cursor = root.walk();
+    let mut stack = vec![root];
+    let mut command_nodes = Vec::new();
+    while let Some(node) = stack.pop() {
+        let kind = node.kind();
+        if node.is_named() {
+            if !ALLOWED_KINDS.contains(&kind) {
+                return None;
+            }
+            if kind == "command" {
+                command_nodes.push(node);
+            }
+        } else {
+            // Reject any punctuation / operator tokens that are not explicitly allowed.
+            if kind.chars().any(|c| "&;|".contains(c)) && !ALLOWED_PUNCT_TOKENS.contains(&kind) {
+                return None;
+            }
+            if !(ALLOWED_PUNCT_TOKENS.contains(&kind) || kind.trim().is_empty()) {
+                // If it's a quote token or operator it's allowed above; we also allow whitespace tokens.
+                // Any other punctuation like parentheses, braces, redirects, backticks, etc are rejected.
+                return None;
+            }
+        }
+        for child in node.children(&mut cursor) {
+            stack.push(child);
+        }
+    }
+
+    let mut commands = Vec::new();
+    for node in command_nodes {
+        if let Some(words) = parse_plain_command_from_node(node, src) {
+            commands.push(words);
+        } else {
+            return None;
+        }
+    }
+    Some(commands)
+}
+
+fn parse_plain_command_from_node(cmd: tree_sitter::Node, src: &str) -> Option<Vec<String>> {
+    if cmd.kind() != "command" {
+        return None;
+    }
+    let mut words = Vec::new();
+    let mut cursor = cmd.walk();
+    for child in cmd.named_children(&mut cursor) {
+        match child.kind() {
+            "command_name" => {
+                let word_node = child.named_child(0)?;
+                if word_node.kind() != "word" {
+                    return None;
+                }
+                words.push(word_node.utf8_text(src.as_bytes()).ok()?.to_owned());
+            }
+            "word" | "number" => {
+                words.push(child.utf8_text(src.as_bytes()).ok()?.to_owned());
+            }
+            "string" => {
+                if child.child_count() == 3
+                    && child.child(0)?.kind() == "\""
+                    && child.child(1)?.kind() == "string_content"
+                    && child.child(2)?.kind() == "\""
+                {
+                    words.push(child.child(1)?.utf8_text(src.as_bytes()).ok()?.to_owned());
+                } else {
+                    return None;
+                }
+            }
+            "raw_string" => {
+                let raw_string = child.utf8_text(src.as_bytes()).ok()?;
+                let stripped = raw_string
+                    .strip_prefix('\'')
+                    .and_then(|s| s.strip_suffix('\''));
+                if let Some(s) = stripped {
+                    words.push(s.to_owned());
+                } else {
+                    return None;
+                }
+            }
+            _ => return None,
+        }
+    }
+    Some(words)
+}
+
+#[cfg(test)]
+mod tests {
+    #![allow(clippy::unwrap_used)]
+    use super::*;
+
+    fn parse_seq(src: &str) -> Option<Vec<Vec<String>>> {
+        let tree = try_parse_bash(src)?;
+        try_parse_word_only_commands_sequence(&tree, src)
+    }
+
+    #[test]
+    fn accepts_single_simple_command() {
+        let cmds = parse_seq("ls -1").unwrap();
+        assert_eq!(cmds, vec![vec!["ls".to_string(), "-1".to_string()]]);
+    }
+
+    #[test]
+    fn accepts_multiple_commands_with_allowed_operators() {
+        let src = "ls && pwd; echo 'hi there' | wc -l";
+        let cmds = parse_seq(src).unwrap();
+        let expected: Vec<Vec<String>> = vec![
+            vec!["wc".to_string(), "-l".to_string()],
+            vec!["echo".to_string(), "hi there".to_string()],
+            vec!["pwd".to_string()],
+            vec!["ls".to_string()],
+        ];
+        assert_eq!(cmds, expected);
+    }
+
+    #[test]
+    fn extracts_double_and_single_quoted_strings() {
+        let cmds = parse_seq("echo \"hello world\"").unwrap();
+        assert_eq!(
+            cmds,
+            vec![vec!["echo".to_string(), "hello world".to_string()]]
+        );
+
+        let cmds2 = parse_seq("echo 'hi there'").unwrap();
+        assert_eq!(
+            cmds2,
+            vec![vec!["echo".to_string(), "hi there".to_string()]]
+        );
+    }
+
+    #[test]
+    fn accepts_numbers_as_words() {
+        let cmds = parse_seq("echo 123 456").unwrap();
+        assert_eq!(
+            cmds,
+            vec![vec![
+                "echo".to_string(),
+                "123".to_string(),
+                "456".to_string()
+            ]]
+        );
+    }
+
+    #[test]
+    fn rejects_parentheses_and_subshells() {
+        assert!(parse_seq("(ls)").is_none());
+        assert!(parse_seq("ls || (pwd && echo hi)").is_none());
+    }
+
+    #[test]
+    fn rejects_redirections_and_unsupported_operators() {
+        assert!(parse_seq("ls > out.txt").is_none());
+        assert!(parse_seq("echo hi & echo bye").is_none());
+    }
+
+    #[test]
+    fn rejects_command_and_process_substitutions_and_expansions() {
+        assert!(parse_seq("echo $(pwd)").is_none());
+        assert!(parse_seq("echo `pwd`").is_none());
+        assert!(parse_seq("echo $HOME").is_none());
+        assert!(parse_seq("echo \"hi $USER\"").is_none());
+    }
+
+    #[test]
+    fn rejects_variable_assignment_prefix() {
+        assert!(parse_seq("FOO=bar ls").is_none());
+    }
+
+    #[test]
+    fn rejects_trailing_operator_parse_error() {
+        assert!(parse_seq("ls &&").is_none());
+    }
+}

codex-rs/core/src/chat_completions.rs

@@ -41,7 +41,7 @@ pub(crate) async fn stream_chat_completions(
 
     for item in &prompt.input {
         match item {
-            ResponseItem::Message { role, content } => {
+            ResponseItem::Message { role, content, .. } => {
                 let mut text = String::new();
                 for c in content {
                     match c {
@@ -58,6 +58,7 @@ pub(crate) async fn stream_chat_completions(
                 name,
                 arguments,
                 call_id,
+                ..
             } => {
                 messages.push(json!({
                     "role": "assistant",
@@ -259,6 +260,7 @@ async fn process_chat_sse<S>(
                     content: vec![ContentItem::OutputText {
                         text: content.to_string(),
                     }],
+                    id: None,
                 };
 
                 let _ = tx_event.send(Ok(ResponseEvent::OutputItemDone(item))).await;
@@ -300,6 +302,7 @@ async fn process_chat_sse<S>(
                     "tool_calls" if fn_call_state.active => {
                         // Build the FunctionCall response item.
                         let item = ResponseItem::FunctionCall {
+                            id: None,
                             name: fn_call_state.name.clone().unwrap_or_else(|| "".to_string()),
                             arguments: fn_call_state.arguments.clone(),
                             call_id: fn_call_state.call_id.clone().unwrap_or_else(String::new),
@@ -402,6 +405,7 @@ where
                 }))) => {
                     if !this.cumulative.is_empty() {
                         let aggregated_item = crate::models::ResponseItem::Message {
+                            id: None,
                             role: "assistant".to_string(),
                             content: vec![crate::models::ContentItem::OutputText {
                                 text: std::mem::take(&mut this.cumulative),

codex-rs/core/src/client.rs

@@ -117,6 +117,15 @@ impl ModelClient {
         let full_instructions = prompt.get_full_instructions(&self.config.model);
         let tools_json = create_tools_json_for_responses_api(prompt, &self.config.model)?;
         let reasoning = create_reasoning_param_for_request(&self.config, self.effort, self.summary);
+
+        // Request encrypted COT if we are not storing responses,
+        // otherwise reasoning items will be referenced by ID
+        let include = if !prompt.store && reasoning.is_some() {
+            vec!["reasoning.encrypted_content".to_string()]
+        } else {
+            vec![]
+        };
+
         let payload = ResponsesApiRequest {
             model: &self.config.model,
             instructions: &full_instructions,
@@ -125,10 +134,10 @@ impl ModelClient {
             tool_choice: "auto",
             parallel_tool_calls: false,
             reasoning,
-            previous_response_id: prompt.prev_id.clone(),
             store: prompt.store,
             // TODO: make this configurable
             stream: true,
+            include,
         };
 
         trace!(

codex-rs/core/src/client_common.rs

@@ -22,8 +22,6 @@ const BASE_INSTRUCTIONS: &str = include_str!("../prompt.md");
 pub struct Prompt {
     /// Conversation context input items.
     pub input: Vec<ResponseItem>,
-    /// Optional previous response ID (when storage is enabled).
-    pub prev_id: Option<String>,
     /// Optional instructions from the user to amend to the built-in agent
     /// instructions.
     pub user_instructions: Option<String>,
@@ -34,11 +32,18 @@ pub struct Prompt {
     /// the "fully qualified" tool name (i.e., prefixed with the server name),
     /// which should be reported to the model in place of Tool::name.
     pub extra_tools: HashMap<String, mcp_types::Tool>,
+
+    /// Optional override for the built-in BASE_INSTRUCTIONS.
+    pub base_instructions_override: Option<String>,
 }
 
 impl Prompt {
     pub(crate) fn get_full_instructions(&self, model: &str) -> Cow<'_, str> {
-        let mut sections: Vec<&str> = vec![BASE_INSTRUCTIONS];
+        let base = self
+            .base_instructions_override
+            .as_deref()
+            .unwrap_or(BASE_INSTRUCTIONS);
+        let mut sections: Vec<&str> = vec![base];
         if let Some(ref user) = self.user_instructions {
             sections.push(user);
         }
@@ -126,11 +131,10 @@ pub(crate) struct ResponsesApiRequest<'a> {
     pub(crate) tool_choice: &'static str,
     pub(crate) parallel_tool_calls: bool,
     pub(crate) reasoning: Option<Reasoning>,
-    #[serde(skip_serializing_if = "Option::is_none")]
-    pub(crate) previous_response_id: Option<String>,
     /// true when using the Responses API.
     pub(crate) store: bool,
     pub(crate) stream: bool,
+    pub(crate) include: Vec<String>,
 }
 
 use crate::config::Config;

codex-rs/core/src/codex.rs

@@ -34,7 +34,6 @@ use tracing::trace;
 use tracing::warn;
 use uuid::Uuid;
 
-use crate::WireApi;
 use crate::client::ModelClient;
 use crate::client_common::Prompt;
 use crate::client_common::ResponseEvent;
@@ -108,13 +107,15 @@ impl Codex {
         let (tx_sub, rx_sub) = async_channel::bounded(64);
         let (tx_event, rx_event) = async_channel::bounded(1600);
 
-        let instructions = get_user_instructions(&config).await;
+        let user_instructions = get_user_instructions(&config).await;
+
         let configure_session = Op::ConfigureSession {
             provider: config.model_provider.clone(),
             model: config.model.clone(),
             model_reasoning_effort: config.model_reasoning_effort,
             model_reasoning_summary: config.model_reasoning_summary,
-            instructions,
+            user_instructions,
+            base_instructions: config.base_instructions.clone(),
             approval_policy: config.approval_policy,
             sandbox_policy: config.sandbox_policy.clone(),
             disable_response_storage: config.disable_response_storage,
@@ -183,11 +184,13 @@ pub(crate) struct Session {
     /// the model as well as sandbox policies are resolved against this path
     /// instead of `std::env::current_dir()`.
     cwd: PathBuf,
-    instructions: Option<String>,
+    base_instructions: Option<String>,
+    user_instructions: Option<String>,
     approval_policy: AskForApproval,
     sandbox_policy: SandboxPolicy,
     shell_environment_policy: ShellEnvironmentPolicy,
     writable_roots: Mutex<Vec<PathBuf>>,
+    disable_response_storage: bool,
 
     /// Manager for external MCP servers/tools.
     mcp_connection_manager: McpConnectionManager,
@@ -216,13 +219,9 @@ impl Session {
 struct State {
     approved_commands: HashSet<Vec<String>>,
     current_task: Option<AgentTask>,
-    /// Call IDs that have been sent from the Responses API but have not been sent back yet.
-    /// You CANNOT send a Responses API follow-up message unless you have sent back the output for all pending calls or else it will 400.
-    pending_call_ids: HashSet<String>,
-    previous_response_id: Option<String>,
     pending_approvals: HashMap<String, oneshot::Sender<ReviewDecision>>,
     pending_input: Vec<ResponseInputItem>,
-    zdr_transcript: Option<ConversationHistory>,
+    history: ConversationHistory,
 }
 
 impl Session {
@@ -254,6 +253,7 @@ impl Session {
     pub async fn request_command_approval(
         &self,
         sub_id: String,
+        call_id: String,
         command: Vec<String>,
         cwd: PathBuf,
         reason: Option<String>,
@@ -262,6 +262,7 @@ impl Session {
         let event = Event {
             id: sub_id.clone(),
             msg: EventMsg::ExecApprovalRequest(ExecApprovalRequestEvent {
+                call_id,
                 command,
                 cwd,
                 reason,
@@ -278,6 +279,7 @@ impl Session {
     pub async fn request_patch_approval(
         &self,
         sub_id: String,
+        call_id: String,
         action: &ApplyPatchAction,
         reason: Option<String>,
         grant_root: Option<PathBuf>,
@@ -286,6 +288,7 @@ impl Session {
         let event = Event {
             id: sub_id.clone(),
             msg: EventMsg::ApplyPatchApprovalRequest(ApplyPatchApprovalRequestEvent {
+                call_id,
                 changes: convert_apply_patch_to_protocol(action),
                 reason,
                 grant_root,
@@ -317,18 +320,11 @@ impl Session {
         debug!("Recording items for conversation: {items:?}");
         self.record_state_snapshot(items).await;
 
-        if let Some(transcript) = self.state.lock().unwrap().zdr_transcript.as_mut() {
-            transcript.record_items(items);
-        }
+        self.state.lock().unwrap().history.record_items(items);
     }
 
     async fn record_state_snapshot(&self, items: &[ResponseItem]) {
-        let snapshot = {
-            let state = self.state.lock().unwrap();
-            crate::rollout::SessionStateSnapshot {
-                previous_response_id: state.previous_response_id.clone(),
-            }
-        };
+        let snapshot = { crate::rollout::SessionStateSnapshot {} };
 
         let recorder = {
             let guard = self.rollout.lock().unwrap();
@@ -430,8 +426,6 @@ impl Session {
     pub fn abort(&self) {
         info!("Aborting existing session");
         let mut state = self.state.lock().unwrap();
-        // Don't clear pending_call_ids because we need to keep track of them to ensure we don't 400 on the next turn.
-        // We will generate a synthetic aborted response for each pending call id.
         state.pending_approvals.clear();
         state.pending_input.clear();
         if let Some(task) = state.current_task.take() {
@@ -476,15 +470,10 @@ impl Drop for Session {
 }
 
 impl State {
-    pub fn partial_clone(&self, retain_zdr_transcript: bool) -> Self {
+    pub fn partial_clone(&self) -> Self {
         Self {
             approved_commands: self.approved_commands.clone(),
-            previous_response_id: self.previous_response_id.clone(),
-            zdr_transcript: if retain_zdr_transcript {
-                self.zdr_transcript.clone()
-            } else {
-                None
-            },
+            history: self.history.clone(),
             ..Default::default()
         }
     }
@@ -577,7 +566,8 @@ async fn submission_loop(
                 model,
                 model_reasoning_effort,
                 model_reasoning_summary,
-                instructions,
+                user_instructions,
+                base_instructions,
                 approval_policy,
                 sandbox_policy,
                 disable_response_storage,
@@ -602,13 +592,11 @@ async fn submission_loop(
                 }
                 // Optionally resume an existing rollout.
                 let mut restored_items: Option<Vec<ResponseItem>> = None;
-                let mut restored_prev_id: Option<String> = None;
                 let rollout_recorder: Option<RolloutRecorder> =
                     if let Some(path) = resume_path.as_ref() {
-                        match RolloutRecorder::resume(path).await {
+                        match RolloutRecorder::resume(path, cwd.clone()).await {
                             Ok((rec, saved)) => {
                                 session_id = saved.session_id;
-                                restored_prev_id = saved.state.previous_response_id;
                                 if !saved.items.is_empty() {
                                     restored_items = Some(saved.items);
                                 }
@@ -625,15 +613,17 @@ async fn submission_loop(
 
                 let rollout_recorder = match rollout_recorder {
                     Some(rec) => Some(rec),
-                    None => match RolloutRecorder::new(&config, session_id, instructions.clone())
-                        .await
-                    {
-                        Ok(r) => Some(r),
-                        Err(e) => {
-                            warn!("failed to initialise rollout recorder: {e}");
-                            None
+                    None => {
+                        match RolloutRecorder::new(&config, session_id, user_instructions.clone())
+                            .await
+                        {
+                            Ok(r) => Some(r),
+                            Err(e) => {
+                                warn!("failed to initialise rollout recorder: {e}");
+                                None
+                            }
                         }
-                    },
+                    }
                 };
 
                 let client = ModelClient::new(
@@ -645,22 +635,13 @@ async fn submission_loop(
                 );
 
                 // abort any current running session and clone its state
-                let retain_zdr_transcript =
-                    record_conversation_history(disable_response_storage, provider.wire_api);
                 let state = match sess.take() {
                     Some(sess) => {
                         sess.abort();
-                        sess.state
-                            .lock()
-                            .unwrap()
-                            .partial_clone(retain_zdr_transcript)
+                        sess.state.lock().unwrap().partial_clone()
                     }
                     None => State {
-                        zdr_transcript: if retain_zdr_transcript {
-                            Some(ConversationHistory::new())
-                        } else {
-                            None
-                        },
+                        history: ConversationHistory::new(),
                         ..Default::default()
                     },
                 };
@@ -699,7 +680,8 @@ async fn submission_loop(
                     client,
                     tx_event: tx_event.clone(),
                     ctrl_c: Arc::clone(&ctrl_c),
-                    instructions,
+                    user_instructions,
+                    base_instructions,
                     approval_policy,
                     sandbox_policy,
                     shell_environment_policy: config.shell_environment_policy.clone(),
@@ -710,18 +692,14 @@ async fn submission_loop(
                     state: Mutex::new(state),
                     rollout: Mutex::new(rollout_recorder),
                     codex_linux_sandbox_exe: config.codex_linux_sandbox_exe.clone(),
+                    disable_response_storage,
                 }));
 
                 // Patch restored state into the newly created session.
                 if let Some(sess_arc) = &sess {
-                    if restored_prev_id.is_some() || restored_items.is_some() {
+                    if restored_items.is_some() {
                         let mut st = sess_arc.state.lock().unwrap();
-                        st.previous_response_id = restored_prev_id;
-                        if let (Some(hist), Some(items)) =
-                            (st.zdr_transcript.as_mut(), restored_items.as_ref())
-                        {
-                            hist.record_items(items.iter());
-                        }
+                        st.history.record_items(restored_items.unwrap().iter());
                     }
                 }
 
@@ -834,6 +812,37 @@ async fn submission_loop(
                     }
                 });
             }
+            Op::Shutdown => {
+                info!("Shutting down Codex instance");
+
+                // Gracefully flush and shutdown rollout recorder on session end so tests
+                // that inspect the rollout file do not race with the background writer.
+                if let Some(sess_arc) = sess {
+                    let recorder_opt = sess_arc.rollout.lock().unwrap().take();
+                    if let Some(rec) = recorder_opt {
+                        if let Err(e) = rec.shutdown().await {
+                            warn!("failed to shutdown rollout recorder: {e}");
+                            let event = Event {
+                                id: sub.id.clone(),
+                                msg: EventMsg::Error(ErrorEvent {
+                                    message: "Failed to shutdown rollout recorder".to_string(),
+                                }),
+                            };
+                            if let Err(e) = tx_event.send(event).await {
+                                warn!("failed to send error message: {e:?}");
+                            }
+                        }
+                    }
+                }
+                let event = Event {
+                    id: sub.id.clone(),
+                    msg: EventMsg::ShutdownComplete,
+                };
+                if let Err(e) = tx_event.send(event).await {
+                    warn!("failed to send Shutdown event: {e}");
+                }
+                break;
+            }
         }
     }
     debug!("Agent loop exited");
@@ -868,14 +877,8 @@ async fn run_task(sess: Arc<Session>, sub_id: String, input: Vec<InputItem>) {
     sess.record_conversation_items(&[initial_input_for_turn.clone().into()])
         .await;
 
-    let mut input_for_next_turn: Vec<ResponseInputItem> = vec![initial_input_for_turn];
     let last_agent_message: Option<String>;
     loop {
-        let mut net_new_turn_input = input_for_next_turn
-            .drain(..)
-            .map(ResponseItem::from)
-            .collect::<Vec<_>>();
-
         // Note that pending_input would be something like a message the user
         // submitted through the UI while the model was running. Though the UI
         // may support this, the model might not.
@@ -892,29 +895,7 @@ async fn run_task(sess: Arc<Session>, sub_id: String, input: Vec<InputItem>) {
         // 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> =
-            if let Some(transcript) = sess.state.lock().unwrap().zdr_transcript.as_mut() {
-                // If we are using Chat/ZDR, we need to send the transcript with
-                // every turn. By induction, `transcript` already contains:
-                // - The `input` that kicked off this task.
-                // - Each `ResponseItem` that was recorded in the previous turn.
-                // - Each response to a `ResponseItem` (in practice, the only
-                //   response type we seem to have is `FunctionCallOutput`).
-                //
-                // The only thing the `transcript` does not contain is the
-                // `pending_input` that was injected while the model was
-                // running. We need to add that to the conversation history
-                // so that the model can see it in the next turn.
-                [transcript.contents(), pending_input].concat()
-            } else {
-                // In practice, net_new_turn_input should contain only:
-                // - User messages
-                // - Outputs for function calls requested by the model
-                net_new_turn_input.extend(pending_input);
-
-                // Responses API path – we can just send the new items and
-                // record the same.
-                net_new_turn_input
-            };
+            [sess.state.lock().unwrap().history.contents(), pending_input].concat();
 
         let turn_input_messages: Vec<String> = turn_input
             .iter()
@@ -990,8 +971,19 @@ async fn run_task(sess: Arc<Session>, sub_id: String, input: Vec<InputItem>) {
                                 },
                             );
                         }
-                        (ResponseItem::Reasoning { .. }, None) => {
-                            // Omit from conversation history.
+                        (
+                            ResponseItem::Reasoning {
+                                id,
+                                summary,
+                                encrypted_content,
+                            },
+                            None,
+                        ) => {
+                            items_to_record_in_conversation_history.push(ResponseItem::Reasoning {
+                                id: id.clone(),
+                                summary: summary.clone(),
+                                encrypted_content: encrypted_content.clone(),
+                            });
                         }
                         _ => {
                             warn!("Unexpected response item: {item:?} with response: {response:?}");
@@ -1020,8 +1012,6 @@ async fn run_task(sess: Arc<Session>, sub_id: String, input: Vec<InputItem>) {
                     });
                     break;
                 }
-
-                input_for_next_turn = responses;
             }
             Err(e) => {
                 info!("Turn error: {e:#}");
@@ -1049,27 +1039,13 @@ async fn run_turn(
     sub_id: String,
     input: Vec<ResponseItem>,
 ) -> CodexResult<Vec<ProcessedResponseItem>> {
-    // Decide whether to use server-side storage (previous_response_id) or disable it
-    let (prev_id, store) = {
-        let state = sess.state.lock().unwrap();
-        let store = state.zdr_transcript.is_none();
-        let prev_id = if store {
-            state.previous_response_id.clone()
-        } else {
-            // When using ZDR, the Responses API may send previous_response_id
-            // back, but trying to use it results in a 400.
-            None
-        };
-        (prev_id, store)
-    };
-
     let extra_tools = sess.mcp_connection_manager.list_all_tools();
     let prompt = Prompt {
         input,
-        prev_id,
-        user_instructions: sess.instructions.clone(),
-        store,
+        user_instructions: sess.user_instructions.clone(),
+        store: !sess.disable_response_storage,
         extra_tools,
+        base_instructions_override: sess.base_instructions.clone(),
     };
 
     let mut retries = 0;
@@ -1141,11 +1117,17 @@ async fn try_run_turn(
     // This usually happens because the user interrupted the model before we responded to one of its tool calls
     // and then the user sent a follow-up message.
     let missing_calls = {
-        sess.state
-            .lock()
-            .unwrap()
-            .pending_call_ids
+        prompt
+            .input
             .iter()
+            .filter_map(|ri| match ri {
+                ResponseItem::FunctionCall { call_id, .. } => Some(call_id),
+                ResponseItem::LocalShellCall {
+                    call_id: Some(call_id),
+                    ..
+                } => Some(call_id),
+                _ => None,
+            })
             .filter_map(|call_id| {
                 if completed_call_ids.contains(&call_id) {
                     None
@@ -1199,31 +1181,14 @@ async fn try_run_turn(
         };
 
         match event {
-            ResponseEvent::Created => {
-                let mut state = sess.state.lock().unwrap();
-                // We successfully created a new response and ensured that all pending calls were included so we can clear the pending call ids.
-                state.pending_call_ids.clear();
-            }
+            ResponseEvent::Created => {}
             ResponseEvent::OutputItemDone(item) => {
-                let call_id = match &item {
-                    ResponseItem::LocalShellCall {
-                        call_id: Some(call_id),
-                        ..
-                    } => Some(call_id),
-                    ResponseItem::FunctionCall { call_id, .. } => Some(call_id),
-                    _ => None,
-                };
-                if let Some(call_id) = call_id {
-                    // We just got a new call id so we need to make sure to respond to it in the next turn.
-                    let mut state = sess.state.lock().unwrap();
-                    state.pending_call_ids.insert(call_id.clone());
-                }
                 let response = handle_response_item(sess, sub_id, item.clone()).await?;
 
                 output.push(ProcessedResponseItem { item, response });
             }
             ResponseEvent::Completed {
-                response_id,
+                response_id: _,
                 token_usage,
             } => {
                 if let Some(token_usage) = token_usage {
@@ -1236,8 +1201,6 @@ async fn try_run_turn(
                         .ok();
                 }
 
-                let mut state = sess.state.lock().unwrap();
-                state.previous_response_id = Some(response_id);
                 return Ok(output);
             }
             ResponseEvent::OutputTextDelta(delta) => {
@@ -1277,7 +1240,7 @@ async fn handle_response_item(
             }
             None
         }
-        ResponseItem::Reasoning { id: _, summary } => {
+        ResponseItem::Reasoning { summary, .. } => {
             for item in summary {
                 let text = match item {
                     ReasoningItemReasoningSummary::SummaryText { text } => text,
@@ -1294,6 +1257,7 @@ async fn handle_response_item(
             name,
             arguments,
             call_id,
+            ..
         } => {
             info!("FunctionCall: {arguments}");
             Some(handle_function_call(sess, sub_id.to_string(), name, arguments, call_id).await)
@@ -1464,6 +1428,7 @@ async fn handle_container_exec_with_params(
             let rx_approve = sess
                 .request_command_approval(
                     sub_id.clone(),
+                    call_id.clone(),
                     params.command.clone(),
                     params.cwd.clone(),
                     None,
@@ -1591,6 +1556,7 @@ async fn handle_sandbox_error(
     let rx_approve = sess
         .request_command_approval(
             sub_id.clone(),
+            call_id.clone(),
             params.command.clone(),
             params.cwd.clone(),
             Some("command failed; retry without sandbox?".to_string()),
@@ -1608,9 +1574,7 @@ async fn handle_sandbox_error(
             sess.notify_background_event(&sub_id, "retrying command without sandbox")
                 .await;
 
-            // Emit a fresh Begin event so progress bars reset.
-            let retry_call_id = format!("{call_id}-retry");
-            sess.notify_exec_command_begin(&sub_id, &retry_call_id, &params)
+            sess.notify_exec_command_begin(&sub_id, &call_id, &params)
                 .await;
 
             // This is an escalated retry; the policy will not be
@@ -1633,14 +1597,8 @@ async fn handle_sandbox_error(
                         duration,
                     } = retry_output;
 
-                    sess.notify_exec_command_end(
-                        &sub_id,
-                        &retry_call_id,
-                        &stdout,
-                        &stderr,
-                        exit_code,
-                    )
-                    .await;
+                    sess.notify_exec_command_end(&sub_id, &call_id, &stdout, &stderr, exit_code)
+                        .await;
 
                     let is_success = exit_code == 0;
                     let content = format_exec_output(
@@ -1704,7 +1662,7 @@ async fn apply_patch(
             // Compute a readable summary of path changes to include in the
             // approval request so the user can make an informed decision.
             let rx_approve = sess
-                .request_patch_approval(sub_id.clone(), &action, None, None)
+                .request_patch_approval(sub_id.clone(), call_id.clone(), &action, None, None)
                 .await;
             match rx_approve.await.unwrap_or_default() {
                 ReviewDecision::Approved | ReviewDecision::ApprovedForSession => false,
@@ -1742,7 +1700,13 @@ async fn apply_patch(
         ));
 
         let rx = sess
-            .request_patch_approval(sub_id.clone(), &action, reason.clone(), Some(root.clone()))
+            .request_patch_approval(
+                sub_id.clone(),
+                call_id.clone(),
+                &action,
+                reason.clone(),
+                Some(root.clone()),
+            )
             .await;
 
         if !matches!(
@@ -1826,6 +1790,7 @@ async fn apply_patch(
                 let rx = sess
                     .request_patch_approval(
                         sub_id.clone(),
+                        call_id.clone(),
                         &action,
                         reason.clone(),
                         Some(root.clone()),
@@ -2084,7 +2049,7 @@ fn format_exec_output(output: &str, exit_code: i32, duration: Duration) -> Strin
 
 fn get_last_assistant_message_from_turn(responses: &[ResponseItem]) -> Option<String> {
     responses.iter().rev().find_map(|item| {
-        if let ResponseItem::Message { role, content } = item {
+        if let ResponseItem::Message { role, content, .. } = item {
             if role == "assistant" {
                 content.iter().rev().find_map(|ci| {
                     if let ContentItem::OutputText { text } = ci {
@@ -2101,15 +2066,3 @@ fn get_last_assistant_message_from_turn(responses: &[ResponseItem]) -> Option<St
         }
     })
 }
-
-/// See [`ConversationHistory`] for details.
-fn record_conversation_history(disable_response_storage: bool, wire_api: WireApi) -> bool {
-    if disable_response_storage {
-        return true;
-    }
-
-    match wire_api {
-        WireApi::Responses => false,
-        WireApi::Chat => true,
-    }
-}

codex-rs/core/src/config.rs

@@ -63,7 +63,10 @@ pub struct Config {
     pub disable_response_storage: bool,
 
     /// User-provided instructions from instructions.md.
-    pub instructions: Option<String>,
+    pub user_instructions: Option<String>,
+
+    /// Base instructions override.
+    pub base_instructions: Option<String>,
 
     /// Optional external notifier command. When set, Codex will spawn this
     /// program after each completed *turn* (i.e. when the agent finishes
@@ -327,6 +330,9 @@ pub struct ConfigToml {
 
     /// Experimental rollout resume path (absolute path to .jsonl; undocumented).
     pub experimental_resume: Option<PathBuf>,
+
+    /// Experimental path to a file whose contents replace the built-in BASE_INSTRUCTIONS.
+    pub experimental_instructions_file: Option<PathBuf>,
 }
 
 impl ConfigToml {
@@ -359,6 +365,7 @@ pub struct ConfigOverrides {
     pub model_provider: Option<String>,
     pub config_profile: Option<String>,
     pub codex_linux_sandbox_exe: Option<PathBuf>,
+    pub base_instructions: Option<String>,
 }
 
 impl Config {
@@ -369,7 +376,7 @@ impl Config {
         overrides: ConfigOverrides,
         codex_home: PathBuf,
     ) -> std::io::Result<Self> {
-        let instructions = Self::load_instructions(Some(&codex_home));
+        let user_instructions = Self::load_instructions(Some(&codex_home));
 
         // Destructure ConfigOverrides fully to ensure all overrides are applied.
         let ConfigOverrides {
@@ -380,6 +387,7 @@ impl Config {
             model_provider,
             config_profile: config_profile_key,
             codex_linux_sandbox_exe,
+            base_instructions,
         } = overrides;
 
         let config_profile = match config_profile_key.as_ref().or(cfg.profile.as_ref()) {
@@ -457,6 +465,10 @@ impl Config {
 
         let experimental_resume = cfg.experimental_resume;
 
+        let base_instructions = base_instructions.or(Self::get_base_instructions(
+            cfg.experimental_instructions_file.as_ref(),
+        ));
+
         let config = Self {
             model,
             model_context_window,
@@ -475,7 +487,8 @@ impl Config {
                 .or(cfg.disable_response_storage)
                 .unwrap_or(false),
             notify: cfg.notify,
-            instructions,
+            user_instructions,
+            base_instructions,
             mcp_servers: cfg.mcp_servers,
             model_providers,
             project_doc_max_bytes: cfg.project_doc_max_bytes.unwrap_or(PROJECT_DOC_MAX_BYTES),
@@ -525,6 +538,15 @@ impl Config {
             }
         })
     }
+
+    fn get_base_instructions(path: Option<&PathBuf>) -> Option<String> {
+        let path = path.as_ref()?;
+
+        std::fs::read_to_string(path)
+            .ok()
+            .map(|s| s.trim().to_string())
+            .filter(|s| !s.is_empty())
+    }
 }
 
 fn default_model() -> String {
@@ -539,7 +561,7 @@ fn default_model() -> String {
 ///   function will Err if the path does not exist.
 /// - If `CODEX_HOME` is not set, this function does not verify that the
 ///   directory exists.
-fn find_codex_home() -> std::io::Result<PathBuf> {
+pub fn find_codex_home() -> std::io::Result<PathBuf> {
     // Honor the `CODEX_HOME` environment variable when it is set to allow users
     // (and tests) to override the default location.
     if let Ok(val) = std::env::var("CODEX_HOME") {
@@ -801,7 +823,7 @@ disable_response_storage = true
                 sandbox_policy: SandboxPolicy::new_read_only_policy(),
                 shell_environment_policy: ShellEnvironmentPolicy::default(),
                 disable_response_storage: false,
-                instructions: None,
+                user_instructions: None,
                 notify: None,
                 cwd: fixture.cwd(),
                 mcp_servers: HashMap::new(),
@@ -818,6 +840,7 @@ disable_response_storage = true
                 model_supports_reasoning_summaries: false,
                 chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
                 experimental_resume: None,
+                base_instructions: None,
             },
             o3_profile_config
         );
@@ -848,7 +871,7 @@ disable_response_storage = true
             sandbox_policy: SandboxPolicy::new_read_only_policy(),
             shell_environment_policy: ShellEnvironmentPolicy::default(),
             disable_response_storage: false,
-            instructions: None,
+            user_instructions: None,
             notify: None,
             cwd: fixture.cwd(),
             mcp_servers: HashMap::new(),
@@ -865,6 +888,7 @@ disable_response_storage = true
             model_supports_reasoning_summaries: false,
             chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
             experimental_resume: None,
+            base_instructions: None,
         };
 
         assert_eq!(expected_gpt3_profile_config, gpt3_profile_config);
@@ -910,7 +934,7 @@ disable_response_storage = true
             sandbox_policy: SandboxPolicy::new_read_only_policy(),
             shell_environment_policy: ShellEnvironmentPolicy::default(),
             disable_response_storage: true,
-            instructions: None,
+            user_instructions: None,
             notify: None,
             cwd: fixture.cwd(),
             mcp_servers: HashMap::new(),
@@ -927,6 +951,7 @@ disable_response_storage = true
             model_supports_reasoning_summaries: false,
             chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
             experimental_resume: None,
+            base_instructions: None,
         };
 
         assert_eq!(expected_zdr_profile_config, zdr_profile_config);

codex-rs/core/src/config_types.rs

@@ -76,20 +76,7 @@ pub enum HistoryPersistence {
 
 /// Collection of settings that are specific to the TUI.
 #[derive(Deserialize, Debug, Clone, PartialEq, Default)]
-pub struct Tui {
-    /// By default, mouse capture is enabled in the TUI so that it is possible
-    /// to scroll the conversation history with a mouse. This comes at the cost
-    /// of not being able to use the mouse to select text in the TUI.
-    /// (Most terminals support a modifier key to allow this. For example,
-    /// text selection works in iTerm if you hold down the `Option` key while
-    /// clicking and dragging.)
-    ///
-    /// Setting this option to `true` disables mouse capture, so scrolling with
-    /// the mouse is not possible, though the keyboard shortcuts e.g. `b` and
-    /// `space` still work. This allows the user to select text in the TUI
-    /// using the mouse without needing to hold down a modifier key.
-    pub disable_mouse_capture: bool,
-}
+pub struct Tui {}
 
 #[derive(Deserialize, Debug, Clone, Copy, PartialEq, Default)]
 #[serde(rename_all = "kebab-case")]

codex-rs/core/src/conversation_history.rs

@@ -1,12 +1,7 @@
 use crate::models::ResponseItem;
 
-/// Transcript of conversation history that is needed:
-/// - for ZDR clients for which previous_response_id is not available, so we
-///   must include the transcript with every API call. This must include each
-///   `function_call` and its corresponding `function_call_output`.
-/// - for clients using the "chat completions" API as opposed to the
-///   "responses" API.
-#[derive(Debug, Clone)]
+/// Transcript of conversation history
+#[derive(Debug, Clone, Default)]
 pub(crate) struct ConversationHistory {
     /// The oldest items are at the beginning of the vector.
     items: Vec<ResponseItem>,
@@ -44,7 +39,8 @@ fn is_api_message(message: &ResponseItem) -> bool {
         ResponseItem::Message { role, .. } => role.as_str() != "system",
         ResponseItem::FunctionCallOutput { .. }
         | ResponseItem::FunctionCall { .. }
-        | ResponseItem::LocalShellCall { .. } => true,
-        ResponseItem::Reasoning { .. } | ResponseItem::Other => false,
+        | ResponseItem::LocalShellCall { .. }
+        | ResponseItem::Reasoning { .. } => true,
+        ResponseItem::Other => false,
     }
 }

codex-rs/core/src/git_info.rs

@@ -0,0 +1,307 @@
+use std::path::Path;
+
+use serde::Deserialize;
+use serde::Serialize;
+use tokio::process::Command;
+use tokio::time::Duration as TokioDuration;
+use tokio::time::timeout;
+
+/// Timeout for git commands to prevent freezing on large repositories
+const GIT_COMMAND_TIMEOUT: TokioDuration = TokioDuration::from_secs(5);
+
+#[derive(Serialize, Deserialize, Clone)]
+pub struct GitInfo {
+    /// Current commit hash (SHA)
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub commit_hash: Option<String>,
+    /// Current branch name
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub branch: Option<String>,
+    /// Repository URL (if available from remote)
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub repository_url: Option<String>,
+}
+
+/// Collect git repository information from the given working directory using command-line git.
+/// Returns None if no git repository is found or if git operations fail.
+/// Uses timeouts to prevent freezing on large repositories.
+/// All git commands (except the initial repo check) run in parallel for better performance.
+pub async fn collect_git_info(cwd: &Path) -> Option<GitInfo> {
+    // Check if we're in a git repository first
+    let is_git_repo = run_git_command_with_timeout(&["rev-parse", "--git-dir"], cwd)
+        .await?
+        .status
+        .success();
+
+    if !is_git_repo {
+        return None;
+    }
+
+    // Run all git info collection commands in parallel
+    let (commit_result, branch_result, url_result) = tokio::join!(
+        run_git_command_with_timeout(&["rev-parse", "HEAD"], cwd),
+        run_git_command_with_timeout(&["rev-parse", "--abbrev-ref", "HEAD"], cwd),
+        run_git_command_with_timeout(&["remote", "get-url", "origin"], cwd)
+    );
+
+    let mut git_info = GitInfo {
+        commit_hash: None,
+        branch: None,
+        repository_url: None,
+    };
+
+    // Process commit hash
+    if let Some(output) = commit_result {
+        if output.status.success() {
+            if let Ok(hash) = String::from_utf8(output.stdout) {
+                git_info.commit_hash = Some(hash.trim().to_string());
+            }
+        }
+    }
+
+    // Process branch name
+    if let Some(output) = branch_result {
+        if output.status.success() {
+            if let Ok(branch) = String::from_utf8(output.stdout) {
+                let branch = branch.trim();
+                if branch != "HEAD" {
+                    git_info.branch = Some(branch.to_string());
+                }
+            }
+        }
+    }
+
+    // Process repository URL
+    if let Some(output) = url_result {
+        if output.status.success() {
+            if let Ok(url) = String::from_utf8(output.stdout) {
+                git_info.repository_url = Some(url.trim().to_string());
+            }
+        }
+    }
+
+    Some(git_info)
+}
+
+/// Run a git command with a timeout to prevent blocking on large repositories
+async fn run_git_command_with_timeout(args: &[&str], cwd: &Path) -> Option<std::process::Output> {
+    let result = timeout(
+        GIT_COMMAND_TIMEOUT,
+        Command::new("git").args(args).current_dir(cwd).output(),
+    )
+    .await;
+
+    match result {
+        Ok(Ok(output)) => Some(output),
+        _ => None, // Timeout or error
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    #![allow(clippy::expect_used)]
+    #![allow(clippy::unwrap_used)]
+
+    use super::*;
+
+    use std::fs;
+    use std::path::PathBuf;
+    use tempfile::TempDir;
+
+    // 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();
+
+        // Initialize git repo
+        Command::new("git")
+            .args(["init"])
+            .current_dir(&repo_path)
+            .output()
+            .await
+            .expect("Failed to init git repo");
+
+        // Configure git user (required for commits)
+        Command::new("git")
+            .args(["config", "user.name", "Test User"])
+            .current_dir(&repo_path)
+            .output()
+            .await
+            .expect("Failed to set git user name");
+
+        Command::new("git")
+            .args(["config", "user.email", "test@example.com"])
+            .current_dir(&repo_path)
+            .output()
+            .await
+            .expect("Failed to set git user email");
+
+        // Create a test file and commit it
+        let test_file = repo_path.join("test.txt");
+        fs::write(&test_file, "test content").expect("Failed to write test file");
+
+        Command::new("git")
+            .args(["add", "."])
+            .current_dir(&repo_path)
+            .output()
+            .await
+            .expect("Failed to add files");
+
+        Command::new("git")
+            .args(["commit", "-m", "Initial commit"])
+            .current_dir(&repo_path)
+            .output()
+            .await
+            .expect("Failed to commit");
+
+        repo_path
+    }
+
+    #[tokio::test]
+    async fn test_collect_git_info_non_git_directory() {
+        let temp_dir = TempDir::new().expect("Failed to create temp dir");
+        let result = collect_git_info(temp_dir.path()).await;
+        assert!(result.is_none());
+    }
+
+    #[tokio::test]
+    async fn test_collect_git_info_git_repository() {
+        let temp_dir = TempDir::new().expect("Failed to create temp dir");
+        let repo_path = create_test_git_repo(&temp_dir).await;
+
+        let git_info = collect_git_info(&repo_path)
+            .await
+            .expect("Should collect git info from repo");
+
+        // Should have commit hash
+        assert!(git_info.commit_hash.is_some());
+        let commit_hash = git_info.commit_hash.unwrap();
+        assert_eq!(commit_hash.len(), 40); // SHA-1 hash should be 40 characters
+        assert!(commit_hash.chars().all(|c| c.is_ascii_hexdigit()));
+
+        // Should have branch (likely "main" or "master")
+        assert!(git_info.branch.is_some());
+        let branch = git_info.branch.unwrap();
+        assert!(branch == "main" || branch == "master");
+
+        // Repository URL might be None for local repos without remote
+        // This is acceptable behavior
+    }
+
+    #[tokio::test]
+    async fn test_collect_git_info_with_remote() {
+        let temp_dir = TempDir::new().expect("Failed to create temp dir");
+        let repo_path = create_test_git_repo(&temp_dir).await;
+
+        // Add a remote origin
+        Command::new("git")
+            .args([
+                "remote",
+                "add",
+                "origin",
+                "https://github.com/example/repo.git",
+            ])
+            .current_dir(&repo_path)
+            .output()
+            .await
+            .expect("Failed to add remote");
+
+        let git_info = collect_git_info(&repo_path)
+            .await
+            .expect("Should collect git info from repo");
+
+        // Should have repository URL
+        assert_eq!(
+            git_info.repository_url,
+            Some("https://github.com/example/repo.git".to_string())
+        );
+    }
+
+    #[tokio::test]
+    async fn test_collect_git_info_detached_head() {
+        let temp_dir = TempDir::new().expect("Failed to create temp dir");
+        let repo_path = create_test_git_repo(&temp_dir).await;
+
+        // Get the current commit hash
+        let output = Command::new("git")
+            .args(["rev-parse", "HEAD"])
+            .current_dir(&repo_path)
+            .output()
+            .await
+            .expect("Failed to get HEAD");
+        let commit_hash = String::from_utf8(output.stdout).unwrap().trim().to_string();
+
+        // Checkout the commit directly (detached HEAD)
+        Command::new("git")
+            .args(["checkout", &commit_hash])
+            .current_dir(&repo_path)
+            .output()
+            .await
+            .expect("Failed to checkout commit");
+
+        let git_info = collect_git_info(&repo_path)
+            .await
+            .expect("Should collect git info from repo");
+
+        // Should have commit hash
+        assert!(git_info.commit_hash.is_some());
+        // Branch should be None for detached HEAD (since rev-parse --abbrev-ref HEAD returns "HEAD")
+        assert!(git_info.branch.is_none());
+    }
+
+    #[tokio::test]
+    async fn test_collect_git_info_with_branch() {
+        let temp_dir = TempDir::new().expect("Failed to create temp dir");
+        let repo_path = create_test_git_repo(&temp_dir).await;
+
+        // Create and checkout a new branch
+        Command::new("git")
+            .args(["checkout", "-b", "feature-branch"])
+            .current_dir(&repo_path)
+            .output()
+            .await
+            .expect("Failed to create branch");
+
+        let git_info = collect_git_info(&repo_path)
+            .await
+            .expect("Should collect git info from repo");
+
+        // Should have the new branch name
+        assert_eq!(git_info.branch, Some("feature-branch".to_string()));
+    }
+
+    #[test]
+    fn test_git_info_serialization() {
+        let git_info = GitInfo {
+            commit_hash: Some("abc123def456".to_string()),
+            branch: Some("main".to_string()),
+            repository_url: Some("https://github.com/example/repo.git".to_string()),
+        };
+
+        let json = serde_json::to_string(&git_info).expect("Should serialize GitInfo");
+        let parsed: serde_json::Value = serde_json::from_str(&json).expect("Should parse JSON");
+
+        assert_eq!(parsed["commit_hash"], "abc123def456");
+        assert_eq!(parsed["branch"], "main");
+        assert_eq!(
+            parsed["repository_url"],
+            "https://github.com/example/repo.git"
+        );
+    }
+
+    #[test]
+    fn test_git_info_serialization_with_nones() {
+        let git_info = GitInfo {
+            commit_hash: None,
+            branch: None,
+            repository_url: None,
+        };
+
+        let json = serde_json::to_string(&git_info).expect("Should serialize GitInfo");
+        let parsed: serde_json::Value = serde_json::from_str(&json).expect("Should parse JSON");
+
+        // Fields with None values should be omitted due to skip_serializing_if
+        assert!(!parsed.as_object().unwrap().contains_key("commit_hash"));
+        assert!(!parsed.as_object().unwrap().contains_key("branch"));
+        assert!(!parsed.as_object().unwrap().contains_key("repository_url"));
+    }
+}

codex-rs/core/src/is_safe_command.rs

@@ -1,31 +1,57 @@
-use tree_sitter::Parser;
-use tree_sitter::Tree;
-use tree_sitter_bash::LANGUAGE as BASH;
+use crate::bash::try_parse_bash;
+use crate::bash::try_parse_word_only_commands_sequence;
 
 pub fn is_known_safe_command(command: &[String]) -> bool {
     if is_safe_to_call_with_exec(command) {
         return true;
     }
 
-    // TODO(mbolin): Also support safe commands that are piped together such
-    // as `cat foo | wc -l`.
-    matches!(
-        command,
-        [bash, flag, script]
-            if bash == "bash"
-            && flag == "-lc"
-            && try_parse_bash(script).and_then(|tree|
-                try_parse_single_word_only_command(&tree, script)).is_some_and(|parsed_bash_command| is_safe_to_call_with_exec(&parsed_bash_command))
-    )
+    // Support `bash -lc "..."` where the script consists solely of one or
+    // more "plain" commands (only bare words / quoted strings) combined with
+    // a conservative allow‑list of shell operators that themselves do not
+    // introduce side effects ( "&&", "||", ";", and "|" ). If every
+    // individual command in the script is itself a known‑safe command, then
+    // the composite expression is considered safe.
+    if let [bash, flag, script] = command {
+        if bash == "bash" && flag == "-lc" {
+            if let Some(tree) = try_parse_bash(script) {
+                if let Some(all_commands) = try_parse_word_only_commands_sequence(&tree, script) {
+                    if !all_commands.is_empty()
+                        && all_commands
+                            .iter()
+                            .all(|cmd| is_safe_to_call_with_exec(cmd))
+                    {
+                        return true;
+                    }
+                }
+            }
+        }
+    }
+
+    false
 }
 
 fn is_safe_to_call_with_exec(command: &[String]) -> bool {
     let cmd0 = command.first().map(String::as_str);
 
     match cmd0 {
-        Some("cat" | "cd" | "echo" | "grep" | "head" | "ls" | "pwd" | "tail" | "wc" | "which") => {
+        #[rustfmt::skip]
+        Some(
+            "cat" |
+            "cd" |
+            "echo" |
+            "false" |
+            "grep" |
+            "head" |
+            "ls" |
+            "nl" |
+            "pwd" |
+            "tail" |
+            "true" |
+            "wc" |
+            "which") => {
             true
-        }
+        },
 
         Some("find") => {
             // Certain options to `find` can delete files, write to files, or
@@ -95,90 +121,7 @@ fn is_safe_to_call_with_exec(command: &[String]) -> bool {
     }
 }
 
-fn try_parse_bash(bash_lc_arg: &str) -> Option<Tree> {
-    let lang = BASH.into();
-    let mut parser = Parser::new();
-    #[expect(clippy::expect_used)]
-    parser.set_language(&lang).expect("load bash grammar");
-
-    let old_tree: Option<&Tree> = None;
-    parser.parse(bash_lc_arg, old_tree)
-}
-
-/// If `tree` represents a single Bash command whose name and every argument is
-/// an ordinary `word`, return those words in order; otherwise, return `None`.
-///
-/// `src` must be the exact source string that was parsed into `tree`, so we can
-/// extract the text for every node.
-pub fn try_parse_single_word_only_command(tree: &Tree, src: &str) -> Option<Vec<String>> {
-    // Any parse error is an immediate rejection.
-    if tree.root_node().has_error() {
-        return None;
-    }
-
-    // (program …) with exactly one statement
-    let root = tree.root_node();
-    if root.kind() != "program" || root.named_child_count() != 1 {
-        return None;
-    }
-
-    let cmd = root.named_child(0)?; // (command …)
-    if cmd.kind() != "command" {
-        return None;
-    }
-
-    let mut words = Vec::new();
-    let mut cursor = cmd.walk();
-
-    for child in cmd.named_children(&mut cursor) {
-        match child.kind() {
-            // The command name node wraps one `word` child.
-            "command_name" => {
-                let word_node = child.named_child(0)?; // make sure it's only a word
-                if word_node.kind() != "word" {
-                    return None;
-                }
-                words.push(word_node.utf8_text(src.as_bytes()).ok()?.to_owned());
-            }
-            // Positional‑argument word (allowed).
-            "word" | "number" => {
-                words.push(child.utf8_text(src.as_bytes()).ok()?.to_owned());
-            }
-            "string" => {
-                if child.child_count() == 3
-                    && child.child(0)?.kind() == "\""
-                    && child.child(1)?.kind() == "string_content"
-                    && child.child(2)?.kind() == "\""
-                {
-                    words.push(child.child(1)?.utf8_text(src.as_bytes()).ok()?.to_owned());
-                } else {
-                    // Anything else means the command is *not* plain words.
-                    return None;
-                }
-            }
-            "concatenation" => {
-                // TODO: Consider things like `'ab\'a'`.
-                return None;
-            }
-            "raw_string" => {
-                // Raw string is a single word, but we need to strip the quotes.
-                let raw_string = child.utf8_text(src.as_bytes()).ok()?;
-                let stripped = raw_string
-                    .strip_prefix('\'')
-                    .and_then(|s| s.strip_suffix('\''));
-                if let Some(stripped) = stripped {
-                    words.push(stripped.to_owned());
-                } else {
-                    return None;
-                }
-            }
-            // Anything else means the command is *not* plain words.
-            _ => return None,
-        }
-    }
-
-    Some(words)
-}
+// (bash parsing helpers implemented in crate::bash)
 
 /* ----------------------------------------------------------
 Example
@@ -216,6 +159,7 @@ fn is_valid_sed_n_arg(arg: Option<&str>) -> bool {
         _ => false,
     }
 }
+
 #[cfg(test)]
 mod tests {
     #![allow(clippy::unwrap_used)]
@@ -232,6 +176,11 @@ mod tests {
         assert!(is_safe_to_call_with_exec(&vec_str(&[
             "sed", "-n", "1,5p", "file.txt"
         ])));
+        assert!(is_safe_to_call_with_exec(&vec_str(&[
+            "nl",
+            "-nrz",
+            "Cargo.toml"
+        ])));
 
         // Safe `find` command (no unsafe options).
         assert!(is_safe_to_call_with_exec(&vec_str(&[
@@ -334,6 +283,30 @@ mod tests {
         ])));
     }
 
+    #[test]
+    fn bash_lc_safe_examples_with_operators() {
+        assert!(is_known_safe_command(&vec_str(&[
+            "bash",
+            "-lc",
+            "grep -R \"Cargo.toml\" -n || true"
+        ])));
+        assert!(is_known_safe_command(&vec_str(&[
+            "bash",
+            "-lc",
+            "ls && pwd"
+        ])));
+        assert!(is_known_safe_command(&vec_str(&[
+            "bash",
+            "-lc",
+            "echo 'hi' ; ls"
+        ])));
+        assert!(is_known_safe_command(&vec_str(&[
+            "bash",
+            "-lc",
+            "ls | wc -l"
+        ])));
+    }
+
     #[test]
     fn bash_lc_unsafe_examples() {
         assert!(
@@ -347,44 +320,29 @@ mod tests {
 
         assert!(
             !is_known_safe_command(&vec_str(&["bash", "-lc", "find . -name file.txt -delete"])),
-            "Unsafe find option should not be auto‑approved."
-        );
-    }
-
-    #[test]
-    fn test_try_parse_single_word_only_command() {
-        let script_with_single_quoted_string = "sed -n '1,5p' file.txt";
-        let parsed_words = try_parse_bash(script_with_single_quoted_string)
-            .and_then(|tree| {
-                try_parse_single_word_only_command(&tree, script_with_single_quoted_string)
-            })
-            .unwrap();
-        assert_eq!(
-            vec![
-                "sed".to_string(),
-                "-n".to_string(),
-                // Ensure the single quotes are properly removed.
-                "1,5p".to_string(),
-                "file.txt".to_string()
-            ],
-            parsed_words,
+            "Unsafe find option should not be auto-approved."
         );
 
-        let script_with_number_arg = "ls -1";
-        let parsed_words = try_parse_bash(script_with_number_arg)
-            .and_then(|tree| try_parse_single_word_only_command(&tree, script_with_number_arg))
-            .unwrap();
-        assert_eq!(vec!["ls", "-1"], parsed_words,);
+        // Disallowed because of unsafe command in sequence.
+        assert!(
+            !is_known_safe_command(&vec_str(&["bash", "-lc", "ls && rm -rf /"])),
+            "Sequence containing unsafe command must be rejected"
+        );
 
-        let script_with_double_quoted_string_with_no_funny_stuff_arg = "grep -R \"Cargo.toml\" -n";
-        let parsed_words = try_parse_bash(script_with_double_quoted_string_with_no_funny_stuff_arg)
-            .and_then(|tree| {
-                try_parse_single_word_only_command(
-                    &tree,
-                    script_with_double_quoted_string_with_no_funny_stuff_arg,
-                )
-            })
-            .unwrap();
-        assert_eq!(vec!["grep", "-R", "Cargo.toml", "-n"], parsed_words);
+        // Disallowed because of parentheses / subshell.
+        assert!(
+            !is_known_safe_command(&vec_str(&["bash", "-lc", "(ls)"])),
+            "Parentheses (subshell) are not provably safe with the current parser"
+        );
+        assert!(
+            !is_known_safe_command(&vec_str(&["bash", "-lc", "ls || (pwd && echo hi)"])),
+            "Nested parentheses are not provably safe with the current parser"
+        );
+
+        // Disallowed redirection.
+        assert!(
+            !is_known_safe_command(&vec_str(&["bash", "-lc", "ls > out.txt"])),
+            "> redirection should be rejected"
+        );
     }
 }

codex-rs/core/src/lib.rs

@@ -5,6 +5,7 @@
 // the TUI or the tracing stack).
 #![deny(clippy::print_stdout, clippy::print_stderr)]
 
+mod bash;
 mod chat_completions;
 mod client;
 mod client_common;
@@ -19,6 +20,7 @@ pub mod error;
 pub mod exec;
 pub mod exec_env;
 mod flags;
+pub mod git_info;
 mod is_safe_command;
 mod mcp_connection_manager;
 mod mcp_tool_call;

codex-rs/core/src/models.rs

@@ -3,6 +3,7 @@ use std::collections::HashMap;
 use base64::Engine;
 use mcp_types::CallToolResult;
 use serde::Deserialize;
+use serde::Deserializer;
 use serde::Serialize;
 use serde::ser::Serializer;
 
@@ -37,12 +38,14 @@ pub enum ContentItem {
 #[serde(tag = "type", rename_all = "snake_case")]
 pub enum ResponseItem {
     Message {
+        id: Option<String>,
         role: String,
         content: Vec<ContentItem>,
     },
     Reasoning {
         id: String,
         summary: Vec<ReasoningItemReasoningSummary>,
+        encrypted_content: Option<String>,
     },
     LocalShellCall {
         /// Set when using the chat completions API.
@@ -53,6 +56,7 @@ pub enum ResponseItem {
         action: LocalShellAction,
     },
     FunctionCall {
+        id: Option<String>,
         name: String,
         // The Responses API returns the function call arguments as a *string* that contains
         // JSON, not as an already‑parsed object. We keep it as a raw string here and let
@@ -78,7 +82,11 @@ pub enum ResponseItem {
 impl From<ResponseInputItem> for ResponseItem {
     fn from(item: ResponseInputItem) -> Self {
         match item {
-            ResponseInputItem::Message { role, content } => Self::Message { role, content },
+            ResponseInputItem::Message { role, content } => Self::Message {
+                role,
+                content,
+                id: None,
+            },
             ResponseInputItem::FunctionCallOutput { call_id, output } => {
                 Self::FunctionCallOutput { call_id, output }
             }
@@ -177,7 +185,7 @@ pub struct ShellToolCallParams {
     pub timeout_ms: Option<u64>,
 }
 
-#[derive(Deserialize, Debug, Clone)]
+#[derive(Debug, Clone)]
 pub struct FunctionCallOutputPayload {
     pub content: String,
     #[expect(dead_code)]
@@ -205,6 +213,19 @@ impl Serialize for FunctionCallOutputPayload {
     }
 }
 
+impl<'de> Deserialize<'de> for FunctionCallOutputPayload {
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        let s = String::deserialize(deserializer)?;
+        Ok(FunctionCallOutputPayload {
+            content: s,
+            success: None,
+        })
+    }
+}
+
 // Implement Display so callers can treat the payload like a plain string when logging or doing
 // trivial substring checks in tests (existing tests call `.contains()` on the output). Display
 // returns the raw `content` field.

codex-rs/core/src/project_doc.rs

@@ -27,16 +27,16 @@ const PROJECT_DOC_SEPARATOR: &str = "\n\n--- project-doc ---\n\n";
 /// string of instructions.
 pub(crate) async fn get_user_instructions(config: &Config) -> Option<String> {
     match find_project_doc(config).await {
-        Ok(Some(project_doc)) => match &config.instructions {
+        Ok(Some(project_doc)) => match &config.user_instructions {
             Some(original_instructions) => Some(format!(
                 "{original_instructions}{PROJECT_DOC_SEPARATOR}{project_doc}"
             )),
             None => Some(project_doc),
         },
-        Ok(None) => config.instructions.clone(),
+        Ok(None) => config.user_instructions.clone(),
         Err(e) => {
             error!("error trying to find project doc: {e:#}");
-            config.instructions.clone()
+            config.user_instructions.clone()
         }
     }
 }
@@ -159,7 +159,7 @@ mod tests {
         config.cwd = root.path().to_path_buf();
         config.project_doc_max_bytes = limit;
 
-        config.instructions = instructions.map(ToOwned::to_owned);
+        config.user_instructions = instructions.map(ToOwned::to_owned);
         config
     }
 

codex-rs/core/src/protocol.rs

@@ -4,13 +4,15 @@
 //! between user and agent.
 
 use std::collections::HashMap;
+use std::fmt;
 use std::path::Path;
 use std::path::PathBuf;
-use std::str::FromStr;
+use std::str::FromStr; // Added for FinalOutput Display implementation
 
 use mcp_types::CallToolResult;
 use serde::Deserialize;
 use serde::Serialize;
+use strum_macros::Display;
 use uuid::Uuid;
 
 use crate::config_types::ReasoningEffort as ReasoningEffortConfig;
@@ -44,8 +46,12 @@ pub enum Op {
         model_reasoning_effort: ReasoningEffortConfig,
         model_reasoning_summary: ReasoningSummaryConfig,
 
-        /// Model instructions
-        instructions: Option<String>,
+        /// Model instructions that are appended to the base instructions.
+        user_instructions: Option<String>,
+
+        /// Base instructions override.
+        base_instructions: Option<String>,
+
         /// When to escalate for approval for execution
         approval_policy: AskForApproval,
         /// How to sandbox commands executed in the system
@@ -112,18 +118,23 @@ pub enum Op {
 
     /// Request a single history entry identified by `log_id` + `offset`.
     GetHistoryEntryRequest { offset: usize, log_id: u64 },
+
+    /// Request to shut down codex instance.
+    Shutdown,
 }
 
 /// Determines the conditions under which the user is consulted to approve
 /// running the command proposed by Codex.
-#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Hash, Serialize, Deserialize)]
+#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Hash, Serialize, Deserialize, Display)]
 #[serde(rename_all = "kebab-case")]
+#[strum(serialize_all = "kebab-case")]
 pub enum AskForApproval {
     /// Under this policy, only "known safe" commands—as determined by
     /// `is_safe_command()`—that **only read files** are auto‑approved.
     /// Everything else will ask the user to approve.
     #[default]
     #[serde(rename = "untrusted")]
+    #[strum(serialize = "untrusted")]
     UnlessTrusted,
 
     /// *All* commands are auto‑approved, but they are expected to run inside a
@@ -322,6 +333,9 @@ pub enum EventMsg {
 
     /// Response to GetHistoryEntryRequest.
     GetHistoryEntryResponse(GetHistoryEntryResponseEvent),
+
+    /// Notification that the agent is shutting down.
+    ShutdownComplete,
 }
 
 // Individual event payload types matching each `EventMsg` variant.
@@ -345,6 +359,36 @@ pub struct TokenUsage {
     pub total_tokens: u64,
 }
 
+#[derive(Debug, Clone, Deserialize, Serialize)]
+pub struct FinalOutput {
+    pub token_usage: TokenUsage,
+}
+
+impl From<TokenUsage> for FinalOutput {
+    fn from(token_usage: TokenUsage) -> Self {
+        Self { token_usage }
+    }
+}
+
+impl fmt::Display for FinalOutput {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        let u = &self.token_usage;
+        write!(
+            f,
+            "Token usage: total={} input={}{} output={}{}",
+            u.total_tokens,
+            u.input_tokens,
+            u.cached_input_tokens
+                .map(|c| format!(" (cached {c})"))
+                .unwrap_or_default(),
+            u.output_tokens,
+            u.reasoning_output_tokens
+                .map(|r| format!(" (reasoning {r})"))
+                .unwrap_or_default()
+        )
+    }
+}
+
 #[derive(Debug, Clone, Deserialize, Serialize)]
 pub struct AgentMessageEvent {
     pub message: String,
@@ -418,6 +462,8 @@ pub struct ExecCommandEndEvent {
 
 #[derive(Debug, Clone, Deserialize, Serialize)]
 pub struct ExecApprovalRequestEvent {
+    /// Identifier for the associated exec call, if available.
+    pub call_id: String,
     /// The command to be executed.
     pub command: Vec<String>,
     /// The command's working directory.
@@ -429,6 +475,8 @@ pub struct ExecApprovalRequestEvent {
 
 #[derive(Debug, Clone, Deserialize, Serialize)]
 pub struct ApplyPatchApprovalRequestEvent {
+    /// Responses API call id for the associated patch apply call, if available.
+    pub call_id: String,
     pub changes: HashMap<PathBuf, FileChange>,
     /// Optional explanatory reason (e.g. request for extra write access).
     #[serde(skip_serializing_if = "Option::is_none")]

codex-rs/core/src/rollout.rs

@@ -14,10 +14,14 @@ use time::macros::format_description;
 use tokio::io::AsyncWriteExt;
 use tokio::sync::mpsc::Sender;
 use tokio::sync::mpsc::{self};
+use tokio::sync::oneshot;
 use tracing::info;
+use tracing::warn;
 use uuid::Uuid;
 
 use crate::config::Config;
+use crate::git_info::GitInfo;
+use crate::git_info::collect_git_info;
 use crate::models::ResponseItem;
 
 const SESSIONS_SUBDIR: &str = "sessions";
@@ -29,11 +33,17 @@ pub struct SessionMeta {
     pub instructions: Option<String>,
 }
 
-#[derive(Serialize, Deserialize, Default, Clone)]
-pub struct SessionStateSnapshot {
-    pub previous_response_id: Option<String>,
+#[derive(Serialize)]
+struct SessionMetaWithGit {
+    #[serde(flatten)]
+    meta: SessionMeta,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    git: Option<GitInfo>,
 }
 
+#[derive(Serialize, Deserialize, Default, Clone)]
+pub struct SessionStateSnapshot {}
+
 #[derive(Serialize, Deserialize, Default, Clone)]
 pub struct SavedSession {
     pub session: SessionMeta,
@@ -58,10 +68,10 @@ pub(crate) struct RolloutRecorder {
     tx: Sender<RolloutCmd>,
 }
 
-#[derive(Clone)]
 enum RolloutCmd {
     AddItems(Vec<ResponseItem>),
     UpdateState(SessionStateSnapshot),
+    Shutdown { ack: oneshot::Sender<()> },
 }
 
 impl RolloutRecorder {
@@ -86,15 +96,12 @@ impl RolloutRecorder {
             .format(timestamp_format)
             .map_err(|e| IoError::other(format!("failed to format timestamp: {e}")))?;
 
-        let meta = SessionMeta {
-            timestamp,
-            id: session_id,
-            instructions,
-        };
+        // Clone the cwd for the spawned task to collect git info asynchronously
+        let cwd = config.cwd.clone();
 
         // A reasonably-sized bounded channel. If the buffer fills up the send
         // future will yield, which is fine – we only need to ensure we do not
-        // perform *blocking* I/O on the caller’s thread.
+        // perform *blocking* I/O on the caller's thread.
         let (tx, rx) = mpsc::channel::<RolloutCmd>(256);
 
         // Spawn a Tokio task that owns the file handle and performs async
@@ -103,7 +110,12 @@ impl RolloutRecorder {
         tokio::task::spawn(rollout_writer(
             tokio::fs::File::from_std(file),
             rx,
-            Some(meta),
+            Some(SessionMeta {
+                timestamp,
+                id: session_id,
+                instructions,
+            }),
+            cwd,
         ));
 
         Ok(Self { tx })
@@ -119,8 +131,9 @@ impl RolloutRecorder {
                 ResponseItem::Message { .. }
                 | ResponseItem::LocalShellCall { .. }
                 | ResponseItem::FunctionCall { .. }
-                | ResponseItem::FunctionCallOutput { .. } => filtered.push(item.clone()),
-                ResponseItem::Reasoning { .. } | ResponseItem::Other => {
+                | ResponseItem::FunctionCallOutput { .. }
+                | ResponseItem::Reasoning { .. } => filtered.push(item.clone()),
+                ResponseItem::Other => {
                     // These should never be serialized.
                     continue;
                 }
@@ -142,7 +155,10 @@ impl RolloutRecorder {
             .map_err(|e| IoError::other(format!("failed to queue rollout state: {e}")))
     }
 
-    pub async fn resume(path: &Path) -> std::io::Result<(Self, SavedSession)> {
+    pub async fn resume(
+        path: &Path,
+        cwd: std::path::PathBuf,
+    ) -> std::io::Result<(Self, SavedSession)> {
         info!("Resuming rollout from {path:?}");
         let text = tokio::fs::read_to_string(path).await?;
         let mut lines = text.lines();
@@ -172,13 +188,17 @@ impl RolloutRecorder {
                 }
                 continue;
             }
-            if let Ok(item) = serde_json::from_value::<ResponseItem>(v.clone()) {
-                match item {
+            match serde_json::from_value::<ResponseItem>(v.clone()) {
+                Ok(item) => match item {
                     ResponseItem::Message { .. }
                     | ResponseItem::LocalShellCall { .. }
                     | ResponseItem::FunctionCall { .. }
-                    | ResponseItem::FunctionCallOutput { .. } => items.push(item),
-                    ResponseItem::Reasoning { .. } | ResponseItem::Other => {}
+                    | ResponseItem::FunctionCallOutput { .. }
+                    | ResponseItem::Reasoning { .. } => items.push(item),
+                    ResponseItem::Other => {}
+                },
+                Err(e) => {
+                    warn!("failed to parse item: {v:?}, error: {e}");
                 }
             }
         }
@@ -196,10 +216,30 @@ impl RolloutRecorder {
             .open(path)?;
 
         let (tx, rx) = mpsc::channel::<RolloutCmd>(256);
-        tokio::task::spawn(rollout_writer(tokio::fs::File::from_std(file), rx, None));
+        tokio::task::spawn(rollout_writer(
+            tokio::fs::File::from_std(file),
+            rx,
+            None,
+            cwd,
+        ));
         info!("Resumed rollout successfully from {path:?}");
         Ok((Self { tx }, saved))
     }
+
+    pub async fn shutdown(&self) -> std::io::Result<()> {
+        let (tx_done, rx_done) = oneshot::channel();
+        match self.tx.send(RolloutCmd::Shutdown { ack: tx_done }).await {
+            Ok(_) => rx_done
+                .await
+                .map_err(|e| IoError::other(format!("failed waiting for rollout shutdown: {e}"))),
+            Err(e) => {
+                warn!("failed to send rollout shutdown command: {e}");
+                Err(IoError::other(format!(
+                    "failed to send rollout shutdown command: {e}"
+                )))
+            }
+        }
+    }
 }
 
 struct LogFileInfo {
@@ -250,15 +290,26 @@ fn create_log_file(config: &Config, session_id: Uuid) -> std::io::Result<LogFile
 async fn rollout_writer(
     mut file: tokio::fs::File,
     mut rx: mpsc::Receiver<RolloutCmd>,
-    meta: Option<SessionMeta>,
+    mut meta: Option<SessionMeta>,
+    cwd: std::path::PathBuf,
 ) {
-    if let Some(meta) = meta {
-        if let Ok(json) = serde_json::to_string(&meta) {
+    // If we have a meta, collect git info asynchronously and write meta first
+    if let Some(session_meta) = meta.take() {
+        let git_info = collect_git_info(&cwd).await;
+        let session_meta_with_git = SessionMetaWithGit {
+            meta: session_meta,
+            git: git_info,
+        };
+
+        // Write the SessionMeta as the first item in the file
+        if let Ok(json) = serde_json::to_string(&session_meta_with_git) {
             let _ = file.write_all(json.as_bytes()).await;
             let _ = file.write_all(b"\n").await;
             let _ = file.flush().await;
         }
     }
+
+    // Process rollout commands
     while let Some(cmd) = rx.recv().await {
         match cmd {
             RolloutCmd::AddItems(items) => {
@@ -267,13 +318,14 @@ async fn rollout_writer(
                         ResponseItem::Message { .. }
                         | ResponseItem::LocalShellCall { .. }
                         | ResponseItem::FunctionCall { .. }
-                        | ResponseItem::FunctionCallOutput { .. } => {
+                        | ResponseItem::FunctionCallOutput { .. }
+                        | ResponseItem::Reasoning { .. } => {
                             if let Ok(json) = serde_json::to_string(&item) {
                                 let _ = file.write_all(json.as_bytes()).await;
                                 let _ = file.write_all(b"\n").await;
                             }
                         }
-                        ResponseItem::Reasoning { .. } | ResponseItem::Other => {}
+                        ResponseItem::Other => {}
                     }
                 }
                 let _ = file.flush().await;
@@ -294,6 +346,9 @@ async fn rollout_writer(
                     let _ = file.flush().await;
                 }
             }
+            RolloutCmd::Shutdown { ack } => {
+                let _ = ack.send(());
+            }
         }
     }
 }

codex-rs/core/tests/cli_stream.rs

@@ -329,6 +329,7 @@ async fn integration_creates_and_checks_session_file() {
         .env("OPENAI_API_KEY", "dummy")
         .env("CODEX_RS_SSE_FIXTURE", &fixture)
         .env("OPENAI_BASE_URL", "http://unused.local");
+
     let output2 = cmd2.output().unwrap();
     assert!(output2.status.success(), "resume codex-cli run failed");
 
@@ -359,3 +360,125 @@ async fn integration_creates_and_checks_session_file() {
         "rollout missing resumed marker"
     );
 }
+
+/// Integration test to verify git info is collected and recorded in session files.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn integration_git_info_unit_test() {
+    // This test verifies git info collection works independently
+    // without depending on the full CLI integration
+
+    // 1. Create temp directory for git repo
+    let temp_dir = TempDir::new().unwrap();
+    let git_repo = temp_dir.path().to_path_buf();
+
+    // 2. Initialize a git repository with some content
+    let init_output = std::process::Command::new("git")
+        .args(["init"])
+        .current_dir(&git_repo)
+        .output()
+        .unwrap();
+    assert!(init_output.status.success(), "git init failed");
+
+    // Configure git user (required for commits)
+    std::process::Command::new("git")
+        .args(["config", "user.name", "Integration Test"])
+        .current_dir(&git_repo)
+        .output()
+        .unwrap();
+
+    std::process::Command::new("git")
+        .args(["config", "user.email", "test@example.com"])
+        .current_dir(&git_repo)
+        .output()
+        .unwrap();
+
+    // Create a test file and commit it
+    let test_file = git_repo.join("test.txt");
+    std::fs::write(&test_file, "integration test content").unwrap();
+
+    std::process::Command::new("git")
+        .args(["add", "."])
+        .current_dir(&git_repo)
+        .output()
+        .unwrap();
+
+    let commit_output = std::process::Command::new("git")
+        .args(["commit", "-m", "Integration test commit"])
+        .current_dir(&git_repo)
+        .output()
+        .unwrap();
+    assert!(commit_output.status.success(), "git commit failed");
+
+    // Create a branch to test branch detection
+    std::process::Command::new("git")
+        .args(["checkout", "-b", "integration-test-branch"])
+        .current_dir(&git_repo)
+        .output()
+        .unwrap();
+
+    // Add a remote to test repository URL detection
+    std::process::Command::new("git")
+        .args([
+            "remote",
+            "add",
+            "origin",
+            "https://github.com/example/integration-test.git",
+        ])
+        .current_dir(&git_repo)
+        .output()
+        .unwrap();
+
+    // 3. Test git info collection directly
+    let git_info = codex_core::git_info::collect_git_info(&git_repo).await;
+
+    // 4. Verify git info is present and contains expected data
+    assert!(git_info.is_some(), "Git info should be collected");
+
+    let git_info = git_info.unwrap();
+
+    // Check that we have a commit hash
+    assert!(
+        git_info.commit_hash.is_some(),
+        "Git info should contain commit_hash"
+    );
+    let commit_hash = git_info.commit_hash.as_ref().unwrap();
+    assert_eq!(commit_hash.len(), 40, "Commit hash should be 40 characters");
+    assert!(
+        commit_hash.chars().all(|c| c.is_ascii_hexdigit()),
+        "Commit hash should be hexadecimal"
+    );
+
+    // Check that we have the correct branch
+    assert!(git_info.branch.is_some(), "Git info should contain branch");
+    let branch = git_info.branch.as_ref().unwrap();
+    assert_eq!(
+        branch, "integration-test-branch",
+        "Branch should match what we created"
+    );
+
+    // Check that we have the repository URL
+    assert!(
+        git_info.repository_url.is_some(),
+        "Git info should contain repository_url"
+    );
+    let repo_url = git_info.repository_url.as_ref().unwrap();
+    assert_eq!(
+        repo_url, "https://github.com/example/integration-test.git",
+        "Repository URL should match what we configured"
+    );
+
+    println!("✅ Git info collection test passed!");
+    println!("   Commit: {commit_hash}");
+    println!("   Branch: {branch}");
+    println!("   Repo: {repo_url}");
+
+    // 5. Test serialization to ensure it works in SessionMeta
+    let serialized = serde_json::to_string(&git_info).unwrap();
+    let deserialized: codex_core::git_info::GitInfo = serde_json::from_str(&serialized).unwrap();
+
+    assert_eq!(git_info.commit_hash, deserialized.commit_hash);
+    assert_eq!(git_info.branch, deserialized.branch);
+    assert_eq!(git_info.repository_url, deserialized.repository_url);
+
+    println!("✅ Git info serialization test passed!");
+}

codex-rs/core/tests/client.rs

@@ -1,5 +1,3 @@
-use std::time::Duration;
-
 use codex_core::Codex;
 use codex_core::ModelProviderInfo;
 use codex_core::exec::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
@@ -7,11 +5,10 @@ use codex_core::protocol::EventMsg;
 use codex_core::protocol::InputItem;
 use codex_core::protocol::Op;
 use codex_core::protocol::SessionConfiguredEvent;
-mod test_support;
+use core_test_support::load_default_config_for_test;
+use core_test_support::load_sse_fixture_with_id;
+use core_test_support::wait_for_event;
 use tempfile::TempDir;
-use test_support::load_default_config_for_test;
-use test_support::load_sse_fixture_with_id;
-use tokio::time::timeout;
 use wiremock::Mock;
 use wiremock::MockServer;
 use wiremock::ResponseTemplate;
@@ -86,21 +83,14 @@ async fn includes_session_id_and_model_headers_in_request() {
         .await
         .unwrap();
 
-    let mut current_session_id = None;
-    // Wait for TaskComplete
-    loop {
-        let ev = timeout(Duration::from_secs(1), codex.next_event())
-            .await
-            .unwrap()
-            .unwrap();
+    let EventMsg::SessionConfigured(SessionConfiguredEvent { session_id, .. }) =
+        wait_for_event(&codex, |ev| matches!(ev, EventMsg::SessionConfigured(_))).await
+    else {
+        unreachable!()
+    };
 
-        if let EventMsg::SessionConfigured(SessionConfiguredEvent { session_id, .. }) = ev.msg {
-            current_session_id = Some(session_id.to_string());
-        }
-        if matches!(ev.msg, EventMsg::TaskComplete(_)) {
-            break;
-        }
-    }
+    let current_session_id = Some(session_id.to_string());
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
 
     // get request from the server
     let request = &server.received_requests().await.unwrap()[0];
@@ -108,6 +98,76 @@ async fn includes_session_id_and_model_headers_in_request() {
     let originator = request.headers.get("originator").unwrap();
 
     assert!(current_session_id.is_some());
-    assert_eq!(request_body.to_str().unwrap(), &current_session_id.unwrap());
+    assert_eq!(
+        request_body.to_str().unwrap(),
+        current_session_id.as_ref().unwrap()
+    );
     assert_eq!(originator.to_str().unwrap(), "codex_cli_rs");
 }
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn includes_base_instructions_override_in_request() {
+    #![allow(clippy::unwrap_used)]
+
+    // Mock server
+    let server = MockServer::start().await;
+
+    // First request – must NOT include `previous_response_id`.
+    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 {
+        name: "openai".into(),
+        base_url: format!("{}/v1", server.uri()),
+        // Environment variable that should exist in the test environment.
+        // ModelClient will return an error if the environment variable for the
+        // provider is not set.
+        env_key: Some("PATH".into()),
+        env_key_instructions: None,
+        wire_api: codex_core::WireApi::Responses,
+        query_params: None,
+        http_headers: None,
+        env_http_headers: None,
+        request_max_retries: Some(0),
+        stream_max_retries: Some(0),
+        stream_idle_timeout_ms: None,
+    };
+
+    let codex_home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&codex_home);
+
+    config.base_instructions = Some("test instructions".to_string());
+    config.model_provider = model_provider;
+
+    let ctrl_c = std::sync::Arc::new(tokio::sync::Notify::new());
+    let (codex, ..) = Codex::spawn(config, 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_body = request.body_json::<serde_json::Value>().unwrap();
+
+    assert!(
+        request_body["instructions"]
+            .as_str()
+            .unwrap()
+            .contains("test instructions")
+    );
+}

codex-rs/core/tests/common/Cargo.toml

@@ -0,0 +1,13 @@
+[package]
+name = "core_test_support"
+version = { workspace = true }
+edition = "2024"
+
+[lib]
+path = "lib.rs"
+
+[dependencies]
+codex-core = { path = "../.." }
+serde_json = "1"
+tempfile = "3"
+tokio = { version = "1", features = ["time"] }

codex-rs/core/tests/common/lib.rs

@@ -1,9 +1,5 @@
 #![allow(clippy::expect_used)]
 
-// Helpers shared by the integration tests.  These are located inside the
-// `tests/` tree on purpose so they never become part of the public API surface
-// of the `codex-core` crate.
-
 use tempfile::TempDir;
 
 use codex_core::config::Config;
@@ -30,7 +26,6 @@ pub fn load_default_config_for_test(codex_home: &TempDir) -> Config {
 /// with only a `type` field results in an event with no `data:` section. This
 /// makes it trivial to extend the fixtures as OpenAI adds new event kinds or
 /// fields.
-#[allow(dead_code)]
 pub fn load_sse_fixture(path: impl AsRef<std::path::Path>) -> String {
     let events: Vec<serde_json::Value> =
         serde_json::from_reader(std::fs::File::open(path).expect("read fixture"))
@@ -55,7 +50,6 @@ pub fn load_sse_fixture(path: impl AsRef<std::path::Path>) -> String {
 /// fixture template with the supplied identifier before parsing. This lets a
 /// single JSON template be reused by multiple tests that each need a unique
 /// `response_id`.
-#[allow(dead_code)]
 pub fn load_sse_fixture_with_id(path: impl AsRef<std::path::Path>, id: &str) -> String {
     let raw = std::fs::read_to_string(path).expect("read fixture template");
     let replaced = raw.replace("__ID__", id);
@@ -76,3 +70,23 @@ pub fn load_sse_fixture_with_id(path: impl AsRef<std::path::Path>, id: &str) ->
         })
         .collect()
 }
+
+pub async fn wait_for_event<F>(
+    codex: &codex_core::Codex,
+    mut predicate: F,
+) -> codex_core::protocol::EventMsg
+where
+    F: FnMut(&codex_core::protocol::EventMsg) -> bool,
+{
+    use tokio::time::Duration;
+    use tokio::time::timeout;
+    loop {
+        let ev = timeout(Duration::from_secs(1), codex.next_event())
+            .await
+            .expect("timeout waiting for event")
+            .expect("stream ended unexpectedly");
+        if predicate(&ev.msg) {
+            return ev.msg;
+        }
+    }
+}

codex-rs/core/tests/live_agent.rs

@@ -26,9 +26,8 @@ use codex_core::protocol::ErrorEvent;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::InputItem;
 use codex_core::protocol::Op;
-mod test_support;
+use core_test_support::load_default_config_for_test;
 use tempfile::TempDir;
-use test_support::load_default_config_for_test;
 use tokio::sync::Notify;
 use tokio::time::timeout;
 

codex-rs/core/tests/previous_response_id.rs

@@ -1,165 +0,0 @@
-use std::time::Duration;
-
-use codex_core::Codex;
-use codex_core::ModelProviderInfo;
-use codex_core::exec::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
-use codex_core::protocol::ErrorEvent;
-use codex_core::protocol::EventMsg;
-use codex_core::protocol::InputItem;
-use codex_core::protocol::Op;
-mod test_support;
-use serde_json::Value;
-use tempfile::TempDir;
-use test_support::load_default_config_for_test;
-use test_support::load_sse_fixture_with_id;
-use tokio::time::timeout;
-use wiremock::Match;
-use wiremock::Mock;
-use wiremock::MockServer;
-use wiremock::Request;
-use wiremock::ResponseTemplate;
-use wiremock::matchers::method;
-use wiremock::matchers::path;
-
-/// Matcher asserting that JSON body has NO `previous_response_id` field.
-struct NoPrevId;
-
-impl Match for NoPrevId {
-    fn matches(&self, req: &Request) -> bool {
-        serde_json::from_slice::<Value>(&req.body)
-            .map(|v| v.get("previous_response_id").is_none())
-            .unwrap_or(false)
-    }
-}
-
-/// Matcher asserting that JSON body HAS a `previous_response_id` field.
-struct HasPrevId;
-
-impl Match for HasPrevId {
-    fn matches(&self, req: &Request) -> bool {
-        serde_json::from_slice::<Value>(&req.body)
-            .map(|v| v.get("previous_response_id").is_some())
-            .unwrap_or(false)
-    }
-}
-
-/// Build minimal SSE stream with completed marker using the JSON fixture.
-fn sse_completed(id: &str) -> String {
-    load_sse_fixture_with_id("tests/fixtures/completed_template.json", id)
-}
-
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn keeps_previous_response_id_between_tasks() {
-    #![allow(clippy::unwrap_used)]
-
-    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
-        println!(
-            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
-        );
-        return;
-    }
-
-    // Mock server
-    let server = MockServer::start().await;
-
-    // First request – must NOT include `previous_response_id`.
-    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"))
-        .and(NoPrevId)
-        .respond_with(first)
-        .expect(1)
-        .mount(&server)
-        .await;
-
-    // Second request – MUST include `previous_response_id`.
-    let second = ResponseTemplate::new(200)
-        .insert_header("content-type", "text/event-stream")
-        .set_body_raw(sse_completed("resp2"), "text/event-stream");
-
-    Mock::given(method("POST"))
-        .and(path("/v1/responses"))
-        .and(HasPrevId)
-        .respond_with(second)
-        .expect(1)
-        .mount(&server)
-        .await;
-
-    // Configure retry behavior explicitly to avoid mutating process-wide
-    // environment variables.
-    let model_provider = ModelProviderInfo {
-        name: "openai".into(),
-        base_url: format!("{}/v1", server.uri()),
-        // Environment variable that should exist in the test environment.
-        // ModelClient will return an error if the environment variable for the
-        // provider is not set.
-        env_key: Some("PATH".into()),
-        env_key_instructions: None,
-        wire_api: codex_core::WireApi::Responses,
-        query_params: None,
-        http_headers: None,
-        env_http_headers: None,
-        // disable retries so we don't get duplicate calls in this test
-        request_max_retries: Some(0),
-        stream_max_retries: Some(0),
-        stream_idle_timeout_ms: None,
-    };
-
-    // Init session
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home);
-    config.model_provider = model_provider;
-    let ctrl_c = std::sync::Arc::new(tokio::sync::Notify::new());
-    let (codex, _init_id, _session_id) = Codex::spawn(config, ctrl_c.clone()).await.unwrap();
-
-    // Task 1 – triggers first request (no previous_response_id)
-    codex
-        .submit(Op::UserInput {
-            items: vec![InputItem::Text {
-                text: "hello".into(),
-            }],
-        })
-        .await
-        .unwrap();
-
-    // Wait for TaskComplete
-    loop {
-        let ev = timeout(Duration::from_secs(1), codex.next_event())
-            .await
-            .unwrap()
-            .unwrap();
-        if matches!(ev.msg, EventMsg::TaskComplete(_)) {
-            break;
-        }
-    }
-
-    // Task 2 – should include `previous_response_id` (triggers second request)
-    codex
-        .submit(Op::UserInput {
-            items: vec![InputItem::Text {
-                text: "again".into(),
-            }],
-        })
-        .await
-        .unwrap();
-
-    // Wait for TaskComplete or error
-    loop {
-        let ev = timeout(Duration::from_secs(1), codex.next_event())
-            .await
-            .unwrap()
-            .unwrap();
-        match ev.msg {
-            EventMsg::TaskComplete(_) => break,
-            EventMsg::Error(ErrorEvent { message }) => {
-                panic!("unexpected error: {message}")
-            }
-            _ => {
-                // Ignore other events.
-            }
-        }
-    }
-}

codex-rs/core/tests/stream_no_completed.rs

@@ -9,11 +9,10 @@ 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;
-mod test_support;
+use core_test_support::load_default_config_for_test;
+use core_test_support::load_sse_fixture;
+use core_test_support::load_sse_fixture_with_id;
 use tempfile::TempDir;
-use test_support::load_default_config_for_test;
-use test_support::load_sse_fixture;
-use test_support::load_sse_fixture_with_id;
 use tokio::time::timeout;
 use wiremock::Mock;
 use wiremock::MockServer;

codex-rs/exec/src/event_processor.rs

@@ -1,15 +1,23 @@
+use std::path::Path;
+
 use codex_common::summarize_sandbox_policy;
 use codex_core::WireApi;
 use codex_core::config::Config;
 use codex_core::model_supports_reasoning_summaries;
 use codex_core::protocol::Event;
 
+pub(crate) enum CodexStatus {
+    Running,
+    InitiateShutdown,
+    Shutdown,
+}
+
 pub(crate) trait EventProcessor {
     /// Print summary of effective configuration and user prompt.
     fn print_config_summary(&mut self, config: &Config, prompt: &str);
 
     /// Handle a single event emitted by the agent.
-    fn process_event(&mut self, event: Event);
+    fn process_event(&mut self, event: Event) -> CodexStatus;
 }
 
 pub(crate) fn create_config_summary_entries(config: &Config) -> Vec<(&'static str, String)> {
@@ -17,7 +25,7 @@ pub(crate) fn create_config_summary_entries(config: &Config) -> Vec<(&'static st
         ("workdir", config.cwd.display().to_string()),
         ("model", config.model.clone()),
         ("provider", config.model_provider_id.clone()),
-        ("approval", format!("{:?}", config.approval_policy)),
+        ("approval", config.approval_policy.to_string()),
         ("sandbox", summarize_sandbox_policy(&config.sandbox_policy)),
     ];
     if config.model_provider.wire_api == WireApi::Responses
@@ -35,3 +43,28 @@ 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."),
+    }
+}
+
+fn write_last_message_file(contents: &str, last_message_path: Option<&Path>) {
+    if let Some(path) = last_message_path {
+        if let Err(e) = std::fs::write(path, contents) {
+            eprintln!("Failed to write last message file {path:?}: {e}");
+        }
+    }
+}

codex-rs/exec/src/event_processor_with_human_output.rs

@@ -15,16 +15,20 @@ use codex_core::protocol::McpToolCallEndEvent;
 use codex_core::protocol::PatchApplyBeginEvent;
 use codex_core::protocol::PatchApplyEndEvent;
 use codex_core::protocol::SessionConfiguredEvent;
+use codex_core::protocol::TaskCompleteEvent;
 use codex_core::protocol::TokenUsage;
 use owo_colors::OwoColorize;
 use owo_colors::Style;
 use shlex::try_join;
 use std::collections::HashMap;
 use std::io::Write;
+use std::path::PathBuf;
 use std::time::Instant;
 
+use crate::event_processor::CodexStatus;
 use crate::event_processor::EventProcessor;
 use crate::event_processor::create_config_summary_entries;
+use crate::event_processor::handle_last_message;
 
 /// This should be configurable. When used in CI, users may not want to impose
 /// a limit so they can see the full transcript.
@@ -54,10 +58,15 @@ pub(crate) struct EventProcessorWithHumanOutput {
     show_agent_reasoning: bool,
     answer_started: bool,
     reasoning_started: bool,
+    last_message_path: Option<PathBuf>,
 }
 
 impl EventProcessorWithHumanOutput {
-    pub(crate) fn create_with_ansi(with_ansi: bool, config: &Config) -> Self {
+    pub(crate) fn create_with_ansi(
+        with_ansi: bool,
+        config: &Config,
+        last_message_path: Option<PathBuf>,
+    ) -> Self {
         let call_id_to_command = HashMap::new();
         let call_id_to_patch = HashMap::new();
         let call_id_to_tool_call = HashMap::new();
@@ -77,6 +86,7 @@ impl EventProcessorWithHumanOutput {
                 show_agent_reasoning: !config.hide_agent_reasoning,
                 answer_started: false,
                 reasoning_started: false,
+                last_message_path,
             }
         } else {
             Self {
@@ -93,6 +103,7 @@ impl EventProcessorWithHumanOutput {
                 show_agent_reasoning: !config.hide_agent_reasoning,
                 answer_started: false,
                 reasoning_started: false,
+                last_message_path,
             }
         }
     }
@@ -158,7 +169,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
         );
     }
 
-    fn process_event(&mut self, event: Event) {
+    fn process_event(&mut self, event: Event) -> CodexStatus {
         let Event { id: _, msg } = event;
         match msg {
             EventMsg::Error(ErrorEvent { message }) => {
@@ -168,9 +179,16 @@ impl EventProcessor for EventProcessorWithHumanOutput {
             EventMsg::BackgroundEvent(BackgroundEventEvent { message }) => {
                 ts_println!(self, "{}", message.style(self.dimmed));
             }
-            EventMsg::TaskStarted | EventMsg::TaskComplete(_) => {
+            EventMsg::TaskStarted => {
                 // Ignore.
             }
+            EventMsg::TaskComplete(TaskCompleteEvent { last_agent_message }) => {
+                handle_last_message(
+                    last_agent_message.as_deref(),
+                    self.last_message_path.as_deref(),
+                );
+                return CodexStatus::InitiateShutdown;
+            }
             EventMsg::TokenCount(TokenUsage { total_tokens, .. }) => {
                 ts_println!(self, "tokens used: {total_tokens}");
             }
@@ -185,7 +203,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
             }
             EventMsg::AgentReasoningDelta(AgentReasoningDeltaEvent { delta }) => {
                 if !self.show_agent_reasoning {
-                    return;
+                    return CodexStatus::Running;
                 }
                 if !self.reasoning_started {
                     ts_println!(
@@ -498,7 +516,9 @@ impl EventProcessor for EventProcessorWithHumanOutput {
             EventMsg::GetHistoryEntryResponse(_) => {
                 // Currently ignored in exec output.
             }
+            EventMsg::ShutdownComplete => return CodexStatus::Shutdown,
         }
+        CodexStatus::Running
     }
 }
 

codex-rs/exec/src/event_processor_with_json_output.rs

@@ -1,18 +1,24 @@
 use std::collections::HashMap;
+use std::path::PathBuf;
 
 use codex_core::config::Config;
 use codex_core::protocol::Event;
 use codex_core::protocol::EventMsg;
+use codex_core::protocol::TaskCompleteEvent;
 use serde_json::json;
 
+use crate::event_processor::CodexStatus;
 use crate::event_processor::EventProcessor;
 use crate::event_processor::create_config_summary_entries;
+use crate::event_processor::handle_last_message;
 
-pub(crate) struct EventProcessorWithJsonOutput;
+pub(crate) struct EventProcessorWithJsonOutput {
+    last_message_path: Option<PathBuf>,
+}
 
 impl EventProcessorWithJsonOutput {
-    pub fn new() -> Self {
-        Self {}
+    pub fn new(last_message_path: Option<PathBuf>) -> Self {
+        Self { last_message_path }
     }
 }
 
@@ -33,15 +39,25 @@ impl EventProcessor for EventProcessorWithJsonOutput {
         println!("{prompt_json}");
     }
 
-    fn process_event(&mut self, event: Event) {
+    fn process_event(&mut self, event: Event) -> CodexStatus {
         match event.msg {
             EventMsg::AgentMessageDelta(_) | EventMsg::AgentReasoningDelta(_) => {
                 // Suppress streaming events in JSON mode.
+                CodexStatus::Running
             }
+            EventMsg::TaskComplete(TaskCompleteEvent { last_agent_message }) => {
+                handle_last_message(
+                    last_agent_message.as_deref(),
+                    self.last_message_path.as_deref(),
+                );
+                CodexStatus::InitiateShutdown
+            }
+            EventMsg::ShutdownComplete => CodexStatus::Shutdown,
             _ => {
                 if let Ok(line) = serde_json::to_string(&event) {
                     println!("{line}");
                 }
+                CodexStatus::Running
             }
         }
     }

codex-rs/exec/src/lib.rs

@@ -5,7 +5,6 @@ mod event_processor_with_json_output;
 
 use std::io::IsTerminal;
 use std::io::Read;
-use std::path::Path;
 use std::path::PathBuf;
 use std::sync::Arc;
 
@@ -28,6 +27,7 @@ use tracing::error;
 use tracing::info;
 use tracing_subscriber::EnvFilter;
 
+use crate::event_processor::CodexStatus;
 use crate::event_processor::EventProcessor;
 
 pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()> {
@@ -110,6 +110,7 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
         cwd: cwd.map(|p| p.canonicalize().unwrap_or(p)),
         model_provider: None,
         codex_linux_sandbox_exe,
+        base_instructions: None,
     };
     // Parse `-c` overrides.
     let cli_kv_overrides = match config_overrides.parse_overrides() {
@@ -122,11 +123,12 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
 
     let config = Config::load_with_cli_overrides(cli_kv_overrides, overrides)?;
     let mut event_processor: Box<dyn EventProcessor> = if json_mode {
-        Box::new(EventProcessorWithJsonOutput::new())
+        Box::new(EventProcessorWithJsonOutput::new(last_message_file.clone()))
     } else {
         Box::new(EventProcessorWithHumanOutput::create_with_ansi(
             stdout_with_ansi,
             &config,
+            last_message_file.clone(),
         ))
     };
 
@@ -223,40 +225,17 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
 
     // Run the loop until the task is complete.
     while let Some(event) = rx.recv().await {
-        let (is_last_event, last_assistant_message) = match &event.msg {
-            EventMsg::TaskComplete(TaskCompleteEvent { last_agent_message }) => {
-                (true, last_agent_message.clone())
+        let shutdown: CodexStatus = event_processor.process_event(event);
+        match shutdown {
+            CodexStatus::Running => continue,
+            CodexStatus::InitiateShutdown => {
+                codex.submit(Op::Shutdown).await?;
+            }
+            CodexStatus::Shutdown => {
+                break;
             }
-            _ => (false, None),
-        };
-        event_processor.process_event(event);
-        if is_last_event {
-            handle_last_message(last_assistant_message, last_message_file.as_deref())?;
-            break;
         }
     }
 
     Ok(())
 }
-
-fn handle_last_message(
-    last_agent_message: Option<String>,
-    last_message_file: Option<&Path>,
-) -> std::io::Result<()> {
-    match (last_agent_message, last_message_file) {
-        (Some(last_agent_message), Some(last_message_file)) => {
-            // Last message and a file to write to.
-            std::fs::write(last_message_file, last_agent_message)?;
-        }
-        (None, Some(last_message_file)) => {
-            eprintln!(
-                "Warning: No last message to write to file: {}",
-                last_message_file.to_string_lossy()
-            );
-        }
-        (_, None) => {
-            // No last message and no file to write to.
-        }
-    }
-    Ok(())
-}

codex-rs/linux-sandbox/Cargo.toml

@@ -17,7 +17,9 @@ workspace = true
 [dependencies]
 anyhow = "1"
 clap = { version = "4", features = ["derive"] }
+codex-common = { path = "../common", features = ["cli"] }
 codex-core = { path = "../core" }
+dotenvy = "0.15.7"
 tokio = { version = "1", features = ["rt-multi-thread"] }
 
 [dev-dependencies]

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

@@ -43,6 +43,10 @@ where
         crate::run_main();
     }
 
+    // This modifies the environment, which is not thread-safe, so do this
+    // before creating any threads/the Tokio runtime.
+    load_dotenv();
+
     // Regular invocation – create a Tokio runtime and execute the provided
     // async entry-point.
     let runtime = tokio::runtime::Runtime::new()?;
@@ -61,3 +65,11 @@ where
 pub fn run_main() -> ! {
     panic!("codex-linux-sandbox is only supported on Linux");
 }
+
+/// Load env vars from ~/.codex/.env and `$(pwd)/.env`.
+fn load_dotenv() {
+    if let Ok(codex_home) = codex_core::config::find_codex_home() {
+        dotenvy::from_path(codex_home.join(".env")).ok();
+    }
+    dotenvy::dotenv().ok();
+}

codex-rs/mcp-server/Cargo.toml

@@ -37,6 +37,7 @@ uuid = { version = "1", features = ["serde", "v4"] }
 
 [dev-dependencies]
 assert_cmd = "2"
+mcp_test_support = { path = "tests/common" }
 pretty_assertions = "1.4.1"
 tempfile = "3"
 tokio-test = "0.4"

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

@@ -14,7 +14,7 @@ use std::path::PathBuf;
 use crate::json_to_toml::json_to_toml;
 
 /// Client-supplied configuration for a `codex` tool-call.
-#[derive(Debug, Clone, Serialize, Deserialize, JsonSchema)]
+#[derive(Debug, Clone, Serialize, Deserialize, JsonSchema, Default)]
 #[serde(rename_all = "kebab-case")]
 pub struct CodexToolCallParam {
     /// The *initial user prompt* to start the Codex conversation.
@@ -46,6 +46,10 @@ pub struct CodexToolCallParam {
     /// CODEX_HOME/config.toml.
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub config: Option<HashMap<String, serde_json::Value>>,
+
+    /// The set of instructions to use instead of the default ones.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub base_instructions: Option<String>,
 }
 
 /// Custom enum mirroring [`AskForApproval`], but has an extra dependency on
@@ -135,6 +139,7 @@ impl CodexToolCallParam {
             approval_policy,
             sandbox,
             config: cli_overrides,
+            base_instructions,
         } = self;
 
         // Build the `ConfigOverrides` recognised by codex-core.
@@ -146,6 +151,7 @@ impl CodexToolCallParam {
             sandbox_mode: sandbox.map(Into::into),
             model_provider: None,
             codex_linux_sandbox_exe,
+            base_instructions,
         };
 
         let cli_overrides = cli_overrides
@@ -162,7 +168,7 @@ impl CodexToolCallParam {
 
 #[derive(Debug, Clone, Serialize, Deserialize, JsonSchema)]
 #[serde(rename_all = "camelCase")]
-pub(crate) struct CodexToolCallReplyParam {
+pub struct CodexToolCallReplyParam {
     /// The *session id* for this conversation.
     pub session_id: String,
 
@@ -268,6 +274,10 @@ mod tests {
                 "description": "The *initial user prompt* to start the Codex conversation.",
                 "type": "string"
               },
+              "base-instructions": {
+                "description": "The set of instructions to use instead of the default ones.",
+                "type": "string"
+              },
             },
             "required": [
               "prompt"

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

@@ -20,6 +20,7 @@ use mcp_types::CallToolResult;
 use mcp_types::ContentBlock;
 use mcp_types::RequestId;
 use mcp_types::TextContent;
+use serde_json::json;
 use tokio::sync::Mutex;
 use uuid::Uuid;
 
@@ -39,6 +40,7 @@ pub async fn run_codex_tool_session(
     config: CodexConfig,
     outgoing: Arc<OutgoingMessageSender>,
     session_map: Arc<Mutex<HashMap<Uuid, Arc<Codex>>>>,
+    running_requests_id_to_codex_uuid: Arc<Mutex<HashMap<RequestId, Uuid>>>,
 ) {
     let (codex, first_event, _ctrl_c, session_id) = match init_codex(config).await {
         Ok(res) => res,
@@ -73,7 +75,10 @@ pub async fn run_codex_tool_session(
         RequestId::String(s) => s.clone(),
         RequestId::Integer(n) => n.to_string(),
     };
-
+    running_requests_id_to_codex_uuid
+        .lock()
+        .await
+        .insert(id.clone(), session_id);
     let submission = Submission {
         id: sub_id.clone(),
         op: Op::UserInput {
@@ -85,9 +90,12 @@ pub async fn run_codex_tool_session(
 
     if let Err(e) = codex.submit_with_id(submission).await {
         tracing::error!("Failed to submit initial prompt: {e}");
+        // unregister the id so we don't keep it in the map
+        running_requests_id_to_codex_uuid.lock().await.remove(&id);
+        return;
     }
 
-    run_codex_tool_session_inner(codex, outgoing, id).await;
+    run_codex_tool_session_inner(codex, outgoing, id, running_requests_id_to_codex_uuid).await;
 }
 
 pub async fn run_codex_tool_session_reply(
@@ -95,7 +103,13 @@ pub async fn run_codex_tool_session_reply(
     outgoing: Arc<OutgoingMessageSender>,
     request_id: RequestId,
     prompt: String,
+    running_requests_id_to_codex_uuid: Arc<Mutex<HashMap<RequestId, Uuid>>>,
+    session_id: Uuid,
 ) {
+    running_requests_id_to_codex_uuid
+        .lock()
+        .await
+        .insert(request_id.clone(), session_id);
     if let Err(e) = codex
         .submit(Op::UserInput {
             items: vec![InputItem::Text { text: prompt }],
@@ -103,15 +117,28 @@ pub async fn run_codex_tool_session_reply(
         .await
     {
         tracing::error!("Failed to submit user input: {e}");
+        // unregister the id so we don't keep it in the map
+        running_requests_id_to_codex_uuid
+            .lock()
+            .await
+            .remove(&request_id);
+        return;
     }
 
-    run_codex_tool_session_inner(codex, outgoing, request_id).await;
+    run_codex_tool_session_inner(
+        codex,
+        outgoing,
+        request_id,
+        running_requests_id_to_codex_uuid,
+    )
+    .await;
 }
 
 async fn run_codex_tool_session_inner(
     codex: Arc<Codex>,
     outgoing: Arc<OutgoingMessageSender>,
     request_id: RequestId,
+    running_requests_id_to_codex_uuid: Arc<Mutex<HashMap<RequestId, Uuid>>>,
 ) {
     let request_id_str = match &request_id {
         RequestId::String(s) => s.clone(),
@@ -129,6 +156,7 @@ async fn run_codex_tool_session_inner(
                     EventMsg::ExecApprovalRequest(ExecApprovalRequestEvent {
                         command,
                         cwd,
+                        call_id,
                         reason: _,
                     }) => {
                         handle_exec_approval_request(
@@ -139,16 +167,27 @@ async fn run_codex_tool_session_inner(
                             request_id.clone(),
                             request_id_str.clone(),
                             event.id.clone(),
+                            call_id,
                         )
                         .await;
                         continue;
                     }
+                    EventMsg::Error(err_event) => {
+                        // Return a response to conclude the tool call when the Codex session reports an error (e.g., interruption).
+                        let result = json!({
+                            "error": err_event.message,
+                        });
+                        outgoing.send_response(request_id.clone(), result).await;
+                        break;
+                    }
                     EventMsg::ApplyPatchApprovalRequest(ApplyPatchApprovalRequestEvent {
+                        call_id,
                         reason,
                         grant_root,
                         changes,
                     }) => {
                         handle_patch_approval_request(
+                            call_id,
                             reason,
                             grant_root,
                             changes,
@@ -178,6 +217,11 @@ async fn run_codex_tool_session_inner(
                         outgoing
                             .send_response(request_id.clone(), result.into())
                             .await;
+                        // unregister the id so we don't keep it in the map
+                        running_requests_id_to_codex_uuid
+                            .lock()
+                            .await
+                            .remove(&request_id);
                         break;
                     }
                     EventMsg::SessionConfigured(_) => {
@@ -192,8 +236,7 @@ async fn run_codex_tool_session_inner(
                     EventMsg::AgentMessage(AgentMessageEvent { .. }) => {
                         // TODO: think how we want to support this in the MCP
                     }
-                    EventMsg::Error(_)
-                    | EventMsg::TaskStarted
+                    EventMsg::TaskStarted
                     | EventMsg::TokenCount(_)
                     | EventMsg::AgentReasoning(_)
                     | EventMsg::McpToolCallBegin(_)
@@ -203,7 +246,8 @@ async fn run_codex_tool_session_inner(
                     | EventMsg::BackgroundEvent(_)
                     | EventMsg::PatchApplyBegin(_)
                     | EventMsg::PatchApplyEnd(_)
-                    | EventMsg::GetHistoryEntryResponse(_) => {
+                    | EventMsg::GetHistoryEntryResponse(_)
+                    | EventMsg::ShutdownComplete => {
                         // For now, we do not do anything extra for these
                         // events. Note that
                         // send(codex_event_to_notification(&event)) above has

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

@@ -32,6 +32,7 @@ pub struct ExecApprovalElicitRequestParams {
     pub codex_elicitation: String,
     pub codex_mcp_tool_call_id: String,
     pub codex_event_id: String,
+    pub codex_call_id: String,
     pub codex_command: Vec<String>,
     pub codex_cwd: PathBuf,
 }
@@ -45,6 +46,7 @@ pub struct ExecApprovalResponse {
     pub decision: ReviewDecision,
 }
 
+#[allow(clippy::too_many_arguments)]
 pub(crate) async fn handle_exec_approval_request(
     command: Vec<String>,
     cwd: PathBuf,
@@ -53,6 +55,7 @@ pub(crate) async fn handle_exec_approval_request(
     request_id: RequestId,
     tool_call_id: String,
     event_id: String,
+    call_id: String,
 ) {
     let escaped_command =
         shlex::try_join(command.iter().map(|s| s.as_str())).unwrap_or_else(|_| command.join(" "));
@@ -71,6 +74,7 @@ pub(crate) async fn handle_exec_approval_request(
         codex_elicitation: "exec-approval".to_string(),
         codex_mcp_tool_call_id: tool_call_id.clone(),
         codex_event_id: event_id.clone(),
+        codex_call_id: call_id,
         codex_command: command,
         codex_cwd: cwd,
     };

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

@@ -27,6 +27,7 @@ use crate::outgoing_message::OutgoingMessage;
 use crate::outgoing_message::OutgoingMessageSender;
 
 pub use crate::codex_tool_config::CodexToolCallParam;
+pub use crate::codex_tool_config::CodexToolCallReplyParam;
 pub use crate::exec_approval::ExecApprovalElicitRequestParams;
 pub use crate::exec_approval::ExecApprovalResponse;
 pub use crate::patch_approval::PatchApprovalElicitRequestParams;
@@ -81,7 +82,7 @@ pub async fn run_main(codex_linux_sandbox_exe: Option<PathBuf>) -> IoResult<()>
                 match msg {
                     JSONRPCMessage::Request(r) => processor.process_request(r).await,
                     JSONRPCMessage::Response(r) => processor.process_response(r).await,
-                    JSONRPCMessage::Notification(n) => processor.process_notification(n),
+                    JSONRPCMessage::Notification(n) => processor.process_notification(n).await,
                     JSONRPCMessage::Error(e) => processor.process_error(e),
                 }
             }

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

@@ -10,6 +10,7 @@ use crate::outgoing_message::OutgoingMessageSender;
 
 use codex_core::Codex;
 use codex_core::config::Config as CodexConfig;
+use codex_core::protocol::Submission;
 use mcp_types::CallToolRequestParams;
 use mcp_types::CallToolResult;
 use mcp_types::ClientRequest;
@@ -35,6 +36,7 @@ pub(crate) struct MessageProcessor {
     initialized: bool,
     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>>>,
 }
 
 impl MessageProcessor {
@@ -49,6 +51,7 @@ impl MessageProcessor {
             initialized: false,
             codex_linux_sandbox_exe,
             session_map: Arc::new(Mutex::new(HashMap::new())),
+            running_requests_id_to_codex_uuid: Arc::new(Mutex::new(HashMap::new())),
         }
     }
 
@@ -116,7 +119,7 @@ impl MessageProcessor {
     }
 
     /// Handle a fire-and-forget JSON-RPC notification.
-    pub(crate) fn process_notification(&mut self, notification: JSONRPCNotification) {
+    pub(crate) async fn process_notification(&mut self, notification: JSONRPCNotification) {
         let server_notification = match ServerNotification::try_from(notification) {
             Ok(n) => n,
             Err(e) => {
@@ -129,7 +132,7 @@ impl MessageProcessor {
         // handler so additional logic can be implemented incrementally.
         match server_notification {
             ServerNotification::CancelledNotification(params) => {
-                self.handle_cancelled_notification(params);
+                self.handle_cancelled_notification(params).await;
             }
             ServerNotification::ProgressNotification(params) => {
                 self.handle_progress_notification(params);
@@ -379,6 +382,7 @@ impl MessageProcessor {
         // Clone outgoing and session map to move into async task.
         let outgoing = self.outgoing.clone();
         let session_map = self.session_map.clone();
+        let running_requests_id_to_codex_uuid = self.running_requests_id_to_codex_uuid.clone();
 
         // Spawn an async task to handle the Codex session so that we do not
         // block the synchronous message-processing loop.
@@ -390,6 +394,7 @@ impl MessageProcessor {
                 config,
                 outgoing,
                 session_map,
+                running_requests_id_to_codex_uuid,
             )
             .await;
         });
@@ -464,13 +469,12 @@ impl MessageProcessor {
 
         // Clone outgoing and session map to move into async task.
         let outgoing = self.outgoing.clone();
+        let running_requests_id_to_codex_uuid = self.running_requests_id_to_codex_uuid.clone();
 
-        // Spawn an async task to handle the Codex session so that we do not
-        // block the synchronous message-processing loop.
-        task::spawn(async move {
+        let codex = {
             let session_map = session_map_mutex.lock().await;
-            let codex = match session_map.get(&session_id) {
-                Some(codex) => codex,
+            match session_map.get(&session_id).cloned() {
+                Some(c) => c,
                 None => {
                     tracing::warn!("Session not found for session_id: {session_id}");
                     let result = CallToolResult {
@@ -482,21 +486,32 @@ impl MessageProcessor {
                         is_error: Some(true),
                         structured_content: None,
                     };
-                    // unwrap_or_default is fine here because we know the result is valid JSON
                     outgoing
                         .send_response(request_id, serde_json::to_value(result).unwrap_or_default())
                         .await;
                     return;
                 }
-            };
+            }
+        };
 
-            crate::codex_tool_runner::run_codex_tool_session_reply(
-                codex.clone(),
-                outgoing,
-                request_id,
-                prompt.clone(),
-            )
-            .await;
+        // Spawn the long-running reply handler.
+        tokio::spawn({
+            let codex = codex.clone();
+            let outgoing = outgoing.clone();
+            let prompt = prompt.clone();
+            let running_requests_id_to_codex_uuid = running_requests_id_to_codex_uuid.clone();
+
+            async move {
+                crate::codex_tool_runner::run_codex_tool_session_reply(
+                    codex,
+                    outgoing,
+                    request_id,
+                    prompt,
+                    running_requests_id_to_codex_uuid,
+                    session_id,
+                )
+                .await;
+            }
         });
     }
 
@@ -518,11 +533,58 @@ impl MessageProcessor {
     // Notification handlers
     // ---------------------------------------------------------------------
 
-    fn handle_cancelled_notification(
+    async fn handle_cancelled_notification(
         &self,
         params: <mcp_types::CancelledNotification as mcp_types::ModelContextProtocolNotification>::Params,
     ) {
-        tracing::info!("notifications/cancelled -> params: {:?}", params);
+        let request_id = params.request_id;
+        // Create a stable string form early for logging and submission id.
+        let request_id_string = match &request_id {
+            RequestId::String(s) => s.clone(),
+            RequestId::Integer(i) => i.to_string(),
+        };
+
+        // Obtain the session_id while holding the first lock, then release.
+        let session_id = {
+            let map_guard = self.running_requests_id_to_codex_uuid.lock().await;
+            match map_guard.get(&request_id) {
+                Some(id) => *id, // Uuid is Copy
+                None => {
+                    tracing::warn!("Session not found for request_id: {}", request_id_string);
+                    return;
+                }
+            }
+        };
+        tracing::info!("session_id: {session_id}");
+
+        // Obtain the Codex Arc while holding the session_map lock, then release.
+        let codex_arc = {
+            let sessions_guard = self.session_map.lock().await;
+            match sessions_guard.get(&session_id) {
+                Some(codex) => Arc::clone(codex),
+                None => {
+                    tracing::warn!("Session not found for session_id: {session_id}");
+                    return;
+                }
+            }
+        };
+
+        // Submit interrupt to Codex.
+        let err = codex_arc
+            .submit_with_id(Submission {
+                id: request_id_string,
+                op: codex_core::protocol::Op::Interrupt,
+            })
+            .await;
+        if let Err(e) = err {
+            tracing::error!("Failed to submit interrupt to Codex: {e}");
+            return;
+        }
+        // unregister the id so we don't keep it in the map
+        self.running_requests_id_to_codex_uuid
+            .lock()
+            .await
+            .remove(&request_id);
     }
 
     fn handle_progress_notification(

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

@@ -27,6 +27,7 @@ pub struct PatchApprovalElicitRequestParams {
     pub codex_elicitation: String,
     pub codex_mcp_tool_call_id: String,
     pub codex_event_id: String,
+    pub codex_call_id: String,
     #[serde(skip_serializing_if = "Option::is_none")]
     pub codex_reason: Option<String>,
     #[serde(skip_serializing_if = "Option::is_none")]
@@ -41,6 +42,7 @@ pub struct PatchApprovalResponse {
 
 #[allow(clippy::too_many_arguments)]
 pub(crate) async fn handle_patch_approval_request(
+    call_id: String,
     reason: Option<String>,
     grant_root: Option<PathBuf>,
     changes: HashMap<PathBuf, FileChange>,
@@ -66,6 +68,7 @@ pub(crate) async fn handle_patch_approval_request(
         codex_elicitation: "patch-approval".to_string(),
         codex_mcp_tool_call_id: tool_call_id.clone(),
         codex_event_id: event_id.clone(),
+        codex_call_id: call_id,
         codex_reason: reason,
         codex_grant_root: grant_root,
         codex_changes: changes,

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

@@ -1,5 +1,3 @@
-mod common;
-
 use std::collections::HashMap;
 use std::env;
 use std::path::Path;
@@ -8,6 +6,7 @@ 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_mcp_server::CodexToolCallParam;
 use codex_mcp_server::ExecApprovalElicitRequestParams;
 use codex_mcp_server::ExecApprovalResponse;
 use codex_mcp_server::PatchApprovalElicitRequestParams;
@@ -25,11 +24,11 @@ use tempfile::TempDir;
 use tokio::time::timeout;
 use wiremock::MockServer;
 
-use crate::common::McpProcess;
-use crate::common::create_apply_patch_sse_response;
-use crate::common::create_final_assistant_message_sse_response;
-use crate::common::create_mock_chat_completions_server;
-use crate::common::create_shell_sse_response;
+use mcp_test_support::McpProcess;
+use mcp_test_support::create_apply_patch_sse_response;
+use mcp_test_support::create_final_assistant_message_sse_response;
+use mcp_test_support::create_mock_chat_completions_server;
+use mcp_test_support::create_shell_sse_response;
 
 const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
 
@@ -76,7 +75,10 @@ async fn shell_command_approval_triggers_elicitation() -> anyhow::Result<()> {
     // In turn, it should reply with a tool call, which the MCP should forward
     // as an elicitation.
     let codex_request_id = mcp_process
-        .send_codex_tool_call(None, "run `git init`")
+        .send_codex_tool_call(CodexToolCallParam {
+            prompt: "run `git init`".to_string(),
+            ..Default::default()
+        })
         .await?;
     let elicitation_request = timeout(
         DEFAULT_READ_TIMEOUT,
@@ -167,6 +169,7 @@ fn create_expected_elicitation_request(
             codex_event_id,
             codex_command: command,
             codex_cwd: workdir.to_path_buf(),
+            codex_call_id: "call1234".to_string(),
         })?),
     })
 }
@@ -209,10 +212,11 @@ async fn patch_approval_triggers_elicitation() -> anyhow::Result<()> {
 
     // Send a "codex" tool request that will trigger the apply_patch command
     let codex_request_id = mcp_process
-        .send_codex_tool_call(
-            Some(cwd.path().to_string_lossy().to_string()),
-            "please modify the test file",
-        )
+        .send_codex_tool_call(CodexToolCallParam {
+            cwd: Some(cwd.path().to_string_lossy().to_string()),
+            prompt: "please modify the test file".to_string(),
+            ..Default::default()
+        })
         .await?;
     let elicitation_request = timeout(
         DEFAULT_READ_TIMEOUT,
@@ -279,6 +283,75 @@ async fn patch_approval_triggers_elicitation() -> anyhow::Result<()> {
     Ok(())
 }
 
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn test_codex_tool_passes_base_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;
+    }
+
+    // Apparently `#[tokio::test]` must return `()`, so we create a helper
+    // function that returns `Result` so we can use `?` in favor of `unwrap`.
+    if let Err(err) = codex_tool_passes_base_instructions().await {
+        panic!("failure: {err}");
+    }
+}
+
+async fn codex_tool_passes_base_instructions() -> anyhow::Result<()> {
+    #![allow(clippy::unwrap_used)]
+
+    let server =
+        create_mock_chat_completions_server(vec![create_final_assistant_message_sse_response(
+            "Enjoy!",
+        )?])
+        .await;
+
+    // Run `codex mcp` with a specific config.toml.
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
+    let mut mcp_process = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp_process.initialize()).await??;
+
+    // Send a "codex" tool request, which should hit the completions endpoint.
+    let codex_request_id = mcp_process
+        .send_codex_tool_call(CodexToolCallParam {
+            prompt: "How are you?".to_string(),
+            base_instructions: Some("You are a helpful assistant.".to_string()),
+            ..Default::default()
+        })
+        .await?;
+
+    let codex_response = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp_process.read_stream_until_response_message(RequestId::Integer(codex_request_id)),
+    )
+    .await??;
+    assert_eq!(
+        JSONRPCResponse {
+            jsonrpc: JSONRPC_VERSION.into(),
+            id: RequestId::Integer(codex_request_id),
+            result: json!({
+                "content": [
+                    {
+                        "text": "Enjoy!",
+                        "type": "text"
+                    }
+                ]
+            }),
+        },
+        codex_response
+    );
+
+    let requests = server.received_requests().await.unwrap();
+    let request = requests[0].body_json::<serde_json::Value>().unwrap();
+    let instructions = request["messages"][0]["content"].as_str().unwrap();
+    assert!(instructions.starts_with("You are a helpful assistant."));
+
+    Ok(())
+}
+
 fn create_expected_patch_approval_elicitation_request(
     elicitation_request_id: RequestId,
     changes: HashMap<PathBuf, FileChange>,
@@ -310,6 +383,7 @@ fn create_expected_patch_approval_elicitation_request(
             codex_reason: reason,
             codex_grant_root: grant_root,
             codex_changes: changes,
+            codex_call_id: "call1234".to_string(),
         })?),
     })
 }

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

@@ -0,0 +1,24 @@
+[package]
+name = "mcp_test_support"
+version = { workspace = true }
+edition = "2024"
+
+[lib]
+path = "lib.rs"
+
+[dependencies]
+anyhow = "1"
+assert_cmd = "2"
+codex-mcp-server = { path = "../.." }
+mcp-types = { path = "../../../mcp-types" }
+pretty_assertions = "1.4.1"
+serde_json = "1"
+shlex = "1.3.0"
+tempfile = "3"
+tokio = { version = "1", features = [
+    "io-std",
+    "macros",
+    "process",
+    "rt-multi-thread",
+] }
+wiremock = "0.6"

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

@@ -12,6 +12,7 @@ use tokio::process::ChildStdout;
 use anyhow::Context;
 use assert_cmd::prelude::*;
 use codex_mcp_server::CodexToolCallParam;
+use codex_mcp_server::CodexToolCallReplyParam;
 use mcp_types::CallToolRequestParams;
 use mcp_types::ClientCapabilities;
 use mcp_types::Implementation;
@@ -141,19 +142,29 @@ impl McpProcess {
     /// correlating notifications.
     pub async fn send_codex_tool_call(
         &mut self,
-        cwd: Option<String>,
-        prompt: &str,
+        params: CodexToolCallParam,
     ) -> anyhow::Result<i64> {
         let codex_tool_call_params = CallToolRequestParams {
             name: "codex".to_string(),
-            arguments: Some(serde_json::to_value(CodexToolCallParam {
-                cwd,
+            arguments: Some(serde_json::to_value(params)?),
+        };
+        self.send_request(
+            mcp_types::CallToolRequest::METHOD,
+            Some(serde_json::to_value(codex_tool_call_params)?),
+        )
+        .await
+    }
+
+    pub async fn send_codex_reply_tool_call(
+        &mut self,
+        session_id: &str,
+        prompt: &str,
+    ) -> anyhow::Result<i64> {
+        let codex_tool_call_params = CallToolRequestParams {
+            name: "codex-reply".to_string(),
+            arguments: Some(serde_json::to_value(CodexToolCallReplyParam {
                 prompt: prompt.to_string(),
-                model: None,
-                profile: None,
-                approval_policy: None,
-                sandbox: None,
-                config: None,
+                session_id: session_id.to_string(),
             })?),
         };
         self.send_request(
@@ -207,7 +218,6 @@ impl McpProcess {
         let message = serde_json::from_str::<JSONRPCMessage>(&line)?;
         Ok(message)
     }
-
     pub async fn read_stream_until_request_message(&mut self) -> anyhow::Result<JSONRPCRequest> {
         loop {
             let message = self.read_jsonrpc_message().await?;
@@ -256,4 +266,56 @@ impl McpProcess {
             }
         }
     }
+
+    pub async fn read_stream_until_configured_response_message(
+        &mut self,
+    ) -> anyhow::Result<String> {
+        loop {
+            let message = self.read_jsonrpc_message().await?;
+            eprint!("message: {message:?}");
+
+            match message {
+                JSONRPCMessage::Notification(notification) => {
+                    if notification.method == "codex/event" {
+                        if let Some(params) = notification.params {
+                            if let Some(msg) = params.get("msg") {
+                                if let Some(msg_type) = msg.get("type") {
+                                    if msg_type == "session_configured" {
+                                        if let Some(session_id) = msg.get("session_id") {
+                                            return Ok(session_id
+                                                .to_string()
+                                                .trim_matches('"')
+                                                .to_string());
+                                        }
+                                    }
+                                }
+                            }
+                        }
+                    }
+                }
+                JSONRPCMessage::Request(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Request: {message:?}");
+                }
+                JSONRPCMessage::Error(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Error: {message:?}");
+                }
+                JSONRPCMessage::Response(_) => {
+                    anyhow::bail!("unexpected JSONRPCMessage::Response: {message:?}");
+                }
+            }
+        }
+    }
+
+    pub async fn send_notification(
+        &mut self,
+        method: &str,
+        params: Option<serde_json::Value>,
+    ) -> anyhow::Result<()> {
+        self.send_jsonrpc_message(JSONRPCMessage::Notification(JSONRPCNotification {
+            jsonrpc: JSONRPC_VERSION.into(),
+            method: method.to_string(),
+            params,
+        }))
+        .await
+    }
 }

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

@@ -0,0 +1,176 @@
+#![cfg(unix)]
+// Support code lives in the `mcp_test_support` crate under tests/common.
+
+use std::path::Path;
+
+use codex_core::exec::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
+use codex_mcp_server::CodexToolCallParam;
+use mcp_types::JSONRPCResponse;
+use mcp_types::RequestId;
+use serde_json::json;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+use mcp_test_support::McpProcess;
+use mcp_test_support::create_mock_chat_completions_server;
+use mcp_test_support::create_shell_sse_response;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn test_shell_command_interruption() {
+    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;
+    }
+
+    if let Err(err) = shell_command_interruption().await {
+        panic!("failure: {err}");
+    }
+}
+
+async fn shell_command_interruption() -> anyhow::Result<()> {
+    // Use a cross-platform blocking command. On Windows plain `sleep` is not guaranteed to exist
+    // (MSYS/GNU coreutils may be absent) and the failure causes the tool call to finish immediately,
+    // which triggers a second model request before the test sends the explicit follow-up. That
+    // prematurely consumes the second mocked SSE response and leads to a third POST (panic: no response for 2).
+    // Powershell Start-Sleep is always available on Windows runners. On Unix we keep using `sleep`.
+    #[cfg(target_os = "windows")]
+    let shell_command = vec![
+        "powershell".to_string(),
+        "-Command".to_string(),
+        "Start-Sleep -Seconds 60".to_string(),
+    ];
+    #[cfg(not(target_os = "windows"))]
+    let shell_command = vec!["sleep".to_string(), "60".to_string()];
+    let workdir_for_shell_function_call = TempDir::new()?;
+
+    // Create mock server with a single SSE response: the long sleep command
+    let server = create_mock_chat_completions_server(vec![
+        create_shell_sse_response(
+            shell_command.clone(),
+            Some(workdir_for_shell_function_call.path()),
+            Some(60_000), // 60 seconds timeout in ms
+            "call_sleep",
+        )?,
+        create_shell_sse_response(
+            shell_command.clone(),
+            Some(workdir_for_shell_function_call.path()),
+            Some(60_000), // 60 seconds timeout in ms
+            "call_sleep",
+        )?,
+    ])
+    .await;
+
+    // Create Codex configuration
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), server.uri())?;
+    let mut mcp_process = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp_process.initialize()).await??;
+
+    // Send codex tool call that triggers "sleep 60"
+    let codex_request_id = mcp_process
+        .send_codex_tool_call(CodexToolCallParam {
+            cwd: None,
+            prompt: "First Run: run `sleep 60`".to_string(),
+            model: None,
+            profile: None,
+            approval_policy: None,
+            sandbox: None,
+            config: None,
+            base_instructions: None,
+        })
+        .await?;
+
+    let session_id = mcp_process
+        .read_stream_until_configured_response_message()
+        .await?;
+
+    // Give the command a moment to start
+    tokio::time::sleep(std::time::Duration::from_secs(1)).await;
+
+    // Send interrupt notification
+    mcp_process
+        .send_notification(
+            "notifications/cancelled",
+            Some(json!({ "requestId": codex_request_id })),
+        )
+        .await?;
+
+    // Expect Codex to return an error or interruption response
+    let codex_response: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp_process.read_stream_until_response_message(RequestId::Integer(codex_request_id)),
+    )
+    .await??;
+
+    assert!(
+        codex_response
+            .result
+            .as_object()
+            .map(|o| o.contains_key("error"))
+            .unwrap_or(false),
+        "Expected an interruption or error result, got: {codex_response:?}"
+    );
+
+    let codex_reply_request_id = mcp_process
+        .send_codex_reply_tool_call(&session_id, "Second Run: run `sleep 60`")
+        .await?;
+
+    // Give the command a moment to start
+    tokio::time::sleep(std::time::Duration::from_secs(1)).await;
+
+    // Send interrupt notification
+    mcp_process
+        .send_notification(
+            "notifications/cancelled",
+            Some(json!({ "requestId": codex_reply_request_id })),
+        )
+        .await?;
+
+    // Expect Codex to return an error or interruption response
+    let codex_response: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp_process.read_stream_until_response_message(RequestId::Integer(codex_reply_request_id)),
+    )
+    .await??;
+
+    assert!(
+        codex_response
+            .result
+            .as_object()
+            .map(|o| o.contains_key("error"))
+            .unwrap_or(false),
+        "Expected an interruption or error result, got: {codex_response:?}"
+    );
+    Ok(())
+}
+
+// ---------------------------------------------------------------------------
+// Helpers
+// ---------------------------------------------------------------------------
+
+fn create_config_toml(codex_home: &Path, server_uri: String) -> 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/tui/Cargo.toml

@@ -42,8 +42,8 @@ ratatui-image = "8.0.0"
 regex-lite = "0.1"
 serde_json = { version = "1", features = ["preserve_order"] }
 shlex = "1.3.0"
-strum = "0.27.1"
-strum_macros = "0.27.1"
+strum = "0.27.2"
+strum_macros = "0.27.2"
 tokio = { version = "1", features = [
     "io-std",
     "macros",
@@ -58,6 +58,7 @@ 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]

codex-rs/tui/src/app.rs

@@ -6,7 +6,6 @@ use crate::get_git_diff::get_git_diff;
 use crate::git_warning_screen::GitWarningOutcome;
 use crate::git_warning_screen::GitWarningScreen;
 use crate::login_screen::LoginScreen;
-use crate::mouse_capture::MouseCapture;
 use crate::scroll_event_helper::ScrollEventHelper;
 use crate::slash_command::SlashCommand;
 use crate::tui;
@@ -197,17 +196,17 @@ impl App<'_> {
         });
     }
 
-    pub(crate) fn run(
-        &mut self,
-        terminal: &mut tui::Tui,
-        mouse_capture: &mut MouseCapture,
-    ) -> Result<()> {
+    pub(crate) fn run(&mut self, terminal: &mut tui::Tui) -> Result<()> {
         // Insert an event to trigger the first render.
         let app_event_tx = self.app_event_tx.clone();
         app_event_tx.send(AppEvent::RequestRedraw);
 
         while let Ok(event) = self.app_event_rx.recv() {
             match event {
+                AppEvent::InsertHistory(lines) => {
+                    crate::insert_history::insert_history_lines(terminal, lines);
+                    self.app_event_tx.send(AppEvent::RequestRedraw);
+                }
                 AppEvent::RequestRedraw => {
                     self.schedule_redraw();
                 }
@@ -223,9 +222,7 @@ impl App<'_> {
                         } => {
                             match &mut self.app_state {
                                 AppState::Chat { widget } => {
-                                    if widget.on_ctrl_c() {
-                                        self.app_event_tx.send(AppEvent::ExitRequest);
-                                    }
+                                    widget.on_ctrl_c();
                                 }
                                 AppState::Login { .. } | AppState::GitWarning { .. } => {
                                     // No-op.
@@ -289,11 +286,6 @@ impl App<'_> {
                         self.app_state = AppState::Chat { widget: new_widget };
                         self.app_event_tx.send(AppEvent::RequestRedraw);
                     }
-                    SlashCommand::ToggleMouseMode => {
-                        if let Err(e) = mouse_capture.toggle() {
-                            tracing::error!("Failed to toggle mouse mode: {e}");
-                        }
-                    }
                     SlashCommand::Quit => {
                         break;
                     }
@@ -334,6 +326,15 @@ impl App<'_> {
         Ok(())
     }
 
+    pub(crate) fn token_usage(&self) -> codex_core::protocol::TokenUsage {
+        match &self.app_state {
+            AppState::Chat { widget } => widget.token_usage().clone(),
+            AppState::Login { .. } | AppState::GitWarning { .. } => {
+                codex_core::protocol::TokenUsage::default()
+            }
+        }
+    }
+
     fn draw_next_frame(&mut self, terminal: &mut tui::Tui) -> Result<()> {
         // TODO: add a throttle to avoid redrawing too often
 

codex-rs/tui/src/app_event.rs

@@ -1,6 +1,7 @@
 use codex_core::protocol::Event;
 use codex_file_search::FileMatch;
 use crossterm::event::KeyEvent;
+use ratatui::text::Line;
 
 use crate::slash_command::SlashCommand;
 
@@ -49,4 +50,6 @@ pub(crate) enum AppEvent {
         query: String,
         matches: Vec<FileMatch>,
     },
+
+    InsertHistory(Vec<Line<'static>>),
 }

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

@@ -50,10 +50,6 @@ impl<'a> BottomPaneView<'a> for ApprovalModalView<'a> {
         self.current.is_complete() && self.queue.is_empty()
     }
 
-    fn calculate_required_height(&self, area: &Rect) -> u16 {
-        self.current.get_height(area)
-    }
-
     fn render(&self, area: Rect, buf: &mut Buffer) {
         (&self.current).render_ref(area, buf);
     }

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

@@ -22,9 +22,6 @@ pub(crate) trait BottomPaneView<'a> {
         false
     }
 
-    /// Height required to render the view.
-    fn calculate_required_height(&self, area: &Rect) -> 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.rs

@@ -22,11 +22,6 @@ use crate::app_event::AppEvent;
 use crate::app_event_sender::AppEventSender;
 use codex_file_search::FileMatch;
 
-/// Minimum number of visible text rows inside the textarea.
-const MIN_TEXTAREA_ROWS: usize = 1;
-/// Rows consumed by the border.
-const BORDER_LINES: u16 = 2;
-
 const BASE_PLACEHOLDER_TEXT: &str = "send a message";
 /// If the pasted content exceeds this number of characters, replace it with a
 /// placeholder in the UI.
@@ -609,17 +604,6 @@ impl ChatComposer<'_> {
         self.dismissed_file_popup_token = None;
     }
 
-    pub fn calculate_required_height(&self, area: &Rect) -> u16 {
-        let rows = self.textarea.lines().len().max(MIN_TEXTAREA_ROWS);
-        let num_popup_rows = match &self.active_popup {
-            ActivePopup::Command(popup) => popup.calculate_required_height(area),
-            ActivePopup::File(popup) => popup.calculate_required_height(area),
-            ActivePopup::None => 0,
-        };
-
-        rows as u16 + BORDER_LINES + num_popup_rows
-    }
-
     fn update_border(&mut self, has_focus: bool) {
         struct BlockState {
             right_title: Line<'static>,

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

@@ -65,10 +65,8 @@ impl BottomPane<'_> {
             if !view.is_complete() {
                 self.active_view = Some(view);
             } else if self.is_task_running {
-                let height = self.composer.calculate_required_height(&Rect::default());
                 self.active_view = Some(Box::new(StatusIndicatorView::new(
                     self.app_event_tx.clone(),
-                    height,
                 )));
             }
             self.request_redraw();
@@ -138,10 +136,8 @@ impl BottomPane<'_> {
         match (running, self.active_view.is_some()) {
             (true, false) => {
                 // Show status indicator overlay.
-                let height = self.composer.calculate_required_height(&Rect::default());
                 self.active_view = Some(Box::new(StatusIndicatorView::new(
                     self.app_event_tx.clone(),
-                    height,
                 )));
                 self.request_redraw();
             }
@@ -203,14 +199,6 @@ impl BottomPane<'_> {
     }
 
     /// Height (terminal rows) required by the current bottom pane.
-    pub fn calculate_required_height(&self, area: &Rect) -> u16 {
-        if let Some(view) = &self.active_view {
-            view.calculate_required_height(area)
-        } else {
-            self.composer.calculate_required_height(area)
-        }
-    }
-
     pub(crate) fn request_redraw(&self) {
         self.app_event_tx.send(AppEvent::RequestRedraw)
     }

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

@@ -1,5 +1,4 @@
 use ratatui::buffer::Buffer;
-use ratatui::layout::Rect;
 use ratatui::widgets::WidgetRef;
 
 use crate::app_event_sender::AppEventSender;
@@ -13,9 +12,9 @@ pub(crate) struct StatusIndicatorView {
 }
 
 impl StatusIndicatorView {
-    pub fn new(app_event_tx: AppEventSender, height: u16) -> Self {
+    pub fn new(app_event_tx: AppEventSender) -> Self {
         Self {
-            view: StatusIndicatorWidget::new(app_event_tx, height),
+            view: StatusIndicatorWidget::new(app_event_tx),
         }
     }
 
@@ -34,11 +33,7 @@ impl BottomPaneView<'_> for StatusIndicatorView {
         true
     }
 
-    fn calculate_required_height(&self, _area: &Rect) -> u16 {
-        self.view.get_height()
-    }
-
-    fn render(&self, area: Rect, buf: &mut Buffer) {
+    fn render(&self, area: ratatui::layout::Rect, buf: &mut Buffer) {
         self.view.render_ref(area, buf);
     }
 }

codex-rs/tui/src/chatwidget.rs

@@ -23,9 +23,6 @@ use codex_core::protocol::TaskCompleteEvent;
 use codex_core::protocol::TokenUsage;
 use crossterm::event::KeyEvent;
 use ratatui::buffer::Buffer;
-use ratatui::layout::Constraint;
-use ratatui::layout::Direction;
-use ratatui::layout::Layout;
 use ratatui::layout::Rect;
 use ratatui::widgets::Widget;
 use ratatui::widgets::WidgetRef;
@@ -52,6 +49,9 @@ pub(crate) struct ChatWidget<'a> {
     initial_user_message: Option<UserMessage>,
     token_usage: TokenUsage,
     reasoning_buffer: String,
+    // Buffer for streaming assistant answer text; we do not surface partial
+    // We wait for the final AgentMessage event and then emit the full text
+    // at once into scrollback so the history contains a single message.
     answer_buffer: String,
 }
 
@@ -187,6 +187,13 @@ impl ChatWidget<'_> {
         }
     }
 
+    /// Emits the last entry's plain lines from conversation_history, if any.
+    fn emit_last_history_entry(&mut self) {
+        if let Some(lines) = self.conversation_history.last_entry_plain_lines() {
+            self.app_event_tx.send(AppEvent::InsertHistory(lines));
+        }
+    }
+
     fn submit_user_message(&mut self, user_message: UserMessage) {
         let UserMessage { text, image_paths } = user_message;
         let mut items: Vec<InputItem> = Vec::new();
@@ -220,7 +227,8 @@ impl ChatWidget<'_> {
 
         // Only show text portion in conversation history for now.
         if !text.is_empty() {
-            self.conversation_history.add_user_message(text);
+            self.conversation_history.add_user_message(text.clone());
+            self.emit_last_history_entry();
         }
         self.conversation_history.scroll_to_bottom();
     }
@@ -232,6 +240,10 @@ impl ChatWidget<'_> {
                 // Record session information at the top of the conversation.
                 self.conversation_history
                     .add_session_info(&self.config, event.clone());
+                // Immediately surface the session banner / settings summary in
+                // scrollback so the user can review configuration (model,
+                // sandbox, approvals, etc.) before interacting.
+                self.emit_last_history_entry();
 
                 // Forward history metadata to the bottom pane so the chat
                 // composer can navigate through past messages.
@@ -247,50 +259,50 @@ impl ChatWidget<'_> {
                 self.request_redraw();
             }
             EventMsg::AgentMessage(AgentMessageEvent { message }) => {
-                // if the answer buffer is empty, this means we haven't received any
-                // delta. Thus, we need to print the message as a new answer.
-                if self.answer_buffer.is_empty() {
-                    self.conversation_history
-                        .add_agent_message(&self.config, message);
+                // Final assistant answer. Prefer the fully provided message
+                // from the event; if it is empty fall back to any accumulated
+                // delta buffer (some providers may only stream deltas and send
+                // an empty final message).
+                let full = if message.is_empty() {
+                    std::mem::take(&mut self.answer_buffer)
                 } else {
+                    self.answer_buffer.clear();
+                    message
+                };
+                if !full.is_empty() {
                     self.conversation_history
-                        .replace_prev_agent_message(&self.config, message);
+                        .add_agent_message(&self.config, full);
+                    self.emit_last_history_entry();
                 }
-                self.answer_buffer.clear();
                 self.request_redraw();
             }
             EventMsg::AgentMessageDelta(AgentMessageDeltaEvent { delta }) => {
-                if self.answer_buffer.is_empty() {
-                    self.conversation_history
-                        .add_agent_message(&self.config, "".to_string());
-                }
-                self.answer_buffer.push_str(&delta.clone());
-                self.conversation_history
-                    .replace_prev_agent_message(&self.config, self.answer_buffer.clone());
-                self.request_redraw();
+                // Buffer only – do not emit partial lines. This avoids cases
+                // where long responses appear truncated if the terminal
+                // wrapped early. The full message is emitted on
+                // AgentMessage.
+                self.answer_buffer.push_str(&delta);
             }
             EventMsg::AgentReasoningDelta(AgentReasoningDeltaEvent { delta }) => {
-                if self.reasoning_buffer.is_empty() {
-                    self.conversation_history
-                        .add_agent_reasoning(&self.config, "".to_string());
-                }
-                self.reasoning_buffer.push_str(&delta.clone());
-                self.conversation_history
-                    .replace_prev_agent_reasoning(&self.config, self.reasoning_buffer.clone());
-                self.request_redraw();
+                // Buffer only – disable incremental reasoning streaming so we
+                // avoid truncated intermediate lines. Full text emitted on
+                // AgentReasoning.
+                self.reasoning_buffer.push_str(&delta);
             }
             EventMsg::AgentReasoning(AgentReasoningEvent { text }) => {
-                // if the reasoning buffer is empty, this means we haven't received any
-                // delta. Thus, we need to print the message as a new reasoning.
-                if self.reasoning_buffer.is_empty() {
-                    self.conversation_history
-                        .add_agent_reasoning(&self.config, "".to_string());
+                // Emit full reasoning text once. Some providers might send
+                // final event with empty text if only deltas were used.
+                let full = if text.is_empty() {
+                    std::mem::take(&mut self.reasoning_buffer)
                 } else {
-                    // else, we rerender one last time.
+                    self.reasoning_buffer.clear();
+                    text
+                };
+                if !full.is_empty() {
                     self.conversation_history
-                        .replace_prev_agent_reasoning(&self.config, text);
+                        .add_agent_reasoning(&self.config, full);
+                    self.emit_last_history_entry();
                 }
-                self.reasoning_buffer.clear();
                 self.request_redraw();
             }
             EventMsg::TaskStarted => {
@@ -310,10 +322,12 @@ impl ChatWidget<'_> {
                     .set_token_usage(self.token_usage.clone(), self.config.model_context_window);
             }
             EventMsg::Error(ErrorEvent { message }) => {
-                self.conversation_history.add_error(message);
+                self.conversation_history.add_error(message.clone());
+                self.emit_last_history_entry();
                 self.bottom_pane.set_task_running(false);
             }
             EventMsg::ExecApprovalRequest(ExecApprovalRequestEvent {
+                call_id: _,
                 command,
                 cwd,
                 reason,
@@ -327,6 +341,7 @@ impl ChatWidget<'_> {
                 self.bottom_pane.push_approval_request(request);
             }
             EventMsg::ApplyPatchApprovalRequest(ApplyPatchApprovalRequestEvent {
+                call_id: _,
                 changes,
                 reason,
                 grant_root,
@@ -344,6 +359,7 @@ impl ChatWidget<'_> {
 
                 self.conversation_history
                     .add_patch_event(PatchEventType::ApprovalRequest, changes);
+                self.emit_last_history_entry();
 
                 self.conversation_history.scroll_to_bottom();
 
@@ -363,6 +379,7 @@ impl ChatWidget<'_> {
             }) => {
                 self.conversation_history
                     .add_active_exec_command(call_id, command);
+                self.emit_last_history_entry();
                 self.request_redraw();
             }
             EventMsg::PatchApplyBegin(PatchApplyBeginEvent {
@@ -374,6 +391,7 @@ impl ChatWidget<'_> {
                 // summary so the user can follow along.
                 self.conversation_history
                     .add_patch_event(PatchEventType::ApplyBegin { auto_approved }, changes);
+                self.emit_last_history_entry();
                 if !auto_approved {
                     self.conversation_history.scroll_to_bottom();
                 }
@@ -397,6 +415,7 @@ impl ChatWidget<'_> {
             }) => {
                 self.conversation_history
                     .add_active_mcp_tool_call(call_id, server, tool, arguments);
+                self.emit_last_history_entry();
                 self.request_redraw();
             }
             EventMsg::McpToolCallEnd(mcp_tool_call_end_event) => {
@@ -417,9 +436,13 @@ impl ChatWidget<'_> {
                 self.bottom_pane
                     .on_history_entry_response(log_id, offset, entry.map(|e| e.text));
             }
+            EventMsg::ShutdownComplete => {
+                self.app_event_tx.send(AppEvent::ExitRequest);
+            }
             event => {
                 self.conversation_history
                     .add_background_event(format!("{event:?}"));
+                self.emit_last_history_entry();
                 self.request_redraw();
             }
         }
@@ -436,7 +459,9 @@ impl ChatWidget<'_> {
     }
 
     pub(crate) fn add_diff_output(&mut self, diff_output: String) {
-        self.conversation_history.add_diff_output(diff_output);
+        self.conversation_history
+            .add_diff_output(diff_output.clone());
+        self.emit_last_history_entry();
         self.request_redraw();
     }
 
@@ -469,6 +494,7 @@ impl ChatWidget<'_> {
             self.reasoning_buffer.clear();
             false
         } else if self.bottom_pane.ctrl_c_quit_hint_visible() {
+            self.submit_op(Op::Shutdown);
             true
         } else {
             self.bottom_pane.show_ctrl_c_quit_hint();
@@ -486,19 +512,18 @@ impl ChatWidget<'_> {
             tracing::error!("failed to submit op: {e}");
         }
     }
+
+    pub(crate) fn token_usage(&self) -> &TokenUsage {
+        &self.token_usage
+    }
 }
 
 impl WidgetRef for &ChatWidget<'_> {
     fn render_ref(&self, area: Rect, buf: &mut Buffer) {
-        let bottom_height = self.bottom_pane.calculate_required_height(&area);
-
-        let chunks = Layout::default()
-            .direction(Direction::Vertical)
-            .constraints([Constraint::Min(0), Constraint::Length(bottom_height)])
-            .split(area);
-
-        self.conversation_history.render(chunks[0], buf);
-        (&self.bottom_pane).render(chunks[1], buf);
+        // In the hybrid inline viewport mode we only draw the interactive
+        // bottom pane; history entries are injected directly into scrollback
+        // via `Terminal::insert_before`.
+        (&self.bottom_pane).render(area, buf);
     }
 }
 

codex-rs/tui/src/conversation_history_widget.rs

@@ -202,14 +202,6 @@ impl ConversationHistoryWidget {
         self.add_to_history(HistoryCell::new_agent_reasoning(config, text));
     }
 
-    pub fn replace_prev_agent_reasoning(&mut self, config: &Config, text: String) {
-        self.replace_last_agent_reasoning(config, text);
-    }
-
-    pub fn replace_prev_agent_message(&mut self, config: &Config, text: String) {
-        self.replace_last_agent_message(config, text);
-    }
-
     pub fn add_background_event(&mut self, message: String) {
         self.add_to_history(HistoryCell::new_background_event(message));
     }
@@ -257,40 +249,10 @@ impl ConversationHistoryWidget {
         });
     }
 
-    pub fn replace_last_agent_reasoning(&mut self, config: &Config, text: String) {
-        if let Some(idx) = self
-            .entries
-            .iter()
-            .rposition(|entry| matches!(entry.cell, HistoryCell::AgentReasoning { .. }))
-        {
-            let width = self.cached_width.get();
-            let entry = &mut self.entries[idx];
-            entry.cell = HistoryCell::new_agent_reasoning(config, text);
-            let height = if width > 0 {
-                entry.cell.height(width)
-            } else {
-                0
-            };
-            entry.line_count.set(height);
-        }
-    }
-
-    pub fn replace_last_agent_message(&mut self, config: &Config, text: String) {
-        if let Some(idx) = self
-            .entries
-            .iter()
-            .rposition(|entry| matches!(entry.cell, HistoryCell::AgentMessage { .. }))
-        {
-            let width = self.cached_width.get();
-            let entry = &mut self.entries[idx];
-            entry.cell = HistoryCell::new_agent_message(config, text);
-            let height = if width > 0 {
-                entry.cell.height(width)
-            } else {
-                0
-            };
-            entry.line_count.set(height);
-        }
+    /// Return the lines for the most recently appended entry (if any) so the
+    /// parent widget can surface them via the new scrollback insertion path.
+    pub(crate) fn last_entry_plain_lines(&self) -> Option<Vec<Line<'static>>> {
+        self.entries.last().map(|e| e.cell.plain_lines())
     }
 
     pub fn record_completed_exec_command(

codex-rs/tui/src/history_cell.rs

@@ -123,6 +123,30 @@ pub(crate) enum HistoryCell {
 const TOOL_CALL_MAX_LINES: usize = 5;
 
 impl HistoryCell {
+    /// Return a cloned, plain representation of the cell's lines suitable for
+    /// one‑shot insertion into the terminal scrollback. Image cells are
+    /// represented with a simple placeholder for now.
+    pub(crate) fn plain_lines(&self) -> Vec<Line<'static>> {
+        match self {
+            HistoryCell::WelcomeMessage { view }
+            | HistoryCell::UserPrompt { view }
+            | HistoryCell::AgentMessage { view }
+            | HistoryCell::AgentReasoning { view }
+            | HistoryCell::BackgroundEvent { view }
+            | HistoryCell::GitDiffOutput { view }
+            | HistoryCell::ErrorEvent { view }
+            | HistoryCell::SessionInfo { view }
+            | HistoryCell::CompletedExecCommand { view }
+            | HistoryCell::CompletedMcpToolCall { view }
+            | HistoryCell::PendingPatch { view }
+            | HistoryCell::ActiveExecCommand { view, .. }
+            | HistoryCell::ActiveMcpToolCall { view, .. } => view.lines.clone(),
+            HistoryCell::CompletedMcpToolCallWithImageOutput { .. } => vec![
+                Line::from("tool result (image output omitted)"),
+                Line::from(""),
+            ],
+        }
+    }
     pub(crate) fn new_session_info(
         config: &Config,
         event: SessionConfiguredEvent,
@@ -156,7 +180,7 @@ impl HistoryCell {
                 ("workdir", config.cwd.display().to_string()),
                 ("model", config.model.clone()),
                 ("provider", config.model_provider_id.clone()),
-                ("approval", format!("{:?}", config.approval_policy)),
+                ("approval", config.approval_policy.to_string()),
                 ("sandbox", summarize_sandbox_policy(&config.sandbox_policy)),
             ];
             if config.model_provider.wire_api == WireApi::Responses

codex-rs/tui/src/insert_history.rs

@@ -0,0 +1,181 @@
+use ratatui::layout::Rect;
+use ratatui::style::Style;
+use ratatui::text::Line;
+use ratatui::text::Span;
+use ratatui::widgets::Paragraph;
+use ratatui::widgets::Widget;
+use unicode_width::UnicodeWidthChar;
+
+/// Insert a batch of history lines into the terminal scrollback above the
+/// inline viewport.
+///
+/// The incoming `lines` are the logical lines supplied by the
+/// `ConversationHistory`. They may contain embedded newlines and arbitrary
+/// runs of whitespace inside individual [`Span`]s. All of that must be
+/// normalised before writing to the backing terminal buffer because the
+/// ratatui [`Paragraph`] widget does not perform soft‑wrapping when used in
+/// conjunction with [`Terminal::insert_before`].
+///
+/// This function performs a minimal wrapping / normalisation pass:
+///
+/// * A terminal width is determined via `Terminal::size()` (falling back to
+///   80 columns if the size probe fails).
+/// * Each logical line is broken into words and whitespace. Consecutive
+///   whitespace is collapsed to a single space; leading whitespace is
+///   discarded.
+/// * Words that do not fit on the current line cause a soft wrap. Extremely
+///   long words (longer than the terminal width) are split character by
+///   character so they still populate the display instead of overflowing the
+///   line.
+/// * Explicit `\n` characters inside a span force a hard line break.
+/// * Empty lines (including a trailing newline at the end of the batch) are
+///   preserved so vertical spacing remains faithful to the logical history.
+///
+/// Finally the physical lines are rendered directly into the terminal's
+/// scrollback region using [`Terminal::insert_before`]. Any backend error is
+/// ignored: failing to insert history is non‑fatal and a subsequent redraw
+/// will eventually repaint a consistent view.
+fn display_width(s: &str) -> usize {
+    s.chars()
+        .map(|c| UnicodeWidthChar::width(c).unwrap_or(0))
+        .sum()
+}
+
+struct LineBuilder {
+    term_width: usize,
+    spans: Vec<Span<'static>>,
+    width: usize,
+}
+
+impl LineBuilder {
+    fn new(term_width: usize) -> Self {
+        Self {
+            term_width,
+            spans: Vec::new(),
+            width: 0,
+        }
+    }
+
+    fn flush_line(&mut self, out: &mut Vec<Line<'static>>) {
+        out.push(Line::from(std::mem::take(&mut self.spans)));
+        self.width = 0;
+    }
+
+    fn push_segment(&mut self, text: String, style: Style) {
+        self.width += display_width(&text);
+        self.spans.push(Span::styled(text, style));
+    }
+
+    fn push_word(&mut self, word: &mut String, style: Style, out: &mut Vec<Line<'static>>) {
+        if word.is_empty() {
+            return;
+        }
+        let w_len = display_width(word);
+        if self.width > 0 && self.width + w_len > self.term_width {
+            self.flush_line(out);
+        }
+        if w_len > self.term_width && self.width == 0 {
+            // Split an overlong word across multiple lines.
+            let mut cur = String::new();
+            let mut cur_w = 0;
+            for ch in word.chars() {
+                let ch_w = UnicodeWidthChar::width(ch).unwrap_or(0);
+                if cur_w + ch_w > self.term_width && cur_w > 0 {
+                    self.push_segment(cur.clone(), style);
+                    self.flush_line(out);
+                    cur.clear();
+                    cur_w = 0;
+                }
+                cur.push(ch);
+                cur_w += ch_w;
+            }
+            if !cur.is_empty() {
+                self.push_segment(cur, style);
+            }
+        } else {
+            self.push_segment(word.clone(), style);
+        }
+        word.clear();
+    }
+
+    fn consume_whitespace(&mut self, ws: &mut String, style: Style, out: &mut Vec<Line<'static>>) {
+        if ws.is_empty() {
+            return;
+        }
+        let space_w = display_width(ws);
+        if self.width > 0 && self.width + space_w > self.term_width {
+            self.flush_line(out);
+        }
+        if self.width > 0 {
+            self.push_segment(" ".to_string(), style);
+        }
+        ws.clear();
+    }
+}
+
+use ratatui::backend::Backend;
+
+pub fn insert_history_lines<B: Backend>(terminal: &mut ratatui::Terminal<B>, lines: Vec<Line<'static>>) {
+    let term_width = terminal.size().map(|a| a.width).unwrap_or(80) as usize;
+    let mut physical: Vec<Line<'static>> = Vec::new();
+
+    for logical in lines.into_iter() {
+        if logical.spans.is_empty() {
+            physical.push(logical);
+            continue;
+        }
+
+        let mut builder = LineBuilder::new(term_width);
+        let mut buf_space = String::new();
+
+        for span in logical.spans.into_iter() {
+            let style = span.style;
+            let mut buf_word = String::new();
+
+            for ch in span.content.chars() {
+                if ch == '\n' {
+                    builder.push_word(&mut buf_word, style, &mut physical);
+                    buf_space.clear();
+                    builder.flush_line(&mut physical);
+                    continue;
+                }
+                if ch.is_whitespace() {
+                    builder.push_word(&mut buf_word, style, &mut physical);
+                    buf_space.push(ch);
+                } else {
+                    builder.consume_whitespace(&mut buf_space, style, &mut physical);
+                    buf_word.push(ch);
+                }
+                if builder.width >= term_width {
+                    builder.flush_line(&mut physical);
+                }
+            }
+            builder.push_word(&mut buf_word, style, &mut physical);
+            // whitespace intentionally left to allow collapsing across spans
+        }
+        if !builder.spans.is_empty() {
+            physical.push(Line::from(std::mem::take(&mut builder.spans)));
+        } else {
+            // Preserve explicit blank line (e.g. due to a trailing newline).
+            physical.push(Line::from(Vec::<Span<'static>>::new()));
+        }
+    }
+
+    let total = physical.len() as u16;
+    terminal
+        .insert_before(total, |buf| {
+            let width = buf.area.width;
+            for (i, line) in physical.into_iter().enumerate() {
+                let area = Rect {
+                    x: 0,
+                    y: i as u16,
+                    width,
+                    height: 1,
+                };
+                Paragraph::new(line).render(area, buf);
+            }
+        })
+        .ok();
+}
+
+// Tests are implemented as integration tests (see tests/insert_history.rs)

codex-rs/tui/src/lib.rs

@@ -33,10 +33,10 @@ mod file_search;
 mod get_git_diff;
 mod git_warning_screen;
 mod history_cell;
+mod insert_history;
 mod log_layer;
 mod login_screen;
 mod markdown;
-mod mouse_capture;
 mod scroll_event_helper;
 mod slash_command;
 mod status_indicator_widget;
@@ -45,9 +45,15 @@ mod text_formatting;
 mod tui;
 mod user_approval_widget;
 
+// Re-export for integration tests
+pub use insert_history::insert_history_lines;
+
 pub use cli::Cli;
 
-pub fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> std::io::Result<()> {
+pub fn run_main(
+    cli: Cli,
+    codex_linux_sandbox_exe: Option<PathBuf>,
+) -> std::io::Result<codex_core::protocol::TokenUsage> {
     let (sandbox_mode, approval_policy) = if cli.full_auto {
         (
             Some(SandboxMode::WorkspaceWrite),
@@ -75,6 +81,7 @@ pub fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> std::io::
             model_provider: None,
             config_profile: cli.config_profile.clone(),
             codex_linux_sandbox_exe,
+            base_instructions: None,
         };
         // Parse `-c` overrides from the CLI.
         let cli_kv_overrides = match cli.config_overrides.parse_overrides() {
@@ -146,24 +153,8 @@ pub fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> std::io::
     // `--allow-no-git-exec` flag.
     let show_git_warning = !cli.skip_git_repo_check && !is_inside_git_repo(&config);
 
-    try_run_ratatui_app(cli, config, show_login_screen, show_git_warning, log_rx);
-    Ok(())
-}
-
-#[expect(
-    clippy::print_stderr,
-    reason = "Resort to stderr in exceptional situations."
-)]
-fn try_run_ratatui_app(
-    cli: Cli,
-    config: Config,
-    show_login_screen: bool,
-    show_git_warning: bool,
-    log_rx: tokio::sync::mpsc::UnboundedReceiver<String>,
-) {
-    if let Err(report) = run_ratatui_app(cli, config, show_login_screen, show_git_warning, log_rx) {
-        eprintln!("Error: {report:?}");
-    }
+    run_ratatui_app(cli, config, show_login_screen, show_git_warning, log_rx)
+        .map_err(|err| std::io::Error::other(err.to_string()))
 }
 
 fn run_ratatui_app(
@@ -172,16 +163,15 @@ fn run_ratatui_app(
     show_login_screen: bool,
     show_git_warning: bool,
     mut log_rx: tokio::sync::mpsc::UnboundedReceiver<String>,
-) -> color_eyre::Result<()> {
+) -> color_eyre::Result<codex_core::protocol::TokenUsage> {
     color_eyre::install()?;
 
-    // Forward panic reports through the tracing stack so that they appear in
-    // the status indicator instead of breaking the alternate screen – the
-    // normal colour‑eyre hook writes to stderr which would corrupt the UI.
+    // Forward panic reports through tracing so they appear in the UI status
+    // line instead of interleaving raw panic output with the interface.
     std::panic::set_hook(Box::new(|info| {
         tracing::error!("panic: {info}");
     }));
-    let (mut terminal, mut mouse_capture) = tui::init(&config)?;
+    let mut terminal = tui::init(&config)?;
     terminal.clear()?;
 
     let Cli { prompt, images, .. } = cli;
@@ -203,10 +193,12 @@ fn run_ratatui_app(
         });
     }
 
-    let app_result = app.run(&mut terminal, &mut mouse_capture);
+    let app_result = app.run(&mut terminal);
+    let usage = app.token_usage();
 
     restore();
-    app_result
+    // ignore error when collecting usage – report underlying error instead
+    app_result.map(|_| usage)
 }
 
 #[expect(

codex-rs/tui/src/main.rs

@@ -20,7 +20,8 @@ fn main() -> anyhow::Result<()> {
             .config_overrides
             .raw_overrides
             .splice(0..0, top_cli.config_overrides.raw_overrides);
-        run_main(inner, codex_linux_sandbox_exe)?;
+        let usage = run_main(inner, codex_linux_sandbox_exe)?;
+        println!("{}", codex_core::protocol::FinalOutput::from(usage));
         Ok(())
     })
 }

codex-rs/tui/src/mouse_capture.rs

@@ -1,69 +0,0 @@
-use crossterm::event::DisableMouseCapture;
-use crossterm::event::EnableMouseCapture;
-use ratatui::crossterm::execute;
-use std::io::Result;
-use std::io::stdout;
-
-pub(crate) struct MouseCapture {
-    mouse_capture_is_active: bool,
-}
-
-impl MouseCapture {
-    pub(crate) fn new_with_capture(mouse_capture_is_active: bool) -> Result<Self> {
-        if mouse_capture_is_active {
-            enable_capture()?;
-        }
-
-        Ok(Self {
-            mouse_capture_is_active,
-        })
-    }
-}
-
-impl MouseCapture {
-    /// Idempotent method to set the mouse capture state.
-    pub fn set_active(&mut self, is_active: bool) -> Result<()> {
-        match (self.mouse_capture_is_active, is_active) {
-            (true, true) => {}
-            (false, false) => {}
-            (true, false) => {
-                disable_capture()?;
-                self.mouse_capture_is_active = false;
-            }
-            (false, true) => {
-                enable_capture()?;
-                self.mouse_capture_is_active = true;
-            }
-        }
-        Ok(())
-    }
-
-    pub(crate) fn toggle(&mut self) -> Result<()> {
-        self.set_active(!self.mouse_capture_is_active)
-    }
-
-    pub(crate) fn disable(&mut self) -> Result<()> {
-        if self.mouse_capture_is_active {
-            disable_capture()?;
-            self.mouse_capture_is_active = false;
-        }
-        Ok(())
-    }
-}
-
-impl Drop for MouseCapture {
-    fn drop(&mut self) {
-        if self.disable().is_err() {
-            // The user is likely shutting down, so ignore any errors so the
-            // shutdown process can complete.
-        }
-    }
-}
-
-fn enable_capture() -> Result<()> {
-    execute!(stdout(), EnableMouseCapture)
-}
-
-fn disable_capture() -> Result<()> {
-    execute!(stdout(), DisableMouseCapture)
-}

codex-rs/tui/src/slash_command.rs

@@ -15,7 +15,6 @@ pub enum SlashCommand {
     New,
     Diff,
     Quit,
-    ToggleMouseMode,
 }
 
 impl SlashCommand {
@@ -23,9 +22,6 @@ impl SlashCommand {
     pub fn description(self) -> &'static str {
         match self {
             SlashCommand::New => "Start a new chat.",
-            SlashCommand::ToggleMouseMode => {
-                "Toggle mouse mode (enable for scrolling, disable for text selection)"
-            }
             SlashCommand::Quit => "Exit the application.",
             SlashCommand::Diff => {
                 "Show git diff of the working directory (including untracked files)"

codex-rs/tui/src/status_indicator_widget.rs

@@ -34,11 +34,6 @@ pub(crate) struct StatusIndicatorWidget {
     /// time).
     text: String,
 
-    /// Height in terminal rows – matches the height of the textarea at the
-    /// moment the task started so the UI does not jump when we toggle between
-    /// input mode and loading mode.
-    height: u16,
-
     frame_idx: Arc<AtomicUsize>,
     running: Arc<AtomicBool>,
     // Keep one sender alive to prevent the channel from closing while the
@@ -50,7 +45,7 @@ pub(crate) struct StatusIndicatorWidget {
 
 impl StatusIndicatorWidget {
     /// Create a new status indicator and start the animation timer.
-    pub(crate) fn new(app_event_tx: AppEventSender, height: u16) -> Self {
+    pub(crate) fn new(app_event_tx: AppEventSender) -> Self {
         let frame_idx = Arc::new(AtomicUsize::new(0));
         let running = Arc::new(AtomicBool::new(true));
 
@@ -72,18 +67,12 @@ impl StatusIndicatorWidget {
 
         Self {
             text: String::from("waiting for logs…"),
-            height: height.max(3),
             frame_idx,
             running,
             _app_event_tx: app_event_tx,
         }
     }
 
-    /// Preferred height in terminal rows.
-    pub(crate) fn get_height(&self) -> u16 {
-        self.height
-    }
-
     /// Update the line that is displayed in the widget.
     pub(crate) fn update_text(&mut self, text: String) {
         self.text = text.replace(['\n', '\r'], " ");

codex-rs/tui/src/tui.rs

@@ -4,31 +4,39 @@ use std::io::stdout;
 
 use codex_core::config::Config;
 use crossterm::event::DisableBracketedPaste;
-use crossterm::event::DisableMouseCapture;
 use crossterm::event::EnableBracketedPaste;
 use ratatui::Terminal;
+use ratatui::TerminalOptions;
+use ratatui::Viewport;
 use ratatui::backend::CrosstermBackend;
 use ratatui::crossterm::execute;
-use ratatui::crossterm::terminal::EnterAlternateScreen;
-use ratatui::crossterm::terminal::LeaveAlternateScreen;
 use ratatui::crossterm::terminal::disable_raw_mode;
 use ratatui::crossterm::terminal::enable_raw_mode;
 
-use crate::mouse_capture::MouseCapture;
-
 /// A type alias for the terminal type used in this application
 pub type Tui = Terminal<CrosstermBackend<Stdout>>;
 
-/// Initialize the terminal
-pub fn init(config: &Config) -> Result<(Tui, MouseCapture)> {
-    execute!(stdout(), EnterAlternateScreen)?;
+/// Initialize the terminal (inline viewport; history stays in normal scrollback)
+pub fn init(_config: &Config) -> Result<Tui> {
     execute!(stdout(), EnableBracketedPaste)?;
-    let mouse_capture = MouseCapture::new_with_capture(!config.tui.disable_mouse_capture)?;
 
     enable_raw_mode()?;
     set_panic_hook();
-    let tui = Terminal::new(CrosstermBackend::new(stdout()))?;
-    Ok((tui, mouse_capture))
+
+    // Reserve a fixed number of lines for the interactive viewport (composer,
+    // status, popups). History is injected above using `insert_before`. This
+    // is an initial step of the refactor – later the height can become
+    // dynamic. For now a conservative default keeps enough room for the
+    // multi‑line composer while not occupying the whole screen.
+    const BOTTOM_VIEWPORT_HEIGHT: u16 = 8;
+    let backend = CrosstermBackend::new(stdout());
+    let tui = Terminal::with_options(
+        backend,
+        TerminalOptions {
+            viewport: Viewport::Inline(BOTTOM_VIEWPORT_HEIGHT),
+        },
+    )?;
+    Ok(tui)
 }
 
 fn set_panic_hook() {
@@ -41,14 +49,7 @@ fn set_panic_hook() {
 
 /// Restore the terminal to its original state
 pub fn restore() -> Result<()> {
-    // We are shutting down, and we cannot reference the `MouseCapture`, so we
-    // categorically disable mouse capture just to be safe.
-    if execute!(stdout(), DisableMouseCapture).is_err() {
-        // It is possible that `DisableMouseCapture` is written more than once
-        // on shutdown, so ignore the error in this case.
-    }
     execute!(stdout(), DisableBracketedPaste)?;
-    execute!(stdout(), LeaveAlternateScreen)?;
     disable_raw_mode()?;
     Ok(())
 }

codex-rs/tui/src/user_approval_widget.rs

@@ -116,10 +116,6 @@ pub(crate) struct UserApprovalWidget<'a> {
     done: bool,
 }
 
-// Number of lines automatically added by ratatui’s [`Block`] when
-// borders are enabled (one at the top, one at the bottom).
-const BORDER_LINES: u16 = 2;
-
 impl UserApprovalWidget<'_> {
     pub(crate) fn new(approval_request: ApprovalRequest, app_event_tx: AppEventSender) -> Self {
         let input = Input::default();
@@ -190,28 +186,6 @@ impl UserApprovalWidget<'_> {
         }
     }
 
-    pub(crate) fn get_height(&self, area: &Rect) -> u16 {
-        let confirmation_prompt_height =
-            self.get_confirmation_prompt_height(area.width - BORDER_LINES);
-
-        match self.mode {
-            Mode::Select => {
-                let num_option_lines = SELECT_OPTIONS.len() as u16;
-                confirmation_prompt_height + num_option_lines + BORDER_LINES
-            }
-            Mode::Input => {
-                //   1. "Give the model feedback ..." prompt
-                //   2. A single‑line input field (we allocate exactly one row;
-                //      the `tui-input` widget will scroll horizontally if the
-                //      text exceeds the width).
-                const INPUT_PROMPT_LINES: u16 = 1;
-                const INPUT_FIELD_LINES: u16 = 1;
-
-                confirmation_prompt_height + INPUT_PROMPT_LINES + INPUT_FIELD_LINES + BORDER_LINES
-            }
-        }
-    }
-
     fn get_confirmation_prompt_height(&self, width: u16) -> u16 {
         // Should cache this for last value of width.
         self.confirmation_prompt.line_count(width) as u16
@@ -333,7 +307,32 @@ impl WidgetRef for &UserApprovalWidget<'_> {
             .borders(Borders::ALL)
             .border_type(BorderType::Rounded);
         let inner = outer.inner(area);
-        let prompt_height = self.get_confirmation_prompt_height(inner.width);
+
+        // Determine how many rows we can allocate for the static confirmation
+        // prompt while *always* keeping enough space for the interactive
+        // response area (select list or input field). When the full prompt
+        // would exceed the available height we truncate it so the response
+        // options never get pushed out of view. This keeps the approval modal
+        // usable even when the overall bottom viewport is small.
+
+        // Full height of the prompt (may be larger than the available area).
+        let full_prompt_height = self.get_confirmation_prompt_height(inner.width);
+
+        // Minimum rows that must remain for the interactive section.
+        let min_response_rows = match self.mode {
+            Mode::Select => SELECT_OPTIONS.len() as u16,
+            // In input mode we need exactly two rows: one for the guidance
+            // prompt and one for the single-line input field.
+            Mode::Input => 2,
+        };
+
+        // Clamp prompt height so confirmation + response never exceed the
+        // available space. `saturating_sub` avoids underflow when the area is
+        // too small even for the minimal layout – in this unlikely case we
+        // fall back to zero-height prompt so at least the options are
+        // visible.
+        let prompt_height = full_prompt_height.min(inner.height.saturating_sub(min_response_rows));
+
         let chunks = Layout::default()
             .direction(Direction::Vertical)
             .constraints([Constraint::Length(prompt_height), Constraint::Min(0)])
@@ -342,8 +341,7 @@ impl WidgetRef for &UserApprovalWidget<'_> {
         let response_chunk = chunks[1];
 
         // Build the inner lines based on the mode. Collect them into a List of
-        // non-wrapping lines rather than a Paragraph because get_height(Rect)
-        // depends on this behavior for its calculation.
+        // non-wrapping lines rather than a Paragraph for predictable layout.
         let lines = match self.mode {
             Mode::Select => SELECT_OPTIONS
                 .iter()

codex-rs/tui/tests/insert_history.rs

@@ -0,0 +1,132 @@
+use codex_tui::insert_history_lines;
+use ratatui::backend::TestBackend;
+use ratatui::text::{Line, Span};
+use ratatui::{Terminal, TerminalOptions, Viewport};
+use ratatui::widgets::Paragraph;
+
+// Helper to initialise a terminal with an inline viewport.
+fn test_terminal(width: u16, height: u16, bottom_height: u16) -> Terminal<TestBackend> {
+    Terminal::with_options(
+        TestBackend::new(width, height),
+        TerminalOptions { viewport: Viewport::Inline(bottom_height) },
+    )
+    .expect("terminal")
+}
+
+// Extract the buffer contents as Strings (one per row) trimming trailing spaces.
+fn buffer_lines(term: &Terminal<TestBackend>) -> Vec<String> {
+    let backend = term.backend();
+    let size = term.size().expect("size");
+    let mut out = Vec::new();
+    for y in 0..size.height {
+        let mut row = String::new();
+        for x in 0..size.width {
+            let cell = backend.buffer().get(x, y);
+            row.push_str(cell.symbol());
+        }
+        out.push(row.trim_end().to_string());
+    }
+    out
+}
+
+#[test]
+fn single_line_passthrough() {
+    let mut term = test_terminal(20, 10, 3); // 7 lines history space
+    insert_history_lines(&mut term, vec![Line::from("hello world")]);
+    let lines = buffer_lines(&term);
+    assert!(lines.iter().any(|l| l.contains("hello world")), "history line visible");
+}
+
+#[test]
+fn explicit_newlines_preserved() {
+    let mut term = test_terminal(20, 10, 3);
+    insert_history_lines(&mut term, vec![Line::from(Span::raw("foo\nbar\n"))]);
+    let lines = buffer_lines(&term);
+    assert!(lines.contains(&"foo".to_string()));
+    assert!(lines.contains(&"bar".to_string()));
+    assert!(lines.iter().filter(|l| l.is_empty()).count() >= 1);
+}
+
+#[test]
+fn whitespace_normalisation() {
+    let mut term = test_terminal(30, 10, 3);
+    insert_history_lines(
+        &mut term,
+        vec![Line::from(vec![Span::raw("   a"), Span::raw("\t\tb"), Span::raw("   c")])],
+    );
+    let joined = buffer_lines(&term).join("\n");
+    assert!(joined.contains("a b c"));
+}
+
+#[test]
+fn soft_wrapping() {
+    let mut term = test_terminal(10, 10, 3);
+    insert_history_lines(&mut term, vec![Line::from("hello world test")]);
+    let lines = buffer_lines(&term);
+    assert!(lines.iter().any(|l| l == "hello"));
+    assert!(lines.iter().any(|l| l == "world test"));
+}
+
+#[test]
+fn overlong_word_splitting() {
+    let mut term = test_terminal(5, 10, 3);
+    insert_history_lines(&mut term, vec![Line::from("abcdefgh")]);
+    let lines = buffer_lines(&term);
+    assert!(lines.iter().any(|l| l == "abcde"));
+    assert!(lines.iter().any(|l| l == "fgh"));
+}
+
+#[test]
+fn whitespace_collapse_across_spans() {
+    let mut term = test_terminal(20, 10, 3);
+    insert_history_lines(&mut term, vec![Line::from(vec![Span::raw("foo "), Span::raw("   bar")])]);
+    let joined = buffer_lines(&term).join("\n");
+    assert!(joined.contains("foo bar"));
+    assert!(!joined.contains("foo   bar"));
+}
+
+#[test]
+fn trailing_newline_preserved() {
+    let mut term = test_terminal(20, 10, 3);
+    insert_history_lines(&mut term, vec![Line::from(Span::raw("xyz\n"))]);
+    let lines = buffer_lines(&term);
+    assert!(lines.contains(&"xyz".to_string()));
+    assert!(lines.iter().filter(|l| l.is_empty()).count() >= 1);
+}
+
+#[test]
+fn wide_unicode_wrapping() {
+    let mut term = test_terminal(6, 10, 3);
+    insert_history_lines(&mut term, vec![Line::from("ＡＢＣＤＥ")]);
+    let lines = buffer_lines(&term);
+    assert!(lines.iter().any(|l| l.contains("Ａ Ｂ Ｃ")));
+    assert!(lines.iter().any(|l| l.contains("Ｄ Ｅ")));
+}
+
+#[test]
+fn sequential_insertions_order() {
+    let mut term = test_terminal(20, 10, 3);
+    insert_history_lines(&mut term, vec![Line::from("first")]);
+    insert_history_lines(&mut term, vec![Line::from("second")]);
+    let lines = buffer_lines(&term);
+    let mut first_idx = None;
+    let mut second_idx = None;
+    for (i, l) in lines.iter().enumerate() {
+        if l.contains("first") { first_idx = Some(i); }
+        if l.contains("second") { second_idx = Some(i); }
+    }
+    let (Some(fi), Some(si)) = (first_idx, second_idx) else { panic!("missing lines") };
+    assert!(fi < si, "expected 'first' above 'second'");
+}
+
+#[test]
+fn integration_bottom_viewport_render() {
+    let mut term = test_terminal(15, 8, 3);
+    insert_history_lines(&mut term, vec![Line::from("history one"), Line::from("history two")]);
+    term.draw(|f| f.render_widget(Paragraph::new("bottom"), f.area())).unwrap();
+    let lines = buffer_lines(&term);
+    assert!(lines.iter().any(|l| l.contains("history one")));
+    assert!(lines.iter().any(|l| l.contains("history two")));
+    assert!(lines.iter().any(|l| l.contains("bottom")));
+}
+