serializing noMatch as an object

Co-authored-by: Michael Bolin <mbolin@openai.com>

README.md

@@ -69,6 +69,39 @@ Codex can access MCP servers. To configure them, refer to the [config docs](./do
 
 Codex CLI supports a rich set of configuration options, with preferences stored in `~/.codex/config.toml`. For full configuration options, see [Configuration](./docs/config.md).
 
+### Execpolicy quickstart
+
+Codex can enforce your own rules-based execution policy before it runs shell commands.
+
+1. Create a policy directory: `mkdir -p ~/.codex/policy`.
+2. Create one or more `.codexpolicy` files into that folder. Codex automatically loads every `.codexpolicy` file in there on startup.
+3. Write `prefix_rule` entries to describe the commands you want to allow, prompt, or block:
+
+```starlark
+prefix_rule(
+    pattern = ["git", ["push", "fetch"]],
+    decision = "prompt",  # allow | prompt | forbidden
+    match = [["git", "push", "origin", "main"]],  # examples that must match
+    not_match = [["git", "status"]],              # examples that must not match
+)
+```
+
+- `pattern` is a list of shell tokens, evaluated from left to right; wrap tokens in a nested list to express alternatives (e.g., match both `push` and `fetch`).
+- `decision` sets the severity; Codex picks the strictest decision when multiple rules match.
+- `match` and `not_match` act as (optional) unit tests. Codex validates them when it loads your policy, so you get feedback if an example has unexpected behavior.
+
+In this example rule, if Codex wants to run commands with the prefix `git push` or `git fetch`, it will first ask for user approval.
+
+Note: If Codex wants to run a command that matches with multiple rules, it will use the strictest decision among the matched rules (forbidden > prompt > allow).
+
+Use the `codex execpolicycheck` subcommand to preview decisions before you save a rule (see the [`execpolicy2` README](./codex-rs/execpolicy2/README.md) for syntax details):
+
+```shell
+codex execpolicycheck --policy ~/.codex/policy/default.codexpolicy git push origin main
+```
+
+Pass multiple `--policy` flags to test how several files combine, and use `--pretty` for formatted JSON output. See the [`codex-rs/execpolicy2` README](./codex-rs/execpolicy2/README.md) for a more detailed walkthrough of the available syntax.
+
 ---
 
 ### Docs & FAQ

codex-rs/Cargo.lock

@@ -849,7 +849,6 @@ dependencies = [
  "codex-login",
  "codex-protocol",
  "codex-utils-json-to-toml",
- "codex-windows-sandbox",
  "core_test_support",
  "mcp-types",
  "opentelemetry-appender-tracing",
@@ -990,6 +989,7 @@ dependencies = [
  "codex-common",
  "codex-core",
  "codex-exec",
+ "codex-execpolicy2",
  "codex-login",
  "codex-mcp-server",
  "codex-process-hardening",
@@ -1086,6 +1086,7 @@ dependencies = [
  "codex-apply-patch",
  "codex-arg0",
  "codex-async-utils",
+ "codex-execpolicy2",
  "codex-file-search",
  "codex-git",
  "codex-keyring-store",
@@ -1183,26 +1184,6 @@ dependencies = [
  "wiremock",
 ]
 
-[[package]]
-name = "codex-exec-server"
-version = "0.0.0"
-dependencies = [
- "anyhow",
- "clap",
- "codex-core",
- "libc",
- "path-absolutize",
- "pretty_assertions",
- "rmcp",
- "serde",
- "serde_json",
- "socket2 0.6.0",
- "tempfile",
- "tokio",
- "tracing",
- "tracing-subscriber",
-]
-
 [[package]]
 name = "codex-execpolicy"
 version = "0.0.0"
@@ -1387,7 +1368,6 @@ dependencies = [
  "codex-app-server-protocol",
  "codex-protocol",
  "eventsource-stream",
- "http",
  "opentelemetry",
  "opentelemetry-otlp",
  "opentelemetry-semantic-conventions",
@@ -1421,7 +1401,6 @@ dependencies = [
  "icu_provider",
  "mcp-types",
  "mime_guess",
- "pretty_assertions",
  "schemars 0.8.22",
  "serde",
  "serde_json",
@@ -1629,7 +1608,6 @@ name = "codex-windows-sandbox"
 version = "0.1.0"
 dependencies = [
  "anyhow",
- "codex-protocol",
  "dirs-next",
  "dunce",
  "rand 0.8.5",
@@ -5198,7 +5176,6 @@ version = "0.23.29"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "2491382039b29b9b11ff08b76ff6c97cf287671dbb74f0be44bda389fffe9bd1"
 dependencies = [
- "log",
  "once_cell",
  "ring",
  "rustls-pki-types",
@@ -6626,10 +6603,8 @@ dependencies = [
  "percent-encoding",
  "pin-project",
  "prost",
- "rustls-native-certs",
  "socket2 0.5.10",
  "tokio",
- "tokio-rustls",
  "tokio-stream",
  "tower",
  "tower-layer",

codex-rs/Cargo.toml

@@ -16,7 +16,6 @@ members = [
     "common",
     "core",
     "exec",
-    "exec-server",
     "execpolicy",
     "execpolicy2",
     "keyring-store",
@@ -67,6 +66,7 @@ codex-chatgpt = { path = "chatgpt" }
 codex-common = { path = "common" }
 codex-core = { path = "core" }
 codex-exec = { path = "exec" }
+codex-execpolicy2 = { path = "execpolicy2" }
 codex-feedback = { path = "feedback" }
 codex-file-search = { path = "file-search" }
 codex-git = { path = "utils/git" }
@@ -177,7 +177,6 @@ sha1 = "0.10.6"
 sha2 = "0.10"
 shlex = "1.3.0"
 similar = "2.7.0"
-socket2 = "0.6.0"
 starlark = "0.13.0"
 strum = "0.27.2"
 strum_macros = "0.27.2"

codex-rs/app-server-protocol/src/export.rs

@@ -61,32 +61,7 @@ pub fn generate_types(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
     Ok(())
 }
 
-#[derive(Clone, Copy, Debug)]
-pub struct GenerateTsOptions {
-    pub generate_indices: bool,
-    pub ensure_headers: bool,
-    pub run_prettier: bool,
-}
-
-impl Default for GenerateTsOptions {
-    fn default() -> Self {
-        Self {
-            generate_indices: true,
-            ensure_headers: true,
-            run_prettier: true,
-        }
-    }
-}
-
 pub fn generate_ts(out_dir: &Path, prettier: Option<&Path>) -> Result<()> {
-    generate_ts_with_options(out_dir, prettier, GenerateTsOptions::default())
-}
-
-pub fn generate_ts_with_options(
-    out_dir: &Path,
-    prettier: Option<&Path>,
-    options: GenerateTsOptions,
-) -> Result<()> {
     let v2_out_dir = out_dir.join("v2");
     ensure_dir(out_dir)?;
     ensure_dir(&v2_out_dir)?;
@@ -99,28 +74,17 @@ pub fn generate_ts_with_options(
     export_server_responses(out_dir)?;
     ServerNotification::export_all_to(out_dir)?;
 
-    if options.generate_indices {
-        generate_index_ts(out_dir)?;
-        generate_index_ts(&v2_out_dir)?;
-    }
+    generate_index_ts(out_dir)?;
+    generate_index_ts(&v2_out_dir)?;
 
     // Ensure our header is present on all TS files (root + subdirs like v2/).
-    let mut ts_files = Vec::new();
-    let should_collect_ts_files =
-        options.ensure_headers || (options.run_prettier && prettier.is_some());
-    if should_collect_ts_files {
-        ts_files = ts_files_in_recursive(out_dir)?;
-    }
-
-    if options.ensure_headers {
-        for file in &ts_files {
-            prepend_header_if_missing(file)?;
-        }
+    let ts_files = ts_files_in_recursive(out_dir)?;
+    for file in &ts_files {
+        prepend_header_if_missing(file)?;
     }
 
     // Optionally run Prettier on all generated TS files.
-    if options.run_prettier
-        && let Some(prettier_bin) = prettier
+    if let Some(prettier_bin) = prettier
         && !ts_files.is_empty()
     {
         let status = Command::new(prettier_bin)
@@ -744,6 +708,7 @@ mod tests {
     use uuid::Uuid;
 
     #[test]
+    #[ignore = "timing out"]
     fn generated_ts_has_no_optional_nullable_fields() -> Result<()> {
         // Assert that there are no types of the form "?: T | null" in the generated TS files.
         let output_dir = std::env::temp_dir().join(format!("codex_ts_types_{}", Uuid::now_v7()));
@@ -759,13 +724,7 @@ mod tests {
 
         let _guard = TempDirGuard(output_dir.clone());
 
-        // Avoid doing more work than necessary to keep the test from timing out.
-        let options = GenerateTsOptions {
-            generate_indices: false,
-            ensure_headers: false,
-            run_prettier: false,
-        };
-        generate_ts_with_options(&output_dir, None, options)?;
+        generate_ts(&output_dir, None)?;
 
         let mut undefined_offenders = Vec::new();
         let mut optional_nullable_offenders = BTreeSet::new();

codex-rs/app-server-protocol/src/protocol/common.rs

@@ -129,10 +129,6 @@ client_request_definitions! {
         params: v2::TurnInterruptParams,
         response: v2::TurnInterruptResponse,
     },
-    ReviewStart => "review/start" {
-        params: v2::ReviewStartParams,
-        response: v2::TurnStartResponse,
-    },
 
     ModelList => "model/list" {
         params: v2::ModelListParams,
@@ -494,9 +490,6 @@ server_notification_definitions! {
     ReasoningSummaryPartAdded => "item/reasoning/summaryPartAdded" (v2::ReasoningSummaryPartAddedNotification),
     ReasoningTextDelta => "item/reasoning/textDelta" (v2::ReasoningTextDeltaNotification),
 
-    /// Notifies the user of world-writable directories on Windows, which cannot be protected by the sandbox.
-    WindowsWorldWritableWarning => "windows/worldWritableWarning" (v2::WindowsWorldWritableWarningNotification),
-
     #[serde(rename = "account/login/completed")]
     #[ts(rename = "account/login/completed")]
     #[strum(serialize = "account/login/completed")]
@@ -531,7 +524,7 @@ mod tests {
         let request = ClientRequest::NewConversation {
             request_id: RequestId::Integer(42),
             params: v1::NewConversationParams {
-                model: Some("arcticfox".to_string()),
+                model: Some("gpt-5.1-codex".to_string()),
                 model_provider: None,
                 profile: None,
                 cwd: None,
@@ -549,7 +542,7 @@ mod tests {
                 "method": "newConversation",
                 "id": 42,
                 "params": {
-                    "model": "arcticfox",
+                    "model": "gpt-5.1-codex",
                     "modelProvider": null,
                     "profile": null,
                     "cwd": null,

codex-rs/app-server-protocol/src/protocol/v2.rs

@@ -402,12 +402,6 @@ pub struct ThreadStartParams {
 #[ts(export_to = "v2/")]
 pub struct ThreadStartResponse {
     pub thread: Thread,
-    pub model: String,
-    pub model_provider: String,
-    pub cwd: PathBuf,
-    pub approval_policy: AskForApproval,
-    pub sandbox: SandboxPolicy,
-    pub reasoning_effort: Option<ReasoningEffort>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Default, Clone, PartialEq, JsonSchema, TS)]
@@ -450,12 +444,6 @@ pub struct ThreadResumeParams {
 #[ts(export_to = "v2/")]
 pub struct ThreadResumeResponse {
     pub thread: Thread,
-    pub model: String,
-    pub model_provider: String,
-    pub cwd: PathBuf,
-    pub approval_policy: AskForApproval,
-    pub sandbox: SandboxPolicy,
-    pub reasoning_effort: Option<ReasoningEffort>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -531,11 +519,9 @@ pub struct AccountUpdatedNotification {
 #[ts(export_to = "v2/")]
 pub struct Turn {
     pub id: String,
-    /// This is currently only populated for resumed threads.
-    /// TODO: properly populate items for all turns.
     pub items: Vec<ThreadItem>,
-    #[serde(flatten)]
     pub status: TurnStatus,
+    pub error: Option<TurnError>,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -546,12 +532,12 @@ pub struct TurnError {
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-#[serde(tag = "status", rename_all = "camelCase")]
-#[ts(tag = "status", export_to = "v2/")]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
 pub enum TurnStatus {
     Completed,
     Interrupted,
-    Failed { error: TurnError },
+    Failed,
     InProgress,
 }
 
@@ -576,45 +562,6 @@ pub struct TurnStartParams {
     pub summary: Option<ReasoningSummary>,
 }
 
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-#[serde(rename_all = "camelCase")]
-#[ts(export_to = "v2/")]
-pub struct ReviewStartParams {
-    pub thread_id: String,
-    pub target: ReviewTarget,
-
-    /// When true, also append the final review message to the original thread.
-    #[serde(default)]
-    pub append_to_original_thread: bool,
-}
-
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-#[serde(tag = "type", rename_all = "camelCase")]
-#[ts(tag = "type", export_to = "v2/")]
-pub enum ReviewTarget {
-    /// Review the working tree: staged, unstaged, and untracked files.
-    UncommittedChanges,
-
-    /// Review changes between the current branch and the given base branch.
-    #[serde(rename_all = "camelCase")]
-    #[ts(rename_all = "camelCase")]
-    BaseBranch { branch: String },
-
-    /// Review the changes introduced by a specific commit.
-    #[serde(rename_all = "camelCase")]
-    #[ts(rename_all = "camelCase")]
-    Commit {
-        sha: String,
-        /// Optional human-readable label (e.g., commit subject) for UIs.
-        title: Option<String>,
-    },
-
-    /// Arbitrary instructions, equivalent to the old free-form prompt.
-    #[serde(rename_all = "camelCase")]
-    #[ts(rename_all = "camelCase")]
-    Custom { instructions: String },
-}
-
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
@@ -867,6 +814,8 @@ pub struct Usage {
 #[ts(export_to = "v2/")]
 pub struct TurnCompletedNotification {
     pub turn: Turn,
+    // TODO: should usage be stored on the Turn object, and we return that instead?
+    pub usage: Usage,
 }
 
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
@@ -934,15 +883,6 @@ pub struct McpToolCallProgressNotification {
     pub message: String,
 }
 
-#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
-#[serde(rename_all = "camelCase")]
-#[ts(export_to = "v2/")]
-pub struct WindowsWorldWritableWarningNotification {
-    pub sample_paths: Vec<String>,
-    pub extra_count: usize,
-    pub failed_scan: bool,
-}
-
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]

codex-rs/app-server-test-client/src/main.rs

@@ -17,20 +17,15 @@ use clap::Parser;
 use clap::Subcommand;
 use codex_app_server_protocol::AddConversationListenerParams;
 use codex_app_server_protocol::AddConversationSubscriptionResponse;
-use codex_app_server_protocol::ApprovalDecision;
 use codex_app_server_protocol::AskForApproval;
 use codex_app_server_protocol::ClientInfo;
 use codex_app_server_protocol::ClientRequest;
-use codex_app_server_protocol::CommandExecutionRequestAcceptSettings;
-use codex_app_server_protocol::CommandExecutionRequestApprovalParams;
-use codex_app_server_protocol::CommandExecutionRequestApprovalResponse;
 use codex_app_server_protocol::GetAccountRateLimitsResponse;
 use codex_app_server_protocol::InitializeParams;
 use codex_app_server_protocol::InitializeResponse;
 use codex_app_server_protocol::InputItem;
 use codex_app_server_protocol::JSONRPCMessage;
 use codex_app_server_protocol::JSONRPCNotification;
-use codex_app_server_protocol::JSONRPCRequest;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::LoginChatGptCompleteNotification;
 use codex_app_server_protocol::LoginChatGptResponse;
@@ -41,17 +36,14 @@ use codex_app_server_protocol::SandboxPolicy;
 use codex_app_server_protocol::SendUserMessageParams;
 use codex_app_server_protocol::SendUserMessageResponse;
 use codex_app_server_protocol::ServerNotification;
-use codex_app_server_protocol::ServerRequest;
 use codex_app_server_protocol::ThreadStartParams;
 use codex_app_server_protocol::ThreadStartResponse;
 use codex_app_server_protocol::TurnStartParams;
 use codex_app_server_protocol::TurnStartResponse;
-use codex_app_server_protocol::TurnStatus;
 use codex_app_server_protocol::UserInput as V2UserInput;
 use codex_protocol::ConversationId;
 use codex_protocol::protocol::Event;
 use codex_protocol::protocol::EventMsg;
-use serde::Serialize;
 use serde::de::DeserializeOwned;
 use serde_json::Value;
 use uuid::Uuid;
@@ -510,9 +502,10 @@ impl CodexClient {
                 ServerNotification::TurnCompleted(payload) => {
                     if payload.turn.id == turn_id {
                         println!("\n< turn/completed notification: {:?}", payload.turn.status);
-                        if let TurnStatus::Failed { error } = &payload.turn.status {
+                        if let Some(error) = payload.turn.error {
                             println!("[turn error] {}", error.message);
                         }
+                        println!("< usage: {:?}", payload.usage);
                         break;
                     }
                 }
@@ -610,8 +603,8 @@ impl CodexClient {
                 JSONRPCMessage::Notification(notification) => {
                     self.pending_notifications.push_back(notification);
                 }
-                JSONRPCMessage::Request(request) => {
-                    self.handle_server_request(request)?;
+                JSONRPCMessage::Request(_) => {
+                    bail!("unexpected request from codex app-server");
                 }
             }
         }
@@ -631,8 +624,8 @@ impl CodexClient {
                     // No outstanding requests, so ignore stray responses/errors for now.
                     continue;
                 }
-                JSONRPCMessage::Request(request) => {
-                    self.handle_server_request(request)?;
+                JSONRPCMessage::Request(_) => {
+                    bail!("unexpected request from codex app-server");
                 }
             }
         }
@@ -668,81 +661,6 @@ impl CodexClient {
     fn request_id(&self) -> RequestId {
         RequestId::String(Uuid::new_v4().to_string())
     }
-
-    fn handle_server_request(&mut self, request: JSONRPCRequest) -> Result<()> {
-        let server_request = ServerRequest::try_from(request)
-            .context("failed to deserialize ServerRequest from JSONRPCRequest")?;
-
-        match server_request {
-            ServerRequest::CommandExecutionRequestApproval { request_id, params } => {
-                self.handle_command_execution_request_approval(request_id, params)?;
-            }
-            other => {
-                bail!("received unsupported server request: {other:?}");
-            }
-        }
-
-        Ok(())
-    }
-
-    fn handle_command_execution_request_approval(
-        &mut self,
-        request_id: RequestId,
-        params: CommandExecutionRequestApprovalParams,
-    ) -> Result<()> {
-        let CommandExecutionRequestApprovalParams {
-            thread_id,
-            turn_id,
-            item_id,
-            reason,
-            risk,
-        } = params;
-
-        println!(
-            "\n< commandExecution approval requested for thread {thread_id}, turn {turn_id}, item {item_id}"
-        );
-        if let Some(reason) = reason.as_deref() {
-            println!("< reason: {reason}");
-        }
-        if let Some(risk) = risk.as_ref() {
-            println!("< risk assessment: {risk:?}");
-        }
-
-        let response = CommandExecutionRequestApprovalResponse {
-            decision: ApprovalDecision::Accept,
-            accept_settings: Some(CommandExecutionRequestAcceptSettings { for_session: false }),
-        };
-        self.send_server_request_response(request_id, &response)?;
-        println!("< approved commandExecution request for item {item_id}");
-        Ok(())
-    }
-
-    fn send_server_request_response<T>(&mut self, request_id: RequestId, response: &T) -> Result<()>
-    where
-        T: Serialize,
-    {
-        let message = JSONRPCMessage::Response(JSONRPCResponse {
-            id: request_id,
-            result: serde_json::to_value(response)?,
-        });
-        self.write_jsonrpc_message(message)
-    }
-
-    fn write_jsonrpc_message(&mut self, message: JSONRPCMessage) -> Result<()> {
-        let payload = serde_json::to_string(&message)?;
-        let pretty = serde_json::to_string_pretty(&message)?;
-        print_multiline_with_prefix("> ", &pretty);
-
-        if let Some(stdin) = self.stdin.as_mut() {
-            writeln!(stdin, "{payload}")?;
-            stdin
-                .flush()
-                .context("failed to flush response to codex app-server")?;
-            return Ok(());
-        }
-
-        bail!("codex app-server stdin closed")
-    }
 }
 
 fn print_multiline_with_prefix(prefix: &str, payload: &str) {

codex-rs/app-server/Cargo.toml

@@ -40,7 +40,6 @@ tracing = { workspace = true, features = ["log"] }
 tracing-subscriber = { workspace = true, features = ["env-filter", "fmt"] }
 opentelemetry-appender-tracing = { workspace = true }
 uuid = { workspace = true, features = ["serde", "v7"] }
-codex-windows-sandbox.workspace = true
 
 [dev-dependencies]
 app_test_support = { workspace = true }

codex-rs/app-server/README.md

@@ -65,7 +65,6 @@ The JSON-RPC API exposes dedicated methods for managing Codex conversations. Thr
 - `thread/archive` — move a thread’s rollout file into the archived directory; returns `{}` on success.
 - `turn/start` — add user input to a thread and begin Codex generation; responds with the initial `turn` object and streams `turn/started`, `item/*`, and `turn/completed` notifications.
 - `turn/interrupt` — request cancellation of an in-flight turn by `(thread_id, turn_id)`; success is an empty `{}` response and the turn finishes with `status: "interrupted"`.
-- `review/start` — kick off Codex’s automated reviewer for a thread; responds like `turn/start` and emits a `item/completed` notification with a `codeReview` item when results are ready.
 
 ### 1) Start or resume a thread
 
@@ -182,58 +181,6 @@ You can cancel a running Turn with `turn/interrupt`.
 
 The server requests cancellations for running subprocesses, then emits a `turn/completed` event with `status: "interrupted"`. Rely on the `turn/completed` to know when Codex-side cleanup is done.
 
-### 6) Request a code review
-
-Use `review/start` to run Codex’s reviewer on the currently checked-out project. The request takes the thread id plus a `target` describing what should be reviewed:
-
-- `{"type":"uncommittedChanges"}` — staged, unstaged, and untracked files.
-- `{"type":"baseBranch","branch":"main"}` — diff against the provided branch’s upstream (see prompt for the exact `git merge-base`/`git diff` instructions Codex will run).
-- `{"type":"commit","sha":"abc1234","title":"Optional subject"}` — review a specific commit.
-- `{"type":"custom","instructions":"Free-form reviewer instructions"}` — fallback prompt equivalent to the legacy manual review request.
-- `appendToOriginalThread` (bool, default `false`) — when `true`, Codex also records a final assistant-style message with the review summary in the original thread. When `false`, only the `codeReview` item is emitted for the review run and no extra message is added to the original thread.
-
-Example request/response:
-
-```json
-{ "method": "review/start", "id": 40, "params": {
-    "threadId": "thr_123",
-    "appendToOriginalThread": true,
-    "target": { "type": "commit", "sha": "1234567deadbeef", "title": "Polish tui colors" }
-} }
-{ "id": 40, "result": { "turn": {
-    "id": "turn_900",
-    "status": "inProgress",
-    "items": [
-        { "type": "userMessage", "id": "turn_900", "content": [ { "type": "text", "text": "Review commit 1234567: Polish tui colors" } ] }
-    ],
-    "error": null
-} } }
-```
-
-Codex streams the usual `turn/started` notification followed by an `item/started`
-with the same `codeReview` item id so clients can show progress:
-
-```json
-{ "method": "item/started", "params": { "item": {
-    "type": "codeReview",
-    "id": "turn_900",
-    "review": "current changes"
-} } }
-```
-
-When the reviewer finishes, the server emits `item/completed` containing the same
-`codeReview` item with the final review text:
-
-```json
-{ "method": "item/completed", "params": { "item": {
-    "type": "codeReview",
-    "id": "turn_900",
-    "review": "Looks solid overall...\n\n- Prefer Stylize helpers — app.rs:10-20\n  ..."
-} } }
-```
-
-The `review` string is plain text that already bundles the overall explanation plus a bullet list for each structured finding (matching `ThreadItem::CodeReview` in the generated schema). Use this notification to render the reviewer output in your client.
-
 ## Auth endpoints
 
 The JSON-RPC auth/account surface exposes request/response methods plus server-initiated notifications (no `id`). Use these to determine auth state, start or cancel logins, logout, and inspect ChatGPT rate limits.

codex-rs/app-server/src/bespoke_event_handling.rs

@@ -1,7 +1,5 @@
 use crate::codex_message_processor::ApiVersion;
 use crate::codex_message_processor::PendingInterrupts;
-use crate::codex_message_processor::TurnSummary;
-use crate::codex_message_processor::TurnSummaryStore;
 use crate::outgoing_message::OutgoingMessageSender;
 use codex_app_server_protocol::AccountRateLimitsUpdatedNotification;
 use codex_app_server_protocol::AgentMessageDeltaNotification;
@@ -28,11 +26,7 @@ use codex_app_server_protocol::SandboxCommandAssessment as V2SandboxCommandAsses
 use codex_app_server_protocol::ServerNotification;
 use codex_app_server_protocol::ServerRequestPayload;
 use codex_app_server_protocol::ThreadItem;
-use codex_app_server_protocol::Turn;
-use codex_app_server_protocol::TurnCompletedNotification;
-use codex_app_server_protocol::TurnError;
 use codex_app_server_protocol::TurnInterruptResponse;
-use codex_app_server_protocol::TurnStatus;
 use codex_core::CodexConversation;
 use codex_core::parse_command::shlex_join;
 use codex_core::protocol::ApplyPatchApprovalRequestEvent;
@@ -44,9 +38,7 @@ use codex_core::protocol::McpToolCallBeginEvent;
 use codex_core::protocol::McpToolCallEndEvent;
 use codex_core::protocol::Op;
 use codex_core::protocol::ReviewDecision;
-use codex_core::review_format::format_review_findings_block;
 use codex_protocol::ConversationId;
-use codex_protocol::protocol::ReviewOutputEvent;
 use std::convert::TryFrom;
 use std::sync::Arc;
 use tokio::sync::oneshot;
@@ -60,14 +52,10 @@ pub(crate) async fn apply_bespoke_event_handling(
     conversation: Arc<CodexConversation>,
     outgoing: Arc<OutgoingMessageSender>,
     pending_interrupts: PendingInterrupts,
-    turn_summary_store: TurnSummaryStore,
     api_version: ApiVersion,
 ) {
     let Event { id: event_id, msg } = event;
     match msg {
-        EventMsg::TaskComplete(_ev) => {
-            handle_turn_complete(conversation_id, event_id, &outgoing, &turn_summary_store).await;
-        }
         EventMsg::ApplyPatchApprovalRequest(ApplyPatchApprovalRequestEvent {
             call_id,
             changes,
@@ -201,20 +189,6 @@ pub(crate) async fn apply_bespoke_event_handling(
                     .await;
             }
         }
-        EventMsg::Error(ev) => {
-            handle_error(conversation_id, ev.message, &turn_summary_store).await;
-        }
-        EventMsg::EnteredReviewMode(review_request) => {
-            let notification = ItemStartedNotification {
-                item: ThreadItem::CodeReview {
-                    id: event_id.clone(),
-                    review: review_request.user_facing_hint,
-                },
-            };
-            outgoing
-                .send_server_notification(ServerNotification::ItemStarted(notification))
-                .await;
-        }
         EventMsg::ItemStarted(item_started_event) => {
             let item: ThreadItem = item_started_event.item.clone().into();
             let notification = ItemStartedNotification { item };
@@ -229,21 +203,6 @@ pub(crate) async fn apply_bespoke_event_handling(
                 .send_server_notification(ServerNotification::ItemCompleted(notification))
                 .await;
         }
-        EventMsg::ExitedReviewMode(review_event) => {
-            let review_text = match review_event.review_output {
-                Some(output) => render_review_output_text(&output),
-                None => REVIEW_FALLBACK_MESSAGE.to_string(),
-            };
-            let notification = ItemCompletedNotification {
-                item: ThreadItem::CodeReview {
-                    id: event_id,
-                    review: review_text,
-                },
-            };
-            outgoing
-                .send_server_notification(ServerNotification::ItemCompleted(notification))
-                .await;
-        }
         EventMsg::ExecCommandBegin(exec_command_begin_event) => {
             let item = ThreadItem::CommandExecution {
                 id: exec_command_begin_event.call_id.clone(),
@@ -339,79 +298,12 @@ pub(crate) async fn apply_bespoke_event_handling(
                     }
                 }
             }
-
-            handle_turn_interrupted(conversation_id, event_id, &outgoing, &turn_summary_store)
-                .await;
         }
 
         _ => {}
     }
 }
 
-async fn emit_turn_completed_with_status(
-    event_id: String,
-    status: TurnStatus,
-    outgoing: &OutgoingMessageSender,
-) {
-    let notification = TurnCompletedNotification {
-        turn: Turn {
-            id: event_id,
-            items: vec![],
-            status,
-        },
-    };
-    outgoing
-        .send_server_notification(ServerNotification::TurnCompleted(notification))
-        .await;
-}
-
-async fn find_and_remove_turn_summary(
-    conversation_id: ConversationId,
-    turn_summary_store: &TurnSummaryStore,
-) -> TurnSummary {
-    let mut map = turn_summary_store.lock().await;
-    map.remove(&conversation_id).unwrap_or_default()
-}
-
-async fn handle_turn_complete(
-    conversation_id: ConversationId,
-    event_id: String,
-    outgoing: &OutgoingMessageSender,
-    turn_summary_store: &TurnSummaryStore,
-) {
-    let turn_summary = find_and_remove_turn_summary(conversation_id, turn_summary_store).await;
-
-    let status = if let Some(message) = turn_summary.last_error_message {
-        TurnStatus::Failed {
-            error: TurnError { message },
-        }
-    } else {
-        TurnStatus::Completed
-    };
-
-    emit_turn_completed_with_status(event_id, status, outgoing).await;
-}
-
-async fn handle_turn_interrupted(
-    conversation_id: ConversationId,
-    event_id: String,
-    outgoing: &OutgoingMessageSender,
-    turn_summary_store: &TurnSummaryStore,
-) {
-    find_and_remove_turn_summary(conversation_id, turn_summary_store).await;
-
-    emit_turn_completed_with_status(event_id, TurnStatus::Interrupted, outgoing).await;
-}
-
-async fn handle_error(
-    conversation_id: ConversationId,
-    message: String,
-    turn_summary_store: &TurnSummaryStore,
-) {
-    let mut map = turn_summary_store.lock().await;
-    map.entry(conversation_id).or_default().last_error_message = Some(message);
-}
-
 async fn on_patch_approval_response(
     event_id: String,
     receiver: oneshot::Receiver<JsonValue>,
@@ -490,28 +382,6 @@ async fn on_exec_approval_response(
     }
 }
 
-const REVIEW_FALLBACK_MESSAGE: &str = "Reviewer failed to output a response.";
-
-fn render_review_output_text(output: &ReviewOutputEvent) -> String {
-    let mut sections = Vec::new();
-    let explanation = output.overall_explanation.trim();
-    if !explanation.is_empty() {
-        sections.push(explanation.to_string());
-    }
-    if !output.findings.is_empty() {
-        let findings = format_review_findings_block(&output.findings, None);
-        let trimmed = findings.trim();
-        if !trimmed.is_empty() {
-            sections.push(trimmed.to_string());
-        }
-    }
-    if sections.is_empty() {
-        REVIEW_FALLBACK_MESSAGE.to_string()
-    } else {
-        sections.join("\n\n")
-    }
-}
-
 async fn on_command_execution_request_approval_response(
     event_id: String,
     receiver: oneshot::Receiver<JsonValue>,
@@ -616,140 +486,13 @@ async fn construct_mcp_tool_call_end_notification(
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::CHANNEL_CAPACITY;
-    use crate::outgoing_message::OutgoingMessage;
-    use crate::outgoing_message::OutgoingMessageSender;
-    use anyhow::Result;
-    use anyhow::anyhow;
-    use anyhow::bail;
     use codex_core::protocol::McpInvocation;
     use mcp_types::CallToolResult;
     use mcp_types::ContentBlock;
     use mcp_types::TextContent;
     use pretty_assertions::assert_eq;
     use serde_json::Value as JsonValue;
-    use std::collections::HashMap;
     use std::time::Duration;
-    use tokio::sync::Mutex;
-    use tokio::sync::mpsc;
-
-    fn new_turn_summary_store() -> TurnSummaryStore {
-        Arc::new(Mutex::new(HashMap::new()))
-    }
-
-    #[tokio::test]
-    async fn test_handle_error_records_message() -> Result<()> {
-        let conversation_id = ConversationId::new();
-        let turn_summary_store = new_turn_summary_store();
-
-        handle_error(conversation_id, "boom".to_string(), &turn_summary_store).await;
-
-        let turn_summary = find_and_remove_turn_summary(conversation_id, &turn_summary_store).await;
-        assert_eq!(turn_summary.last_error_message, Some("boom".to_string()));
-        Ok(())
-    }
-
-    #[tokio::test]
-    async fn test_handle_turn_complete_emits_completed_without_error() -> Result<()> {
-        let conversation_id = ConversationId::new();
-        let event_id = "complete1".to_string();
-        let (tx, mut rx) = mpsc::channel(CHANNEL_CAPACITY);
-        let outgoing = Arc::new(OutgoingMessageSender::new(tx));
-        let turn_summary_store = new_turn_summary_store();
-
-        handle_turn_complete(
-            conversation_id,
-            event_id.clone(),
-            &outgoing,
-            &turn_summary_store,
-        )
-        .await;
-
-        let msg = rx
-            .recv()
-            .await
-            .ok_or_else(|| anyhow!("should send one notification"))?;
-        match msg {
-            OutgoingMessage::AppServerNotification(ServerNotification::TurnCompleted(n)) => {
-                assert_eq!(n.turn.id, event_id);
-                assert_eq!(n.turn.status, TurnStatus::Completed);
-            }
-            other => bail!("unexpected message: {other:?}"),
-        }
-        assert!(rx.try_recv().is_err(), "no extra messages expected");
-        Ok(())
-    }
-
-    #[tokio::test]
-    async fn test_handle_turn_interrupted_emits_interrupted_with_error() -> Result<()> {
-        let conversation_id = ConversationId::new();
-        let event_id = "interrupt1".to_string();
-        let turn_summary_store = new_turn_summary_store();
-        handle_error(conversation_id, "oops".to_string(), &turn_summary_store).await;
-        let (tx, mut rx) = mpsc::channel(CHANNEL_CAPACITY);
-        let outgoing = Arc::new(OutgoingMessageSender::new(tx));
-
-        handle_turn_interrupted(
-            conversation_id,
-            event_id.clone(),
-            &outgoing,
-            &turn_summary_store,
-        )
-        .await;
-
-        let msg = rx
-            .recv()
-            .await
-            .ok_or_else(|| anyhow!("should send one notification"))?;
-        match msg {
-            OutgoingMessage::AppServerNotification(ServerNotification::TurnCompleted(n)) => {
-                assert_eq!(n.turn.id, event_id);
-                assert_eq!(n.turn.status, TurnStatus::Interrupted);
-            }
-            other => bail!("unexpected message: {other:?}"),
-        }
-        assert!(rx.try_recv().is_err(), "no extra messages expected");
-        Ok(())
-    }
-
-    #[tokio::test]
-    async fn test_handle_turn_complete_emits_failed_with_error() -> Result<()> {
-        let conversation_id = ConversationId::new();
-        let event_id = "complete_err1".to_string();
-        let turn_summary_store = new_turn_summary_store();
-        handle_error(conversation_id, "bad".to_string(), &turn_summary_store).await;
-        let (tx, mut rx) = mpsc::channel(CHANNEL_CAPACITY);
-        let outgoing = Arc::new(OutgoingMessageSender::new(tx));
-
-        handle_turn_complete(
-            conversation_id,
-            event_id.clone(),
-            &outgoing,
-            &turn_summary_store,
-        )
-        .await;
-
-        let msg = rx
-            .recv()
-            .await
-            .ok_or_else(|| anyhow!("should send one notification"))?;
-        match msg {
-            OutgoingMessage::AppServerNotification(ServerNotification::TurnCompleted(n)) => {
-                assert_eq!(n.turn.id, event_id);
-                assert_eq!(
-                    n.turn.status,
-                    TurnStatus::Failed {
-                        error: TurnError {
-                            message: "bad".to_string(),
-                        }
-                    }
-                );
-            }
-            other => bail!("unexpected message: {other:?}"),
-        }
-        assert!(rx.try_recv().is_err(), "no extra messages expected");
-        Ok(())
-    }
 
     #[tokio::test]
     async fn test_construct_mcp_tool_call_begin_notification_with_args() {
@@ -779,105 +522,6 @@ mod tests {
         assert_eq!(notification, expected);
     }
 
-    #[tokio::test]
-    async fn test_handle_turn_complete_emits_error_multiple_turns() -> Result<()> {
-        // Conversation A will have two turns; Conversation B will have one turn.
-        let conversation_a = ConversationId::new();
-        let conversation_b = ConversationId::new();
-        let turn_summary_store = new_turn_summary_store();
-
-        let (tx, mut rx) = mpsc::channel(CHANNEL_CAPACITY);
-        let outgoing = Arc::new(OutgoingMessageSender::new(tx));
-
-        // Turn 1 on conversation A
-        let a_turn1 = "a_turn1".to_string();
-        handle_error(conversation_a, "a1".to_string(), &turn_summary_store).await;
-        handle_turn_complete(
-            conversation_a,
-            a_turn1.clone(),
-            &outgoing,
-            &turn_summary_store,
-        )
-        .await;
-
-        // Turn 1 on conversation B
-        let b_turn1 = "b_turn1".to_string();
-        handle_error(conversation_b, "b1".to_string(), &turn_summary_store).await;
-        handle_turn_complete(
-            conversation_b,
-            b_turn1.clone(),
-            &outgoing,
-            &turn_summary_store,
-        )
-        .await;
-
-        // Turn 2 on conversation A
-        let a_turn2 = "a_turn2".to_string();
-        handle_turn_complete(
-            conversation_a,
-            a_turn2.clone(),
-            &outgoing,
-            &turn_summary_store,
-        )
-        .await;
-
-        // Verify: A turn 1
-        let msg = rx
-            .recv()
-            .await
-            .ok_or_else(|| anyhow!("should send first notification"))?;
-        match msg {
-            OutgoingMessage::AppServerNotification(ServerNotification::TurnCompleted(n)) => {
-                assert_eq!(n.turn.id, a_turn1);
-                assert_eq!(
-                    n.turn.status,
-                    TurnStatus::Failed {
-                        error: TurnError {
-                            message: "a1".to_string(),
-                        }
-                    }
-                );
-            }
-            other => bail!("unexpected message: {other:?}"),
-        }
-
-        // Verify: B turn 1
-        let msg = rx
-            .recv()
-            .await
-            .ok_or_else(|| anyhow!("should send second notification"))?;
-        match msg {
-            OutgoingMessage::AppServerNotification(ServerNotification::TurnCompleted(n)) => {
-                assert_eq!(n.turn.id, b_turn1);
-                assert_eq!(
-                    n.turn.status,
-                    TurnStatus::Failed {
-                        error: TurnError {
-                            message: "b1".to_string(),
-                        }
-                    }
-                );
-            }
-            other => bail!("unexpected message: {other:?}"),
-        }
-
-        // Verify: A turn 2
-        let msg = rx
-            .recv()
-            .await
-            .ok_or_else(|| anyhow!("should send third notification"))?;
-        match msg {
-            OutgoingMessage::AppServerNotification(ServerNotification::TurnCompleted(n)) => {
-                assert_eq!(n.turn.id, a_turn2);
-                assert_eq!(n.turn.status, TurnStatus::Completed);
-            }
-            other => bail!("unexpected message: {other:?}"),
-        }
-
-        assert!(rx.try_recv().is_err(), "no extra messages expected");
-        Ok(())
-    }
-
     #[tokio::test]
     async fn test_construct_mcp_tool_call_begin_notification_without_args() {
         let begin_event = McpToolCallBeginEvent {

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

@@ -60,8 +60,6 @@ use codex_app_server_protocol::RemoveConversationSubscriptionResponse;
 use codex_app_server_protocol::RequestId;
 use codex_app_server_protocol::ResumeConversationParams;
 use codex_app_server_protocol::ResumeConversationResponse;
-use codex_app_server_protocol::ReviewStartParams;
-use codex_app_server_protocol::ReviewTarget;
 use codex_app_server_protocol::SandboxMode;
 use codex_app_server_protocol::SendUserMessageParams;
 use codex_app_server_protocol::SendUserMessageResponse;
@@ -111,15 +109,12 @@ use codex_core::config_loader::load_config_as_toml;
 use codex_core::default_client::get_codex_user_agent;
 use codex_core::exec::ExecParams;
 use codex_core::exec_env::create_env;
-use codex_core::features::Feature;
 use codex_core::find_conversation_path_by_id_str;
 use codex_core::get_platform_sandbox;
 use codex_core::git_info::git_diff_to_remote;
 use codex_core::parse_cursor;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::Op;
-use codex_core::protocol::ReviewRequest;
-use codex_core::protocol::SessionConfiguredEvent;
 use codex_core::read_head_for_summary;
 use codex_feedback::CodexFeedback;
 use codex_login::ServerOptions as LoginServerOptions;
@@ -156,14 +151,6 @@ use uuid::Uuid;
 type PendingInterruptQueue = Vec<(RequestId, ApiVersion)>;
 pub(crate) type PendingInterrupts = Arc<Mutex<HashMap<ConversationId, PendingInterruptQueue>>>;
 
-/// Per-conversation accumulation of the latest states e.g. error message while a turn runs.
-#[derive(Default, Clone)]
-pub(crate) struct TurnSummary {
-    pub(crate) last_error_message: Option<String>,
-}
-
-pub(crate) type TurnSummaryStore = Arc<Mutex<HashMap<ConversationId, TurnSummary>>>;
-
 // Duration before a ChatGPT login attempt is abandoned.
 const LOGIN_CHATGPT_TIMEOUT: Duration = Duration::from_secs(10 * 60);
 struct ActiveLogin {
@@ -188,7 +175,6 @@ pub(crate) struct CodexMessageProcessor {
     active_login: Arc<Mutex<Option<ActiveLogin>>>,
     // Queue of pending interrupt requests per conversation. We reply when TurnAborted arrives.
     pending_interrupts: PendingInterrupts,
-    turn_summary_store: TurnSummaryStore,
     pending_fuzzy_searches: Arc<Mutex<HashMap<String, Arc<AtomicBool>>>>,
     feedback: CodexFeedback,
 }
@@ -241,97 +227,11 @@ impl CodexMessageProcessor {
             conversation_listeners: HashMap::new(),
             active_login: Arc::new(Mutex::new(None)),
             pending_interrupts: Arc::new(Mutex::new(HashMap::new())),
-            turn_summary_store: Arc::new(Mutex::new(HashMap::new())),
             pending_fuzzy_searches: Arc::new(Mutex::new(HashMap::new())),
             feedback,
         }
     }
 
-    fn review_request_from_target(
-        target: ReviewTarget,
-        append_to_original_thread: bool,
-    ) -> Result<(ReviewRequest, String), JSONRPCErrorError> {
-        fn invalid_request(message: String) -> JSONRPCErrorError {
-            JSONRPCErrorError {
-                code: INVALID_REQUEST_ERROR_CODE,
-                message,
-                data: None,
-            }
-        }
-
-        match target {
-            // TODO(jif) those messages will be extracted in a follow-up PR.
-            ReviewTarget::UncommittedChanges => Ok((
-                ReviewRequest {
-                    prompt: "Review the current code changes (staged, unstaged, and untracked files) and provide prioritized findings.".to_string(),
-                    user_facing_hint: "current changes".to_string(),
-                    append_to_original_thread,
-                },
-                "Review uncommitted changes".to_string(),
-            )),
-            ReviewTarget::BaseBranch { branch } => {
-                let branch = branch.trim().to_string();
-                if branch.is_empty() {
-                    return Err(invalid_request("branch must not be empty".to_string()));
-                }
-                let prompt = format!("Review the code changes against the base branch '{branch}'. Start by finding the merge diff between the current branch and {branch}'s upstream e.g. (`git merge-base HEAD \"$(git rev-parse --abbrev-ref \"{branch}@{{upstream}}\")\"`), then run `git diff` against that SHA to see what changes we would merge into the {branch} branch. Provide prioritized, actionable findings.");
-                let hint = format!("changes against '{branch}'");
-                let display = format!("Review changes against base branch '{branch}'");
-                Ok((
-                    ReviewRequest {
-                        prompt,
-                        user_facing_hint: hint,
-                        append_to_original_thread,
-                    },
-                    display,
-                ))
-            }
-            ReviewTarget::Commit { sha, title } => {
-                let sha = sha.trim().to_string();
-                if sha.is_empty() {
-                    return Err(invalid_request("sha must not be empty".to_string()));
-                }
-                let brief_title = title
-                    .map(|t| t.trim().to_string())
-                    .filter(|t| !t.is_empty());
-                let prompt = if let Some(title) = brief_title.clone() {
-                    format!("Review the code changes introduced by commit {sha} (\"{title}\"). Provide prioritized, actionable findings.")
-                } else {
-                    format!("Review the code changes introduced by commit {sha}. Provide prioritized, actionable findings.")
-                };
-                let short_sha = sha.chars().take(7).collect::<String>();
-                let hint = format!("commit {short_sha}");
-                let display = if let Some(title) = brief_title {
-                    format!("Review commit {short_sha}: {title}")
-                } else {
-                    format!("Review commit {short_sha}")
-                };
-                Ok((
-                    ReviewRequest {
-                        prompt,
-                        user_facing_hint: hint,
-                        append_to_original_thread,
-                    },
-                    display,
-                ))
-            }
-            ReviewTarget::Custom { instructions } => {
-                let trimmed = instructions.trim().to_string();
-                if trimmed.is_empty() {
-                    return Err(invalid_request("instructions must not be empty".to_string()));
-                }
-                Ok((
-                    ReviewRequest {
-                        prompt: trimmed.clone(),
-                        user_facing_hint: trimmed.clone(),
-                        append_to_original_thread,
-                    },
-                    trimmed,
-                ))
-            }
-        }
-    }
-
     pub async fn process_request(&mut self, request: ClientRequest) {
         match request {
             ClientRequest::Initialize { .. } => {
@@ -363,9 +263,6 @@ impl CodexMessageProcessor {
             ClientRequest::TurnInterrupt { request_id, params } => {
                 self.turn_interrupt(request_id, params).await;
             }
-            ClientRequest::ReviewStart { request_id, params } => {
-                self.review_start(request_id, params).await;
-            }
             ClientRequest::NewConversation { request_id, params } => {
                 // Do not tokio::spawn() to process new_conversation()
                 // asynchronously because we need to ensure the conversation is
@@ -1250,17 +1147,7 @@ impl CodexMessageProcessor {
             ..Default::default()
         };
 
-        // Persist windows sandbox feature.
-        // TODO: persist default config in general.
-        let mut cli_overrides = cli_overrides.unwrap_or_default();
-        if cfg!(target_os = "windows") && self.config.features.enabled(Feature::WindowsSandbox) {
-            cli_overrides.insert(
-                "features.enable_experimental_windows_sandbox".to_string(),
-                serde_json::json!(true),
-            );
-        }
-
-        let config = match derive_config_from_params(overrides, Some(cli_overrides)).await {
+        let config = match derive_config_from_params(overrides, cli_overrides).await {
             Ok(config) => config,
             Err(err) => {
                 let error = JSONRPCErrorError {
@@ -1325,12 +1212,8 @@ impl CodexMessageProcessor {
 
         match self.conversation_manager.new_conversation(config).await {
             Ok(new_conv) => {
-                let NewConversation {
-                    conversation_id,
-                    session_configured,
-                    ..
-                } = new_conv;
-                let rollout_path = session_configured.rollout_path.clone();
+                let conversation_id = new_conv.conversation_id;
+                let rollout_path = new_conv.session_configured.rollout_path.clone();
                 let fallback_provider = self.config.model_provider_id.as_str();
 
                 // A bit hacky, but the summary contains a lot of useful information for the thread
@@ -1355,22 +1238,8 @@ impl CodexMessageProcessor {
                     }
                 };
 
-                let SessionConfiguredEvent {
-                    model,
-                    model_provider_id,
-                    cwd,
-                    approval_policy,
-                    sandbox_policy,
-                    ..
-                } = session_configured;
                 let response = ThreadStartResponse {
                     thread: thread.clone(),
-                    model,
-                    model_provider: model_provider_id,
-                    cwd,
-                    approval_policy: approval_policy.into(),
-                    sandbox: sandbox_policy.into(),
-                    reasoning_effort: session_configured.reasoning_effort,
                 };
 
                 // Auto-attach a conversation listener when starting a thread.
@@ -1683,15 +1552,7 @@ impl CodexMessageProcessor {
                         return;
                     }
                 };
-                let response = ThreadResumeResponse {
-                    thread,
-                    model: session_configured.model,
-                    model_provider: session_configured.model_provider_id,
-                    cwd: session_configured.cwd,
-                    approval_policy: session_configured.approval_policy.into(),
-                    sandbox: session_configured.sandbox_policy.into(),
-                    reasoning_effort: session_configured.reasoning_effort,
-                };
+                let response = ThreadResumeResponse { thread };
                 self.outgoing.send_response(request_id, response).await;
             }
             Err(err) => {
@@ -2411,6 +2272,9 @@ impl CodexMessageProcessor {
             }
         };
 
+        // Keep a copy of v2 inputs for the notification payload.
+        let v2_inputs_for_notif = params.input.clone();
+
         // Map v2 input items to core input items.
         let mapped_items: Vec<CoreInputItem> = params
             .input
@@ -2450,8 +2314,12 @@ impl CodexMessageProcessor {
             Ok(turn_id) => {
                 let turn = Turn {
                     id: turn_id.clone(),
-                    items: vec![],
+                    items: vec![ThreadItem::UserMessage {
+                        id: turn_id,
+                        content: v2_inputs_for_notif,
+                    }],
                     status: TurnStatus::InProgress,
+                    error: None,
                 };
 
                 let response = TurnStartResponse { turn: turn.clone() };
@@ -2474,64 +2342,6 @@ impl CodexMessageProcessor {
         }
     }
 
-    async fn review_start(&self, request_id: RequestId, params: ReviewStartParams) {
-        let ReviewStartParams {
-            thread_id,
-            target,
-            append_to_original_thread,
-        } = params;
-        let (_, conversation) = match self.conversation_from_thread_id(&thread_id).await {
-            Ok(v) => v,
-            Err(error) => {
-                self.outgoing.send_error(request_id, error).await;
-                return;
-            }
-        };
-
-        let (review_request, display_text) =
-            match Self::review_request_from_target(target, append_to_original_thread) {
-                Ok(value) => value,
-                Err(err) => {
-                    self.outgoing.send_error(request_id, err).await;
-                    return;
-                }
-            };
-
-        let turn_id = conversation.submit(Op::Review { review_request }).await;
-
-        match turn_id {
-            Ok(turn_id) => {
-                let mut items = Vec::new();
-                if !display_text.is_empty() {
-                    items.push(ThreadItem::UserMessage {
-                        id: turn_id.clone(),
-                        content: vec![V2UserInput::Text { text: display_text }],
-                    });
-                }
-                let turn = Turn {
-                    id: turn_id.clone(),
-                    items,
-                    status: TurnStatus::InProgress,
-                };
-                let response = TurnStartResponse { turn: turn.clone() };
-                self.outgoing.send_response(request_id, response).await;
-
-                let notif = TurnStartedNotification { turn };
-                self.outgoing
-                    .send_server_notification(ServerNotification::TurnStarted(notif))
-                    .await;
-            }
-            Err(err) => {
-                let error = JSONRPCErrorError {
-                    code: INTERNAL_ERROR_CODE,
-                    message: format!("failed to start review: {err}"),
-                    data: None,
-                };
-                self.outgoing.send_error(request_id, error).await;
-            }
-        }
-    }
-
     async fn turn_interrupt(&mut self, request_id: RequestId, params: TurnInterruptParams) {
         let TurnInterruptParams { thread_id, .. } = params;
 
@@ -2631,7 +2441,6 @@ impl CodexMessageProcessor {
 
         let outgoing_for_task = self.outgoing.clone();
         let pending_interrupts = self.pending_interrupts.clone();
-        let turn_summary_store = self.turn_summary_store.clone();
         let api_version_for_task = api_version;
         tokio::spawn(async move {
             loop {
@@ -2688,7 +2497,6 @@ impl CodexMessageProcessor {
                             conversation.clone(),
                             outgoing_for_task.clone(),
                             pending_interrupts.clone(),
-                            turn_summary_store.clone(),
                             api_version_for_task,
                         )
                         .await;

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

@@ -6,19 +6,17 @@ use crate::outgoing_message::OutgoingMessageSender;
 use codex_app_server_protocol::ClientInfo;
 use codex_app_server_protocol::ClientRequest;
 use codex_app_server_protocol::InitializeResponse;
+
 use codex_app_server_protocol::JSONRPCError;
 use codex_app_server_protocol::JSONRPCErrorError;
 use codex_app_server_protocol::JSONRPCNotification;
 use codex_app_server_protocol::JSONRPCRequest;
 use codex_app_server_protocol::JSONRPCResponse;
-use codex_app_server_protocol::ServerNotification;
-use codex_app_server_protocol::WindowsWorldWritableWarningNotification;
 use codex_core::AuthManager;
 use codex_core::ConversationManager;
 use codex_core::config::Config;
 use codex_core::default_client::USER_AGENT_SUFFIX;
 use codex_core::default_client::get_codex_user_agent;
-use codex_core::features::Feature;
 use codex_feedback::CodexFeedback;
 use codex_protocol::protocol::SessionSource;
 use std::sync::Arc;
@@ -26,7 +24,6 @@ use std::sync::Arc;
 pub(crate) struct MessageProcessor {
     outgoing: Arc<OutgoingMessageSender>,
     codex_message_processor: CodexMessageProcessor,
-    config: Arc<Config>,
     initialized: bool,
 }
 
@@ -54,14 +51,13 @@ impl MessageProcessor {
             conversation_manager,
             outgoing.clone(),
             codex_linux_sandbox_exe,
-            config.clone(),
+            config,
             feedback,
         );
 
         Self {
             outgoing,
             codex_message_processor,
-            config,
             initialized: false,
         }
     }
@@ -122,8 +118,6 @@ impl MessageProcessor {
                     self.outgoing.send_response(request_id, response).await;
 
                     self.initialized = true;
-                    self.handle_windows_world_writable_warning().await;
-
                     return;
                 }
             }
@@ -162,47 +156,4 @@ impl MessageProcessor {
     pub(crate) fn process_error(&mut self, err: JSONRPCError) {
         tracing::error!("<- error: {:?}", err);
     }
-
-    /// On Windows, when using the experimental sandbox, we need to warn the user about world-writable directories.
-    async fn handle_windows_world_writable_warning(&self) {
-        if !cfg!(windows) {
-            return;
-        }
-
-        if !self.config.features.enabled(Feature::WindowsSandbox) {
-            return;
-        }
-
-        if !matches!(
-            self.config.sandbox_policy,
-            codex_protocol::protocol::SandboxPolicy::WorkspaceWrite { .. }
-                | codex_protocol::protocol::SandboxPolicy::ReadOnly
-        ) {
-            return;
-        }
-
-        if self
-            .config
-            .notices
-            .hide_world_writable_warning
-            .unwrap_or(false)
-        {
-            return;
-        }
-
-        // This function is stubbed out to return None on non-Windows platforms
-        if let Some((sample_paths, extra_count, failed_scan)) =
-            codex_windows_sandbox::world_writable_warning_details(self.config.codex_home.as_path())
-        {
-            self.outgoing
-                .send_server_notification(ServerNotification::WindowsWorldWritableWarning(
-                    WindowsWorldWritableWarningNotification {
-                        sample_paths,
-                        extra_count,
-                        failed_scan,
-                    },
-                ))
-                .await;
-        }
-    }
 }

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

@@ -35,7 +35,6 @@ use codex_app_server_protocol::NewConversationParams;
 use codex_app_server_protocol::RemoveConversationListenerParams;
 use codex_app_server_protocol::RequestId;
 use codex_app_server_protocol::ResumeConversationParams;
-use codex_app_server_protocol::ReviewStartParams;
 use codex_app_server_protocol::SendUserMessageParams;
 use codex_app_server_protocol::SendUserTurnParams;
 use codex_app_server_protocol::ServerRequest;
@@ -378,15 +377,6 @@ impl McpProcess {
         self.send_request("turn/interrupt", params).await
     }
 
-    /// Send a `review/start` JSON-RPC request (v2).
-    pub async fn send_review_start_request(
-        &mut self,
-        params: ReviewStartParams,
-    ) -> anyhow::Result<i64> {
-        let params = Some(serde_json::to_value(params)?);
-        self.send_request("review/start", params).await
-    }
-
     /// Send a `cancelLoginChatGpt` JSON-RPC request.
     pub async fn send_cancel_login_chat_gpt_request(
         &mut self,

codex-rs/app-server/tests/suite/config.rs

@@ -27,7 +27,7 @@ fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
     std::fs::write(
         config_toml,
         r#"
-model = "arcticfox"
+model = "gpt-5.1-codex"
 approval_policy = "on-request"
 sandbox_mode = "workspace-write"
 model_reasoning_summary = "detailed"
@@ -87,7 +87,7 @@ async fn get_config_toml_parses_all_fields() -> Result<()> {
             }),
             forced_chatgpt_workspace_id: Some("12345678-0000-0000-0000-000000000000".into()),
             forced_login_method: Some(ForcedLoginMethod::Chatgpt),
-            model: Some("arcticfox".into()),
+            model: Some("gpt-5.1-codex".into()),
             model_reasoning_effort: Some(ReasoningEffort::High),
             model_reasoning_summary: Some(ReasoningSummary::Detailed),
             model_verbosity: Some(Verbosity::Medium),

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

@@ -57,7 +57,7 @@ fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
     std::fs::write(
         config_toml,
         r#"
-model = "arcticfox"
+model = "gpt-5.1-codex"
 model_reasoning_effort = "medium"
 "#,
     )

codex-rs/app-server/tests/suite/v2/mod.rs

@@ -1,7 +1,6 @@
 mod account;
 mod model_list;
 mod rate_limits;
-mod review;
 mod thread_archive;
 mod thread_list;
 mod thread_resume;

codex-rs/app-server/tests/suite/v2/model_list.rs

@@ -45,33 +45,6 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
     } = to_response::<ModelListResponse>(response)?;
 
     let expected_models = vec![
-        Model {
-            id: "arcticfox".to_string(),
-            model: "arcticfox".to_string(),
-            display_name: "arcticfox".to_string(),
-            description: "Latest Codex-optimized flagship for deep and fast reasoning.".to_string(),
-            supported_reasoning_efforts: vec![
-                ReasoningEffortOption {
-                    reasoning_effort: ReasoningEffort::Low,
-                    description: "Fast responses with lighter reasoning".to_string(),
-                },
-                ReasoningEffortOption {
-                    reasoning_effort: ReasoningEffort::Medium,
-                    description: "Balances speed and reasoning depth for everyday tasks"
-                        .to_string(),
-                },
-                ReasoningEffortOption {
-                    reasoning_effort: ReasoningEffort::High,
-                    description: "Maximizes reasoning depth for complex problems".to_string(),
-                },
-                ReasoningEffortOption {
-                    reasoning_effort: ReasoningEffort::XHigh,
-                    description: "Extra high reasoning depth for complex problems".to_string(),
-                },
-            ],
-            default_reasoning_effort: ReasoningEffort::Medium,
-            is_default: true,
-        },
         Model {
             id: "gpt-5.1-codex".to_string(),
             model: "gpt-5.1-codex".to_string(),
@@ -93,7 +66,7 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
                 },
             ],
             default_reasoning_effort: ReasoningEffort::Medium,
-            is_default: false,
+            is_default: true,
         },
         Model {
             id: "gpt-5.1-codex-mini".to_string(),
@@ -174,7 +147,7 @@ async fn list_models_pagination_works() -> Result<()> {
     } = to_response::<ModelListResponse>(first_response)?;
 
     assert_eq!(first_items.len(), 1);
-    assert_eq!(first_items[0].id, "arcticfox");
+    assert_eq!(first_items[0].id, "gpt-5.1-codex");
     let next_cursor = first_cursor.ok_or_else(|| anyhow!("cursor for second page"))?;
 
     let second_request = mcp
@@ -196,7 +169,7 @@ async fn list_models_pagination_works() -> Result<()> {
     } = to_response::<ModelListResponse>(second_response)?;
 
     assert_eq!(second_items.len(), 1);
-    assert_eq!(second_items[0].id, "gpt-5.1-codex");
+    assert_eq!(second_items[0].id, "gpt-5.1-codex-mini");
     let third_cursor = second_cursor.ok_or_else(|| anyhow!("cursor for third page"))?;
 
     let third_request = mcp
@@ -218,30 +191,8 @@ async fn list_models_pagination_works() -> Result<()> {
     } = to_response::<ModelListResponse>(third_response)?;
 
     assert_eq!(third_items.len(), 1);
-    assert_eq!(third_items[0].id, "gpt-5.1-codex-mini");
-    let fourth_cursor = third_cursor.ok_or_else(|| anyhow!("cursor for fourth page"))?;
-
-    let fourth_request = mcp
-        .send_list_models_request(ModelListParams {
-            limit: Some(1),
-            cursor: Some(fourth_cursor.clone()),
-        })
-        .await?;
-
-    let fourth_response: JSONRPCResponse = timeout(
-        DEFAULT_TIMEOUT,
-        mcp.read_stream_until_response_message(RequestId::Integer(fourth_request)),
-    )
-    .await??;
-
-    let ModelListResponse {
-        data: fourth_items,
-        next_cursor: fourth_cursor,
-    } = to_response::<ModelListResponse>(fourth_response)?;
-
-    assert_eq!(fourth_items.len(), 1);
-    assert_eq!(fourth_items[0].id, "gpt-5.1");
-    assert!(fourth_cursor.is_none());
+    assert_eq!(third_items[0].id, "gpt-5.1");
+    assert!(third_cursor.is_none());
     Ok(())
 }
 

codex-rs/app-server/tests/suite/v2/review.rs

@@ -1,279 +0,0 @@
-use anyhow::Result;
-use app_test_support::McpProcess;
-use app_test_support::create_final_assistant_message_sse_response;
-use app_test_support::create_mock_chat_completions_server_unchecked;
-use app_test_support::to_response;
-use codex_app_server_protocol::ItemCompletedNotification;
-use codex_app_server_protocol::ItemStartedNotification;
-use codex_app_server_protocol::JSONRPCError;
-use codex_app_server_protocol::JSONRPCNotification;
-use codex_app_server_protocol::JSONRPCResponse;
-use codex_app_server_protocol::RequestId;
-use codex_app_server_protocol::ReviewStartParams;
-use codex_app_server_protocol::ReviewTarget;
-use codex_app_server_protocol::ThreadItem;
-use codex_app_server_protocol::ThreadStartParams;
-use codex_app_server_protocol::ThreadStartResponse;
-use codex_app_server_protocol::TurnStartResponse;
-use codex_app_server_protocol::TurnStatus;
-use serde_json::json;
-use tempfile::TempDir;
-use tokio::time::timeout;
-
-const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(10);
-const INVALID_REQUEST_ERROR_CODE: i64 = -32600;
-
-#[tokio::test]
-async fn review_start_runs_review_turn_and_emits_code_review_item() -> Result<()> {
-    let review_payload = json!({
-        "findings": [
-            {
-                "title": "Prefer Stylize helpers",
-                "body": "Use .dim()/.bold() chaining instead of manual Style.",
-                "confidence_score": 0.9,
-                "priority": 1,
-                "code_location": {
-                    "absolute_file_path": "/tmp/file.rs",
-                    "line_range": {"start": 10, "end": 20}
-                }
-            }
-        ],
-        "overall_correctness": "good",
-        "overall_explanation": "Looks solid overall with minor polish suggested.",
-        "overall_confidence_score": 0.75
-    })
-    .to_string();
-    let responses = vec![create_final_assistant_message_sse_response(
-        &review_payload,
-    )?];
-    let server = create_mock_chat_completions_server_unchecked(responses).await;
-
-    let codex_home = TempDir::new()?;
-    create_config_toml(codex_home.path(), &server.uri())?;
-
-    let mut mcp = McpProcess::new(codex_home.path()).await?;
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
-
-    let thread_id = start_default_thread(&mut mcp).await?;
-
-    let review_req = mcp
-        .send_review_start_request(ReviewStartParams {
-            thread_id: thread_id.clone(),
-            append_to_original_thread: true,
-            target: ReviewTarget::Commit {
-                sha: "1234567deadbeef".to_string(),
-                title: Some("Tidy UI colors".to_string()),
-            },
-        })
-        .await?;
-    let review_resp: JSONRPCResponse = timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_response_message(RequestId::Integer(review_req)),
-    )
-    .await??;
-    let TurnStartResponse { turn } = to_response::<TurnStartResponse>(review_resp)?;
-    let turn_id = turn.id.clone();
-    assert_eq!(turn.status, TurnStatus::InProgress);
-    assert_eq!(turn.items.len(), 1);
-    match &turn.items[0] {
-        ThreadItem::UserMessage { content, .. } => {
-            assert_eq!(content.len(), 1);
-            assert!(matches!(
-                &content[0],
-                codex_app_server_protocol::UserInput::Text { .. }
-            ));
-        }
-        other => panic!("expected user message, got {other:?}"),
-    }
-
-    let _started: JSONRPCNotification = timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_notification_message("turn/started"),
-    )
-    .await??;
-    let item_started: JSONRPCNotification = timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_notification_message("item/started"),
-    )
-    .await??;
-    let started: ItemStartedNotification =
-        serde_json::from_value(item_started.params.expect("params must be present"))?;
-    match started.item {
-        ThreadItem::CodeReview { id, review } => {
-            assert_eq!(id, turn_id);
-            assert_eq!(review, "commit 1234567");
-        }
-        other => panic!("expected code review item, got {other:?}"),
-    }
-
-    let mut review_body: Option<String> = None;
-    for _ in 0..5 {
-        let review_notif: JSONRPCNotification = timeout(
-            DEFAULT_READ_TIMEOUT,
-            mcp.read_stream_until_notification_message("item/completed"),
-        )
-        .await??;
-        let completed: ItemCompletedNotification =
-            serde_json::from_value(review_notif.params.expect("params must be present"))?;
-        match completed.item {
-            ThreadItem::CodeReview { id, review } => {
-                assert_eq!(id, turn_id);
-                review_body = Some(review);
-                break;
-            }
-            ThreadItem::UserMessage { .. } => continue,
-            other => panic!("unexpected item/completed payload: {other:?}"),
-        }
-    }
-
-    let review = review_body.expect("did not observe a code review item");
-    assert!(review.contains("Prefer Stylize helpers"));
-    assert!(review.contains("/tmp/file.rs:10-20"));
-
-    Ok(())
-}
-
-#[tokio::test]
-async fn review_start_rejects_empty_base_branch() -> Result<()> {
-    let server = create_mock_chat_completions_server_unchecked(vec![]).await;
-    let codex_home = TempDir::new()?;
-    create_config_toml(codex_home.path(), &server.uri())?;
-
-    let mut mcp = McpProcess::new(codex_home.path()).await?;
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
-    let thread_id = start_default_thread(&mut mcp).await?;
-
-    let request_id = mcp
-        .send_review_start_request(ReviewStartParams {
-            thread_id,
-            append_to_original_thread: true,
-            target: ReviewTarget::BaseBranch {
-                branch: "   ".to_string(),
-            },
-        })
-        .await?;
-    let error: JSONRPCError = timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
-    )
-    .await??;
-    assert_eq!(error.error.code, INVALID_REQUEST_ERROR_CODE);
-    assert!(
-        error.error.message.contains("branch must not be empty"),
-        "unexpected message: {}",
-        error.error.message
-    );
-
-    Ok(())
-}
-
-#[tokio::test]
-async fn review_start_rejects_empty_commit_sha() -> Result<()> {
-    let server = create_mock_chat_completions_server_unchecked(vec![]).await;
-    let codex_home = TempDir::new()?;
-    create_config_toml(codex_home.path(), &server.uri())?;
-
-    let mut mcp = McpProcess::new(codex_home.path()).await?;
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
-    let thread_id = start_default_thread(&mut mcp).await?;
-
-    let request_id = mcp
-        .send_review_start_request(ReviewStartParams {
-            thread_id,
-            append_to_original_thread: true,
-            target: ReviewTarget::Commit {
-                sha: "\t".to_string(),
-                title: None,
-            },
-        })
-        .await?;
-    let error: JSONRPCError = timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
-    )
-    .await??;
-    assert_eq!(error.error.code, INVALID_REQUEST_ERROR_CODE);
-    assert!(
-        error.error.message.contains("sha must not be empty"),
-        "unexpected message: {}",
-        error.error.message
-    );
-
-    Ok(())
-}
-
-#[tokio::test]
-async fn review_start_rejects_empty_custom_instructions() -> Result<()> {
-    let server = create_mock_chat_completions_server_unchecked(vec![]).await;
-    let codex_home = TempDir::new()?;
-    create_config_toml(codex_home.path(), &server.uri())?;
-
-    let mut mcp = McpProcess::new(codex_home.path()).await?;
-    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
-    let thread_id = start_default_thread(&mut mcp).await?;
-
-    let request_id = mcp
-        .send_review_start_request(ReviewStartParams {
-            thread_id,
-            append_to_original_thread: true,
-            target: ReviewTarget::Custom {
-                instructions: "\n\n".to_string(),
-            },
-        })
-        .await?;
-    let error: JSONRPCError = timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_error_message(RequestId::Integer(request_id)),
-    )
-    .await??;
-    assert_eq!(error.error.code, INVALID_REQUEST_ERROR_CODE);
-    assert!(
-        error
-            .error
-            .message
-            .contains("instructions must not be empty"),
-        "unexpected message: {}",
-        error.error.message
-    );
-
-    Ok(())
-}
-
-async fn start_default_thread(mcp: &mut McpProcess) -> Result<String> {
-    let thread_req = mcp
-        .send_thread_start_request(ThreadStartParams {
-            model: Some("mock-model".to_string()),
-            ..Default::default()
-        })
-        .await?;
-    let thread_resp: JSONRPCResponse = timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
-    )
-    .await??;
-    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
-    Ok(thread.id)
-}
-
-fn create_config_toml(codex_home: &std::path::Path, server_uri: &str) -> std::io::Result<()> {
-    let config_toml = codex_home.join("config.toml");
-    std::fs::write(
-        config_toml,
-        format!(
-            r#"
-model = "mock-model"
-approval_policy = "never"
-sandbox_mode = "read-only"
-
-model_provider = "mock_provider"
-
-[model_providers.mock_provider]
-name = "Mock provider"
-base_url = "{server_uri}/v1"
-wire_api = "chat"
-request_max_retries = 0
-stream_max_retries = 0
-"#
-        ),
-    )
-}

codex-rs/app-server/tests/suite/v2/thread_archive.rs

@@ -35,7 +35,7 @@ async fn thread_archive_moves_rollout_into_archived_directory() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
     )
     .await??;
-    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
     assert!(!thread.id.is_empty());
 
     // Locate the rollout path recorded for this thread id.

codex-rs/app-server/tests/suite/v2/thread_resume.rs

@@ -27,7 +27,7 @@ async fn thread_resume_returns_original_thread() -> Result<()> {
     // Start a thread.
     let start_id = mcp
         .send_thread_start_request(ThreadStartParams {
-            model: Some("arcticfox".to_string()),
+            model: Some("gpt-5.1-codex".to_string()),
             ..Default::default()
         })
         .await?;
@@ -36,7 +36,7 @@ async fn thread_resume_returns_original_thread() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
     )
     .await??;
-    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
 
     // Resume it via v2 API.
     let resume_id = mcp
@@ -50,9 +50,8 @@ async fn thread_resume_returns_original_thread() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
     )
     .await??;
-    let ThreadResumeResponse {
-        thread: resumed, ..
-    } = to_response::<ThreadResumeResponse>(resume_resp)?;
+    let ThreadResumeResponse { thread: resumed } =
+        to_response::<ThreadResumeResponse>(resume_resp)?;
     assert_eq!(resumed, thread);
 
     Ok(())
@@ -69,7 +68,7 @@ async fn thread_resume_prefers_path_over_thread_id() -> Result<()> {
 
     let start_id = mcp
         .send_thread_start_request(ThreadStartParams {
-            model: Some("arcticfox".to_string()),
+            model: Some("gpt-5.1-codex".to_string()),
             ..Default::default()
         })
         .await?;
@@ -78,7 +77,7 @@ async fn thread_resume_prefers_path_over_thread_id() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
     )
     .await??;
-    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
 
     let thread_path = thread.path.clone();
     let resume_id = mcp
@@ -94,9 +93,8 @@ async fn thread_resume_prefers_path_over_thread_id() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
     )
     .await??;
-    let ThreadResumeResponse {
-        thread: resumed, ..
-    } = to_response::<ThreadResumeResponse>(resume_resp)?;
+    let ThreadResumeResponse { thread: resumed } =
+        to_response::<ThreadResumeResponse>(resume_resp)?;
     assert_eq!(resumed, thread);
 
     Ok(())
@@ -114,7 +112,7 @@ async fn thread_resume_supports_history_and_overrides() -> Result<()> {
     // Start a thread.
     let start_id = mcp
         .send_thread_start_request(ThreadStartParams {
-            model: Some("arcticfox".to_string()),
+            model: Some("gpt-5.1-codex".to_string()),
             ..Default::default()
         })
         .await?;
@@ -123,7 +121,7 @@ async fn thread_resume_supports_history_and_overrides() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
     )
     .await??;
-    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
 
     let history_text = "Hello from history";
     let history = vec![ResponseItem::Message {
@@ -149,13 +147,10 @@ async fn thread_resume_supports_history_and_overrides() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
     )
     .await??;
-    let ThreadResumeResponse {
-        thread: resumed,
-        model_provider,
-        ..
-    } = to_response::<ThreadResumeResponse>(resume_resp)?;
+    let ThreadResumeResponse { thread: resumed } =
+        to_response::<ThreadResumeResponse>(resume_resp)?;
     assert!(!resumed.id.is_empty());
-    assert_eq!(model_provider, "mock_provider");
+    assert_eq!(resumed.model_provider, "mock_provider");
     assert_eq!(resumed.preview, history_text);
 
     Ok(())

codex-rs/app-server/tests/suite/v2/thread_start.rs

@@ -40,17 +40,13 @@ async fn thread_start_creates_thread_and_emits_started() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(req_id)),
     )
     .await??;
-    let ThreadStartResponse {
-        thread,
-        model_provider,
-        ..
-    } = to_response::<ThreadStartResponse>(resp)?;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(resp)?;
     assert!(!thread.id.is_empty(), "thread id should not be empty");
     assert!(
         thread.preview.is_empty(),
         "new threads should start with an empty preview"
     );
-    assert_eq!(model_provider, "mock_provider");
+    assert_eq!(thread.model_provider, "mock_provider");
     assert!(
         thread.created_at > 0,
         "created_at should be a positive UNIX timestamp"

codex-rs/app-server/tests/suite/v2/turn_interrupt.rs

@@ -5,17 +5,14 @@ use app_test_support::McpProcess;
 use app_test_support::create_mock_chat_completions_server;
 use app_test_support::create_shell_sse_response;
 use app_test_support::to_response;
-use codex_app_server_protocol::JSONRPCNotification;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::RequestId;
 use codex_app_server_protocol::ThreadStartParams;
 use codex_app_server_protocol::ThreadStartResponse;
-use codex_app_server_protocol::TurnCompletedNotification;
 use codex_app_server_protocol::TurnInterruptParams;
 use codex_app_server_protocol::TurnInterruptResponse;
 use codex_app_server_protocol::TurnStartParams;
 use codex_app_server_protocol::TurnStartResponse;
-use codex_app_server_protocol::TurnStatus;
 use codex_app_server_protocol::UserInput as V2UserInput;
 use tempfile::TempDir;
 use tokio::time::timeout;
@@ -65,7 +62,7 @@ async fn turn_interrupt_aborts_running_turn() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
     )
     .await??;
-    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(thread_resp)?;
 
     // Start a turn that triggers a long-running command.
     let turn_req = mcp
@@ -102,18 +99,7 @@ async fn turn_interrupt_aborts_running_turn() -> Result<()> {
     .await??;
     let _resp: TurnInterruptResponse = to_response::<TurnInterruptResponse>(interrupt_resp)?;
 
-    let completed_notif: JSONRPCNotification = timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_notification_message("turn/completed"),
-    )
-    .await??;
-    let completed: TurnCompletedNotification = serde_json::from_value(
-        completed_notif
-            .params
-            .expect("turn/completed params must be present"),
-    )?;
-    assert_eq!(completed.turn.status, TurnStatus::Interrupted);
-
+    // No fields to assert on; successful deserialization confirms proper response shape.
     Ok(())
 }
 

codex-rs/app-server/tests/suite/v2/turn_start.rs

@@ -14,11 +14,9 @@ use codex_app_server_protocol::ServerRequest;
 use codex_app_server_protocol::ThreadItem;
 use codex_app_server_protocol::ThreadStartParams;
 use codex_app_server_protocol::ThreadStartResponse;
-use codex_app_server_protocol::TurnCompletedNotification;
 use codex_app_server_protocol::TurnStartParams;
 use codex_app_server_protocol::TurnStartResponse;
 use codex_app_server_protocol::TurnStartedNotification;
-use codex_app_server_protocol::TurnStatus;
 use codex_app_server_protocol::UserInput as V2UserInput;
 use codex_core::protocol_config_types::ReasoningEffort;
 use codex_core::protocol_config_types::ReasoningSummary;
@@ -59,7 +57,7 @@ async fn turn_start_emits_notifications_and_accepts_model_override() -> Result<(
         mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
     )
     .await??;
-    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(thread_resp)?;
 
     // Start a turn with only input and thread_id set (no overrides).
     let turn_req = mcp
@@ -120,17 +118,13 @@ async fn turn_start_emits_notifications_and_accepts_model_override() -> Result<(
     )
     .await??;
 
-    let completed_notif: JSONRPCNotification = timeout(
+    // And we should ultimately get a task_complete without having to add a
+    // legacy conversation listener explicitly (auto-attached by thread/start).
+    let _task_complete: JSONRPCNotification = timeout(
         DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_notification_message("turn/completed"),
+        mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
     .await??;
-    let completed: TurnCompletedNotification = serde_json::from_value(
-        completed_notif
-            .params
-            .expect("turn/completed params must be present"),
-    )?;
-    assert_eq!(completed.turn.status, TurnStatus::Completed);
 
     Ok(())
 }
@@ -163,7 +157,7 @@ async fn turn_start_accepts_local_image_input() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
     )
     .await??;
-    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(thread_resp)?;
 
     let image_path = codex_home.path().join("image.png");
     // No need to actually write the file; we just exercise the input path.
@@ -239,7 +233,7 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
     )
     .await??;
-    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
 
     // turn/start — expect CommandExecutionRequestApproval request from server
     let first_turn_id = mcp
@@ -280,11 +274,6 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
     .await??;
-    timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_notification_message("turn/completed"),
-    )
-    .await??;
 
     // Second turn with approval_policy=never should not elicit approval
     let second_turn_id = mcp
@@ -313,11 +302,6 @@ async fn turn_start_exec_approval_toggle_v2() -> Result<()> {
         mcp.read_stream_until_notification_message("codex/event/task_complete"),
     )
     .await??;
-    timeout(
-        DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_notification_message("turn/completed"),
-    )
-    .await??;
 
     Ok(())
 }
@@ -378,7 +362,7 @@ async fn turn_start_updates_sandbox_and_cwd_between_turns_v2() -> Result<()> {
         mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
     )
     .await??;
-    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+    let ThreadStartResponse { thread } = to_response::<ThreadStartResponse>(start_resp)?;
 
     // first turn with workspace-write sandbox and first_cwd
     let first_turn = mcp

codex-rs/cli/Cargo.toml

@@ -26,6 +26,7 @@ codex-cloud-tasks = { path = "../cloud-tasks" }
 codex-common = { workspace = true, features = ["cli"] }
 codex-core = { workspace = true }
 codex-exec = { workspace = true }
+codex-execpolicy2 = { workspace = true }
 codex-login = { workspace = true }
 codex-mcp-server = { workspace = true }
 codex-process-hardening = { workspace = true }

codex-rs/cli/src/debug_sandbox.rs

@@ -138,7 +138,11 @@ async fn run_command_under_sandbox(
         {
             use codex_windows_sandbox::run_windows_sandbox_capture;
 
-            let policy_str = serde_json::to_string(&config.sandbox_policy)?;
+            let policy_str = match &config.sandbox_policy {
+                codex_core::protocol::SandboxPolicy::DangerFullAccess => "workspace-write",
+                codex_core::protocol::SandboxPolicy::ReadOnly => "read-only",
+                codex_core::protocol::SandboxPolicy::WorkspaceWrite { .. } => "workspace-write",
+            };
 
             let sandbox_cwd = sandbox_policy_cwd.clone();
             let cwd_clone = cwd.clone();
@@ -149,7 +153,7 @@ async fn run_command_under_sandbox(
             // Preflight audit is invoked elsewhere at the appropriate times.
             let res = tokio::task::spawn_blocking(move || {
                 run_windows_sandbox_capture(
-                    policy_str.as_str(),
+                    policy_str,
                     &sandbox_cwd,
                     base_dir.as_path(),
                     command_vec,

codex-rs/cli/src/main.rs

@@ -18,6 +18,7 @@ use codex_cli::login::run_logout;
 use codex_cloud_tasks::Cli as CloudTasksCli;
 use codex_common::CliConfigOverrides;
 use codex_exec::Cli as ExecCli;
+use codex_execpolicy2::ExecPolicyCheckCommand;
 use codex_responses_api_proxy::Args as ResponsesApiProxyArgs;
 use codex_tui::AppExitInfo;
 use codex_tui::Cli as TuiCli;
@@ -112,6 +113,10 @@ enum Subcommand {
     #[clap(hide = true, name = "stdio-to-uds")]
     StdioToUds(StdioToUdsCommand),
 
+    /// Check execpolicy files against a command.
+    #[clap(name = "execpolicycheck")]
+    ExecPolicyCheck(ExecPolicyCheckCommand),
+
     /// Inspect feature flags.
     Features(FeaturesCli),
 }
@@ -134,10 +139,6 @@ struct ResumeCommand {
     #[arg(long = "last", default_value_t = false, conflicts_with = "session_id")]
     last: bool,
 
-    /// Show all sessions (disables cwd filtering and shows CWD column).
-    #[arg(long = "all", default_value_t = false)]
-    all: bool,
-
     #[clap(flatten)]
     config_overrides: TuiCli,
 }
@@ -327,6 +328,12 @@ fn run_update_action(action: UpdateAction) -> anyhow::Result<()> {
     Ok(())
 }
 
+fn run_execpolicycheck(cmd: ExecPolicyCheckCommand) -> anyhow::Result<()> {
+    let json = cmd.run()?;
+    println!("{json}");
+    Ok(())
+}
+
 #[derive(Debug, Default, Parser, Clone)]
 struct FeatureToggles {
     /// Enable a feature (repeatable). Equivalent to `-c features.<name>=true`.
@@ -452,7 +459,6 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
         Some(Subcommand::Resume(ResumeCommand {
             session_id,
             last,
-            all,
             config_overrides,
         })) => {
             interactive = finalize_resume_interactive(
@@ -460,7 +466,6 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
                 root_config_overrides.clone(),
                 session_id,
                 last,
-                all,
                 config_overrides,
             );
             let exit_info = codex_tui::run_main(interactive, codex_linux_sandbox_exe).await?;
@@ -565,6 +570,7 @@ async fn cli_main(codex_linux_sandbox_exe: Option<PathBuf>) -> anyhow::Result<()
             tokio::task::spawn_blocking(move || codex_stdio_to_uds::run(socket_path.as_path()))
                 .await??;
         }
+        Some(Subcommand::ExecPolicyCheck(cmd)) => run_execpolicycheck(cmd)?,
         Some(Subcommand::Features(FeaturesCli { sub })) => match sub {
             FeaturesSubcommand::List => {
                 // Respect root-level `-c` overrides plus top-level flags like `--profile`.
@@ -617,7 +623,6 @@ fn finalize_resume_interactive(
     root_config_overrides: CliConfigOverrides,
     session_id: Option<String>,
     last: bool,
-    show_all: bool,
     resume_cli: TuiCli,
 ) -> TuiCli {
     // Start with the parsed interactive CLI so resume shares the same
@@ -626,7 +631,6 @@ fn finalize_resume_interactive(
     interactive.resume_picker = resume_session_id.is_none() && !last;
     interactive.resume_last = last;
     interactive.resume_session_id = resume_session_id;
-    interactive.resume_show_all = show_all;
 
     // Merge resume-scoped flags and overrides with highest precedence.
     merge_resume_cli_flags(&mut interactive, resume_cli);
@@ -710,21 +714,13 @@ mod tests {
         let Subcommand::Resume(ResumeCommand {
             session_id,
             last,
-            all,
             config_overrides: resume_cli,
         }) = subcommand.expect("resume present")
         else {
             unreachable!()
         };
 
-        finalize_resume_interactive(
-            interactive,
-            root_overrides,
-            session_id,
-            last,
-            all,
-            resume_cli,
-        )
+        finalize_resume_interactive(interactive, root_overrides, session_id, last, resume_cli)
     }
 
     fn sample_exit_info(conversation: Option<&str>) -> AppExitInfo {
@@ -791,7 +787,6 @@ mod tests {
         assert!(interactive.resume_picker);
         assert!(!interactive.resume_last);
         assert_eq!(interactive.resume_session_id, None);
-        assert!(!interactive.resume_show_all);
     }
 
     #[test]
@@ -800,7 +795,6 @@ mod tests {
         assert!(!interactive.resume_picker);
         assert!(interactive.resume_last);
         assert_eq!(interactive.resume_session_id, None);
-        assert!(!interactive.resume_show_all);
     }
 
     #[test]
@@ -809,14 +803,6 @@ mod tests {
         assert!(!interactive.resume_picker);
         assert!(!interactive.resume_last);
         assert_eq!(interactive.resume_session_id.as_deref(), Some("1234"));
-        assert!(!interactive.resume_show_all);
-    }
-
-    #[test]
-    fn resume_all_flag_sets_show_all() {
-        let interactive = finalize_from_args(["codex", "resume", "--all"].as_ref());
-        assert!(interactive.resume_picker);
-        assert!(interactive.resume_show_all);
     }
 
     #[test]

codex-rs/codex-backend-openapi-models/src/models/credit_status_details.rs

@@ -1,52 +0,0 @@
-/*
- * codex-backend
- *
- * codex-backend
- *
- * The version of the OpenAPI document: 0.0.1
- *
- * Generated by: https://openapi-generator.tech
- */
-use serde::Deserialize;
-use serde::Serialize;
-
-#[derive(Clone, Default, Debug, PartialEq, Serialize, Deserialize)]
-pub struct CreditStatusDetails {
-    #[serde(rename = "has_credits")]
-    pub has_credits: bool,
-    #[serde(rename = "unlimited")]
-    pub unlimited: bool,
-    #[serde(
-        rename = "balance",
-        default,
-        with = "::serde_with::rust::double_option",
-        skip_serializing_if = "Option::is_none"
-    )]
-    pub balance: Option<Option<String>>,
-    #[serde(
-        rename = "approx_local_messages",
-        default,
-        with = "::serde_with::rust::double_option",
-        skip_serializing_if = "Option::is_none"
-    )]
-    pub approx_local_messages: Option<Option<Vec<serde_json::Value>>>,
-    #[serde(
-        rename = "approx_cloud_messages",
-        default,
-        with = "::serde_with::rust::double_option",
-        skip_serializing_if = "Option::is_none"
-    )]
-    pub approx_cloud_messages: Option<Option<Vec<serde_json::Value>>>,
-}
-
-impl CreditStatusDetails {
-    pub fn new(has_credits: bool, unlimited: bool) -> CreditStatusDetails {
-        CreditStatusDetails {
-            has_credits,
-            unlimited,
-            balance: None,
-            approx_local_messages: None,
-            approx_cloud_messages: None,
-        }
-    }
-}

codex-rs/codex-backend-openapi-models/src/models/mod.rs

@@ -32,6 +32,3 @@ pub use self::rate_limit_status_details::RateLimitStatusDetails;
 
 pub mod rate_limit_window_snapshot;
 pub use self::rate_limit_window_snapshot::RateLimitWindowSnapshot;
-
-pub mod credit_status_details;
-pub use self::credit_status_details::CreditStatusDetails;

codex-rs/codex-backend-openapi-models/src/models/rate_limit_status_payload.rs

@@ -23,13 +23,6 @@ pub struct RateLimitStatusPayload {
         skip_serializing_if = "Option::is_none"
     )]
     pub rate_limit: Option<Option<Box<models::RateLimitStatusDetails>>>,
-    #[serde(
-        rename = "credits",
-        default,
-        with = "::serde_with::rust::double_option",
-        skip_serializing_if = "Option::is_none"
-    )]
-    pub credits: Option<Option<Box<models::CreditStatusDetails>>>,
 }
 
 impl RateLimitStatusPayload {
@@ -37,15 +30,12 @@ impl RateLimitStatusPayload {
         RateLimitStatusPayload {
             plan_type,
             rate_limit: None,
-            credits: None,
         }
     }
 }
 
 #[derive(Clone, Copy, Debug, Eq, PartialEq, Ord, PartialOrd, Hash, Serialize, Deserialize)]
 pub enum PlanType {
-    #[serde(rename = "guest")]
-    Guest,
     #[serde(rename = "free")]
     Free,
     #[serde(rename = "go")]
@@ -54,8 +44,6 @@ pub enum PlanType {
     Plus,
     #[serde(rename = "pro")]
     Pro,
-    #[serde(rename = "free_workspace")]
-    FreeWorkspace,
     #[serde(rename = "team")]
     Team,
     #[serde(rename = "business")]
@@ -64,8 +52,6 @@ pub enum PlanType {
     Education,
     #[serde(rename = "quorum")]
     Quorum,
-    #[serde(rename = "k12")]
-    K12,
     #[serde(rename = "enterprise")]
     Enterprise,
     #[serde(rename = "edu")]
@@ -74,6 +60,6 @@ pub enum PlanType {
 
 impl Default for PlanType {
     fn default() -> PlanType {
-        Self::Guest
+        Self::Free
     }
 }

codex-rs/codex-backend-openapi-models/src/models/rate_limit_window_snapshot.rs

@@ -7,6 +7,7 @@
  *
  * Generated by: https://openapi-generator.tech
  */
+
 use serde::Deserialize;
 use serde::Serialize;
 

codex-rs/common/src/model_presets.rs

@@ -4,9 +4,6 @@ use codex_app_server_protocol::AuthMode;
 use codex_core::protocol_config_types::ReasoningEffort;
 use once_cell::sync::Lazy;
 
-pub const HIDE_GPT5_1_MIGRATION_PROMPT_CONFIG: &str = "hide_gpt5_1_migration_prompt";
-pub const HIDE_ARCTICFOX_MIGRATION_PROMPT_CONFIG: &str = "hide_arcticfox_migration_prompt";
-
 /// A reasoning effort option that can be surfaced for a model.
 #[derive(Debug, Clone, Copy)]
 pub struct ReasoningEffortPreset {
@@ -20,7 +17,6 @@ pub struct ReasoningEffortPreset {
 pub struct ModelUpgrade {
     pub id: &'static str,
     pub reasoning_effort_mapping: Option<HashMap<ReasoningEffort, ReasoningEffort>>,
-    pub migration_config_key: &'static str,
 }
 
 /// Metadata describing a Codex-supported model.
@@ -42,40 +38,10 @@ pub struct ModelPreset {
     pub is_default: bool,
     /// recommended upgrade model
     pub upgrade: Option<ModelUpgrade>,
-    /// Whether this preset should appear in the picker UI.
-    pub show_in_picker: bool,
 }
 
 static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
     vec![
-        ModelPreset {
-            id: "arcticfox",
-            model: "arcticfox",
-            display_name: "arcticfox",
-            description: "Latest Codex-optimized flagship for deep and fast reasoning.",
-            default_reasoning_effort: ReasoningEffort::Medium,
-            supported_reasoning_efforts: &[
-                ReasoningEffortPreset {
-                    effort: ReasoningEffort::Low,
-                    description: "Fast responses with lighter reasoning",
-                },
-                ReasoningEffortPreset {
-                    effort: ReasoningEffort::Medium,
-                    description: "Balances speed and reasoning depth for everyday tasks",
-                },
-                ReasoningEffortPreset {
-                    effort: ReasoningEffort::High,
-                    description: "Maximizes reasoning depth for complex problems",
-                },
-                ReasoningEffortPreset {
-                    effort: ReasoningEffort::XHigh,
-                    description: "Extra high reasoning depth for complex problems",
-                },
-            ],
-            is_default: true,
-            upgrade: None,
-            show_in_picker: true,
-        },
         ModelPreset {
             id: "gpt-5.1-codex",
             model: "gpt-5.1-codex",
@@ -96,13 +62,8 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     description: "Maximizes reasoning depth for complex or ambiguous problems",
                 },
             ],
-            is_default: false,
-            upgrade: Some(ModelUpgrade {
-                id: "arcticfox",
-                reasoning_effort_mapping: None,
-                migration_config_key: HIDE_ARCTICFOX_MIGRATION_PROMPT_CONFIG,
-            }),
-            show_in_picker: true,
+            is_default: true,
+            upgrade: None,
         },
         ModelPreset {
             id: "gpt-5.1-codex-mini",
@@ -122,7 +83,6 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
             ],
             is_default: false,
             upgrade: None,
-            show_in_picker: true,
         },
         ModelPreset {
             id: "gpt-5.1",
@@ -146,7 +106,6 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
             ],
             is_default: false,
             upgrade: None,
-            show_in_picker: true,
         },
         // Deprecated models.
         ModelPreset {
@@ -171,11 +130,9 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
             ],
             is_default: false,
             upgrade: Some(ModelUpgrade {
-                id: "arcticfox",
+                id: "gpt-5.1-codex",
                 reasoning_effort_mapping: None,
-                migration_config_key: HIDE_ARCTICFOX_MIGRATION_PROMPT_CONFIG,
             }),
-            show_in_picker: false,
         },
         ModelPreset {
             id: "gpt-5-codex-mini",
@@ -197,9 +154,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
             upgrade: Some(ModelUpgrade {
                 id: "gpt-5.1-codex-mini",
                 reasoning_effort_mapping: None,
-                migration_config_key: HIDE_GPT5_1_MIGRATION_PROMPT_CONFIG,
             }),
-            show_in_picker: false,
         },
         ModelPreset {
             id: "gpt-5",
@@ -232,20 +187,16 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     ReasoningEffort::Minimal,
                     ReasoningEffort::Low,
                 )])),
-                migration_config_key: HIDE_GPT5_1_MIGRATION_PROMPT_CONFIG,
             }),
-            show_in_picker: false,
         },
     ]
 });
 
-pub fn builtin_model_presets(auth_mode: Option<AuthMode>) -> Vec<ModelPreset> {
+pub fn builtin_model_presets(_auth_mode: Option<AuthMode>) -> Vec<ModelPreset> {
+    // leave auth mode for later use
     PRESETS
         .iter()
-        .filter(|preset| match auth_mode {
-            Some(AuthMode::ApiKey) => preset.show_in_picker && preset.id != "arcticfox",
-            _ => preset.show_in_picker,
-        })
+        .filter(|preset| preset.upgrade.is_none())
         .cloned()
         .collect()
 }
@@ -257,17 +208,10 @@ pub fn all_model_presets() -> &'static Vec<ModelPreset> {
 #[cfg(test)]
 mod tests {
     use super::*;
-    use codex_app_server_protocol::AuthMode;
 
     #[test]
     fn only_one_default_model_is_configured() {
         let default_models = PRESETS.iter().filter(|preset| preset.is_default).count();
         assert!(default_models == 1);
     }
-
-    #[test]
-    fn arcticfox_hidden_for_api_key_auth() {
-        let presets = builtin_model_presets(Some(AuthMode::ApiKey));
-        assert!(presets.iter().all(|preset| preset.id != "arcticfox"));
-    }
 }

codex-rs/core/Cargo.toml

@@ -22,6 +22,7 @@ chrono = { workspace = true, features = ["serde"] }
 codex-app-server-protocol = { workspace = true }
 codex-apply-patch = { workspace = true }
 codex-async-utils = { workspace = true }
+codex-execpolicy2 = { workspace = true }
 codex-file-search = { workspace = true }
 codex-git = { workspace = true }
 codex-keyring-store = { workspace = true }

codex-rs/core/arcticfox_prompt.md

@@ -1,117 +0,0 @@
-You are Codex, based on GPT-5. You are running as a coding agent in the Codex CLI on a user's computer.
-
-## General
-
-- When searching for text or files, prefer using `rg` or `rg --files` respectively because `rg` is much faster than alternatives like `grep`. (If the `rg` command is not found, then use alternatives.)
-
-## Editing constraints
-
-- Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them.
-- Add succinct code comments that explain what is going on if code is not self-explanatory. You should not add comments like "Assigns the value to the variable", but a brief comment might be useful ahead of a complex code block that the user would otherwise have to spend time parsing out. Usage of these comments should be rare.
-- Try to use apply_patch for single file edits, but it is fine to explore other options to make the edit if it does not work well. Do not use apply_patch for changes that are auto-generated (i.e. generating package.json or running a lint or format command like gofmt) or when scripting is more efficient (such as search and replacing a string across a codebase).
-- You may be in a dirty git worktree.
-    * NEVER revert existing changes you did not make unless explicitly requested, since these changes were made by the user.
-    * If asked to make a commit or code edits and there are unrelated changes to your work or changes that you didn't make in those files, don't revert those changes.
-    * If the changes are in files you've touched recently, you should read carefully and understand how you can work with the changes rather than reverting them.
-    * If the changes are in unrelated files, just ignore them and don't revert them.
-- Do not amend a commit unless explicitly requested to do so.
-- While you are working, you might notice unexpected changes that you didn't make. If this happens, STOP IMMEDIATELY and ask the user how they would like to proceed.
-- **NEVER** use destructive commands like `git reset --hard` or `git checkout --` unless specifically requested or approved by the user.
-
-## Plan tool
-
-When using the planning tool:
-- Skip using the planning tool for straightforward tasks (roughly the easiest 25%).
-- Do not make single-step plans.
-- When you made a plan, update it after having performed one of the sub-tasks that you shared on the plan.
-
-## Codex CLI harness, sandboxing, and approvals
-
-The Codex CLI harness supports several different configurations for sandboxing and escalation approvals that the user can choose from.
-
-Filesystem sandboxing defines which files can be read or written. The options for `sandbox_mode` are:
-- **read-only**: The sandbox only permits reading files.
-- **workspace-write**: The sandbox permits reading files, and editing files in `cwd` and `writable_roots`. Editing files in other directories requires approval.
-- **danger-full-access**: No filesystem sandboxing - all commands are permitted.
-
-Network sandboxing defines whether network can be accessed without approval. Options for `network_access` are:
-- **restricted**: Requires approval
-- **enabled**: No approval needed
-
-Approvals are your mechanism to get user consent to run shell commands without the sandbox. Possible configuration options for `approval_policy` are
-- **untrusted**: The harness will escalate most commands for user approval, apart from a limited allowlist of safe "read" commands.
-- **on-failure**: The harness will allow all commands to run in the sandbox (if enabled), and failures will be escalated to the user for approval to run again without the sandbox.
-- **on-request**: Commands will be run in the sandbox by default, and you can specify in your tool call if you want to escalate a command to run without sandboxing. (Note that this mode is not always available. If it is, you'll see parameters for it in the `shell` command description.)
-- **never**: This is a non-interactive mode where you may NEVER ask the user for approval to run commands. Instead, you must always persist and work around constraints to solve the task for the user. You MUST do your utmost best to finish the task and validate your work before yielding. If this mode is paired with `danger-full-access`, take advantage of it to deliver the best outcome for the user. Further, in this mode, your default testing philosophy is overridden: Even if you don't see local patterns for testing, you may add tests and scripts to validate your work. Just remove them before yielding.
-
-When you are running with `approval_policy == on-request`, and sandboxing enabled, here are scenarios where you'll need to request approval:
-- You need to run a command that writes to a directory that requires it (e.g. running tests that write to /var)
-- You need to run a GUI app (e.g., open/xdg-open/osascript) to open browsers or files.
-- You are running sandboxed and need to run a command that requires network access (e.g. installing packages)
-- If you run a command that is important to solving the user's query, but it fails because of sandboxing, rerun the command with approval. ALWAYS proceed to use the `with_escalated_permissions` and `justification` parameters - do not message the user before requesting approval for the command.
-- You are about to take a potentially destructive action such as an `rm` or `git reset` that the user did not explicitly ask for
-- (for all of these, you should weigh alternative paths that do not require approval)
-
-When `sandbox_mode` is set to read-only, you'll need to request approval for any command that isn't a read.
-
-You will be told what filesystem sandboxing, network sandboxing, and approval mode are active in a developer or user message. If you are not told about this, assume that you are running with workspace-write, network sandboxing enabled, and approval on-failure.
-
-Although they introduce friction to the user because your work is paused until the user responds, you should leverage them when necessary to accomplish important work. If the completing the task requires escalated permissions, Do not let these settings or the sandbox deter you from attempting to accomplish the user's task unless it is set to "never", in which case never ask for approvals.
-
-When requesting approval to execute a command that will require escalated privileges:
-  - Provide the `with_escalated_permissions` parameter with the boolean value true
-  - Include a short, 1 sentence explanation for why you need to enable `with_escalated_permissions` in the justification parameter
-
-## Special user requests
-
-- If the user makes a simple request (such as asking for the time) which you can fulfill by running a terminal command (such as `date`), you should do so.
-- If the user asks for a "review", default to a code review mindset: prioritise identifying bugs, risks, behavioural regressions, and missing tests. Findings must be the primary focus of the response - keep summaries or overviews brief and only after enumerating the issues. Present findings first (ordered by severity with file/line references), follow with open questions or assumptions, and offer a change-summary only as a secondary detail. If no findings are discovered, state that explicitly and mention any residual risks or testing gaps.
-
-## Frontend tasks
-When doing frontend design tasks, avoid collapsing into "AI slop" or safe, average-looking layouts.
-Aim for interfaces that feel intentional, bold, and a bit surprising.
-- Typography: Use expressive, purposeful fonts and avoid default stacks (Inter, Roboto, Arial, system).
-- Color & Look: Choose a clear visual direction; define CSS variables; avoid purple-on-white defaults. No purple bias or dark mode bias.
-- Motion: Use a few meaningful animations (page-load, staggered reveals) instead of generic micro-motions.
-- Background: Don't rely on flat, single-color backgrounds; use gradients, shapes, or subtle patterns to build atmosphere.
-- Overall: Avoid boilerplate layouts and interchangeable UI patterns. Vary themes, type families, and visual languages across outputs.
-- Ensure the page loads properly on both desktop and mobile
-
-Exception: If working within an existing website or design system, preserve the established patterns, structure, and visual language.
-
-## Presenting your work and final message
-
-You are producing plain text that will later be styled by the CLI. Follow these rules exactly. Formatting should make results easy to scan, but not feel mechanical. Use judgment to decide how much structure adds value.
-
-- Default: be very concise; friendly coding teammate tone.
-- Ask only when needed; suggest ideas; mirror the user's style.
-- For substantial work, summarize clearly; follow final‑answer formatting.
-- Skip heavy formatting for simple confirmations.
-- Don't dump large files you've written; reference paths only.
-- No "save/copy this file" - User is on the same machine.
-- Offer logical next steps (tests, commits, build) briefly; add verify steps if you couldn't do something.
-- For code changes:
-  * Lead with a quick explanation of the change, and then give more details on the context covering where and why a change was made. Do not start this explanation with "summary", just jump right in.
-  * If there are natural next steps the user may want to take, suggest them at the end of your response. Do not make suggestions if there are no natural next steps.
-  * When suggesting multiple options, use numeric lists for the suggestions so the user can quickly respond with a single number.
-- The user does not command execution outputs. When asked to show the output of a command (e.g. `git show`), relay the important details in your answer or summarize the key lines so the user understands the result.
-
-### Final answer structure and style guidelines
-
-- Plain text; CLI handles styling. Use structure only when it helps scanability.
-- Headers: optional; short Title Case (1-3 words) wrapped in **…**; no blank line before the first bullet; add only if they truly help.
-- Bullets: use - ; merge related points; keep to one line when possible; 4–6 per list ordered by importance; keep phrasing consistent.
-- Monospace: backticks for commands/paths/env vars/code ids and inline examples; use for literal keyword bullets; never combine with **.
-- Code samples or multi-line snippets should be wrapped in fenced code blocks; include an info string as often as possible.
-- Structure: group related bullets; order sections general → specific → supporting; for subsections, start with a bolded keyword bullet, then items; match complexity to the task.
-- Tone: collaborative, concise, factual; present tense, active voice; self‑contained; no "above/below"; parallel wording.
-- Don'ts: no nested bullets/hierarchies; no ANSI codes; don't cram unrelated keywords; keep keyword lists short—wrap/reformat if long; avoid naming formatting styles in answers.
-- Adaptation: code explanations → precise, structured with code refs; simple tasks → lead with outcome; big changes → logical walkthrough + rationale + next actions; casual one-offs → plain sentences, no headers/bullets.
-- File References: When referencing files in your response follow the below rules:
-  * Use inline code to make file paths clickable.
-  * Each reference should have a stand alone path. Even if it's the same file.
-  * Accepted: absolute, workspace‑relative, a/ or b/ diff prefixes, or bare filename/suffix.
-  * Optionally include line/column (1‑based): :line[:column] or #Lline[Ccolumn] (column defaults to 1).
-  * Do not use URIs like file://, vscode://, or https://.
-  * Do not provide range of lines
-  * Examples: src/app.ts, src/app.ts:42, b/server/index.js#L10, C:\repo\project\main.rs:12:5

codex-rs/core/src/client_common.rs

@@ -136,7 +136,7 @@ fn reserialize_shell_outputs(items: &mut [ResponseItem]) {
 }
 
 fn is_shell_tool_name(name: &str) -> bool {
-    matches!(name, "shell" | "container.exec")
+    matches!(name, "shell" | "container.exec" | "shell_command")
 }
 
 #[derive(Deserialize)]
@@ -165,9 +165,11 @@ fn build_structured_output(parsed: &ExecOutputJson) -> String {
     ));
 
     let mut output = parsed.output.clone();
-    if let Some((stripped, total_lines)) = strip_total_output_header(&parsed.output) {
+    if let Some(total_lines) = extract_total_output_lines(&parsed.output) {
         sections.push(format!("Total output lines: {total_lines}"));
-        output = stripped.to_string();
+        if let Some(stripped) = strip_total_output_header(&output) {
+            output = stripped.to_string();
+        }
     }
 
     sections.push("Output:".to_string());
@@ -176,12 +178,19 @@ fn build_structured_output(parsed: &ExecOutputJson) -> String {
     sections.join("\n")
 }
 
-fn strip_total_output_header(output: &str) -> Option<(&str, u32)> {
+fn extract_total_output_lines(output: &str) -> Option<u32> {
+    let marker_start = output.find("[... omitted ")?;
+    let marker = &output[marker_start..];
+    let (_, after_of) = marker.split_once(" of ")?;
+    let (total_segment, _) = after_of.split_once(' ')?;
+    total_segment.parse::<u32>().ok()
+}
+
+fn strip_total_output_header(output: &str) -> Option<&str> {
     let after_prefix = output.strip_prefix("Total output lines: ")?;
-    let (total_segment, remainder) = after_prefix.split_once('\n')?;
-    let total_lines = total_segment.parse::<u32>().ok()?;
+    let (_, remainder) = after_prefix.split_once('\n')?;
     let remainder = remainder.strip_prefix('\n').unwrap_or(remainder);
-    Some((remainder, total_lines))
+    Some(remainder)
 }
 
 #[derive(Debug)]
@@ -422,7 +431,7 @@ mod tests {
                 expects_apply_patch_instructions: false,
             },
             InstructionsTestCase {
-                slug: "arcticfox",
+                slug: "gpt-5.1-codex",
                 expects_apply_patch_instructions: false,
             },
         ];

codex-rs/core/src/codex.rs

@@ -7,9 +7,6 @@ use std::sync::atomic::AtomicU64;
 use crate::AuthManager;
 use crate::client_common::REVIEW_PROMPT;
 use crate::compact;
-use crate::compact::run_inline_auto_compact_task;
-use crate::compact::should_use_remote_compact_task;
-use crate::compact_remote::run_inline_remote_auto_compact_task;
 use crate::features::Feature;
 use crate::function_tool::FunctionCallError;
 use crate::parse_command::parse_command;
@@ -121,6 +118,7 @@ use crate::user_instructions::UserInstructions;
 use crate::user_notification::UserNotification;
 use crate::util::backoff;
 use codex_async_utils::OrCancelExt;
+use codex_execpolicy2::Policy as ExecPolicy;
 use codex_otel::otel_event_manager::OtelEventManager;
 use codex_protocol::config_types::ReasoningEffort as ReasoningEffortConfig;
 use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
@@ -166,6 +164,10 @@ impl Codex {
 
         let user_instructions = get_user_instructions(&config).await;
 
+        let exec_policy = crate::exec_policy::exec_policy_for(&config.features, &config.codex_home)
+            .await
+            .map_err(|err| CodexErr::Fatal(format!("failed to load execpolicy: {err}")))?;
+
         let config = Arc::new(config);
 
         let session_configuration = SessionConfiguration {
@@ -182,6 +184,7 @@ impl Codex {
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             features: config.features.clone(),
+            exec_policy,
             session_source,
         };
 
@@ -279,6 +282,7 @@ pub(crate) struct TurnContext {
     pub(crate) final_output_json_schema: Option<Value>,
     pub(crate) codex_linux_sandbox_exe: Option<PathBuf>,
     pub(crate) tool_call_gate: Arc<ReadinessFlag>,
+    pub(crate) exec_policy: Arc<ExecPolicy>,
     pub(crate) truncation_policy: TruncationPolicy,
 }
 
@@ -296,6 +300,7 @@ impl TurnContext {
     }
 }
 
+#[allow(dead_code)]
 #[derive(Clone)]
 pub(crate) struct SessionConfiguration {
     /// Provider identifier ("openai", "openrouter", ...).
@@ -335,6 +340,8 @@ pub(crate) struct SessionConfiguration {
 
     /// Set of feature flags for this session
     features: Features,
+    /// Execpolicy policy, applied only when enabled by feature flag.
+    exec_policy: Arc<ExecPolicy>,
 
     // TODO(pakrym): Remove config from here
     original_config_do_not_use: Arc<Config>,
@@ -435,6 +442,7 @@ impl Session {
             final_output_json_schema: None,
             codex_linux_sandbox_exe: config.codex_linux_sandbox_exe.clone(),
             tool_call_gate: Arc::new(ReadinessFlag::new()),
+            exec_policy: session_configuration.exec_policy.clone(),
             truncation_policy: TruncationPolicy::new(&per_turn_config),
         }
     }
@@ -583,10 +591,6 @@ impl Session {
             msg: EventMsg::SessionConfigured(SessionConfiguredEvent {
                 session_id: conversation_id,
                 model: session_configuration.model.clone(),
-                model_provider_id: config.model_provider_id.clone(),
-                approval_policy: session_configuration.approval_policy,
-                sandbox_policy: session_configuration.sandbox_policy.clone(),
-                cwd: session_configuration.cwd.clone(),
                 reasoning_effort: session_configuration.model_reasoning_effort,
                 history_log_id,
                 history_entry_count,
@@ -1085,14 +1089,11 @@ impl Session {
         self.send_token_count_event(turn_context).await;
     }
 
-    pub(crate) async fn recompute_token_usage(&self, turn_context: &TurnContext) {
-        let Some(estimated_total_tokens) = self
-            .clone_history()
-            .await
-            .estimate_token_count(turn_context)
-        else {
-            return;
-        };
+    pub(crate) async fn override_last_token_usage_estimate(
+        &self,
+        turn_context: &TurnContext,
+        estimated_total_tokens: i64,
+    ) {
         {
             let mut state = self.state.lock().await;
             let mut info = state.token_info().unwrap_or(TokenUsageInfo {
@@ -1335,10 +1336,7 @@ impl Session {
 }
 
 async fn submission_loop(sess: Arc<Session>, config: Arc<Config>, rx_sub: Receiver<Submission>) {
-    // Seed with context in case there is an OverrideTurnContext first.
-    let mut previous_context: Option<Arc<TurnContext>> =
-        Some(sess.new_turn(SessionSettingsUpdate::default()).await);
-
+    let mut previous_context: Option<Arc<TurnContext>> = None;
     // To break out of this loop, send Op::Shutdown.
     while let Ok(sub) = rx_sub.recv().await {
         debug!(?sub, "Submission");
@@ -1443,7 +1441,6 @@ mod handlers {
     use codex_protocol::protocol::ReviewRequest;
     use codex_protocol::protocol::TurnAbortReason;
 
-    use codex_protocol::user_input::UserInput;
     use std::sync::Arc;
     use tracing::info;
     use tracing::warn;
@@ -1652,14 +1649,8 @@ mod handlers {
             .new_turn_with_sub_id(sub_id, SessionSettingsUpdate::default())
             .await;
 
-        sess.spawn_task(
-            Arc::clone(&turn_context),
-            vec![UserInput::Text {
-                text: turn_context.compact_prompt().to_string(),
-            }],
-            CompactTask,
-        )
-        .await;
+        sess.spawn_task(Arc::clone(&turn_context), vec![], CompactTask)
+            .await;
     }
 
     pub async fn shutdown(sess: &Arc<Session>, sub_id: String) -> bool {
@@ -1785,6 +1776,7 @@ async fn spawn_review_thread(
         final_output_json_schema: None,
         codex_linux_sandbox_exe: parent_turn_context.codex_linux_sandbox_exe.clone(),
         tool_call_gate: Arc::new(ReadinessFlag::new()),
+        exec_policy: parent_turn_context.exec_policy.clone(),
         truncation_policy: TruncationPolicy::new(&per_turn_config),
     };
 
@@ -1793,12 +1785,7 @@ async fn spawn_review_thread(
         text: review_prompt,
     }];
     let tc = Arc::new(review_turn_context);
-    sess.spawn_task(
-        tc.clone(),
-        input,
-        ReviewTask::new(review_request.append_to_original_thread),
-    )
-    .await;
+    sess.spawn_task(tc.clone(), input, ReviewTask).await;
 
     // Announce entering review mode so UIs can switch modes.
     sess.send_event(&tc, EventMsg::EnteredReviewMode(review_request))
@@ -1899,12 +1886,7 @@ pub(crate) async fn run_task(
 
                 // as long as compaction works well in getting us way below the token limit, we shouldn't worry about being in an infinite loop.
                 if token_limit_reached {
-                    if should_use_remote_compact_task(&sess).await {
-                        run_inline_remote_auto_compact_task(sess.clone(), turn_context.clone())
-                            .await;
-                    } else {
-                        run_inline_auto_compact_task(sess.clone(), turn_context.clone()).await;
-                    }
+                    compact::run_inline_auto_compact_task(sess.clone(), turn_context.clone()).await;
                     continue;
                 }
 
@@ -1988,14 +1970,12 @@ async fn run_turn(
     let mut base_instructions = turn_context.base_instructions.clone();
     if parallel_tool_calls {
         static INSTRUCTIONS: &str = include_str!("../templates/parallel/instructions.md");
-        if let Some(family) =
-            find_family_for_model(&sess.state.lock().await.session_configuration.model)
-        {
-            let mut new_instructions = base_instructions.unwrap_or(family.base_instructions);
-            new_instructions.push_str(INSTRUCTIONS);
-            base_instructions = Some(new_instructions);
-        }
+        static INSERTION_SPOT: &str = "## Editing constraints";
+        base_instructions
+            .as_mut()
+            .map(|base| base.replace(INSERTION_SPOT, INSTRUCTIONS));
     }
+
     let prompt = Prompt {
         input,
         tools: router.specs(),
@@ -2487,9 +2467,8 @@ mod tests {
             duration: StdDuration::from_secs(1),
             timed_out: true,
         };
-        let (_, turn_context) = make_session_and_context();
 
-        let out = format_exec_output_str(&exec, turn_context.truncation_policy);
+        let out = format_exec_output_str(&exec);
 
         assert_eq!(
             out,
@@ -2605,6 +2584,7 @@ mod tests {
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             features: Features::default(),
+            exec_policy: Arc::new(codex_execpolicy2::PolicyParser::new().build()),
             session_source: SessionSource::Exec,
         };
 
@@ -2682,6 +2662,7 @@ mod tests {
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             features: Features::default(),
+            exec_policy: Arc::new(codex_execpolicy2::PolicyParser::new().build()),
             session_source: SessionSource::Exec,
         };
 
@@ -2812,8 +2793,7 @@ mod tests {
         let input = vec![UserInput::Text {
             text: "start review".to_string(),
         }];
-        sess.spawn_task(Arc::clone(&tc), input, ReviewTask::new(true))
-            .await;
+        sess.spawn_task(Arc::clone(&tc), input, ReviewTask).await;
 
         sess.abort_all_tasks(TurnAbortReason::Interrupted).await;
 

codex-rs/core/src/command_safety/is_dangerous_command.rs

@@ -1,6 +1,8 @@
 use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::SandboxPolicy;
 
+use crate::sandboxing::SandboxPermissions;
+
 use crate::bash::parse_shell_lc_plain_commands;
 use crate::is_safe_command::is_known_safe_command;
 
@@ -8,7 +10,7 @@ pub fn requires_initial_appoval(
     policy: AskForApproval,
     sandbox_policy: &SandboxPolicy,
     command: &[String],
-    with_escalated_permissions: bool,
+    sandbox_permissions: SandboxPermissions,
 ) -> bool {
     if is_known_safe_command(command) {
         return false;
@@ -24,8 +26,7 @@ pub fn requires_initial_appoval(
             // In restricted sandboxes (ReadOnly/WorkspaceWrite), do not prompt for
             // non‑escalated, non‑dangerous commands — let the sandbox enforce
             // restrictions (e.g., block network/write) without a user prompt.
-            let wants_escalation: bool = with_escalated_permissions;
-            if wants_escalation {
+            if sandbox_permissions.requires_escalated_permissions() {
                 return true;
             }
             command_might_be_dangerous(command)

codex-rs/core/src/compact.rs

@@ -7,7 +7,6 @@ use crate::codex::TurnContext;
 use crate::codex::get_last_assistant_message_from_turn;
 use crate::error::CodexErr;
 use crate::error::Result as CodexResult;
-use crate::features::Feature;
 use crate::protocol::AgentMessageEvent;
 use crate::protocol::CompactedItem;
 use crate::protocol::ErrorEvent;
@@ -19,7 +18,6 @@ use crate::truncate::TruncationPolicy;
 use crate::truncate::approx_token_count;
 use crate::truncate::truncate_text;
 use crate::util::backoff;
-use codex_app_server_protocol::AuthMode;
 use codex_protocol::items::TurnItem;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseInputItem;
@@ -33,22 +31,12 @@ pub const SUMMARIZATION_PROMPT: &str = include_str!("../templates/compact/prompt
 pub const SUMMARY_PREFIX: &str = include_str!("../templates/compact/summary_prefix.md");
 const COMPACT_USER_MESSAGE_MAX_TOKENS: usize = 20_000;
 
-pub(crate) async fn should_use_remote_compact_task(session: &Session) -> bool {
-    session
-        .services
-        .auth_manager
-        .auth()
-        .is_some_and(|auth| auth.mode == AuthMode::ChatGPT)
-        && session.enabled(Feature::RemoteCompaction).await
-}
-
 pub(crate) async fn run_inline_auto_compact_task(
     sess: Arc<Session>,
     turn_context: Arc<TurnContext>,
 ) {
     let prompt = turn_context.compact_prompt().to_string();
     let input = vec![UserInput::Text { text: prompt }];
-
     run_compact_task_inner(sess, turn_context, input).await;
 }
 
@@ -56,12 +44,13 @@ pub(crate) async fn run_compact_task(
     sess: Arc<Session>,
     turn_context: Arc<TurnContext>,
     input: Vec<UserInput>,
-) {
+) -> Option<String> {
     let start_event = EventMsg::TaskStarted(TaskStartedEvent {
         model_context_window: turn_context.client.get_model_context_window(),
     });
     sess.send_event(&turn_context, start_event).await;
     run_compact_task_inner(sess.clone(), turn_context, input).await;
+    None
 }
 
 async fn run_compact_task_inner(
@@ -171,7 +160,15 @@ async fn run_compact_task_inner(
         .collect();
     new_history.extend(ghost_snapshots);
     sess.replace_history(new_history).await;
-    sess.recompute_token_usage(&turn_context).await;
+
+    if let Some(estimated_tokens) = sess
+        .clone_history()
+        .await
+        .estimate_token_count(&turn_context)
+    {
+        sess.override_last_token_usage_estimate(&turn_context, estimated_tokens)
+            .await;
+    }
 
     let rollout_item = RolloutItem::Compacted(CompactedItem {
         message: summary_text.clone(),

codex-rs/core/src/compact_remote.rs

@@ -10,38 +10,52 @@ use crate::protocol::ErrorEvent;
 use crate::protocol::EventMsg;
 use crate::protocol::RolloutItem;
 use crate::protocol::TaskStartedEvent;
+use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ResponseItem;
+use codex_protocol::user_input::UserInput;
 
-pub(crate) async fn run_inline_remote_auto_compact_task(
+pub(crate) async fn run_remote_compact_task(
     sess: Arc<Session>,
     turn_context: Arc<TurnContext>,
-) {
-    run_remote_compact_task_inner(&sess, &turn_context).await;
-}
-
-pub(crate) async fn run_remote_compact_task(sess: Arc<Session>, turn_context: Arc<TurnContext>) {
+    input: Vec<UserInput>,
+) -> Option<String> {
     let start_event = EventMsg::TaskStarted(TaskStartedEvent {
         model_context_window: turn_context.client.get_model_context_window(),
     });
     sess.send_event(&turn_context, start_event).await;
 
-    run_remote_compact_task_inner(&sess, &turn_context).await;
-}
-
-async fn run_remote_compact_task_inner(sess: &Arc<Session>, turn_context: &Arc<TurnContext>) {
-    if let Err(err) = run_remote_compact_task_inner_impl(sess, turn_context).await {
-        let event = EventMsg::Error(ErrorEvent {
-            message: format!("Error running remote compact task: {err}"),
-        });
-        sess.send_event(turn_context, event).await;
+    match run_remote_compact_task_inner(&sess, &turn_context, input).await {
+        Ok(()) => {
+            let event = EventMsg::AgentMessage(AgentMessageEvent {
+                message: "Compact task completed".to_string(),
+            });
+            sess.send_event(&turn_context, event).await;
+        }
+        Err(err) => {
+            let event = EventMsg::Error(ErrorEvent {
+                message: err.to_string(),
+            });
+            sess.send_event(&turn_context, event).await;
+        }
     }
+
+    None
 }
 
-async fn run_remote_compact_task_inner_impl(
+async fn run_remote_compact_task_inner(
     sess: &Arc<Session>,
     turn_context: &Arc<TurnContext>,
+    input: Vec<UserInput>,
 ) -> CodexResult<()> {
     let mut history = sess.clone_history().await;
+    if !input.is_empty() {
+        let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
+        history.record_items(
+            &[initial_input_for_turn.into()],
+            turn_context.truncation_policy,
+        );
+    }
+
     let prompt = Prompt {
         input: history.get_history_for_prompt(),
         tools: vec![],
@@ -66,7 +80,15 @@ async fn run_remote_compact_task_inner_impl(
         new_history.extend(ghost_snapshots);
     }
     sess.replace_history(new_history.clone()).await;
-    sess.recompute_token_usage(turn_context).await;
+
+    if let Some(estimated_tokens) = sess
+        .clone_history()
+        .await
+        .estimate_token_count(turn_context.as_ref())
+    {
+        sess.override_last_token_usage_estimate(turn_context.as_ref(), estimated_tokens)
+            .await;
+    }
 
     let compacted_item = CompactedItem {
         message: String::new(),
@@ -74,11 +96,5 @@ async fn run_remote_compact_task_inner_impl(
     };
     sess.persist_rollout_items(&[RolloutItem::Compacted(compacted_item)])
         .await;
-
-    let event = EventMsg::AgentMessage(AgentMessageEvent {
-        message: "Compact task completed".to_string(),
-    });
-    sess.send_event(turn_context, event).await;
-
     Ok(())
 }

codex-rs/core/src/config/edit.rs

@@ -845,36 +845,6 @@ hide_gpt5_1_migration_prompt = true
         assert_eq!(contents, expected);
     }
 
-    #[test]
-    fn blocking_set_hide_arcticfox_migration_prompt_preserves_table() {
-        let tmp = tempdir().expect("tmpdir");
-        let codex_home = tmp.path();
-        std::fs::write(
-            codex_home.join(CONFIG_TOML_FILE),
-            r#"[notice]
-existing = "value"
-"#,
-        )
-        .expect("seed");
-        apply_blocking(
-            codex_home,
-            None,
-            &[ConfigEdit::SetNoticeHideModelMigrationPrompt(
-                "hide_arcticfox_migration_prompt".to_string(),
-                true,
-            )],
-        )
-        .expect("persist");
-
-        let contents =
-            std::fs::read_to_string(codex_home.join(CONFIG_TOML_FILE)).expect("read config");
-        let expected = r#"[notice]
-existing = "value"
-hide_arcticfox_migration_prompt = true
-"#;
-        assert_eq!(contents, expected);
-    }
-
     #[test]
     fn blocking_replace_mcp_servers_round_trips() {
         let tmp = tempdir().expect("tmpdir");

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

@@ -62,11 +62,11 @@ pub mod profile;
 pub mod types;
 
 #[cfg(target_os = "windows")]
-pub const OPENAI_DEFAULT_MODEL: &str = "arcticfox";
+pub const OPENAI_DEFAULT_MODEL: &str = "gpt-5.1";
 #[cfg(not(target_os = "windows"))]
-pub const OPENAI_DEFAULT_MODEL: &str = "arcticfox";
-const OPENAI_DEFAULT_REVIEW_MODEL: &str = "arcticfox";
-pub const GPT_5_CODEX_MEDIUM_MODEL: &str = "arcticfox";
+pub const OPENAI_DEFAULT_MODEL: &str = "gpt-5.1-codex";
+const OPENAI_DEFAULT_REVIEW_MODEL: &str = "gpt-5.1-codex";
+pub const GPT_5_CODEX_MEDIUM_MODEL: &str = "gpt-5.1-codex";
 
 /// Maximum number of bytes of the documentation that will be embedded. Larger
 /// files are *silently truncated* to this size so we do not take up too much of
@@ -81,7 +81,7 @@ pub struct Config {
     /// Optional override of model selection.
     pub model: String,
 
-    /// Model used specifically for review sessions. Defaults to "arcticfox".
+    /// Model used specifically for review sessions. Defaults to "gpt-5.1-codex".
     pub review_model: String,
 
     pub model_family: ModelFamily,

codex-rs/core/src/config/types.rs

@@ -282,14 +282,6 @@ pub enum OtelHttpProtocol {
     Json,
 }
 
-#[derive(Deserialize, Debug, Clone, PartialEq, Default)]
-#[serde(rename_all = "kebab-case")]
-pub struct OtelTlsConfig {
-    pub ca_certificate: Option<PathBuf>,
-    pub client_certificate: Option<PathBuf>,
-    pub client_private_key: Option<PathBuf>,
-}
-
 /// Which OTEL exporter to use.
 #[derive(Deserialize, Debug, Clone, PartialEq)]
 #[serde(rename_all = "kebab-case")]
@@ -297,18 +289,12 @@ pub enum OtelExporterKind {
     None,
     OtlpHttp {
         endpoint: String,
-        #[serde(default)]
         headers: HashMap<String, String>,
         protocol: OtelHttpProtocol,
-        #[serde(default)]
-        tls: Option<OtelTlsConfig>,
     },
     OtlpGrpc {
         endpoint: String,
-        #[serde(default)]
         headers: HashMap<String, String>,
-        #[serde(default)]
-        tls: Option<OtelTlsConfig>,
     },
 }
 
@@ -378,8 +364,6 @@ pub struct Notice {
     pub hide_rate_limit_model_nudge: Option<bool>,
     /// Tracks whether the user has seen the model migration prompt
     pub hide_gpt5_1_migration_prompt: Option<bool>,
-    /// Tracks whether the user has seen the arcticfox migration prompt
-    pub hide_arcticfox_migration_prompt: Option<bool>,
 }
 
 impl Notice {

codex-rs/core/src/context_manager/history.rs

@@ -145,17 +145,13 @@ impl ContextManager {
     }
 
     fn process_item(&self, item: &ResponseItem, policy: TruncationPolicy) -> ResponseItem {
-        let policy_with_serialization_budget = policy.mul(1.2);
         match item {
             ResponseItem::FunctionCallOutput { call_id, output } => {
-                let truncated =
-                    truncate_text(output.content.as_str(), policy_with_serialization_budget);
-                let truncated_items = output.content_items.as_ref().map(|items| {
-                    truncate_function_output_items_with_policy(
-                        items,
-                        policy_with_serialization_budget,
-                    )
-                });
+                let truncated = truncate_text(output.content.as_str(), policy);
+                let truncated_items = output
+                    .content_items
+                    .as_ref()
+                    .map(|items| truncate_function_output_items_with_policy(items, policy));
                 ResponseItem::FunctionCallOutput {
                     call_id: call_id.clone(),
                     output: FunctionCallOutputPayload {
@@ -166,7 +162,7 @@ impl ContextManager {
                 }
             }
             ResponseItem::CustomToolCallOutput { call_id, output } => {
-                let truncated = truncate_text(output, policy_with_serialization_budget);
+                let truncated = truncate_text(output, policy);
                 ResponseItem::CustomToolCallOutput {
                     call_id: call_id.clone(),
                     output: truncated,

codex-rs/core/src/context_manager/history_tests.rs

@@ -12,8 +12,8 @@ use codex_protocol::models::ReasoningItemReasoningSummary;
 use pretty_assertions::assert_eq;
 use regex_lite::Regex;
 
+const EXEC_FORMAT_MAX_LINES: usize = 256;
 const EXEC_FORMAT_MAX_BYTES: usize = 10_000;
-const EXEC_FORMAT_MAX_TOKENS: usize = 2_500;
 
 fn assistant_msg(text: &str) -> ResponseItem {
     ResponseItem::Message {
@@ -56,10 +56,6 @@ fn reasoning_msg(text: &str) -> ResponseItem {
     }
 }
 
-fn truncate_exec_output(content: &str) -> String {
-    truncate::truncate_text(content, TruncationPolicy::Tokens(EXEC_FORMAT_MAX_TOKENS))
-}
-
 #[test]
 fn filters_non_api_messages() {
     let mut h = ContextManager::default();
@@ -232,7 +228,7 @@ fn normalization_retains_local_shell_outputs() {
         ResponseItem::FunctionCallOutput {
             call_id: "shell-1".to_string(),
             output: FunctionCallOutputPayload {
-                content: "Total output lines: 1\n\nok".to_string(),
+                content: "ok".to_string(),
                 ..Default::default()
             },
         },
@@ -334,8 +330,8 @@ fn record_items_respects_custom_token_limit() {
     assert!(stored.content.contains("tokens truncated"));
 }
 
-fn assert_truncated_message_matches(message: &str, line: &str, expected_removed: usize) {
-    let pattern = truncated_message_pattern(line);
+fn assert_truncated_message_matches(message: &str, line: &str, total_lines: usize) {
+    let pattern = truncated_message_pattern(line, total_lines);
     let regex = Regex::new(&pattern).unwrap_or_else(|err| {
         panic!("failed to compile regex {pattern}: {err}");
     });
@@ -351,18 +347,23 @@ fn assert_truncated_message_matches(message: &str, line: &str, expected_removed:
         "body exceeds byte limit: {} bytes",
         body.len()
     );
-    let removed: usize = captures
-        .name("removed")
-        .expect("missing removed capture")
-        .as_str()
-        .parse()
-        .unwrap_or_else(|err| panic!("invalid removed tokens: {err}"));
-    assert_eq!(removed, expected_removed, "mismatched removed token count");
 }
 
-fn truncated_message_pattern(line: &str) -> String {
+fn truncated_message_pattern(line: &str, total_lines: usize) -> String {
+    let head_lines = EXEC_FORMAT_MAX_LINES / 2;
+    let tail_lines = EXEC_FORMAT_MAX_LINES - head_lines;
+    let head_take = head_lines.min(total_lines);
+    let tail_take = tail_lines.min(total_lines.saturating_sub(head_take));
+    let omitted = total_lines.saturating_sub(head_take + tail_take);
     let escaped_line = regex_lite::escape(line);
-    format!(r"(?s)^(?P<body>{escaped_line}.*?)(?:\r?)?…(?P<removed>\d+) tokens truncated…(?:.*)?$")
+    if omitted == 0 {
+        return format!(
+            r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} removed \d+ bytes to fit {EXEC_FORMAT_MAX_BYTES} byte limit \.{{3}}]\n\n.*)$",
+        );
+    }
+    format!(
+        r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} omitted {omitted} of {total_lines} lines \.{{3}}]\n\n.*)$",
+    )
 }
 
 #[test]
@@ -370,18 +371,27 @@ fn format_exec_output_truncates_large_error() {
     let line = "very long execution error line that should trigger truncation\n";
     let large_error = line.repeat(2_500); // way beyond both byte and line limits
 
-    let truncated = truncate_exec_output(&large_error);
+    let truncated = truncate::truncate_with_line_bytes_budget(&large_error, EXEC_FORMAT_MAX_BYTES);
 
-    assert_truncated_message_matches(&truncated, line, 36250);
+    let total_lines = large_error.lines().count();
+    assert_truncated_message_matches(&truncated, line, total_lines);
     assert_ne!(truncated, large_error);
 }
 
 #[test]
 fn format_exec_output_marks_byte_truncation_without_omitted_lines() {
-    let long_line = "a".repeat(EXEC_FORMAT_MAX_BYTES + 10000);
-    let truncated = truncate_exec_output(&long_line);
+    let long_line = "a".repeat(EXEC_FORMAT_MAX_BYTES + 50);
+    let truncated = truncate::truncate_with_line_bytes_budget(&long_line, EXEC_FORMAT_MAX_BYTES);
+
     assert_ne!(truncated, long_line);
-    assert_truncated_message_matches(&truncated, "a", 2500);
+    let removed_bytes = long_line.len().saturating_sub(EXEC_FORMAT_MAX_BYTES);
+    let marker_line = format!(
+        "[... removed {removed_bytes} bytes to fit {EXEC_FORMAT_MAX_BYTES} byte limit ...]"
+    );
+    assert!(
+        truncated.contains(&marker_line),
+        "missing byte truncation marker: {truncated}"
+    );
     assert!(
         !truncated.contains("omitted"),
         "line omission marker should not appear when no lines were dropped: {truncated}"
@@ -391,25 +401,34 @@ fn format_exec_output_marks_byte_truncation_without_omitted_lines() {
 #[test]
 fn format_exec_output_returns_original_when_within_limits() {
     let content = "example output\n".repeat(10);
-    assert_eq!(truncate_exec_output(&content), content);
+
+    assert_eq!(
+        truncate::truncate_with_line_bytes_budget(&content, EXEC_FORMAT_MAX_BYTES),
+        content
+    );
 }
 
 #[test]
 fn format_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
-    let total_lines = 2_000;
-    let filler = "x".repeat(64);
+    let total_lines = EXEC_FORMAT_MAX_LINES + 100;
     let content: String = (0..total_lines)
-        .map(|idx| format!("line-{idx}-{filler}\n"))
+        .map(|idx| format!("line-{idx}\n"))
         .collect();
 
-    let truncated = truncate_exec_output(&content);
-    assert_truncated_message_matches(&truncated, "line-0-", 34_723);
+    let truncated = truncate::truncate_with_line_bytes_budget(&content, EXEC_FORMAT_MAX_BYTES);
+    let omitted = total_lines - EXEC_FORMAT_MAX_LINES;
+    let expected_marker = format!("[... omitted {omitted} of {total_lines} lines ...]");
+
     assert!(
-        truncated.contains("line-0-"),
+        truncated.contains(&expected_marker),
+        "missing omitted marker: {truncated}"
+    );
+    assert!(
+        truncated.contains("line-0\n"),
         "expected head line to remain: {truncated}"
     );
 
-    let last_line = format!("line-{}-", total_lines - 1);
+    let last_line = format!("line-{}\n", total_lines - 1);
     assert!(
         truncated.contains(&last_line),
         "expected tail line to remain: {truncated}"
@@ -418,15 +437,22 @@ fn format_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
 
 #[test]
 fn format_exec_output_prefers_line_marker_when_both_limits_exceeded() {
-    let total_lines = 300;
+    let total_lines = EXEC_FORMAT_MAX_LINES + 42;
     let long_line = "x".repeat(256);
     let content: String = (0..total_lines)
         .map(|idx| format!("line-{idx}-{long_line}\n"))
         .collect();
 
-    let truncated = truncate_exec_output(&content);
+    let truncated = truncate::truncate_with_line_bytes_budget(&content, EXEC_FORMAT_MAX_BYTES);
 
-    assert_truncated_message_matches(&truncated, "line-0-", 17_423);
+    assert!(
+        truncated.contains("[... omitted 42 of 298 lines ...]"),
+        "expected omitted marker when line count exceeds limit: {truncated}"
+    );
+    assert!(
+        !truncated.contains("byte limit"),
+        "line omission marker should take precedence over byte marker: {truncated}"
+    );
 }
 
 //TODO(aibrahim): run CI in release mode.

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

@@ -1,4 +1,5 @@
 mod history;
 mod normalize;
 
+pub(crate) use crate::truncate::truncate_with_line_bytes_budget;
 pub(crate) use history::ContextManager;

codex-rs/core/src/exec.rs

@@ -182,11 +182,12 @@ async fn exec_windows_sandbox(
         ..
     } = params;
 
-    let policy_str = serde_json::to_string(sandbox_policy).map_err(|err| {
-        CodexErr::Io(io::Error::other(format!(
-            "failed to serialize Windows sandbox policy: {err}"
-        )))
-    })?;
+    let policy_str = match sandbox_policy {
+        SandboxPolicy::DangerFullAccess => "workspace-write",
+        SandboxPolicy::ReadOnly => "read-only",
+        SandboxPolicy::WorkspaceWrite { .. } => "workspace-write",
+    };
+
     let sandbox_cwd = cwd.clone();
     let codex_home = find_codex_home().map_err(|err| {
         CodexErr::Io(io::Error::other(format!(
@@ -195,7 +196,7 @@ async fn exec_windows_sandbox(
     })?;
     let spawn_res = tokio::task::spawn_blocking(move || {
         run_windows_sandbox_capture(
-            policy_str.as_str(),
+            policy_str,
             &sandbox_cwd,
             codex_home.as_ref(),
             command,
@@ -443,9 +444,7 @@ async fn exec(
     stdout_stream: Option<StdoutStream>,
 ) -> Result<RawExecToolCallOutput> {
     #[cfg(target_os = "windows")]
-    if sandbox == SandboxType::WindowsRestrictedToken
-        && !matches!(sandbox_policy, SandboxPolicy::DangerFullAccess)
-    {
+    if sandbox == SandboxType::WindowsRestrictedToken {
         return exec_windows_sandbox(params, sandbox_policy).await;
     }
     let timeout = params.timeout_duration();

codex-rs/core/src/exec_policy.rs

@@ -0,0 +1,365 @@
+use std::io::ErrorKind;
+use std::path::Path;
+use std::path::PathBuf;
+use std::sync::Arc;
+
+use crate::command_safety::is_dangerous_command::requires_initial_appoval;
+use codex_execpolicy2::Decision;
+use codex_execpolicy2::Evaluation;
+use codex_execpolicy2::Policy;
+use codex_execpolicy2::PolicyParser;
+use codex_protocol::protocol::AskForApproval;
+use codex_protocol::protocol::SandboxPolicy;
+use thiserror::Error;
+use tokio::fs;
+
+use crate::bash::parse_shell_lc_plain_commands;
+use crate::features::Feature;
+use crate::features::Features;
+use crate::sandboxing::SandboxPermissions;
+use crate::tools::sandboxing::ApprovalRequirement;
+
+const FORBIDDEN_REASON: &str = "execpolicy forbids this command";
+const PROMPT_REASON: &str = "execpolicy requires approval for this command";
+const POLICY_DIR_NAME: &str = "policy";
+const POLICY_EXTENSION: &str = "codexpolicy";
+
+#[derive(Debug, Error)]
+pub enum ExecPolicyError {
+    #[error("failed to read execpolicy files from {dir}: {source}")]
+    ReadDir {
+        dir: PathBuf,
+        source: std::io::Error,
+    },
+
+    #[error("failed to read execpolicy file {path}: {source}")]
+    ReadFile {
+        path: PathBuf,
+        source: std::io::Error,
+    },
+
+    #[error("failed to parse execpolicy file {path}: {source}")]
+    ParsePolicy {
+        path: String,
+        source: codex_execpolicy2::Error,
+    },
+}
+
+pub(crate) async fn exec_policy_for(
+    features: &Features,
+    codex_home: &Path,
+) -> Result<Arc<Policy>, ExecPolicyError> {
+    if !features.enabled(Feature::ExecPolicy) {
+        return Ok(Arc::new(Policy::empty()));
+    }
+
+    let policy_dir = codex_home.join(POLICY_DIR_NAME);
+    let policy_paths = collect_policy_files(&policy_dir).await?;
+
+    let mut parser = PolicyParser::new();
+    for policy_path in &policy_paths {
+        let contents =
+            fs::read_to_string(policy_path)
+                .await
+                .map_err(|source| ExecPolicyError::ReadFile {
+                    path: policy_path.clone(),
+                    source,
+                })?;
+        let identifier = policy_path.to_string_lossy().to_string();
+        parser
+            .parse(&identifier, &contents)
+            .map_err(|source| ExecPolicyError::ParsePolicy {
+                path: identifier,
+                source,
+            })?;
+    }
+
+    let policy = Arc::new(parser.build());
+    tracing::debug!(
+        "loaded execpolicy from {} files in {}",
+        policy_paths.len(),
+        policy_dir.display()
+    );
+
+    Ok(policy)
+}
+
+fn evaluate_with_policy(
+    policy: &Policy,
+    command: &[String],
+    approval_policy: AskForApproval,
+) -> Option<ApprovalRequirement> {
+    let commands = parse_shell_lc_plain_commands(command).unwrap_or_else(|| vec![command.to_vec()]);
+    let evaluation = policy.check_multiple(commands.iter());
+
+    match evaluation {
+        Evaluation::Match { decision, .. } => match decision {
+            Decision::Forbidden => Some(ApprovalRequirement::Forbidden {
+                reason: FORBIDDEN_REASON.to_string(),
+            }),
+            Decision::Prompt => {
+                let reason = PROMPT_REASON.to_string();
+                if matches!(approval_policy, AskForApproval::Never) {
+                    Some(ApprovalRequirement::Forbidden { reason })
+                } else {
+                    Some(ApprovalRequirement::NeedsApproval {
+                        reason: Some(reason),
+                    })
+                }
+            }
+            Decision::Allow => Some(ApprovalRequirement::Skip),
+        },
+        Evaluation::NoMatch { .. } => None,
+    }
+}
+
+pub(crate) fn create_approval_requirement_for_command(
+    policy: &Policy,
+    command: &[String],
+    approval_policy: AskForApproval,
+    sandbox_policy: &SandboxPolicy,
+    sandbox_permissions: SandboxPermissions,
+) -> ApprovalRequirement {
+    if let Some(requirement) = evaluate_with_policy(policy, command, approval_policy) {
+        return requirement;
+    }
+
+    if requires_initial_appoval(
+        approval_policy,
+        sandbox_policy,
+        command,
+        sandbox_permissions,
+    ) {
+        ApprovalRequirement::NeedsApproval { reason: None }
+    } else {
+        ApprovalRequirement::Skip
+    }
+}
+
+async fn collect_policy_files(dir: &Path) -> Result<Vec<PathBuf>, ExecPolicyError> {
+    let mut read_dir = match fs::read_dir(dir).await {
+        Ok(read_dir) => read_dir,
+        Err(err) if err.kind() == ErrorKind::NotFound => return Ok(Vec::new()),
+        Err(source) => {
+            return Err(ExecPolicyError::ReadDir {
+                dir: dir.to_path_buf(),
+                source,
+            });
+        }
+    };
+
+    let mut policy_paths = Vec::new();
+    while let Some(entry) =
+        read_dir
+            .next_entry()
+            .await
+            .map_err(|source| ExecPolicyError::ReadDir {
+                dir: dir.to_path_buf(),
+                source,
+            })?
+    {
+        let path = entry.path();
+        let file_type = entry
+            .file_type()
+            .await
+            .map_err(|source| ExecPolicyError::ReadDir {
+                dir: dir.to_path_buf(),
+                source,
+            })?;
+
+        if path
+            .extension()
+            .and_then(|ext| ext.to_str())
+            .is_some_and(|ext| ext == POLICY_EXTENSION)
+            && file_type.is_file()
+        {
+            policy_paths.push(path);
+        }
+    }
+
+    policy_paths.sort();
+
+    Ok(policy_paths)
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::features::Feature;
+    use crate::features::Features;
+    use codex_protocol::protocol::AskForApproval;
+    use codex_protocol::protocol::SandboxPolicy;
+    use pretty_assertions::assert_eq;
+    use std::fs;
+    use tempfile::tempdir;
+
+    #[tokio::test]
+    async fn returns_empty_policy_when_feature_disabled() {
+        let mut features = Features::with_defaults();
+        features.disable(Feature::ExecPolicy);
+        let temp_dir = tempdir().expect("create temp dir");
+
+        let policy = exec_policy_for(&features, temp_dir.path())
+            .await
+            .expect("policy result");
+
+        let commands = [vec!["rm".to_string()]];
+        assert!(matches!(
+            policy.check_multiple(commands.iter()),
+            Evaluation::NoMatch { .. }
+        ));
+        assert!(!temp_dir.path().join(POLICY_DIR_NAME).exists());
+    }
+
+    #[tokio::test]
+    async fn collect_policy_files_returns_empty_when_dir_missing() {
+        let temp_dir = tempdir().expect("create temp dir");
+
+        let policy_dir = temp_dir.path().join(POLICY_DIR_NAME);
+        let files = collect_policy_files(&policy_dir)
+            .await
+            .expect("collect policy files");
+
+        assert!(files.is_empty());
+    }
+
+    #[tokio::test]
+    async fn loads_policies_from_policy_subdirectory() {
+        let temp_dir = tempdir().expect("create temp dir");
+        let policy_dir = temp_dir.path().join(POLICY_DIR_NAME);
+        fs::create_dir_all(&policy_dir).expect("create policy dir");
+        fs::write(
+            policy_dir.join("deny.codexpolicy"),
+            r#"prefix_rule(pattern=["rm"], decision="forbidden")"#,
+        )
+        .expect("write policy file");
+
+        let policy = exec_policy_for(&Features::with_defaults(), temp_dir.path())
+            .await
+            .expect("policy result");
+        let command = [vec!["rm".to_string()]];
+        assert!(matches!(
+            policy.check_multiple(command.iter()),
+            Evaluation::Match { .. }
+        ));
+    }
+
+    #[tokio::test]
+    async fn ignores_policies_outside_policy_dir() {
+        let temp_dir = tempdir().expect("create temp dir");
+        fs::write(
+            temp_dir.path().join("root.codexpolicy"),
+            r#"prefix_rule(pattern=["ls"], decision="prompt")"#,
+        )
+        .expect("write policy file");
+
+        let policy = exec_policy_for(&Features::with_defaults(), temp_dir.path())
+            .await
+            .expect("policy result");
+        let command = [vec!["ls".to_string()]];
+        assert!(matches!(
+            policy.check_multiple(command.iter()),
+            Evaluation::NoMatch { .. }
+        ));
+    }
+
+    #[test]
+    fn evaluates_bash_lc_inner_commands() {
+        let policy_src = r#"
+prefix_rule(pattern=["rm"], decision="forbidden")
+"#;
+        let mut parser = PolicyParser::new();
+        parser
+            .parse("test.codexpolicy", policy_src)
+            .expect("parse policy");
+        let policy = parser.build();
+
+        let forbidden_script = vec![
+            "bash".to_string(),
+            "-lc".to_string(),
+            "rm -rf /tmp".to_string(),
+        ];
+
+        let requirement =
+            evaluate_with_policy(&policy, &forbidden_script, AskForApproval::OnRequest)
+                .expect("expected match for forbidden command");
+
+        assert_eq!(
+            requirement,
+            ApprovalRequirement::Forbidden {
+                reason: FORBIDDEN_REASON.to_string()
+            }
+        );
+    }
+
+    #[test]
+    fn approval_requirement_prefers_execpolicy_match() {
+        let policy_src = r#"prefix_rule(pattern=["rm"], decision="prompt")"#;
+        let mut parser = PolicyParser::new();
+        parser
+            .parse("test.codexpolicy", policy_src)
+            .expect("parse policy");
+        let policy = parser.build();
+        let command = vec!["rm".to_string()];
+
+        let requirement = create_approval_requirement_for_command(
+            &policy,
+            &command,
+            AskForApproval::OnRequest,
+            &SandboxPolicy::DangerFullAccess,
+            SandboxPermissions::UseDefault,
+        );
+
+        assert_eq!(
+            requirement,
+            ApprovalRequirement::NeedsApproval {
+                reason: Some(PROMPT_REASON.to_string())
+            }
+        );
+    }
+
+    #[test]
+    fn approval_requirement_respects_approval_policy() {
+        let policy_src = r#"prefix_rule(pattern=["rm"], decision="prompt")"#;
+        let mut parser = PolicyParser::new();
+        parser
+            .parse("test.codexpolicy", policy_src)
+            .expect("parse policy");
+        let policy = parser.build();
+        let command = vec!["rm".to_string()];
+
+        let requirement = create_approval_requirement_for_command(
+            &policy,
+            &command,
+            AskForApproval::Never,
+            &SandboxPolicy::DangerFullAccess,
+            SandboxPermissions::UseDefault,
+        );
+
+        assert_eq!(
+            requirement,
+            ApprovalRequirement::Forbidden {
+                reason: PROMPT_REASON.to_string()
+            }
+        );
+    }
+
+    #[test]
+    fn approval_requirement_falls_back_to_heuristics() {
+        let command = vec!["python".to_string()];
+
+        let empty_policy = Policy::empty();
+        let requirement = create_approval_requirement_for_command(
+            &empty_policy,
+            &command,
+            AskForApproval::UnlessTrusted,
+            &SandboxPolicy::ReadOnly,
+            SandboxPermissions::UseDefault,
+        );
+
+        assert_eq!(
+            requirement,
+            ApprovalRequirement::NeedsApproval { reason: None }
+        );
+    }
+}

codex-rs/core/src/features.rs

@@ -42,6 +42,8 @@ pub enum Feature {
     ViewImageTool,
     /// Allow the model to request web searches.
     WebSearchRequest,
+    /// Gate the execpolicy enforcement for shell/unified exec.
+    ExecPolicy,
     /// Enable the model-based risk assessments for sandboxed commands.
     SandboxCommandAssessment,
     /// Enable Windows sandbox (restricted token) on Windows.
@@ -260,12 +262,6 @@ pub const FEATURES: &[FeatureSpec] = &[
         stage: Stage::Stable,
         default_enabled: true,
     },
-    FeatureSpec {
-        id: Feature::ViewImageTool,
-        key: "view_image_tool",
-        stage: Stage::Stable,
-        default_enabled: true,
-    },
     // Unstable features.
     FeatureSpec {
         id: Feature::UnifiedExec,
@@ -291,12 +287,24 @@ pub const FEATURES: &[FeatureSpec] = &[
         stage: Stage::Beta,
         default_enabled: false,
     },
+    FeatureSpec {
+        id: Feature::ViewImageTool,
+        key: "view_image_tool",
+        stage: Stage::Stable,
+        default_enabled: true,
+    },
     FeatureSpec {
         id: Feature::WebSearchRequest,
         key: "web_search_request",
         stage: Stage::Stable,
         default_enabled: false,
     },
+    FeatureSpec {
+        id: Feature::ExecPolicy,
+        key: "exec_policy",
+        stage: Stage::Experimental,
+        default_enabled: true,
+    },
     FeatureSpec {
         id: Feature::SandboxCommandAssessment,
         key: "experimental_sandbox_command_assessment",
@@ -313,7 +321,7 @@ pub const FEATURES: &[FeatureSpec] = &[
         id: Feature::RemoteCompaction,
         key: "remote_compaction",
         stage: Stage::Experimental,
-        default_enabled: true,
+        default_enabled: false,
     },
     FeatureSpec {
         id: Feature::ParallelToolCalls,

codex-rs/core/src/lib.rs

@@ -25,6 +25,7 @@ mod environment_context;
 pub mod error;
 pub mod exec;
 pub mod exec_env;
+mod exec_policy;
 pub mod features;
 mod flags;
 pub mod git_info;

codex-rs/core/src/model_family.rs

@@ -12,7 +12,6 @@ const BASE_INSTRUCTIONS: &str = include_str!("../prompt.md");
 
 const GPT_5_CODEX_INSTRUCTIONS: &str = include_str!("../gpt_5_codex_prompt.md");
 const GPT_5_1_INSTRUCTIONS: &str = include_str!("../gpt_5_1_prompt.md");
-const ARCTICFOX_INSTRUCTIONS: &str = include_str!("../arcticfox_prompt.md");
 
 /// A model family is a group of models that share certain characteristics.
 #[derive(Debug, Clone, PartialEq, Eq, Hash)]
@@ -150,7 +149,6 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
                 "test_sync_tool".to_string(),
             ],
             supports_parallel_tool_calls: true,
-            shell_type: ConfigShellToolType::ShellCommand,
             support_verbosity: true,
             truncation_policy: TruncationPolicy::Tokens(10_000),
         )
@@ -168,23 +166,11 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
                 "list_dir".to_string(),
                 "read_file".to_string(),
             ],
-            shell_type: ConfigShellToolType::ShellCommand,
+            shell_type: if cfg!(windows) { ConfigShellToolType::ShellCommand } else { ConfigShellToolType::Default },
             supports_parallel_tool_calls: true,
             support_verbosity: true,
             truncation_policy: TruncationPolicy::Tokens(10_000),
         )
-    } else if slug.starts_with("arcticfox") {
-        model_family!(
-            slug, slug,
-            supports_reasoning_summaries: true,
-            reasoning_summary_format: ReasoningSummaryFormat::Experimental,
-            base_instructions: ARCTICFOX_INSTRUCTIONS.to_string(),
-            apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
-            shell_type: ConfigShellToolType::ShellCommand,
-            supports_parallel_tool_calls: true,
-            support_verbosity: false,
-            truncation_policy: TruncationPolicy::Tokens(10_000),
-        )
 
     // Production models.
     } else if slug.starts_with("gpt-5-codex")
@@ -197,19 +183,7 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
             reasoning_summary_format: ReasoningSummaryFormat::Experimental,
             base_instructions: GPT_5_CODEX_INSTRUCTIONS.to_string(),
             apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
-            shell_type: ConfigShellToolType::ShellCommand,
-            supports_parallel_tool_calls: true,
-            support_verbosity: false,
-            truncation_policy: TruncationPolicy::Tokens(10_000),
-        )
-    } else if slug.starts_with("arcticfox") {
-        model_family!(
-            slug, slug,
-            supports_reasoning_summaries: true,
-            reasoning_summary_format: ReasoningSummaryFormat::Experimental,
-            base_instructions: BASE_INSTRUCTIONS.to_string(),
-            apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
-            shell_type: ConfigShellToolType::ShellCommand,
+            shell_type: if cfg!(windows) { ConfigShellToolType::ShellCommand } else { ConfigShellToolType::Default },
             supports_parallel_tool_calls: true,
             support_verbosity: false,
             truncation_policy: TruncationPolicy::Tokens(10_000),
@@ -224,7 +198,6 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
             base_instructions: GPT_5_1_INSTRUCTIONS.to_string(),
             default_reasoning_effort: Some(ReasoningEffort::Medium),
             truncation_policy: TruncationPolicy::Bytes(10_000),
-            shell_type: ConfigShellToolType::ShellCommand,
             supports_parallel_tool_calls: true,
         )
     } else if slug.starts_with("gpt-5") {
@@ -232,7 +205,6 @@ pub fn find_family_for_model(slug: &str) -> Option<ModelFamily> {
             slug, "gpt-5",
             supports_reasoning_summaries: true,
             needs_special_apply_patch_instructions: true,
-            shell_type: ConfigShellToolType::Default,
             support_verbosity: true,
             truncation_policy: TruncationPolicy::Bytes(10_000),
         )

codex-rs/core/src/openai_model_info.rs

@@ -70,10 +70,7 @@ pub(crate) fn get_model_info(model_family: &ModelFamily) -> Option<ModelInfo> {
         // https://platform.openai.com/docs/models/gpt-3.5-turbo
         "gpt-3.5-turbo" => Some(ModelInfo::new(16_385, 4_096)),
 
-        _ if slug.starts_with("gpt-5-codex")
-            || slug.starts_with("gpt-5.1-codex")
-            || slug.starts_with("arcticfox") =>
-        {
+        _ if slug.starts_with("gpt-5-codex") || slug.starts_with("gpt-5.1-codex") => {
             Some(ModelInfo::new(CONTEXT_WINDOW_272K, MAX_OUTPUT_TOKENS_128K))
         }
 

codex-rs/core/src/otel_init.rs

@@ -5,7 +5,6 @@ use crate::default_client::originator;
 use codex_otel::config::OtelExporter;
 use codex_otel::config::OtelHttpProtocol;
 use codex_otel::config::OtelSettings;
-use codex_otel::config::OtelTlsConfig as OtelTlsSettings;
 use codex_otel::otel_provider::OtelProvider;
 use std::error::Error;
 
@@ -22,7 +21,6 @@ pub fn build_provider(
             endpoint,
             headers,
             protocol,
-            tls,
         } => {
             let protocol = match protocol {
                 Protocol::Json => OtelHttpProtocol::Json,
@@ -36,28 +34,14 @@ pub fn build_provider(
                     .map(|(k, v)| (k.clone(), v.clone()))
                     .collect(),
                 protocol,
-                tls: tls.as_ref().map(|config| OtelTlsSettings {
-                    ca_certificate: config.ca_certificate.clone(),
-                    client_certificate: config.client_certificate.clone(),
-                    client_private_key: config.client_private_key.clone(),
-                }),
             }
         }
-        Kind::OtlpGrpc {
-            endpoint,
-            headers,
-            tls,
-        } => OtelExporter::OtlpGrpc {
+        Kind::OtlpGrpc { endpoint, headers } => OtelExporter::OtlpGrpc {
             endpoint: endpoint.clone(),
             headers: headers
                 .iter()
                 .map(|(k, v)| (k.clone(), v.clone()))
                 .collect(),
-            tls: tls.as_ref().map(|config| OtelTlsSettings {
-                ca_certificate: config.ca_certificate.clone(),
-                client_certificate: config.client_certificate.clone(),
-                client_private_key: config.client_private_key.clone(),
-            }),
         },
     };
 

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

@@ -26,6 +26,28 @@ use std::collections::HashMap;
 use std::path::Path;
 use std::path::PathBuf;
 
+#[derive(Clone, Copy, Debug, Eq, PartialEq)]
+pub enum SandboxPermissions {
+    UseDefault,
+    RequireEscalated,
+}
+
+impl SandboxPermissions {
+    pub fn requires_escalated_permissions(self) -> bool {
+        matches!(self, SandboxPermissions::RequireEscalated)
+    }
+}
+
+impl From<bool> for SandboxPermissions {
+    fn from(with_escalated_permissions: bool) -> Self {
+        if with_escalated_permissions {
+            SandboxPermissions::RequireEscalated
+        } else {
+            SandboxPermissions::UseDefault
+        }
+    }
+}
+
 #[derive(Clone, Debug)]
 pub struct CommandSpec {
     pub program: String,

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

@@ -3,8 +3,10 @@ use std::sync::Arc;
 use super::SessionTask;
 use super::SessionTaskContext;
 use crate::codex::TurnContext;
+use crate::features::Feature;
 use crate::state::TaskKind;
 use async_trait::async_trait;
+use codex_app_server_protocol::AuthMode;
 use codex_protocol::user_input::UserInput;
 use tokio_util::sync::CancellationToken;
 
@@ -25,12 +27,16 @@ impl SessionTask for CompactTask {
         _cancellation_token: CancellationToken,
     ) -> Option<String> {
         let session = session.clone_session();
-        if crate::compact::should_use_remote_compact_task(&session).await {
-            crate::compact_remote::run_remote_compact_task(session, ctx).await
+        if session
+            .services
+            .auth_manager
+            .auth()
+            .is_some_and(|auth| auth.mode == AuthMode::ChatGPT)
+            && session.enabled(Feature::RemoteCompaction).await
+        {
+            crate::compact_remote::run_remote_compact_task(session, ctx, input).await
         } else {
             crate::compact::run_compact_task(session, ctx, input).await
         }
-
-        None
     }
 }

codex-rs/core/src/tasks/ghost_snapshot.rs

@@ -1,14 +1,10 @@
 use crate::codex::TurnContext;
-use crate::protocol::EventMsg;
-use crate::protocol::WarningEvent;
 use crate::state::TaskKind;
 use crate::tasks::SessionTask;
 use crate::tasks::SessionTaskContext;
 use async_trait::async_trait;
 use codex_git::CreateGhostCommitOptions;
-use codex_git::GhostSnapshotReport;
 use codex_git::GitToolingError;
-use codex_git::capture_ghost_snapshot_report;
 use codex_git::create_ghost_commit;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::user_input::UserInput;
@@ -43,27 +39,6 @@ impl SessionTask for GhostSnapshotTask {
                 _ = cancellation_token.cancelled() => true,
                 _ = async {
                     let repo_path = ctx_for_task.cwd.clone();
-                    // First, compute a snapshot report so we can warn about
-                    // large untracked directories before running the heavier
-                    // snapshot logic.
-                    if let Ok(Ok(report)) = tokio::task::spawn_blocking({
-                        let repo_path = repo_path.clone();
-                        move || {
-                            let options = CreateGhostCommitOptions::new(&repo_path);
-                            capture_ghost_snapshot_report(&options)
-                        }
-                    })
-                    .await
-                        && let Some(message) = format_large_untracked_warning(&report) {
-                                session
-                                    .session
-                                    .send_event(
-                                        &ctx_for_task,
-                                        EventMsg::Warning(WarningEvent { message }),
-                                    )
-                                    .await;
-                            }
-
                     // Required to run in a dedicated blocking pool.
                     match tokio::task::spawn_blocking(move || {
                         let options = CreateGhostCommitOptions::new(&repo_path);
@@ -81,18 +56,23 @@ impl SessionTask for GhostSnapshotTask {
                                 .await;
                             info!("ghost commit captured: {}", ghost_commit.id());
                         }
-                        Ok(Err(err)) => match err {
-                            GitToolingError::NotAGitRepository { .. } => info!(
+                        Ok(Err(err)) => {
+                            warn!(
                                 sub_id = ctx_for_task.sub_id.as_str(),
-                                "skipping ghost snapshot because current directory is not a Git repository"
-                            ),
-                            _ => {
-                                warn!(
-                                    sub_id = ctx_for_task.sub_id.as_str(),
-                                    "failed to capture ghost snapshot: {err}"
-                                );
-                            }
-                        },
+                                "failed to capture ghost snapshot: {err}"
+                            );
+                            let message = match err {
+                                GitToolingError::NotAGitRepository { .. } => {
+                                    "Snapshots disabled: current directory is not a Git repository."
+                                        .to_string()
+                                }
+                                _ => format!("Snapshots disabled after ghost snapshot error: {err}."),
+                            };
+                            session
+                                .session
+                                .notify_background_event(&ctx_for_task, message)
+                                .await;
+                        }
                         Err(err) => {
                             warn!(
                                 sub_id = ctx_for_task.sub_id.as_str(),
@@ -128,22 +108,3 @@ impl GhostSnapshotTask {
         Self { token }
     }
 }
-
-fn format_large_untracked_warning(report: &GhostSnapshotReport) -> Option<String> {
-    if report.large_untracked_dirs.is_empty() {
-        return None;
-    }
-    const MAX_DIRS: usize = 3;
-    let mut parts: Vec<String> = Vec::new();
-    for dir in report.large_untracked_dirs.iter().take(MAX_DIRS) {
-        parts.push(format!("{} ({} files)", dir.path.display(), dir.file_count));
-    }
-    if report.large_untracked_dirs.len() > MAX_DIRS {
-        let remaining = report.large_untracked_dirs.len() - MAX_DIRS;
-        parts.push(format!("{remaining} more"));
-    }
-    Some(format!(
-        "Repository snapshot encountered large untracked directories: {}. This can slow Codex; consider adding these paths to .gitignore or disabling undo in your config.",
-        parts.join(", ")
-    ))
-}

codex-rs/core/src/tasks/review.rs

@@ -23,18 +23,8 @@ use codex_protocol::user_input::UserInput;
 use super::SessionTask;
 use super::SessionTaskContext;
 
-#[derive(Clone, Copy)]
-pub(crate) struct ReviewTask {
-    append_to_original_thread: bool,
-}
-
-impl ReviewTask {
-    pub(crate) fn new(append_to_original_thread: bool) -> Self {
-        Self {
-            append_to_original_thread,
-        }
-    }
-}
+#[derive(Clone, Copy, Default)]
+pub(crate) struct ReviewTask;
 
 #[async_trait]
 impl SessionTask for ReviewTask {
@@ -62,25 +52,13 @@ impl SessionTask for ReviewTask {
             None => None,
         };
         if !cancellation_token.is_cancelled() {
-            exit_review_mode(
-                session.clone_session(),
-                output.clone(),
-                ctx.clone(),
-                self.append_to_original_thread,
-            )
-            .await;
+            exit_review_mode(session.clone_session(), output.clone(), ctx.clone()).await;
         }
         None
     }
 
     async fn abort(&self, session: Arc<SessionTaskContext>, ctx: Arc<TurnContext>) {
-        exit_review_mode(
-            session.clone_session(),
-            None,
-            ctx,
-            self.append_to_original_thread,
-        )
-        .await;
+        exit_review_mode(session.clone_session(), None, ctx).await;
     }
 }
 
@@ -197,35 +175,32 @@ pub(crate) async fn exit_review_mode(
     session: Arc<Session>,
     review_output: Option<ReviewOutputEvent>,
     ctx: Arc<TurnContext>,
-    append_to_original_thread: bool,
 ) {
-    if append_to_original_thread {
-        let user_message = if let Some(out) = review_output.clone() {
-            let mut findings_str = String::new();
-            let text = out.overall_explanation.trim();
-            if !text.is_empty() {
-                findings_str.push_str(text);
-            }
-            if !out.findings.is_empty() {
-                let block = format_review_findings_block(&out.findings, None);
-                findings_str.push_str(&format!("\n{block}"));
-            }
-            crate::client_common::REVIEW_EXIT_SUCCESS_TMPL.replace("{results}", &findings_str)
-        } else {
-            crate::client_common::REVIEW_EXIT_INTERRUPTED_TMPL.to_string()
-        };
+    let user_message = if let Some(out) = review_output.clone() {
+        let mut findings_str = String::new();
+        let text = out.overall_explanation.trim();
+        if !text.is_empty() {
+            findings_str.push_str(text);
+        }
+        if !out.findings.is_empty() {
+            let block = format_review_findings_block(&out.findings, None);
+            findings_str.push_str(&format!("\n{block}"));
+        }
+        crate::client_common::REVIEW_EXIT_SUCCESS_TMPL.replace("{results}", &findings_str)
+    } else {
+        crate::client_common::REVIEW_EXIT_INTERRUPTED_TMPL.to_string()
+    };
 
-        session
-            .record_conversation_items(
-                &ctx,
-                &[ResponseItem::Message {
-                    id: None,
-                    role: "user".to_string(),
-                    content: vec![ContentItem::InputText { text: user_message }],
-                }],
-            )
-            .await;
-    }
+    session
+        .record_conversation_items(
+            &ctx,
+            &[ResponseItem::Message {
+                id: None,
+                role: "user".to_string(),
+                content: vec![ContentItem::InputText { text: user_message }],
+            }],
+        )
+        .await;
     session
         .send_event(
             ctx.as_ref(),

codex-rs/core/src/tasks/user_shell.rs

@@ -122,11 +122,7 @@ impl SessionTask for UserShellCommandTask {
                     duration: Duration::ZERO,
                     timed_out: false,
                 };
-                let output_items = [user_shell_command_record_item(
-                    &raw_command,
-                    &exec_output,
-                    &turn_context,
-                )];
+                let output_items = [user_shell_command_record_item(&raw_command, &exec_output)];
                 session
                     .record_conversation_items(turn_context.as_ref(), &output_items)
                     .await;
@@ -168,19 +164,12 @@ impl SessionTask for UserShellCommandTask {
                             aggregated_output: output.aggregated_output.text.clone(),
                             exit_code: output.exit_code,
                             duration: output.duration,
-                            formatted_output: format_exec_output_str(
-                                &output,
-                                turn_context.truncation_policy,
-                            ),
+                            formatted_output: format_exec_output_str(&output),
                         }),
                     )
                     .await;
 
-                let output_items = [user_shell_command_record_item(
-                    &raw_command,
-                    &output,
-                    &turn_context,
-                )];
+                let output_items = [user_shell_command_record_item(&raw_command, &output)];
                 session
                     .record_conversation_items(turn_context.as_ref(), &output_items)
                     .await;
@@ -212,18 +201,11 @@ impl SessionTask for UserShellCommandTask {
                             aggregated_output: exec_output.aggregated_output.text.clone(),
                             exit_code: exec_output.exit_code,
                             duration: exec_output.duration,
-                            formatted_output: format_exec_output_str(
-                                &exec_output,
-                                turn_context.truncation_policy,
-                            ),
+                            formatted_output: format_exec_output_str(&exec_output),
                         }),
                     )
                     .await;
-                let output_items = [user_shell_command_record_item(
-                    &raw_command,
-                    &exec_output,
-                    &turn_context,
-                )];
+                let output_items = [user_shell_command_record_item(&raw_command, &exec_output)];
                 session
                     .record_conversation_items(turn_context.as_ref(), &output_items)
                     .await;

codex-rs/core/src/tools/events.rs

@@ -88,7 +88,6 @@ pub(crate) enum ToolEmitter {
         cwd: PathBuf,
         source: ExecCommandSource,
         parsed_cmd: Vec<ParsedCommand>,
-        freeform: bool,
     },
     ApplyPatch {
         changes: HashMap<PathBuf, FileChange>,
@@ -104,19 +103,13 @@ pub(crate) enum ToolEmitter {
 }
 
 impl ToolEmitter {
-    pub fn shell(
-        command: Vec<String>,
-        cwd: PathBuf,
-        source: ExecCommandSource,
-        freeform: bool,
-    ) -> Self {
+    pub fn shell(command: Vec<String>, cwd: PathBuf, source: ExecCommandSource) -> Self {
         let parsed_cmd = parse_command(&command);
         Self::Shell {
             command,
             cwd,
             source,
             parsed_cmd,
-            freeform,
         }
     }
 
@@ -151,7 +144,6 @@ impl ToolEmitter {
                     cwd,
                     source,
                     parsed_cmd,
-                    ..
                 },
                 stage,
             ) => {
@@ -242,19 +234,6 @@ impl ToolEmitter {
         self.emit(ctx, ToolEventStage::Begin).await;
     }
 
-    fn format_exec_output_for_model(
-        &self,
-        output: &ExecToolCallOutput,
-        ctx: ToolEventCtx<'_>,
-    ) -> String {
-        match self {
-            Self::Shell { freeform: true, .. } => {
-                super::format_exec_output_for_model_freeform(output, ctx.turn.truncation_policy)
-            }
-            _ => super::format_exec_output_for_model_structured(output, ctx.turn.truncation_policy),
-        }
-    }
-
     pub async fn finish(
         &self,
         ctx: ToolEventCtx<'_>,
@@ -262,7 +241,7 @@ impl ToolEmitter {
     ) -> Result<String, FunctionCallError> {
         let (event, result) = match out {
             Ok(output) => {
-                let content = self.format_exec_output_for_model(&output, ctx);
+                let content = super::format_exec_output_for_model(&output);
                 let exit_code = output.exit_code;
                 let event = ToolEventStage::Success(output);
                 let result = if exit_code == 0 {
@@ -274,7 +253,7 @@ impl ToolEmitter {
             }
             Err(ToolError::Codex(CodexErr::Sandbox(SandboxErr::Timeout { output })))
             | Err(ToolError::Codex(CodexErr::Sandbox(SandboxErr::Denied { output }))) => {
-                let response = self.format_exec_output_for_model(&output, ctx);
+                let response = super::format_exec_output_for_model(&output);
                 let event = ToolEventStage::Failure(ToolEventFailure::Output(*output));
                 let result = Err(FunctionCallError::RespondToModel(response));
                 (event, result)
@@ -363,7 +342,7 @@ async fn emit_exec_stage(
                 aggregated_output: output.aggregated_output.text.clone(),
                 exit_code: output.exit_code,
                 duration: output.duration,
-                formatted_output: format_exec_output_str(&output, ctx.turn.truncation_policy),
+                formatted_output: format_exec_output_str(&output),
             };
             emit_exec_end(ctx, exec_input, exec_result).await;
         }

codex-rs/core/src/tools/handlers/shell.rs

@@ -9,9 +9,11 @@ use crate::apply_patch::convert_apply_patch_to_protocol;
 use crate::codex::TurnContext;
 use crate::exec::ExecParams;
 use crate::exec_env::create_env;
+use crate::exec_policy::create_approval_requirement_for_command;
 use crate::function_tool::FunctionCallError;
 use crate::is_safe_command::is_known_safe_command;
 use crate::protocol::ExecCommandSource;
+use crate::sandboxing::SandboxPermissions;
 use crate::tools::context::ToolInvocation;
 use crate::tools::context::ToolOutput;
 use crate::tools::context::ToolPayload;
@@ -117,7 +119,6 @@ impl ToolHandler for ShellHandler {
                     turn,
                     tracker,
                     call_id,
-                    false,
                 )
                 .await
             }
@@ -130,7 +131,6 @@ impl ToolHandler for ShellHandler {
                     turn,
                     tracker,
                     call_id,
-                    false,
                 )
                 .await
             }
@@ -178,7 +178,6 @@ impl ToolHandler for ShellCommandHandler {
             turn,
             tracker,
             call_id,
-            true,
         )
         .await
     }
@@ -192,7 +191,6 @@ impl ShellHandler {
         turn: Arc<TurnContext>,
         tracker: crate::tools::context::SharedTurnDiffTracker,
         call_id: String,
-        freeform: bool,
     ) -> Result<ToolOutput, FunctionCallError> {
         // Approval policy guard for explicit escalation in non-OnRequest modes.
         if exec_params.with_escalated_permissions.unwrap_or(false)
@@ -286,12 +284,8 @@ impl ShellHandler {
         }
 
         let source = ExecCommandSource::Agent;
-        let emitter = ToolEmitter::shell(
-            exec_params.command.clone(),
-            exec_params.cwd.clone(),
-            source,
-            freeform,
-        );
+        let emitter =
+            ToolEmitter::shell(exec_params.command.clone(), exec_params.cwd.clone(), source);
         let event_ctx = ToolEventCtx::new(session.as_ref(), turn.as_ref(), &call_id, None);
         emitter.begin(event_ctx).await;
 
@@ -302,6 +296,13 @@ impl ShellHandler {
             env: exec_params.env.clone(),
             with_escalated_permissions: exec_params.with_escalated_permissions,
             justification: exec_params.justification.clone(),
+            approval_requirement: create_approval_requirement_for_command(
+                &turn.exec_policy,
+                &exec_params.command,
+                turn.approval_policy,
+                &turn.sandbox_policy,
+                SandboxPermissions::from(exec_params.with_escalated_permissions.unwrap_or(false)),
+            ),
         };
         let mut orchestrator = ToolOrchestrator::new();
         let mut runtime = ShellRuntime::new();

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

@@ -9,10 +9,8 @@ pub mod runtimes;
 pub mod sandboxing;
 pub mod spec;
 
+use crate::context_manager::truncate_with_line_bytes_budget;
 use crate::exec::ExecToolCallOutput;
-use crate::truncate::TruncationPolicy;
-use crate::truncate::formatted_truncate_text;
-use crate::truncate::truncate_text;
 pub use router::ToolRouter;
 use serde::Serialize;
 
@@ -22,12 +20,12 @@ pub(crate) const TELEMETRY_PREVIEW_MAX_LINES: usize = 64; // lines
 pub(crate) const TELEMETRY_PREVIEW_TRUNCATION_NOTICE: &str =
     "[... telemetry preview truncated ...]";
 
+// TODO(aibrahim): migrate shell tool to use truncate text and respect config value
+const SHELL_OUTPUT_MAX_BYTES: usize = 10_000;
+
 /// Format the combined exec output for sending back to the model.
 /// Includes exit code and duration metadata; truncates large bodies safely.
-pub fn format_exec_output_for_model_structured(
-    exec_output: &ExecToolCallOutput,
-    truncation_policy: TruncationPolicy,
-) -> String {
+pub fn format_exec_output_for_model(exec_output: &ExecToolCallOutput) -> String {
     let ExecToolCallOutput {
         exit_code,
         duration,
@@ -49,7 +47,7 @@ pub fn format_exec_output_for_model_structured(
     // round to 1 decimal place
     let duration_seconds = ((duration.as_secs_f32()) * 10.0).round() / 10.0;
 
-    let formatted_output = format_exec_output_str(exec_output, truncation_policy);
+    let formatted_output = format_exec_output_str(exec_output);
 
     let payload = ExecOutput {
         output: &formatted_output,
@@ -63,35 +61,7 @@ pub fn format_exec_output_for_model_structured(
     serde_json::to_string(&payload).expect("serialize ExecOutput")
 }
 
-pub fn format_exec_output_for_model_freeform(
-    exec_output: &ExecToolCallOutput,
-    truncation_policy: TruncationPolicy,
-) -> String {
-    // round to 1 decimal place
-    let duration_seconds = ((exec_output.duration.as_secs_f32()) * 10.0).round() / 10.0;
-
-    let total_lines = exec_output.aggregated_output.text.lines().count();
-
-    let formatted_output = truncate_text(&exec_output.aggregated_output.text, truncation_policy);
-
-    let mut sections = Vec::new();
-
-    sections.push(format!("Exit code: {}", exec_output.exit_code));
-    sections.push(format!("Wall time: {duration_seconds} seconds"));
-    if total_lines != formatted_output.lines().count() {
-        sections.push(format!("Total output lines: {total_lines}"));
-    }
-
-    sections.push("Output:".to_string());
-    sections.push(formatted_output);
-
-    sections.join("\n")
-}
-
-pub fn format_exec_output_str(
-    exec_output: &ExecToolCallOutput,
-    truncation_policy: TruncationPolicy,
-) -> String {
+pub fn format_exec_output_str(exec_output: &ExecToolCallOutput) -> String {
     let ExecToolCallOutput {
         aggregated_output, ..
     } = exec_output;
@@ -108,5 +78,5 @@ pub fn format_exec_output_str(
     };
 
     // Truncate for model consumption before serialization.
-    formatted_truncate_text(&body, truncation_policy)
+    truncate_with_line_bytes_budget(&body, SHELL_OUTPUT_MAX_BYTES)
 }

codex-rs/core/src/tools/orchestrator.rs

@@ -11,11 +11,13 @@ use crate::error::get_error_message_ui;
 use crate::exec::ExecToolCallOutput;
 use crate::sandboxing::SandboxManager;
 use crate::tools::sandboxing::ApprovalCtx;
+use crate::tools::sandboxing::ApprovalRequirement;
 use crate::tools::sandboxing::ProvidesSandboxRetryData;
 use crate::tools::sandboxing::SandboxAttempt;
 use crate::tools::sandboxing::ToolCtx;
 use crate::tools::sandboxing::ToolError;
 use crate::tools::sandboxing::ToolRuntime;
+use crate::tools::sandboxing::default_approval_requirement;
 use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::ReviewDecision;
 
@@ -49,40 +51,52 @@ impl ToolOrchestrator {
         let otel_cfg = codex_otel::otel_event_manager::ToolDecisionSource::Config;
 
         // 1) Approval
-        let needs_initial_approval =
-            tool.wants_initial_approval(req, approval_policy, &turn_ctx.sandbox_policy);
         let mut already_approved = false;
 
-        if needs_initial_approval {
-            let mut risk = None;
-
-            if let Some(metadata) = req.sandbox_retry_data() {
-                risk = tool_ctx
-                    .session
-                    .assess_sandbox_command(turn_ctx, &tool_ctx.call_id, &metadata.command, None)
-                    .await;
+        let requirement = tool.approval_requirement(req).unwrap_or_else(|| {
+            default_approval_requirement(approval_policy, &turn_ctx.sandbox_policy)
+        });
+        match requirement {
+            ApprovalRequirement::Skip => {
+                otel.tool_decision(otel_tn, otel_ci, ReviewDecision::Approved, otel_cfg);
             }
+            ApprovalRequirement::Forbidden { reason } => {
+                return Err(ToolError::Rejected(reason));
+            }
+            ApprovalRequirement::NeedsApproval { reason } => {
+                let mut risk = None;
 
-            let approval_ctx = ApprovalCtx {
-                session: tool_ctx.session,
-                turn: turn_ctx,
-                call_id: &tool_ctx.call_id,
-                retry_reason: None,
-                risk,
-            };
-            let decision = tool.start_approval_async(req, approval_ctx).await;
-
-            otel.tool_decision(otel_tn, otel_ci, decision, otel_user.clone());
-
-            match decision {
-                ReviewDecision::Denied | ReviewDecision::Abort => {
-                    return Err(ToolError::Rejected("rejected by user".to_string()));
+                if let Some(metadata) = req.sandbox_retry_data() {
+                    risk = tool_ctx
+                        .session
+                        .assess_sandbox_command(
+                            turn_ctx,
+                            &tool_ctx.call_id,
+                            &metadata.command,
+                            None,
+                        )
+                        .await;
                 }
-                ReviewDecision::Approved | ReviewDecision::ApprovedForSession => {}
+
+                let approval_ctx = ApprovalCtx {
+                    session: tool_ctx.session,
+                    turn: turn_ctx,
+                    call_id: &tool_ctx.call_id,
+                    retry_reason: reason,
+                    risk,
+                };
+                let decision = tool.start_approval_async(req, approval_ctx).await;
+
+                otel.tool_decision(otel_tn, otel_ci, decision, otel_user.clone());
+
+                match decision {
+                    ReviewDecision::Denied | ReviewDecision::Abort => {
+                        return Err(ToolError::Rejected("rejected by user".to_string()));
+                    }
+                    ReviewDecision::Approved | ReviewDecision::ApprovedForSession => {}
+                }
+                already_approved = true;
             }
-            already_approved = true;
-        } else {
-            otel.tool_decision(otel_tn, otel_ci, ReviewDecision::Approved, otel_cfg);
         }
 
         // 2) First attempt under the selected sandbox.

codex-rs/core/src/tools/runtimes/shell.rs

@@ -4,13 +4,12 @@ Runtime: shell
 Executes shell requests under the orchestrator: asks for approval when needed,
 builds a CommandSpec, and runs it under the current SandboxAttempt.
 */
-use crate::command_safety::is_dangerous_command::requires_initial_appoval;
 use crate::exec::ExecToolCallOutput;
-use crate::protocol::SandboxPolicy;
 use crate::sandboxing::execute_env;
 use crate::tools::runtimes::build_command_spec;
 use crate::tools::sandboxing::Approvable;
 use crate::tools::sandboxing::ApprovalCtx;
+use crate::tools::sandboxing::ApprovalRequirement;
 use crate::tools::sandboxing::ProvidesSandboxRetryData;
 use crate::tools::sandboxing::SandboxAttempt;
 use crate::tools::sandboxing::SandboxRetryData;
@@ -20,7 +19,6 @@ use crate::tools::sandboxing::ToolCtx;
 use crate::tools::sandboxing::ToolError;
 use crate::tools::sandboxing::ToolRuntime;
 use crate::tools::sandboxing::with_cached_approval;
-use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::ReviewDecision;
 use futures::future::BoxFuture;
 use std::path::PathBuf;
@@ -33,6 +31,7 @@ pub struct ShellRequest {
     pub env: std::collections::HashMap<String, String>,
     pub with_escalated_permissions: Option<bool>,
     pub justification: Option<String>,
+    pub approval_requirement: ApprovalRequirement,
 }
 
 impl ProvidesSandboxRetryData for ShellRequest {
@@ -114,18 +113,8 @@ impl Approvable<ShellRequest> for ShellRuntime {
         })
     }
 
-    fn wants_initial_approval(
-        &self,
-        req: &ShellRequest,
-        policy: AskForApproval,
-        sandbox_policy: &SandboxPolicy,
-    ) -> bool {
-        requires_initial_appoval(
-            policy,
-            sandbox_policy,
-            &req.command,
-            req.with_escalated_permissions.unwrap_or(false),
-        )
+    fn approval_requirement(&self, req: &ShellRequest) -> Option<ApprovalRequirement> {
+        Some(req.approval_requirement.clone())
     }
 
     fn wants_escalated_first_attempt(&self, req: &ShellRequest) -> bool {

codex-rs/core/src/tools/runtimes/unified_exec.rs

@@ -1,4 +1,3 @@
-use crate::command_safety::is_dangerous_command::requires_initial_appoval;
 /*
 Runtime: unified exec
 
@@ -10,6 +9,7 @@ use crate::error::SandboxErr;
 use crate::tools::runtimes::build_command_spec;
 use crate::tools::sandboxing::Approvable;
 use crate::tools::sandboxing::ApprovalCtx;
+use crate::tools::sandboxing::ApprovalRequirement;
 use crate::tools::sandboxing::ProvidesSandboxRetryData;
 use crate::tools::sandboxing::SandboxAttempt;
 use crate::tools::sandboxing::SandboxRetryData;
@@ -22,9 +22,7 @@ use crate::tools::sandboxing::with_cached_approval;
 use crate::unified_exec::UnifiedExecError;
 use crate::unified_exec::UnifiedExecSession;
 use crate::unified_exec::UnifiedExecSessionManager;
-use codex_protocol::protocol::AskForApproval;
 use codex_protocol::protocol::ReviewDecision;
-use codex_protocol::protocol::SandboxPolicy;
 use futures::future::BoxFuture;
 use std::collections::HashMap;
 use std::path::PathBuf;
@@ -36,6 +34,7 @@ pub struct UnifiedExecRequest {
     pub env: HashMap<String, String>,
     pub with_escalated_permissions: Option<bool>,
     pub justification: Option<String>,
+    pub approval_requirement: ApprovalRequirement,
 }
 
 impl ProvidesSandboxRetryData for UnifiedExecRequest {
@@ -65,6 +64,7 @@ impl UnifiedExecRequest {
         env: HashMap<String, String>,
         with_escalated_permissions: Option<bool>,
         justification: Option<String>,
+        approval_requirement: ApprovalRequirement,
     ) -> Self {
         Self {
             command,
@@ -72,6 +72,7 @@ impl UnifiedExecRequest {
             env,
             with_escalated_permissions,
             justification,
+            approval_requirement,
         }
     }
 }
@@ -129,18 +130,8 @@ impl Approvable<UnifiedExecRequest> for UnifiedExecRuntime<'_> {
         })
     }
 
-    fn wants_initial_approval(
-        &self,
-        req: &UnifiedExecRequest,
-        policy: AskForApproval,
-        sandbox_policy: &SandboxPolicy,
-    ) -> bool {
-        requires_initial_appoval(
-            policy,
-            sandbox_policy,
-            &req.command,
-            req.with_escalated_permissions.unwrap_or(false),
-        )
+    fn approval_requirement(&self, req: &UnifiedExecRequest) -> Option<ApprovalRequirement> {
+        Some(req.approval_requirement.clone())
     }
 
     fn wants_escalated_first_attempt(&self, req: &UnifiedExecRequest) -> bool {

codex-rs/core/src/tools/sandboxing.rs

@@ -86,6 +86,37 @@ pub(crate) struct ApprovalCtx<'a> {
     pub risk: Option<SandboxCommandAssessment>,
 }
 
+// Specifies what tool orchestrator should do with a given tool call.
+#[derive(Clone, Debug, PartialEq, Eq)]
+pub(crate) enum ApprovalRequirement {
+    /// No approval required for this tool call
+    Skip,
+    /// Approval required for this tool call
+    NeedsApproval { reason: Option<String> },
+    /// Execution forbidden for this tool call
+    Forbidden { reason: String },
+}
+
+/// - Never, OnFailure: do not ask
+/// - OnRequest: ask unless sandbox policy is DangerFullAccess
+/// - UnlessTrusted: always ask
+pub(crate) fn default_approval_requirement(
+    policy: AskForApproval,
+    sandbox_policy: &SandboxPolicy,
+) -> ApprovalRequirement {
+    let needs_approval = match policy {
+        AskForApproval::Never | AskForApproval::OnFailure => false,
+        AskForApproval::OnRequest => !matches!(sandbox_policy, SandboxPolicy::DangerFullAccess),
+        AskForApproval::UnlessTrusted => true,
+    };
+
+    if needs_approval {
+        ApprovalRequirement::NeedsApproval { reason: None }
+    } else {
+        ApprovalRequirement::Skip
+    }
+}
+
 pub(crate) trait Approvable<Req> {
     type ApprovalKey: Hash + Eq + Clone + Debug + Serialize;
 
@@ -106,22 +137,11 @@ pub(crate) trait Approvable<Req> {
         matches!(policy, AskForApproval::Never)
     }
 
-    /// Decide whether an initial user approval should be requested before the
-    /// first attempt. Defaults to the orchestrator's behavior (pre‑refactor):
-    /// - Never, OnFailure: do not ask
-    /// - OnRequest: ask unless sandbox policy is DangerFullAccess
-    /// - UnlessTrusted: always ask
-    fn wants_initial_approval(
-        &self,
-        _req: &Req,
-        policy: AskForApproval,
-        sandbox_policy: &SandboxPolicy,
-    ) -> bool {
-        match policy {
-            AskForApproval::Never | AskForApproval::OnFailure => false,
-            AskForApproval::OnRequest => !matches!(sandbox_policy, SandboxPolicy::DangerFullAccess),
-            AskForApproval::UnlessTrusted => true,
-        }
+    /// Override the default approval requirement. Return `Some(_)` to specify
+    /// a custom requirement, or `None` to fall back to
+    /// policy-based default.
+    fn approval_requirement(&self, _req: &Req) -> Option<ApprovalRequirement> {
+        None
     }
 
     /// Decide we can request an approval for no-sandbox execution.

codex-rs/core/src/tools/spec.rs

@@ -1292,7 +1292,11 @@ mod tests {
             "gpt-5-codex",
             &Features::with_defaults(),
             &[
-                "shell_command",
+                if cfg!(windows) {
+                    "shell_command"
+                } else {
+                    "shell"
+                },
                 "list_mcp_resources",
                 "list_mcp_resource_templates",
                 "read_mcp_resource",
@@ -1309,7 +1313,11 @@ mod tests {
             "gpt-5.1-codex",
             &Features::with_defaults(),
             &[
-                "shell_command",
+                if cfg!(windows) {
+                    "shell_command"
+                } else {
+                    "shell"
+                },
                 "list_mcp_resources",
                 "list_mcp_resource_templates",
                 "read_mcp_resource",
@@ -1384,7 +1392,11 @@ mod tests {
             "gpt-5.1-codex-mini",
             &Features::with_defaults(),
             &[
-                "shell_command",
+                if cfg!(windows) {
+                    "shell_command"
+                } else {
+                    "shell"
+                },
                 "list_mcp_resources",
                 "list_mcp_resource_templates",
                 "read_mcp_resource",
@@ -1395,29 +1407,13 @@ mod tests {
         );
     }
 
-    #[test]
-    fn test_gpt_5_defaults() {
-        assert_model_tools(
-            "gpt-5",
-            &Features::with_defaults(),
-            &[
-                "shell",
-                "list_mcp_resources",
-                "list_mcp_resource_templates",
-                "read_mcp_resource",
-                "update_plan",
-                "view_image",
-            ],
-        );
-    }
-
     #[test]
     fn test_gpt_5_1_defaults() {
         assert_model_tools(
             "gpt-5.1",
             &Features::with_defaults(),
             &[
-                "shell_command",
+                "shell",
                 "list_mcp_resources",
                 "list_mcp_resource_templates",
                 "read_mcp_resource",

codex-rs/core/src/truncate.rs

@@ -2,8 +2,11 @@
 //! and suffix on UTF-8 boundaries, and helpers for line/token‑based truncation
 //! used across the core crate.
 
-use crate::config::Config;
 use codex_protocol::models::FunctionCallOutputContentItem;
+use codex_utils_string::take_bytes_at_char_boundary;
+use codex_utils_string::take_last_bytes_at_char_boundary;
+
+use crate::config::Config;
 
 const APPROX_BYTES_PER_TOKEN: usize = 4;
 
@@ -14,18 +17,6 @@ pub enum TruncationPolicy {
 }
 
 impl TruncationPolicy {
-    /// Scale the underlying budget by `multiplier`, rounding up to avoid under-budgeting.
-    pub fn mul(self, multiplier: f64) -> Self {
-        match self {
-            TruncationPolicy::Bytes(bytes) => {
-                TruncationPolicy::Bytes((bytes as f64 * multiplier).ceil() as usize)
-            }
-            TruncationPolicy::Tokens(tokens) => {
-                TruncationPolicy::Tokens((tokens as f64 * multiplier).ceil() as usize)
-            }
-        }
-    }
-
     pub fn new(config: &Config) -> Self {
         let config_token_limit = config.tool_output_token_limit;
 
@@ -74,20 +65,34 @@ impl TruncationPolicy {
     }
 }
 
-pub(crate) fn formatted_truncate_text(content: &str, policy: TruncationPolicy) -> String {
-    if content.len() <= policy.byte_budget() {
+/// Format a block of exec/tool output for model consumption, truncating by
+/// lines and bytes while preserving head and tail segments.
+pub(crate) fn truncate_with_line_bytes_budget(content: &str, bytes_budget: usize) -> String {
+    // TODO(aibrahim): to be removed
+    let lines_budget = 256;
+    // Head+tail truncation for the model: show the beginning and end with an elision.
+    // Clients still receive full streams; only this formatted summary is capped.
+    let total_lines = content.lines().count();
+    if content.len() <= bytes_budget && total_lines <= lines_budget {
         return content.to_string();
     }
-    let total_lines = content.lines().count();
-    let result = truncate_text(content, policy);
-    format!("Total output lines: {total_lines}\n\n{result}")
+    let output = truncate_formatted_exec_output(content, total_lines, bytes_budget, lines_budget);
+    format!("Total output lines: {total_lines}\n\n{output}")
 }
 
 pub(crate) fn truncate_text(content: &str, policy: TruncationPolicy) -> String {
     match policy {
-        TruncationPolicy::Bytes(_) => truncate_with_byte_estimate(content, policy),
-        TruncationPolicy::Tokens(_) => {
-            let (truncated, _) = truncate_with_token_budget(content, policy);
+        TruncationPolicy::Bytes(bytes) => truncate_with_byte_estimate(
+            content,
+            bytes,
+            TruncationSource::Policy(TruncationPolicy::Bytes(bytes)),
+        ),
+        TruncationPolicy::Tokens(tokens) => {
+            let (truncated, _) = truncate_with_token_budget(
+                content,
+                tokens,
+                TruncationSource::Policy(TruncationPolicy::Tokens(tokens)),
+            );
             truncated
         }
     }
@@ -157,18 +162,24 @@ pub(crate) fn truncate_function_output_items_with_policy(
 /// preserving the beginning and the end. Returns the possibly truncated string
 /// and `Some(original_token_count)` if truncation occurred; otherwise returns
 /// the original string and `None`.
-fn truncate_with_token_budget(s: &str, policy: TruncationPolicy) -> (String, Option<u64>) {
+fn truncate_with_token_budget(
+    s: &str,
+    max_tokens: usize,
+    source: TruncationSource,
+) -> (String, Option<u64>) {
     if s.is_empty() {
         return (String::new(), None);
     }
-    let max_tokens = policy.token_budget();
 
     let byte_len = s.len();
-    if max_tokens > 0 && byte_len <= approx_bytes_for_tokens(max_tokens) {
-        return (s.to_string(), None);
+    if max_tokens > 0 {
+        let small_threshold = approx_bytes_for_tokens(max_tokens / 4);
+        if small_threshold > 0 && byte_len <= small_threshold {
+            return (s.to_string(), None);
+        }
     }
 
-    let truncated = truncate_with_byte_estimate(s, policy);
+    let truncated = truncate_with_byte_estimate(s, approx_bytes_for_tokens(max_tokens), source);
     let approx_total_usize = approx_token_count(s);
     let approx_total = u64::try_from(approx_total_usize).unwrap_or(u64::MAX);
     if truncated == s {
@@ -181,19 +192,14 @@ fn truncate_with_token_budget(s: &str, policy: TruncationPolicy) -> (String, Opt
 /// Truncate a string using a byte budget derived from the token budget, without
 /// performing any real tokenization. This keeps the logic purely byte-based and
 /// uses a bytes placeholder in the truncated output.
-fn truncate_with_byte_estimate(s: &str, policy: TruncationPolicy) -> String {
+fn truncate_with_byte_estimate(s: &str, max_bytes: usize, source: TruncationSource) -> String {
     if s.is_empty() {
         return String::new();
     }
-    let total_chars = s.chars().count();
-    let max_bytes = policy.byte_budget();
 
     if max_bytes == 0 {
         // No budget to show content; just report that everything was truncated.
-        let marker = format_truncation_marker(
-            policy,
-            removed_units_for_source(policy, s.len(), total_chars),
-        );
+        let marker = format_truncation_marker(source, removed_units_for_source(source, s.len()));
         return marker;
     }
 
@@ -202,32 +208,127 @@ fn truncate_with_byte_estimate(s: &str, policy: TruncationPolicy) -> String {
     }
 
     let total_bytes = s.len();
+    let removed_bytes = total_bytes.saturating_sub(max_bytes);
+    let marker = format_truncation_marker(source, removed_units_for_source(source, removed_bytes));
+    let marker_len = marker.len();
 
-    let (left_budget, right_budget) = split_budget(max_bytes);
+    if marker_len >= max_bytes {
+        let truncated_marker = truncate_on_boundary(&marker, max_bytes);
+        return truncated_marker.to_string();
+    }
+
+    let keep_budget = max_bytes - marker_len;
+    let (left_budget, right_budget) = split_budget(keep_budget);
     let prefix_end = pick_prefix_end(s, left_budget);
     let mut suffix_start = pick_suffix_start(s, right_budget);
     if suffix_start < prefix_end {
         suffix_start = prefix_end;
     }
 
-    let left_chars = s[..prefix_end].chars().count();
-    let right_chars = s[suffix_start..].chars().count();
-    let removed_chars = total_chars
-        .saturating_sub(left_chars)
-        .saturating_sub(right_chars);
+    let mut out = assemble_truncated_output(&s[..prefix_end], &s[suffix_start..], &marker);
 
-    let marker = format_truncation_marker(
-        policy,
-        removed_units_for_source(policy, total_bytes.saturating_sub(max_bytes), removed_chars),
-    );
+    if out.len() > max_bytes {
+        let boundary = truncate_on_boundary(&out, max_bytes);
+        out.truncate(boundary.len());
+    }
 
-    assemble_truncated_output(&s[..prefix_end], &s[suffix_start..], &marker)
+    out
 }
 
-fn format_truncation_marker(policy: TruncationPolicy, removed_count: u64) -> String {
-    match policy {
-        TruncationPolicy::Tokens(_) => format!("…{removed_count} tokens truncated…"),
-        TruncationPolicy::Bytes(_) => format!("…{removed_count} chars truncated…"),
+fn truncate_formatted_exec_output(
+    content: &str,
+    total_lines: usize,
+    limit_bytes: usize,
+    limit_lines: usize,
+) -> String {
+    error_on_double_truncation(content);
+    let head_lines: usize = limit_lines / 2;
+    let tail_lines: usize = limit_lines - head_lines; // 128
+    let head_bytes: usize = limit_bytes / 2;
+    let segments: Vec<&str> = content.split_inclusive('\n').collect();
+    let head_take = head_lines.min(segments.len());
+    let tail_take = tail_lines.min(segments.len().saturating_sub(head_take));
+    let omitted = segments.len().saturating_sub(head_take + tail_take);
+
+    let head_slice_end: usize = segments
+        .iter()
+        .take(head_take)
+        .map(|segment| segment.len())
+        .sum();
+    let tail_slice_start: usize = if tail_take == 0 {
+        content.len()
+    } else {
+        content.len()
+            - segments
+                .iter()
+                .rev()
+                .take(tail_take)
+                .map(|segment| segment.len())
+                .sum::<usize>()
+    };
+    let head_slice = &content[..head_slice_end];
+    let tail_slice = &content[tail_slice_start..];
+    let truncated_by_bytes = content.len() > limit_bytes;
+    // this is a bit wrong. We are counting metadata lines and not just shell output lines.
+    let marker = if omitted > 0 {
+        let marker_text = format_truncation_marker(
+            TruncationSource::LineOmission { total_lines },
+            u64::try_from(omitted).unwrap_or(u64::MAX),
+        );
+        Some(format!("\n{marker_text}\n\n"))
+    } else if truncated_by_bytes {
+        let removed_bytes =
+            u64::try_from(content.len().saturating_sub(limit_bytes)).unwrap_or(u64::MAX);
+        let marker_text =
+            format_truncation_marker(TruncationSource::ByteLimit { limit_bytes }, removed_bytes);
+        Some(format!("\n{marker_text}\n\n"))
+    } else {
+        None
+    };
+
+    let marker_len = marker.as_ref().map_or(0, String::len);
+    let base_head_budget = head_bytes.min(limit_bytes);
+    let head_budget = base_head_budget.min(limit_bytes.saturating_sub(marker_len));
+    let head_part = take_bytes_at_char_boundary(head_slice, head_budget);
+    let mut result = String::with_capacity(limit_bytes.min(content.len()));
+
+    result.push_str(head_part);
+    if let Some(marker_text) = marker.as_ref() {
+        result.push_str(marker_text);
+    }
+
+    let remaining = limit_bytes.saturating_sub(result.len());
+    if remaining == 0 {
+        return result;
+    }
+
+    let tail_part = take_last_bytes_at_char_boundary(tail_slice, remaining);
+    result.push_str(tail_part);
+
+    result
+}
+
+#[derive(Clone, Copy)]
+pub enum TruncationSource {
+    Policy(TruncationPolicy),
+    LineOmission { total_lines: usize },
+    ByteLimit { limit_bytes: usize },
+}
+
+fn format_truncation_marker(source: TruncationSource, removed_count: u64) -> String {
+    match source {
+        TruncationSource::Policy(TruncationPolicy::Tokens(_)) => {
+            format!("[…{removed_count} tokens truncated…]")
+        }
+        TruncationSource::Policy(TruncationPolicy::Bytes(_)) => {
+            format!("[…{removed_count} bytes truncated…]")
+        }
+        TruncationSource::LineOmission { total_lines } => {
+            format!("[... omitted {removed_count} of {total_lines} lines ...]")
+        }
+        TruncationSource::ByteLimit { limit_bytes } => {
+            format!("[... removed {removed_count} bytes to fit {limit_bytes} byte limit ...]")
+        }
     }
 }
 
@@ -236,14 +337,12 @@ fn split_budget(budget: usize) -> (usize, usize) {
     (left, budget - left)
 }
 
-fn removed_units_for_source(
-    policy: TruncationPolicy,
-    removed_bytes: usize,
-    removed_chars: usize,
-) -> u64 {
-    match policy {
-        TruncationPolicy::Tokens(_) => approx_tokens_from_byte_count(removed_bytes),
-        TruncationPolicy::Bytes(_) => u64::try_from(removed_chars).unwrap_or(u64::MAX),
+fn removed_units_for_source(source: TruncationSource, removed_bytes: usize) -> u64 {
+    match source {
+        TruncationSource::Policy(TruncationPolicy::Tokens(_)) => {
+            approx_tokens_from_byte_count(removed_bytes)
+        }
+        _ => u64::try_from(removed_bytes).unwrap_or(u64::MAX),
     }
 }
 
@@ -251,6 +350,7 @@ fn assemble_truncated_output(prefix: &str, suffix: &str, marker: &str) -> String
     let mut out = String::with_capacity(prefix.len() + marker.len() + suffix.len() + 1);
     out.push_str(prefix);
     out.push_str(marker);
+    out.push('\n');
     out.push_str(suffix);
     out
 }
@@ -305,130 +405,203 @@ fn pick_suffix_start(s: &str, right_budget: usize) -> usize {
     idx
 }
 
+fn error_on_double_truncation(content: &str) {
+    if content.contains("Total output lines:") && content.contains("omitted") {
+        tracing::error!(
+            "FunctionCallOutput content was already truncated before ContextManager::record_items; this would cause double truncation {content}"
+        );
+    }
+}
+
 #[cfg(test)]
 mod tests {
+    use crate::config::OPENAI_DEFAULT_MODEL;
+    use crate::model_family::derive_default_model_family;
+    use crate::model_family::find_family_for_model;
 
     use super::TruncationPolicy;
+    use super::TruncationSource;
     use super::approx_token_count;
-    use super::formatted_truncate_text;
     use super::truncate_function_output_items_with_policy;
-    use super::truncate_text;
+    use super::truncate_with_line_bytes_budget;
     use super::truncate_with_token_budget;
     use codex_protocol::models::FunctionCallOutputContentItem;
     use pretty_assertions::assert_eq;
+    use regex_lite::Regex;
 
-    #[test]
-    fn truncate_bytes_less_than_placeholder_returns_placeholder() {
-        let content = "example output";
+    const MODEL_FORMAT_MAX_LINES: usize = 256;
 
-        assert_eq!(
-            "Total output lines: 1\n\n…13 chars truncated…t",
-            formatted_truncate_text(content, TruncationPolicy::Bytes(1)),
-        );
+    fn model_format_max_bytes() -> usize {
+        find_family_for_model(OPENAI_DEFAULT_MODEL)
+            .unwrap_or_else(|| derive_default_model_family(OPENAI_DEFAULT_MODEL))
+            .truncation_policy
+            .byte_budget()
+    }
+
+    fn truncated_message_pattern(line: &str, total_lines: usize) -> String {
+        let head_lines = MODEL_FORMAT_MAX_LINES / 2;
+        let tail_lines = MODEL_FORMAT_MAX_LINES - head_lines;
+        let head_take = head_lines.min(total_lines);
+        let tail_take = tail_lines.min(total_lines.saturating_sub(head_take));
+        let omitted = total_lines.saturating_sub(head_take + tail_take);
+        let escaped_line = regex_lite::escape(line);
+        if omitted == 0 {
+            return format!(
+                r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} removed \d+ bytes to fit {max_bytes} byte limit \.{{3}}]\n\n.*)$",
+                max_bytes = model_format_max_bytes(),
+            );
+        }
+        format!(
+            r"(?s)^Total output lines: {total_lines}\n\n(?P<body>{escaped_line}.*\n\[\.{{3}} omitted {omitted} of {total_lines} lines \.{{3}}]\n\n.*)$",
+        )
     }
 
     #[test]
-    fn truncate_tokens_less_than_placeholder_returns_placeholder() {
-        let content = "example output";
-
-        assert_eq!(
-            "Total output lines: 1\n\nex…3 tokens truncated…ut",
-            formatted_truncate_text(content, TruncationPolicy::Tokens(1)),
-        );
-    }
-
-    #[test]
-    fn truncate_tokens_under_limit_returns_original() {
-        let content = "example output";
-
-        assert_eq!(
-            content,
-            formatted_truncate_text(content, TruncationPolicy::Tokens(10)),
-        );
-    }
-
-    #[test]
-    fn truncate_bytes_under_limit_returns_original() {
-        let content = "example output";
-
-        assert_eq!(
-            content,
-            formatted_truncate_text(content, TruncationPolicy::Bytes(20)),
-        );
-    }
-
-    #[test]
-    fn truncate_tokens_over_limit_returns_truncated() {
-        let content = "this is an example of a long output that should be truncated";
-
-        assert_eq!(
-            "Total output lines: 1\n\nthis is an…10 tokens truncated… truncated",
-            formatted_truncate_text(content, TruncationPolicy::Tokens(5)),
-        );
-    }
-
-    #[test]
-    fn truncate_bytes_over_limit_returns_truncated() {
-        let content = "this is an example of a long output that should be truncated";
-
-        assert_eq!(
-            "Total output lines: 1\n\nthis is an exam…30 chars truncated…ld be truncated",
-            formatted_truncate_text(content, TruncationPolicy::Bytes(30)),
-        );
-    }
-
-    #[test]
-    fn truncate_bytes_reports_original_line_count_when_truncated() {
-        let content =
-            "this is an example of a long output that should be truncated\nalso some other line";
-
-        assert_eq!(
-            "Total output lines: 2\n\nthis is an exam…51 chars truncated…some other line",
-            formatted_truncate_text(content, TruncationPolicy::Bytes(30)),
-        );
-    }
-
-    #[test]
-    fn truncate_tokens_reports_original_line_count_when_truncated() {
-        let content =
-            "this is an example of a long output that should be truncated\nalso some other line";
-
-        assert_eq!(
-            "Total output lines: 2\n\nthis is an example o…11 tokens truncated…also some other line",
-            formatted_truncate_text(content, TruncationPolicy::Tokens(10)),
-        );
-    }
-
-    #[test]
-    fn truncate_with_token_budget_returns_original_when_under_limit() {
+    fn truncate_middle_returns_original_when_under_limit() {
         let s = "short output";
         let limit = 100;
-        let (out, original) = truncate_with_token_budget(s, TruncationPolicy::Tokens(limit));
+        let source = TruncationSource::Policy(TruncationPolicy::Tokens(limit));
+        let (out, original) = truncate_with_token_budget(s, limit, source);
         assert_eq!(out, s);
         assert_eq!(original, None);
     }
 
     #[test]
-    fn truncate_with_token_budget_reports_truncation_at_zero_limit() {
+    fn truncate_middle_reports_truncation_at_zero_limit() {
         let s = "abcdef";
-        let (out, original) = truncate_with_token_budget(s, TruncationPolicy::Tokens(0));
-        assert_eq!(out, "…2 tokens truncated…");
-        assert_eq!(original, Some(2));
+        let source = TruncationSource::Policy(TruncationPolicy::Tokens(0));
+        let (out, original) = truncate_with_token_budget(s, 0, source);
+        assert_eq!(out, "[…2 tokens truncated…]");
+        assert_eq!(original, Some(approx_token_count(s) as u64));
     }
 
     #[test]
-    fn truncate_middle_tokens_handles_utf8_content() {
-        let s = "😀😀😀😀😀😀😀😀😀😀\nsecond line with text\n";
-        let (out, tokens) = truncate_with_token_budget(s, TruncationPolicy::Tokens(8));
-        assert_eq!(out, "😀😀😀😀…8 tokens truncated…");
-        assert_eq!(tokens, Some(16));
+    fn truncate_middle_enforces_token_budget() {
+        let s = "alpha beta gamma delta epsilon zeta eta theta iota kappa";
+        let max_tokens = 12;
+        let source = TruncationSource::Policy(TruncationPolicy::Tokens(max_tokens));
+        let (out, original) = truncate_with_token_budget(s, max_tokens, source);
+        assert!(out.contains("tokens truncated"));
+        assert_eq!(original, Some(approx_token_count(s) as u64));
+        assert!(out.len() < s.len(), "truncated output should be shorter");
     }
 
     #[test]
-    fn truncate_middle_bytes_handles_utf8_content() {
+    fn truncate_middle_handles_utf8_content() {
         let s = "😀😀😀😀😀😀😀😀😀😀\nsecond line with text\n";
-        let out = truncate_text(s, TruncationPolicy::Bytes(20));
-        assert_eq!(out, "😀😀…31 chars truncated…");
+        let max_tokens = 8;
+        let source = TruncationSource::Policy(TruncationPolicy::Tokens(max_tokens));
+        let (out, tokens) = truncate_with_token_budget(s, max_tokens, source);
+
+        assert!(out.contains("tokens truncated"));
+        assert!(!out.contains('\u{fffd}'));
+        assert_eq!(tokens, Some(approx_token_count(s) as u64));
+        assert!(out.len() < s.len(), "UTF-8 content should be shortened");
+    }
+
+    #[test]
+    fn format_exec_output_truncates_large_error() {
+        let line = "very long execution error line that should trigger truncation\n";
+        let large_error = line.repeat(2_500); // way beyond both byte and line limits
+
+        let truncated = truncate_with_line_bytes_budget(&large_error, model_format_max_bytes());
+
+        let total_lines = large_error.lines().count();
+        let pattern = truncated_message_pattern(line, total_lines);
+        let regex = Regex::new(&pattern).unwrap_or_else(|err| {
+            panic!("failed to compile regex {pattern}: {err}");
+        });
+        let captures = regex
+            .captures(&truncated)
+            .unwrap_or_else(|| panic!("message failed to match pattern {pattern}: {truncated}"));
+        let body = captures
+            .name("body")
+            .expect("missing body capture")
+            .as_str();
+        assert!(
+            body.len() <= model_format_max_bytes(),
+            "body exceeds byte limit: {} bytes",
+            body.len()
+        );
+        assert_ne!(truncated, large_error);
+    }
+
+    #[test]
+    fn format_exec_output_marks_byte_truncation_without_omitted_lines() {
+        let max_bytes = model_format_max_bytes();
+        let long_line = "a".repeat(max_bytes + 50);
+        let truncated = truncate_with_line_bytes_budget(&long_line, max_bytes);
+
+        assert_ne!(truncated, long_line);
+        let removed_bytes = long_line.len().saturating_sub(max_bytes);
+        let marker_line =
+            format!("[... removed {removed_bytes} bytes to fit {max_bytes} byte limit ...]");
+        assert!(
+            truncated.contains(&marker_line),
+            "missing byte truncation marker: {truncated}"
+        );
+        assert!(
+            !truncated.contains("omitted"),
+            "line omission marker should not appear when no lines were dropped: {truncated}"
+        );
+    }
+
+    #[test]
+    fn format_exec_output_returns_original_when_within_limits() {
+        let content = "example output\n".repeat(10);
+
+        assert_eq!(
+            truncate_with_line_bytes_budget(&content, model_format_max_bytes()),
+            content
+        );
+    }
+
+    #[test]
+    fn format_exec_output_reports_omitted_lines_and_keeps_head_and_tail() {
+        let total_lines = MODEL_FORMAT_MAX_LINES + 100;
+        let content: String = (0..total_lines)
+            .map(|idx| format!("line-{idx}\n"))
+            .collect();
+
+        let truncated = truncate_with_line_bytes_budget(&content, model_format_max_bytes());
+
+        let omitted = total_lines - MODEL_FORMAT_MAX_LINES;
+        let expected_marker = format!("[... omitted {omitted} of {total_lines} lines ...]");
+
+        assert!(
+            truncated.contains(&expected_marker),
+            "missing omitted marker: {truncated}"
+        );
+        assert!(
+            truncated.contains("line-0\n"),
+            "expected head line to remain: {truncated}"
+        );
+
+        let last_line = format!("line-{}\n", total_lines - 1);
+        assert!(
+            truncated.contains(&last_line),
+            "expected tail line to remain: {truncated}"
+        );
+    }
+
+    #[test]
+    fn format_exec_output_prefers_line_marker_when_both_limits_exceeded() {
+        let total_lines = MODEL_FORMAT_MAX_LINES + 42;
+        let long_line = "x".repeat(256);
+        let content: String = (0..total_lines)
+            .map(|idx| format!("line-{idx}-{long_line}\n"))
+            .collect();
+
+        let truncated = truncate_with_line_bytes_budget(&content, model_format_max_bytes());
+
+        assert!(
+            truncated.contains("[... omitted 42 of 298 lines ...]"),
+            "expected omitted marker when line count exceeds limit: {truncated}"
+        );
+        assert!(
+            !truncated.contains("byte limit"),
+            "line omission marker should take precedence over byte marker: {truncated}"
+        );
     }
 
     #[test]

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

@@ -15,7 +15,7 @@ use crate::exec::SandboxType;
 use crate::exec::StreamOutput;
 use crate::exec::is_likely_sandbox_denied;
 use crate::truncate::TruncationPolicy;
-use crate::truncate::formatted_truncate_text;
+use crate::truncate::truncate_text;
 use codex_utils_pty::ExecCommandSession;
 use codex_utils_pty::SpawnedPty;
 
@@ -167,7 +167,7 @@ impl UnifiedExecSession {
         };
 
         if is_likely_sandbox_denied(self.sandbox_type(), &exec_output) {
-            let snippet = formatted_truncate_text(
+            let snippet = truncate_text(
                 &aggregated_text,
                 TruncationPolicy::Tokens(UNIFIED_EXEC_OUTPUT_MAX_TOKENS),
             );

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

@@ -11,10 +11,12 @@ use crate::codex::TurnContext;
 use crate::exec::ExecToolCallOutput;
 use crate::exec::StreamOutput;
 use crate::exec_env::create_env;
+use crate::exec_policy::create_approval_requirement_for_command;
 use crate::protocol::BackgroundEventEvent;
 use crate::protocol::EventMsg;
 use crate::protocol::ExecCommandSource;
 use crate::sandboxing::ExecEnv;
+use crate::sandboxing::SandboxPermissions;
 use crate::tools::events::ToolEmitter;
 use crate::tools::events::ToolEventCtx;
 use crate::tools::events::ToolEventFailure;
@@ -25,7 +27,7 @@ use crate::tools::runtimes::unified_exec::UnifiedExecRuntime;
 use crate::tools::sandboxing::ToolCtx;
 use crate::truncate::TruncationPolicy;
 use crate::truncate::approx_token_count;
-use crate::truncate::formatted_truncate_text;
+use crate::truncate::truncate_text;
 
 use super::ExecCommandRequest;
 use super::SessionEntry;
@@ -72,7 +74,7 @@ impl UnifiedExecSessionManager {
         let wall_time = Instant::now().saturating_duration_since(start);
 
         let text = String::from_utf8_lossy(&collected).to_string();
-        let output = formatted_truncate_text(&text, TruncationPolicy::Tokens(max_tokens));
+        let output = truncate_text(&text, TruncationPolicy::Tokens(max_tokens));
         let chunk_id = generate_chunk_id();
         let has_exited = session.has_exited();
         let stored_id = self
@@ -179,7 +181,7 @@ impl UnifiedExecSessionManager {
         let wall_time = Instant::now().saturating_duration_since(start);
 
         let text = String::from_utf8_lossy(&collected).to_string();
-        let output = formatted_truncate_text(&text, TruncationPolicy::Tokens(max_tokens));
+        let output = truncate_text(&text, TruncationPolicy::Tokens(max_tokens));
         let original_token_count = approx_token_count(&text);
         let chunk_id = generate_chunk_id();
 
@@ -449,6 +451,13 @@ impl UnifiedExecSessionManager {
             create_env(&context.turn.shell_environment_policy),
             with_escalated_permissions,
             justification,
+            create_approval_requirement_for_command(
+                &context.turn.exec_policy,
+                command,
+                context.turn.approval_policy,
+                &context.turn.sandbox_policy,
+                SandboxPermissions::from(with_escalated_permissions.unwrap_or(false)),
+            ),
         );
         let tool_ctx = ToolCtx {
             session: context.session.as_ref(),

codex-rs/core/src/user_shell_command.rs

@@ -3,7 +3,6 @@ use std::time::Duration;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseItem;
 
-use crate::codex::TurnContext;
 use crate::exec::ExecToolCallOutput;
 use crate::tools::format_exec_output_str;
 
@@ -21,11 +20,7 @@ fn format_duration_line(duration: Duration) -> String {
     format!("Duration: {duration_seconds:.4} seconds")
 }
 
-fn format_user_shell_command_body(
-    command: &str,
-    exec_output: &ExecToolCallOutput,
-    turn_context: &TurnContext,
-) -> String {
+fn format_user_shell_command_body(command: &str, exec_output: &ExecToolCallOutput) -> String {
     let mut sections = Vec::new();
     sections.push("<command>".to_string());
     sections.push(command.to_string());
@@ -34,33 +29,25 @@ fn format_user_shell_command_body(
     sections.push(format!("Exit code: {}", exec_output.exit_code));
     sections.push(format_duration_line(exec_output.duration));
     sections.push("Output:".to_string());
-    sections.push(format_exec_output_str(
-        exec_output,
-        turn_context.truncation_policy,
-    ));
+    sections.push(format_exec_output_str(exec_output));
     sections.push("</result>".to_string());
     sections.join("\n")
 }
 
-pub fn format_user_shell_command_record(
-    command: &str,
-    exec_output: &ExecToolCallOutput,
-    turn_context: &TurnContext,
-) -> String {
-    let body = format_user_shell_command_body(command, exec_output, turn_context);
+pub fn format_user_shell_command_record(command: &str, exec_output: &ExecToolCallOutput) -> String {
+    let body = format_user_shell_command_body(command, exec_output);
     format!("{USER_SHELL_COMMAND_OPEN}\n{body}\n{USER_SHELL_COMMAND_CLOSE}")
 }
 
 pub fn user_shell_command_record_item(
     command: &str,
     exec_output: &ExecToolCallOutput,
-    turn_context: &TurnContext,
 ) -> ResponseItem {
     ResponseItem::Message {
         id: None,
         role: "user".to_string(),
         content: vec![ContentItem::InputText {
-            text: format_user_shell_command_record(command, exec_output, turn_context),
+            text: format_user_shell_command_record(command, exec_output),
         }],
     }
 }
@@ -68,7 +55,6 @@ pub fn user_shell_command_record_item(
 #[cfg(test)]
 mod tests {
     use super::*;
-    use crate::codex::make_session_and_context;
     use crate::exec::StreamOutput;
     use pretty_assertions::assert_eq;
 
@@ -90,8 +76,7 @@ mod tests {
             duration: Duration::from_secs(1),
             timed_out: false,
         };
-        let (_, turn_context) = make_session_and_context();
-        let item = user_shell_command_record_item("echo hi", &exec_output, &turn_context);
+        let item = user_shell_command_record_item("echo hi", &exec_output);
         let ResponseItem::Message { content, .. } = item else {
             panic!("expected message");
         };
@@ -114,8 +99,7 @@ mod tests {
             duration: Duration::from_millis(120),
             timed_out: false,
         };
-        let (_, turn_context) = make_session_and_context();
-        let record = format_user_shell_command_record("false", &exec_output, &turn_context);
+        let record = format_user_shell_command_record("false", &exec_output);
         assert_eq!(
             record,
             "<user_shell_command>\n<command>\nfalse\n</command>\n<result>\nExit code: 42\nDuration: 0.1200 seconds\nOutput:\ncombined output wins\n</result>\n</user_shell_command>"

codex-rs/core/templates/parallel/instructions.md

@@ -1,4 +1,3 @@
-
 ## Exploration and reading files
 
 - **Think first.** Before any tool call, decide ALL files/resources you will need.
@@ -11,3 +10,5 @@
 * Always maximize parallelism. Never read files one-by-one unless logically unavoidable.
 * This concern every read/list/search operations including, but not only, `cat`, `rg`, `sed`, `ls`, `git show`, `nl`, `wc`, ...
 * Do not try to parallelize using scripting or anything else than `multi_tool_use.parallel`.
+
+## Editing constraints

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

@@ -460,13 +460,6 @@ pub fn ev_apply_patch_function_call(call_id: &str, patch: &str) -> Value {
     })
 }
 
-pub fn ev_shell_command_call(call_id: &str, command: &str) -> Value {
-    let args = serde_json::json!({ "command": command });
-    let arguments = serde_json::to_string(&args).expect("serialize shell arguments");
-
-    ev_function_call(call_id, "shell_command", &arguments)
-}
-
 pub fn ev_apply_patch_shell_call(call_id: &str, patch: &str) -> Value {
     let args = serde_json::json!({ "command": ["apply_patch", patch] });
     let arguments = serde_json::to_string(&args).expect("serialize apply_patch arguments");

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

@@ -1155,7 +1155,7 @@ async fn token_count_includes_rate_limits_snapshot() {
                     "reasoning_output_tokens": 0,
                     "total_tokens": 123
                 },
-                // Default model is arcticfox in tests → 95% usable context window
+                // Default model is gpt-5.1-codex in tests → 95% usable context window
                 "model_context_window": 258400
             },
             "rate_limits": {

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

@@ -70,7 +70,6 @@ async fn codex_delegate_forwards_exec_approval_and_proceeds_on_approval() {
             review_request: ReviewRequest {
                 prompt: "Please review".to_string(),
                 user_facing_hint: "review".to_string(),
-                append_to_original_thread: true,
             },
         })
         .await
@@ -146,7 +145,6 @@ async fn codex_delegate_forwards_patch_approval_and_proceeds_on_decision() {
             review_request: ReviewRequest {
                 prompt: "Please review".to_string(),
                 user_facing_hint: "review".to_string(),
-                append_to_original_thread: true,
             },
         })
         .await
@@ -201,7 +199,6 @@ async fn codex_delegate_ignores_legacy_deltas() {
             review_request: ReviewRequest {
                 prompt: "Please review".to_string(),
                 user_facing_hint: "review".to_string(),
-                append_to_original_thread: true,
             },
         })
         .await

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

@@ -191,19 +191,22 @@ async fn summarize_context_three_requests_and_instructions() {
     let body2_str = body2.to_string();
     let input2 = body2.get("input").and_then(|v| v.as_array()).unwrap();
     let has_compact_prompt = body_contains_text(&body2_str, SUMMARIZATION_PROMPT);
-    assert!(
-        has_compact_prompt,
-        "compaction request should include the summarize trigger"
-    );
-    // The last item is the user message created from the injected input.
-    let last2 = input2.last().unwrap();
-    assert_eq!(last2.get("type").unwrap().as_str().unwrap(), "message");
-    assert_eq!(last2.get("role").unwrap().as_str().unwrap(), "user");
-    let text2 = last2["content"][0]["text"].as_str().unwrap();
-    assert_eq!(
-        text2, SUMMARIZATION_PROMPT,
-        "expected summarize trigger, got `{text2}`"
-    );
+    if has_compact_prompt {
+        // The last item is the user message created from the injected input.
+        let last2 = input2.last().unwrap();
+        assert_eq!(last2.get("type").unwrap().as_str().unwrap(), "message");
+        assert_eq!(last2.get("role").unwrap().as_str().unwrap(), "user");
+        let text2 = last2["content"][0]["text"].as_str().unwrap();
+        assert_eq!(
+            text2, SUMMARIZATION_PROMPT,
+            "expected summarize trigger, got `{text2}`"
+        );
+    } else {
+        assert!(
+            !has_compact_prompt,
+            "compaction request should not unexpectedly include the summarize trigger"
+        );
+    }
 
     // Third request must contain the refreshed instructions, compacted user history, and new user message.
     let input3 = body3.get("input").and_then(|v| v.as_array()).unwrap();

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

@@ -13,13 +13,10 @@ use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::user_input::UserInput;
 use core_test_support::responses;
-use core_test_support::responses::mount_sse_once;
-use core_test_support::responses::sse;
 use core_test_support::skip_if_no_network;
 use core_test_support::test_codex::TestCodexHarness;
 use core_test_support::test_codex::test_codex;
 use core_test_support::wait_for_event;
-use core_test_support::wait_for_event_match;
 use pretty_assertions::assert_eq;
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
@@ -128,72 +125,6 @@ async fn remote_compact_replaces_history_for_followups() -> Result<()> {
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn remote_compact_runs_automatically() -> Result<()> {
-    skip_if_no_network!(Ok(()));
-
-    let harness = TestCodexHarness::with_builder(
-        test_codex()
-            .with_auth(CodexAuth::create_dummy_chatgpt_auth_for_testing())
-            .with_config(|config| {
-                config.features.enable(Feature::RemoteCompaction);
-            }),
-    )
-    .await?;
-    let codex = harness.test().codex.clone();
-
-    mount_sse_once(
-        harness.server(),
-        sse(vec![
-            responses::ev_shell_command_call("m1", "echo 'hi'"),
-            responses::ev_completed_with_tokens("resp-1", 100000000), // over token limit
-        ]),
-    )
-    .await;
-    let responses_mock = mount_sse_once(
-        harness.server(),
-        responses::sse(vec![
-            responses::ev_assistant_message("m2", "AFTER_COMPACT_REPLY"),
-            responses::ev_completed("resp-2"),
-        ]),
-    )
-    .await;
-
-    let compacted_history = vec![ResponseItem::Message {
-        id: None,
-        role: "user".to_string(),
-        content: vec![ContentItem::InputText {
-            text: "REMOTE_COMPACTED_SUMMARY".to_string(),
-        }],
-    }];
-    let compact_mock = responses::mount_compact_json_once(
-        harness.server(),
-        serde_json::json!({ "output": compacted_history.clone() }),
-    )
-    .await;
-
-    codex
-        .submit(Op::UserInput {
-            items: vec![UserInput::Text {
-                text: "hello remote compact".into(),
-            }],
-        })
-        .await?;
-    let message = wait_for_event_match(&codex, |ev| match ev {
-        EventMsg::AgentMessage(ev) => Some(ev.message.clone()),
-        _ => None,
-    })
-    .await;
-    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
-
-    assert_eq!(message, "Compact task completed");
-    assert_eq!(compact_mock.requests().len(), 1);
-    let follow_up_body = responses_mock.single_request().body_json().to_string();
-    assert!(follow_up_body.contains("REMOTE_COMPACTED_SUMMARY"));
-
-    Ok(())
-}
-
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn remote_compact_persists_replacement_history_in_rollout() -> Result<()> {
     skip_if_no_network!(Ok(()));

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

@@ -0,0 +1,101 @@
+#![allow(clippy::unwrap_used, clippy::expect_used)]
+
+use anyhow::Result;
+use codex_core::protocol::AskForApproval;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::Op;
+use codex_core::protocol::SandboxPolicy;
+use codex_protocol::config_types::ReasoningSummary;
+use codex_protocol::user_input::UserInput;
+use core_test_support::responses::ev_assistant_message;
+use core_test_support::responses::ev_completed;
+use core_test_support::responses::ev_function_call;
+use core_test_support::responses::ev_response_created;
+use core_test_support::responses::mount_sse_once;
+use core_test_support::responses::sse;
+use core_test_support::responses::start_mock_server;
+use core_test_support::test_codex::test_codex;
+use core_test_support::wait_for_event;
+use serde_json::json;
+use std::fs;
+
+#[tokio::test]
+async fn execpolicy_blocks_shell_invocation() -> Result<()> {
+    let mut builder = test_codex().with_config(|config| {
+        let policy_path = config.codex_home.join("policy").join("policy.codexpolicy");
+        fs::create_dir_all(
+            policy_path
+                .parent()
+                .expect("policy directory must have a parent"),
+        )
+        .expect("create policy directory");
+        fs::write(
+            &policy_path,
+            r#"prefix_rule(pattern=["echo"], decision="forbidden")"#,
+        )
+        .expect("write policy file");
+    });
+    let server = start_mock_server().await;
+    let test = builder.build(&server).await?;
+
+    let call_id = "shell-forbidden";
+    let args = json!({
+        "command": ["echo", "blocked"],
+        "timeout_ms": 1_000,
+    });
+
+    mount_sse_once(
+        &server,
+        sse(vec![
+            ev_response_created("resp-1"),
+            ev_function_call(call_id, "shell", &serde_json::to_string(&args)?),
+            ev_completed("resp-1"),
+        ]),
+    )
+    .await;
+    mount_sse_once(
+        &server,
+        sse(vec![
+            ev_assistant_message("msg-1", "done"),
+            ev_completed("resp-2"),
+        ]),
+    )
+    .await;
+
+    let session_model = test.session_configured.model.clone();
+    test.codex
+        .submit(Op::UserTurn {
+            items: vec![UserInput::Text {
+                text: "run shell command".into(),
+            }],
+            final_output_json_schema: None,
+            cwd: test.cwd_path().to_path_buf(),
+            approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::DangerFullAccess,
+            model: session_model,
+            effort: None,
+            summary: ReasoningSummary::Auto,
+        })
+        .await?;
+
+    let EventMsg::ExecCommandEnd(end) = wait_for_event(&test.codex, |event| {
+        matches!(event, EventMsg::ExecCommandEnd(_))
+    })
+    .await
+    else {
+        unreachable!()
+    };
+    wait_for_event(&test.codex, |event| {
+        matches!(event, EventMsg::TaskComplete(_))
+    })
+    .await;
+
+    assert!(
+        end.aggregated_output
+            .contains("execpolicy forbids this command"),
+        "unexpected output: {}",
+        end.aggregated_output
+    );
+
+    Ok(())
+}

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

@@ -28,6 +28,7 @@ mod compact_remote;
 mod compact_resume_fork;
 mod deprecation_notice;
 mod exec;
+mod exec_policy;
 mod fork_conversation;
 mod grep_files;
 mod items;

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

@@ -102,11 +102,29 @@ async fn model_selects_expected_tools() {
         "codex-mini-latest should expose the local shell tool",
     );
 
+    let o3_tools = collect_tool_identifiers_for_model("o3").await;
+    assert_eq!(
+        o3_tools,
+        vec![
+            "shell".to_string(),
+            "list_mcp_resources".to_string(),
+            "list_mcp_resource_templates".to_string(),
+            "read_mcp_resource".to_string(),
+            "update_plan".to_string()
+        ],
+        "o3 should expose the generic shell tool",
+    );
+
     let gpt5_codex_tools = collect_tool_identifiers_for_model("gpt-5-codex").await;
     assert_eq!(
         gpt5_codex_tools,
         vec![
-            "shell_command".to_string(),
+            if cfg!(windows) {
+                "shell_command"
+            } else {
+                "shell"
+            }
+            .to_string(),
             "list_mcp_resources".to_string(),
             "list_mcp_resource_templates".to_string(),
             "read_mcp_resource".to_string(),
@@ -120,7 +138,12 @@ async fn model_selects_expected_tools() {
     assert_eq!(
         gpt51_codex_tools,
         vec![
-            "shell_command".to_string(),
+            if cfg!(windows) {
+                "shell_command"
+            } else {
+                "shell"
+            }
+            .to_string(),
             "list_mcp_resources".to_string(),
             "list_mcp_resource_templates".to_string(),
             "read_mcp_resource".to_string(),
@@ -130,24 +153,11 @@ async fn model_selects_expected_tools() {
         "gpt-5.1-codex should expose the apply_patch tool",
     );
 
-    let gpt5_tools = collect_tool_identifiers_for_model("gpt-5").await;
-    assert_eq!(
-        gpt5_tools,
-        vec![
-            "shell".to_string(),
-            "list_mcp_resources".to_string(),
-            "list_mcp_resource_templates".to_string(),
-            "read_mcp_resource".to_string(),
-            "update_plan".to_string(),
-        ],
-        "gpt-5 should expose the apply_patch tool",
-    );
-
     let gpt51_tools = collect_tool_identifiers_for_model("gpt-5.1").await;
     assert_eq!(
         gpt51_tools,
         vec![
-            "shell_command".to_string(),
+            "shell".to_string(),
             "list_mcp_resources".to_string(),
             "list_mcp_resource_templates".to_string(),
             "read_mcp_resource".to_string(),

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

@@ -1,9 +1,9 @@
 #![allow(clippy::unwrap_used)]
 
+use codex_core::config::OPENAI_DEFAULT_MODEL;
 use codex_core::features::Feature;
 use codex_core::model_family::find_family_for_model;
 use codex_core::protocol::AskForApproval;
-use codex_core::protocol::ENVIRONMENT_CONTEXT_OPEN_TAG;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::Op;
 use codex_core::protocol::SandboxPolicy;
@@ -19,6 +19,7 @@ use core_test_support::skip_if_no_network;
 use core_test_support::test_codex::TestCodex;
 use core_test_support::test_codex::test_codex;
 use core_test_support::wait_for_event;
+use std::collections::HashMap;
 use tempfile::TempDir;
 
 fn text_user_input(text: String) -> serde_json::Value {
@@ -155,15 +156,61 @@ async fn prompt_tools_are_consistent_across_requests() -> anyhow::Result<()> {
         .await?;
     wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
 
-    let expected_tools_names = vec![
-        "shell_command",
-        "list_mcp_resources",
-        "list_mcp_resource_templates",
-        "read_mcp_resource",
-        "update_plan",
-        "apply_patch",
-        "view_image",
-    ];
+    // our internal implementation is responsible for keeping tools in sync
+    // with the OpenAI schema, so we just verify the tool presence here
+    let tools_by_model: HashMap<&'static str, Vec<&'static str>> = HashMap::from([
+        (
+            "gpt-5.1",
+            vec![
+                "shell",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "view_image",
+            ],
+        ),
+        (
+            "gpt-5.1",
+            vec![
+                "shell",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "apply_patch",
+                "view_image",
+            ],
+        ),
+        (
+            "gpt-5.1-codex",
+            vec![
+                "shell",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "apply_patch",
+                "view_image",
+            ],
+        ),
+        (
+            "gpt-5.1-codex",
+            vec![
+                "shell",
+                "list_mcp_resources",
+                "list_mcp_resource_templates",
+                "read_mcp_resource",
+                "update_plan",
+                "apply_patch",
+                "view_image",
+            ],
+        ),
+    ]);
+    let expected_tools_names = tools_by_model
+        .get(OPENAI_DEFAULT_MODEL)
+        .unwrap_or_else(|| panic!("expected tools to be defined for model {OPENAI_DEFAULT_MODEL}"))
+        .as_slice();
     let body0 = req1.single_request().body_json();
 
     let expected_instructions = if expected_tools_names.contains(&"apply_patch") {
@@ -180,14 +227,14 @@ async fn prompt_tools_are_consistent_across_requests() -> anyhow::Result<()> {
         body0["instructions"],
         serde_json::json!(expected_instructions),
     );
-    assert_tool_names(&body0, &expected_tools_names);
+    assert_tool_names(&body0, expected_tools_names);
 
     let body1 = req2.single_request().body_json();
     assert_eq!(
         body1["instructions"],
         serde_json::json!(expected_instructions),
     );
-    assert_tool_names(&body1, &expected_tools_names);
+    assert_tool_names(&body1, expected_tools_names);
 
     Ok(())
 }
@@ -373,89 +420,6 @@ async fn overrides_turn_context_but_keeps_cached_prefix_and_key_constant() -> an
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn override_before_first_turn_emits_environment_context() -> anyhow::Result<()> {
-    skip_if_no_network!(Ok(()));
-
-    let server = start_mock_server().await;
-    let req = mount_sse_once(&server, sse_completed("resp-1")).await;
-
-    let TestCodex { codex, .. } = test_codex().build(&server).await?;
-
-    codex
-        .submit(Op::OverrideTurnContext {
-            cwd: None,
-            approval_policy: Some(AskForApproval::Never),
-            sandbox_policy: None,
-            model: None,
-            effort: None,
-            summary: None,
-        })
-        .await?;
-
-    codex
-        .submit(Op::UserInput {
-            items: vec![UserInput::Text {
-                text: "first message".into(),
-            }],
-        })
-        .await?;
-
-    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
-
-    let body = req.single_request().body_json();
-    let input = body["input"]
-        .as_array()
-        .expect("input array must be present");
-    assert!(
-        !input.is_empty(),
-        "expected at least environment context and user message"
-    );
-
-    let env_msg = &input[1];
-    let env_text = env_msg["content"][0]["text"]
-        .as_str()
-        .expect("environment context text");
-    assert!(
-        env_text.starts_with(ENVIRONMENT_CONTEXT_OPEN_TAG),
-        "second entry should be environment context, got: {env_text}"
-    );
-    assert!(
-        env_text.contains("<approval_policy>never</approval_policy>"),
-        "environment context should reflect overridden approval policy: {env_text}"
-    );
-
-    let env_count = input
-        .iter()
-        .filter(|msg| {
-            msg["content"]
-                .as_array()
-                .and_then(|content| {
-                    content.iter().find(|item| {
-                        item["type"].as_str() == Some("input_text")
-                            && item["text"]
-                                .as_str()
-                                .map(|text| text.starts_with(ENVIRONMENT_CONTEXT_OPEN_TAG))
-                                .unwrap_or(false)
-                    })
-                })
-                .is_some()
-        })
-        .count();
-    assert_eq!(
-        env_count, 2,
-        "environment context should appear exactly twice, found {env_count}"
-    );
-
-    let user_msg = &input[2];
-    let user_text = user_msg["content"][0]["text"]
-        .as_str()
-        .expect("user message text");
-    assert_eq!(user_text, "first message");
-
-    Ok(())
-}
-
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn per_turn_overrides_keep_cached_prefix_and_key_constant() -> anyhow::Result<()> {
     skip_if_no_network!(Ok(()));

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

@@ -82,7 +82,6 @@ async fn review_op_emits_lifecycle_and_review_output() {
             review_request: ReviewRequest {
                 prompt: "Please review my changes".to_string(),
                 user_facing_hint: "my changes".to_string(),
-                append_to_original_thread: true,
             },
         })
         .await
@@ -179,7 +178,6 @@ async fn review_op_with_plain_text_emits_review_fallback() {
             review_request: ReviewRequest {
                 prompt: "Plain text review".to_string(),
                 user_facing_hint: "plain text review".to_string(),
-                append_to_original_thread: true,
             },
         })
         .await
@@ -238,7 +236,6 @@ async fn review_filters_agent_message_related_events() {
             review_request: ReviewRequest {
                 prompt: "Filter streaming events".to_string(),
                 user_facing_hint: "Filter streaming events".to_string(),
-                append_to_original_thread: true,
             },
         })
         .await
@@ -323,7 +320,6 @@ async fn review_does_not_emit_agent_message_on_structured_output() {
             review_request: ReviewRequest {
                 prompt: "check structured".to_string(),
                 user_facing_hint: "check structured".to_string(),
-                append_to_original_thread: true,
             },
         })
         .await
@@ -377,7 +373,6 @@ async fn review_uses_custom_review_model_from_config() {
             review_request: ReviewRequest {
                 prompt: "use custom model".to_string(),
                 user_facing_hint: "use custom model".to_string(),
-                append_to_original_thread: true,
             },
         })
         .await
@@ -495,7 +490,6 @@ async fn review_input_isolated_from_parent_history() {
             review_request: ReviewRequest {
                 prompt: review_prompt.clone(),
                 user_facing_hint: review_prompt.clone(),
-                append_to_original_thread: true,
             },
         })
         .await
@@ -608,7 +602,6 @@ async fn review_history_does_not_leak_into_parent_session() {
             review_request: ReviewRequest {
                 prompt: "Start a review".to_string(),
                 user_facing_hint: "Start a review".to_string(),
-                append_to_original_thread: true,
             },
         })
         .await

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

@@ -101,6 +101,7 @@ fn shell_responses(
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 #[test_case(ShellModelOutput::Shell)]
+#[test_case(ShellModelOutput::ShellCommand)]
 #[test_case(ShellModelOutput::LocalShell)]
 async fn shell_output_stays_json_without_freeform_apply_patch(
     output_type: ShellModelOutput,
@@ -212,6 +213,7 @@ freeform shell
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 #[test_case(ShellModelOutput::Shell)]
+#[test_case(ShellModelOutput::ShellCommand)]
 #[test_case(ShellModelOutput::LocalShell)]
 async fn shell_output_preserves_fixture_json_without_serialization(
     output_type: ShellModelOutput,
@@ -407,6 +409,7 @@ $"#;
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 #[test_case(ShellModelOutput::Shell)]
+#[test_case(ShellModelOutput::ShellCommand)]
 #[test_case(ShellModelOutput::LocalShell)]
 async fn shell_output_reserializes_truncated_content(output_type: ShellModelOutput) -> Result<()> {
     skip_if_no_network!(Ok(()));
@@ -416,7 +419,6 @@ async fn shell_output_reserializes_truncated_content(output_type: ShellModelOutp
         config.model = "gpt-5.1-codex".to_string();
         config.model_family =
             find_family_for_model("gpt-5.1-codex").expect("gpt-5.1-codex is a model family");
-        config.tool_output_token_limit = Some(200);
         if matches!(output_type, ShellModelOutput::ShellCommand) {
             config.features.enable(Feature::ShellCommandTool);
         }
@@ -457,7 +459,10 @@ Output:
 4
 5
 6
-.*…45 tokens truncated….*
+.*
+\[\.{3} omitted \d+ of 400 lines \.{3}\]
+
+.*
 396
 397
 398
@@ -755,7 +760,7 @@ Output:
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn shell_command_output_is_freeform() -> Result<()> {
+async fn shell_command_output_is_structured() -> Result<()> {
     skip_if_no_network!(Ok(()));
 
     let server = start_mock_server().await;
@@ -807,114 +812,6 @@ shell command
     Ok(())
 }
 
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn shell_command_output_is_not_truncated_under_10k_bytes() -> Result<()> {
-    skip_if_no_network!(Ok(()));
-
-    let server = start_mock_server().await;
-    let mut builder = test_codex()
-        .with_model("gpt-5.1")
-        .with_config(move |config| {
-            config.features.enable(Feature::ShellCommandTool);
-        });
-    let test = builder.build(&server).await?;
-
-    let call_id = "shell-command";
-    let args = json!({
-        "command": "perl -e 'print \"1\" x 10000'",
-        "timeout_ms": 1000,
-    });
-    let responses = vec![
-        sse(vec![
-            json!({"type": "response.created", "response": {"id": "resp-1"}}),
-            ev_function_call(call_id, "shell_command", &serde_json::to_string(&args)?),
-            ev_completed("resp-1"),
-        ]),
-        sse(vec![
-            ev_assistant_message("msg-1", "shell_command done"),
-            ev_completed("resp-2"),
-        ]),
-    ];
-    let mock = mount_sse_sequence(&server, responses).await;
-
-    test.submit_turn_with_policy(
-        "run the shell_command script in the user's shell",
-        SandboxPolicy::DangerFullAccess,
-    )
-    .await?;
-
-    let req = mock
-        .last_request()
-        .expect("shell_command output request recorded");
-    let output_item = req.function_call_output(call_id);
-    let output = output_item
-        .get("output")
-        .and_then(Value::as_str)
-        .expect("shell_command output string");
-
-    let expected_pattern = r"(?s)^Exit code: 0
-Wall time: [0-9]+(?:\.[0-9]+)? seconds
-Output:
-1{10000}$";
-    assert_regex_match(expected_pattern, output);
-
-    Ok(())
-}
-
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn shell_command_output_is_not_truncated_over_10k_bytes() -> Result<()> {
-    skip_if_no_network!(Ok(()));
-
-    let server = start_mock_server().await;
-    let mut builder = test_codex()
-        .with_model("gpt-5.1")
-        .with_config(move |config| {
-            config.features.enable(Feature::ShellCommandTool);
-        });
-    let test = builder.build(&server).await?;
-
-    let call_id = "shell-command";
-    let args = json!({
-        "command": "perl -e 'print \"1\" x 10001'",
-        "timeout_ms": 1000,
-    });
-    let responses = vec![
-        sse(vec![
-            json!({"type": "response.created", "response": {"id": "resp-1"}}),
-            ev_function_call(call_id, "shell_command", &serde_json::to_string(&args)?),
-            ev_completed("resp-1"),
-        ]),
-        sse(vec![
-            ev_assistant_message("msg-1", "shell_command done"),
-            ev_completed("resp-2"),
-        ]),
-    ];
-    let mock = mount_sse_sequence(&server, responses).await;
-
-    test.submit_turn_with_policy(
-        "run the shell_command script in the user's shell",
-        SandboxPolicy::DangerFullAccess,
-    )
-    .await?;
-
-    let req = mock
-        .last_request()
-        .expect("shell_command output request recorded");
-    let output_item = req.function_call_output(call_id);
-    let output = output_item
-        .get("output")
-        .and_then(Value::as_str)
-        .expect("shell_command output string");
-
-    let expected_pattern = r"(?s)^Exit code: 0
-Wall time: [0-9]+(?:\.[0-9]+)? seconds
-Output:
-1*…1 chars truncated…1*$";
-    assert_regex_match(expected_pattern, output);
-
-    Ok(())
-}
-
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn local_shell_call_output_is_structured() -> Result<()> {
     skip_if_no_network!(Ok(()));

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

@@ -98,160 +98,6 @@ async fn truncate_function_error_trims_respond_to_model() -> Result<()> {
     Ok(())
 }
 
-// Verifies that a standard tool call (shell) exceeding the model formatting
-// limits is truncated before being sent back to the model.
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn tool_call_output_configured_limit_chars_type() -> Result<()> {
-    skip_if_no_network!(Ok(()));
-
-    let server = start_mock_server().await;
-
-    // Use a model that exposes the generic shell tool.
-    let mut builder = test_codex().with_model("gpt-5.1").with_config(|config| {
-        config.tool_output_token_limit = Some(100_000);
-    });
-
-    let fixture = builder.build(&server).await?;
-
-    let call_id = "shell-too-large";
-    let args = if cfg!(windows) {
-        serde_json::json!({
-            "command": [
-                "powershell",
-                "-Command",
-                "for ($i=1; $i -le 100000; $i++) { Write-Output $i }"
-            ],
-            "timeout_ms": 5_000,
-        })
-    } else {
-        serde_json::json!({
-            "command": ["/bin/sh", "-c", "seq 1 100000"],
-            "timeout_ms": 5_000,
-        })
-    };
-
-    // First response: model tells us to run the tool; second: complete the turn.
-    mount_sse_once(
-        &server,
-        sse(vec![
-            responses::ev_response_created("resp-1"),
-            responses::ev_function_call(call_id, "shell", &serde_json::to_string(&args)?),
-            responses::ev_completed("resp-1"),
-        ]),
-    )
-    .await;
-    let mock2 = mount_sse_once(
-        &server,
-        sse(vec![
-            responses::ev_assistant_message("msg-1", "done"),
-            responses::ev_completed("resp-2"),
-        ]),
-    )
-    .await;
-
-    fixture
-        .submit_turn_with_policy("trigger big shell output", SandboxPolicy::DangerFullAccess)
-        .await?;
-
-    // Inspect what we sent back to the model; it should contain a truncated
-    // function_call_output for the shell call.
-    let output = mock2
-        .single_request()
-        .function_call_output_text(call_id)
-        .context("function_call_output present for shell call")?;
-    let output = output.replace("\r\n", "\n");
-
-    // Expect plain text (not JSON) containing the entire shell output.
-    assert!(
-        serde_json::from_str::<Value>(&output).is_err(),
-        "expected truncated shell output to be plain text"
-    );
-
-    assert_eq!(output.len(), 400094, "we should be almost 100k tokens");
-
-    assert!(
-        !output.contains("tokens truncated"),
-        "shell output should not contain tokens truncated marker: {output}"
-    );
-
-    Ok(())
-}
-
-// Verifies that a standard tool call (shell) exceeding the model formatting
-// limits is truncated before being sent back to the model.
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn tool_call_output_exceeds_limit_truncated_chars_limit() -> Result<()> {
-    skip_if_no_network!(Ok(()));
-
-    let server = start_mock_server().await;
-
-    // Use a model that exposes the generic shell tool.
-    let mut builder = test_codex().with_model("gpt-5.1");
-
-    let fixture = builder.build(&server).await?;
-
-    let call_id = "shell-too-large";
-    let args = if cfg!(windows) {
-        serde_json::json!({
-            "command": [
-                "powershell",
-                "-Command",
-                "for ($i=1; $i -le 100000; $i++) { Write-Output $i }"
-            ],
-            "timeout_ms": 5_000,
-        })
-    } else {
-        serde_json::json!({
-            "command": ["/bin/sh", "-c", "seq 1 100000"],
-            "timeout_ms": 5_000,
-        })
-    };
-
-    // First response: model tells us to run the tool; second: complete the turn.
-    mount_sse_once(
-        &server,
-        sse(vec![
-            responses::ev_response_created("resp-1"),
-            responses::ev_function_call(call_id, "shell", &serde_json::to_string(&args)?),
-            responses::ev_completed("resp-1"),
-        ]),
-    )
-    .await;
-    let mock2 = mount_sse_once(
-        &server,
-        sse(vec![
-            responses::ev_assistant_message("msg-1", "done"),
-            responses::ev_completed("resp-2"),
-        ]),
-    )
-    .await;
-
-    fixture
-        .submit_turn_with_policy("trigger big shell output", SandboxPolicy::DangerFullAccess)
-        .await?;
-
-    // Inspect what we sent back to the model; it should contain a truncated
-    // function_call_output for the shell call.
-    let output = mock2
-        .single_request()
-        .function_call_output_text(call_id)
-        .context("function_call_output present for shell call")?;
-    let output = output.replace("\r\n", "\n");
-
-    // Expect plain text (not JSON) containing the entire shell output.
-    assert!(
-        serde_json::from_str::<Value>(&output).is_err(),
-        "expected truncated shell output to be plain text"
-    );
-
-    assert_eq!(output.len(), 9976); // ~10k characters
-    let truncated_pattern = r#"(?s)^Exit code: 0\nWall time: 0 seconds\nTotal output lines: 100000\n.*?…578898 chars truncated….*$"#;
-
-    assert_regex_match(truncated_pattern, &output);
-
-    Ok(())
-}
-
 // Verifies that a standard tool call (shell) exceeding the model formatting
 // limits is truncated before being sent back to the model.
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
@@ -274,13 +120,13 @@ async fn tool_call_output_exceeds_limit_truncated_for_model() -> Result<()> {
             "command": [
                 "powershell",
                 "-Command",
-                "for ($i=1; $i -le 100000; $i++) { Write-Output $i }"
+                "for ($i=1; $i -le 400; $i++) { Write-Output $i }"
             ],
             "timeout_ms": 5_000,
         })
     } else {
         serde_json::json!({
-            "command": ["/bin/sh", "-c", "seq 1 100000"],
+            "command": ["/bin/sh", "-c", "seq 1 400"],
             "timeout_ms": 5_000,
         })
     };
@@ -316,14 +162,14 @@ async fn tool_call_output_exceeds_limit_truncated_for_model() -> Result<()> {
         .context("function_call_output present for shell call")?;
     let output = output.replace("\r\n", "\n");
 
-    // Expect plain text (not JSON) containing the entire shell output.
+    // Expect plain text (not JSON) with truncation markers and line elision.
     assert!(
         serde_json::from_str::<Value>(&output).is_err(),
         "expected truncated shell output to be plain text"
     );
     let truncated_pattern = r#"(?s)^Exit code: 0
-Wall time: [0-9]+(?:\.[0-9]+)? seconds
-Total output lines: 100000
+Wall time: .* seconds
+Total output lines: 400
 Output:
 1
 2
@@ -331,9 +177,15 @@ Output:
 4
 5
 6
-.*…137224 tokens truncated.*
-99999
-100000
+.*
+\[\.{3} omitted 144 of 400 lines \.{3}\]
+
+.*
+396
+397
+398
+399
+400
 $"#;
     assert_regex_match(truncated_pattern, &output);
 
@@ -359,13 +211,13 @@ async fn tool_call_output_truncated_only_once() -> Result<()> {
             "command": [
                 "powershell",
                 "-Command",
-                "for ($i=1; $i -le 10000; $i++) { Write-Output $i }"
+                "for ($i=1; $i -le 2000; $i++) { Write-Output $i }"
             ],
             "timeout_ms": 5_000,
         })
     } else {
         serde_json::json!({
-            "command": ["/bin/sh", "-c", "seq 1 10000"],
+            "command": ["/bin/sh", "-c", "seq 1 2000"],
             "timeout_ms": 5_000,
         })
     };
@@ -397,11 +249,11 @@ async fn tool_call_output_truncated_only_once() -> Result<()> {
         .function_call_output_text(call_id)
         .context("function_call_output present for shell call")?;
 
-    let truncation_markers = output.matches("tokens truncated").count();
+    let total_line_headers = output.matches("Total output lines:").count();
 
     assert_eq!(
-        truncation_markers, 1,
-        "shell output should carry only one truncation marker: {output}"
+        total_line_headers, 1,
+        "shell output should carry only one truncation header: {output}"
     );
 
     Ok(())
@@ -469,7 +321,6 @@ async fn mcp_tool_call_output_exceeds_limit_truncated_for_model() -> Result<()>
                 disabled_tools: None,
             },
         );
-        config.tool_output_token_limit = Some(500);
     });
     let fixture = builder.build(&server).await?;
 
@@ -486,6 +337,12 @@ async fn mcp_tool_call_output_exceeds_limit_truncated_for_model() -> Result<()>
         .function_call_output_text(call_id)
         .context("function_call_output present for rmcp call")?;
 
+    // Expect plain text with token-based truncation marker; the original JSON body
+    // is truncated in the middle of the echo string.
+    assert!(
+        serde_json::from_str::<Value>(&output).is_err(),
+        "expected truncated MCP output to be plain text"
+    );
     assert!(
         !output.contains("Total output lines:"),
         "MCP output should not include line-based truncation header: {output}"
@@ -493,7 +350,6 @@ async fn mcp_tool_call_output_exceeds_limit_truncated_for_model() -> Result<()>
 
     let truncated_pattern = r#"(?s)^\{"echo":\s*"ECHOING: long-message-with-newlines-.*tokens truncated.*long-message-with-newlines-.*$"#;
     assert_regex_match(truncated_pattern, &output);
-    assert!(output.len() < 2500, "{}", output.len());
 
     Ok(())
 }
@@ -645,9 +501,7 @@ async fn token_policy_marker_reports_tokens() -> Result<()> {
         .function_call_output_text(call_id)
         .context("shell output present")?;
 
-    let pattern = r#"(?s)^\{"output":"Total output lines: 150\\n\\n1\\n2\\n3\\n4\\n5\\n.*?…\d+ tokens truncated…7\\n138\\n139\\n140\\n141\\n142\\n143\\n144\\n145\\n146\\n147\\n148\\n149\\n150\\n","metadata":\{"exit_code":0,"duration_seconds":0\.0\}\}$"#;
-
-    assert_regex_match(pattern, &output);
+    assert_regex_match(r"\[\u{2026}127 tokens truncated\u{2026}]", &output);
 
     Ok(())
 }
@@ -698,16 +552,14 @@ async fn byte_policy_marker_reports_bytes() -> Result<()> {
         .function_call_output_text(call_id)
         .context("shell output present")?;
 
-    let pattern = r#"(?s)^\{"output":"Total output lines: 150\\n\\n1\\n2\\n3\\n4\\n5.*?…\d+ chars truncated…7\\n138\\n139\\n140\\n141\\n142\\n143\\n144\\n145\\n146\\n147\\n148\\n149\\n150\\n","metadata":\{"exit_code":0,"duration_seconds":0\.0\}\}$"#;
-
-    assert_regex_match(pattern, &output);
+    assert_regex_match(r"\[\u{2026}505 bytes truncated\u{2026}]", &output);
 
     Ok(())
 }
 
-// Shell tool output should remain intact when the config opts into a large token budget.
+// Overriding config with a large token budget should avoid truncation.
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn shell_tool_output_not_truncated_with_custom_limit() -> Result<()> {
+async fn large_budget_avoids_truncation() -> Result<()> {
     skip_if_no_network!(Ok(()));
 
     let server = start_mock_server().await;
@@ -724,7 +576,6 @@ async fn shell_tool_output_not_truncated_with_custom_limit() -> Result<()> {
         "command": ["/bin/sh", "-c", "seq 1 1000"],
         "timeout_ms": 5_000,
     });
-    let expected_body: String = (1..=1000).map(|i| format!("{i}\n")).collect();
 
     mount_sse_once(
         &server,
@@ -756,10 +607,6 @@ async fn shell_tool_output_not_truncated_with_custom_limit() -> Result<()> {
         .function_call_output_text(call_id)
         .context("shell output present")?;
 
-    assert!(
-        output.ends_with(&expected_body),
-        "expected entire shell output when budget increased: {output}"
-    );
     assert!(
         !output.contains("truncated"),
         "output should remain untruncated with ample budget"
@@ -767,101 +614,3 @@ async fn shell_tool_output_not_truncated_with_custom_limit() -> Result<()> {
 
     Ok(())
 }
-
-// MCP server output should also remain intact when the config increases the token limit.
-#[tokio::test(flavor = "multi_thread", worker_threads = 1)]
-async fn mcp_tool_call_output_not_truncated_with_custom_limit() -> Result<()> {
-    skip_if_no_network!(Ok(()));
-
-    let server = start_mock_server().await;
-
-    let call_id = "rmcp-untruncated";
-    let server_name = "rmcp";
-    let tool_name = format!("mcp__{server_name}__echo");
-    let large_msg = "a".repeat(80_000);
-    let args_json = serde_json::json!({ "message": large_msg });
-
-    mount_sse_once(
-        &server,
-        sse(vec![
-            responses::ev_response_created("resp-1"),
-            responses::ev_function_call(call_id, &tool_name, &args_json.to_string()),
-            responses::ev_completed("resp-1"),
-        ]),
-    )
-    .await;
-    let mock2 = mount_sse_once(
-        &server,
-        sse(vec![
-            responses::ev_assistant_message("msg-1", "rmcp echo tool completed."),
-            responses::ev_completed("resp-2"),
-        ]),
-    )
-    .await;
-
-    let rmcp_test_server_bin = CargoBuild::new()
-        .package("codex-rmcp-client")
-        .bin("test_stdio_server")
-        .run()?
-        .path()
-        .to_string_lossy()
-        .into_owned();
-
-    let mut builder = test_codex().with_config(move |config| {
-        config.features.enable(Feature::RmcpClient);
-        config.tool_output_token_limit = Some(50_000);
-        config.mcp_servers.insert(
-            server_name.to_string(),
-            codex_core::config::types::McpServerConfig {
-                transport: codex_core::config::types::McpServerTransportConfig::Stdio {
-                    command: rmcp_test_server_bin,
-                    args: Vec::new(),
-                    env: None,
-                    env_vars: Vec::new(),
-                    cwd: None,
-                },
-                enabled: true,
-                startup_timeout_sec: Some(std::time::Duration::from_secs(10)),
-                tool_timeout_sec: None,
-                enabled_tools: None,
-                disabled_tools: None,
-            },
-        );
-    });
-    let fixture = builder.build(&server).await?;
-
-    fixture
-        .submit_turn_with_policy(
-            "call the rmcp echo tool with a very large message",
-            SandboxPolicy::ReadOnly,
-        )
-        .await?;
-
-    let output = mock2
-        .single_request()
-        .function_call_output_text(call_id)
-        .context("function_call_output present for rmcp call")?;
-
-    let parsed: Value = serde_json::from_str(&output)?;
-    assert_eq!(
-        output.len(),
-        80031,
-        "parsed MCP output should retain its serialized length"
-    );
-    let expected_echo = format!("ECHOING: {large_msg}");
-    let echo_str = parsed["echo"]
-        .as_str()
-        .context("echo field should be a string in rmcp echo output")?;
-    assert_eq!(
-        echo_str.len(),
-        expected_echo.len(),
-        "echo length should match"
-    );
-    assert_eq!(echo_str, expected_echo);
-    assert!(
-        !output.contains("truncated"),
-        "output should not include truncation markers when limit is raised: {output}"
-    );
-
-    Ok(())
-}

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

@@ -1586,7 +1586,7 @@ PY
     let large_output = outputs.get(call_id).expect("missing large output summary");
 
     let output_text = large_output.output.replace("\r\n", "\n");
-    let truncated_pattern = r"(?s)^Total output lines: \d+\n\n(token token \n){5,}.*…\d+ tokens truncated…(token token \n){5,}$";
+    let truncated_pattern = r#"(?s)^(token token \n){5,}.*\[\u{2026}\d+ tokens truncated\u{2026}]\n(token token \n){5,}$"#;
     assert_regex_match(truncated_pattern, &output_text);
 
     let original_tokens = large_output

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

@@ -207,16 +207,10 @@ async fn user_shell_command_history_is_persisted_and_shared_with_model() -> anyh
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-#[cfg(not(target_os = "windows"))] // TODO: unignore on windows
 async fn user_shell_command_output_is_truncated_in_history() -> anyhow::Result<()> {
     let server = responses::start_mock_server().await;
-    let builder = core_test_support::test_codex::test_codex();
-    let test = builder
-        .with_config(|config| {
-            config.tool_output_token_limit = Some(100);
-        })
-        .build(&server)
-        .await?;
+    let mut builder = core_test_support::test_codex::test_codex();
+    let test = builder.build(&server).await?;
 
     #[cfg(windows)]
     let command = r#"for ($i=1; $i -le 400; $i++) { Write-Output $i }"#.to_string();
@@ -255,9 +249,10 @@ async fn user_shell_command_output_is_truncated_in_history() -> anyhow::Result<(
         .expect("command message recorded in request");
     let command_message = command_message.replace("\r\n", "\n");
 
-    let head = (1..=69).map(|i| format!("{i}\n")).collect::<String>();
-    let tail = (352..=400).map(|i| format!("{i}\n")).collect::<String>();
-    let truncated_body = format!("Total output lines: 400\n\n{head}…273 tokens truncated…{tail}");
+    let head = (1..=128).map(|i| format!("{i}\n")).collect::<String>();
+    let tail = (273..=400).map(|i| format!("{i}\n")).collect::<String>();
+    let truncated_body =
+        format!("Total output lines: 400\n\n{head}\n[... omitted 144 of 400 lines ...]\n\n{tail}");
     let escaped_command = escape(&command);
     let escaped_truncated_body = escape(&truncated_body);
     let expected_pattern = format!(
@@ -275,7 +270,6 @@ async fn user_shell_command_is_truncated_only_once() -> anyhow::Result<()> {
     let server = start_mock_server().await;
 
     let mut builder = test_codex().with_config(|config| {
-        config.tool_output_token_limit = Some(100);
         config.model = "gpt-5.1-codex".to_string();
         config.model_family =
             find_family_for_model("gpt-5-codex").expect("gpt-5-codex is a model family");

codex-rs/exec-server/Cargo.toml

@@ -1,47 +0,0 @@
-[package]
-edition = "2024"
-name = "codex-exec-server"
-version = { workspace = true }
-
-[[bin]]
-name = "codex-exec-server"
-path = "src/main.rs"
-
-[lints]
-workspace = true
-
-[dependencies]
-anyhow = { workspace = true }
-clap = { workspace = true, features = ["derive"] }
-codex-core = { workspace = true }
-libc = { workspace = true }
-path-absolutize = { workspace = true }
-rmcp = { workspace = true, default-features = false, features = [
-    "auth",
-    "elicitation",
-    "base64",
-    "client",
-    "macros",
-    "schemars",
-    "server",
-    "transport-child-process",
-    "transport-streamable-http-client-reqwest",
-    "transport-streamable-http-server",
-    "transport-io",
-] }
-serde = { workspace = true, features = ["derive"] }
-serde_json = { workspace = true }
-socket2 = { workspace = true }
-tokio = { workspace = true, features = [
-    "io-std",
-    "macros",
-    "process",
-    "rt-multi-thread",
-    "signal",
-] }
-tracing = { workspace = true }
-tracing-subscriber = { workspace = true, features = ["env-filter", "fmt"] }
-
-[dev-dependencies]
-pretty_assertions = { workspace = true }
-tempfile = { workspace = true }

codex-rs/exec-server/src/main.rs

@@ -1,11 +0,0 @@
-#[cfg(target_os = "windows")]
-fn main() {
-    eprintln!("codex-exec-server is not implemented on Windows targets");
-    std::process::exit(1);
-}
-
-#[cfg(not(target_os = "windows"))]
-mod posix;
-
-#[cfg(not(target_os = "windows"))]
-pub use posix::main;

codex-rs/exec-server/src/posix.rs

@@ -1,164 +0,0 @@
-//! This is an MCP that implements an alternative `shell` tool with fine-grained privilege
-//! escalation based on a per-exec() policy.
-//!
-//! We spawn Bash process inside a sandbox. The Bash we spawn is patched to allow us to intercept
-//! every exec() call it makes by invoking a wrapper program and passing in the arguments it would
-//! have passed to exec(). The Bash process (and its descendants) inherit a communication socket
-//! from us, and we give its fd number in the CODEX_ESCALATE_SOCKET environment variable.
-//!
-//! When we intercept an exec() call, we send a message over the socket back to the main
-//! MCP process. The MCP process can then decide whether to allow the exec() call to proceed
-//! or to escalate privileges and run the requested command with elevated permissions. In the
-//! latter case, we send a message back to the child requesting that it forward its open FDs to us.
-//! We then execute the requested command on its behalf, patching in the forwarded FDs.
-//!
-//!
-//! ### The privilege escalation flow
-//!
-//! Child  MCP   Bash   Escalate Helper
-//!         |
-//!         o----->o
-//!         |      |
-//!         |      o--(exec)-->o
-//!         |      |           |
-//!         |o<-(EscalateReq)--o
-//!         ||     |           |
-//!         |o--(Escalate)---->o
-//!         ||     |           |
-//!         |o<---------(fds)--o
-//!         ||     |           |
-//!   o<-----o     |           |
-//!   |     ||     |           |
-//!   x----->o     |           |
-//!         ||     |           |
-//!         |x--(exit code)--->o
-//!         |      |           |
-//!         |      o<--(exit)--x
-//!         |      |
-//!         o<-----x
-//!
-//! ### The non-escalation flow
-//!
-//!  MCP   Bash   Escalate Helper   Child
-//!   |
-//!   o----->o
-//!   |      |
-//!   |      o--(exec)-->o
-//!   |      |           |
-//!   |o<-(EscalateReq)--o
-//!   ||     |           |
-//!   |o-(Run)---------->o
-//!   |      |           |
-//!   |      |           x--(exec)-->o
-//!   |      |                       |
-//!   |      o<--------------(exit)--x
-//!   |      |
-//!   o<-----x
-//!
-use std::path::Path;
-
-use clap::Parser;
-use clap::Subcommand;
-use tracing_subscriber::EnvFilter;
-use tracing_subscriber::{self};
-
-use crate::posix::escalate_protocol::EscalateAction;
-use crate::posix::escalate_server::EscalateServer;
-
-mod escalate_client;
-mod escalate_protocol;
-mod escalate_server;
-mod mcp;
-mod socket;
-
-fn dummy_exec_policy(file: &Path, argv: &[String], _workdir: &Path) -> EscalateAction {
-    // TODO: execpolicy
-    if file == Path::new("/opt/homebrew/bin/gh")
-        && let [_, arg1, arg2, ..] = argv
-        && arg1 == "issue"
-        && arg2 == "list"
-    {
-        return EscalateAction::Escalate;
-    }
-    EscalateAction::Run
-}
-
-#[derive(Parser)]
-#[command(version)]
-pub struct Cli {
-    #[command(subcommand)]
-    subcommand: Option<Commands>,
-}
-
-#[derive(Subcommand)]
-enum Commands {
-    Escalate(EscalateArgs),
-    ShellExec(ShellExecArgs),
-}
-
-/// Invoked from within the sandbox to (potentially) escalate permissions.
-#[derive(Parser, Debug)]
-struct EscalateArgs {
-    file: String,
-
-    #[arg(trailing_var_arg = true)]
-    argv: Vec<String>,
-}
-
-impl EscalateArgs {
-    /// This is the escalate client. It talks to the escalate server to determine whether to exec()
-    /// the command directly or to proxy to the escalate server.
-    async fn run(self) -> anyhow::Result<i32> {
-        let EscalateArgs { file, argv } = self;
-        escalate_client::run(file, argv).await
-    }
-}
-
-/// Debugging command to emulate an MCP "shell" tool call.
-#[derive(Parser, Debug)]
-struct ShellExecArgs {
-    command: String,
-}
-
-#[tokio::main]
-pub async fn main() -> anyhow::Result<()> {
-    let cli = Cli::parse();
-    tracing_subscriber::fmt()
-        .with_env_filter(EnvFilter::from_default_env())
-        .with_writer(std::io::stderr)
-        .with_ansi(false)
-        .init();
-
-    match cli.subcommand {
-        Some(Commands::Escalate(args)) => {
-            std::process::exit(args.run().await?);
-        }
-        Some(Commands::ShellExec(args)) => {
-            let bash_path = mcp::get_bash_path()?;
-            let escalate_server = EscalateServer::new(bash_path, dummy_exec_policy);
-            let result = escalate_server
-                .exec(
-                    args.command.clone(),
-                    std::env::vars().collect(),
-                    std::env::current_dir()?,
-                    None,
-                )
-                .await?;
-            println!("{result:?}");
-            std::process::exit(result.exit_code);
-        }
-        None => {
-            let bash_path = mcp::get_bash_path()?;
-
-            tracing::info!("Starting MCP server");
-            let service = mcp::serve(bash_path, dummy_exec_policy)
-                .await
-                .inspect_err(|e| {
-                    tracing::error!("serving error: {:?}", e);
-                })?;
-
-            service.waiting().await?;
-            Ok(())
-        }
-    }
-}

codex-rs/exec-server/src/posix/escalate_client.rs

@@ -1,102 +0,0 @@
-use std::io;
-use std::os::fd::AsRawFd;
-use std::os::fd::FromRawFd as _;
-use std::os::fd::OwnedFd;
-
-use anyhow::Context as _;
-
-use crate::posix::escalate_protocol::BASH_EXEC_WRAPPER_ENV_VAR;
-use crate::posix::escalate_protocol::ESCALATE_SOCKET_ENV_VAR;
-use crate::posix::escalate_protocol::EscalateAction;
-use crate::posix::escalate_protocol::EscalateRequest;
-use crate::posix::escalate_protocol::EscalateResponse;
-use crate::posix::escalate_protocol::SuperExecMessage;
-use crate::posix::escalate_protocol::SuperExecResult;
-use crate::posix::socket::AsyncDatagramSocket;
-use crate::posix::socket::AsyncSocket;
-
-fn get_escalate_client() -> anyhow::Result<AsyncDatagramSocket> {
-    // TODO: we should defensively require only calling this once, since AsyncSocket will take ownership of the fd.
-    let client_fd = std::env::var(ESCALATE_SOCKET_ENV_VAR)?.parse::<i32>()?;
-    if client_fd < 0 {
-        return Err(anyhow::anyhow!(
-            "{ESCALATE_SOCKET_ENV_VAR} is not a valid file descriptor: {client_fd}"
-        ));
-    }
-    Ok(unsafe { AsyncDatagramSocket::from_raw_fd(client_fd) }?)
-}
-
-pub(crate) async fn run(file: String, argv: Vec<String>) -> anyhow::Result<i32> {
-    let handshake_client = get_escalate_client()?;
-    let (server, client) = AsyncSocket::pair()?;
-    const HANDSHAKE_MESSAGE: [u8; 1] = [0];
-    handshake_client
-        .send_with_fds(&HANDSHAKE_MESSAGE, &[server.into_inner().into()])
-        .await
-        .context("failed to send handshake datagram")?;
-    let env = std::env::vars()
-        .filter(|(k, _)| {
-            !matches!(
-                k.as_str(),
-                ESCALATE_SOCKET_ENV_VAR | BASH_EXEC_WRAPPER_ENV_VAR
-            )
-        })
-        .collect();
-    client
-        .send(EscalateRequest {
-            file: file.clone().into(),
-            argv: argv.clone(),
-            workdir: std::env::current_dir()?,
-            env,
-        })
-        .await
-        .context("failed to send EscalateRequest")?;
-    let message = client.receive::<EscalateResponse>().await?;
-    match message.action {
-        EscalateAction::Escalate => {
-            // TODO: maybe we should send ALL open FDs (except the escalate client)?
-            let fds_to_send = [
-                unsafe { OwnedFd::from_raw_fd(io::stdin().as_raw_fd()) },
-                unsafe { OwnedFd::from_raw_fd(io::stdout().as_raw_fd()) },
-                unsafe { OwnedFd::from_raw_fd(io::stderr().as_raw_fd()) },
-            ];
-
-            // TODO: also forward signals over the super-exec socket
-
-            client
-                .send_with_fds(
-                    SuperExecMessage {
-                        fds: fds_to_send.iter().map(AsRawFd::as_raw_fd).collect(),
-                    },
-                    &fds_to_send,
-                )
-                .await
-                .context("failed to send SuperExecMessage")?;
-            let SuperExecResult { exit_code } = client.receive::<SuperExecResult>().await?;
-            Ok(exit_code)
-        }
-        EscalateAction::Run => {
-            // We avoid std::process::Command here because we want to be as transparent as
-            // possible. std::os::unix::process::CommandExt has .exec() but it does some funky
-            // stuff with signal masks and dup2() on its standard FDs, which we don't want.
-            use std::ffi::CString;
-            let file = CString::new(file).context("NUL in file")?;
-
-            let argv_cstrs: Vec<CString> = argv
-                .iter()
-                .map(|s| CString::new(s.as_str()).context("NUL in argv"))
-                .collect::<Result<Vec<_>, _>>()?;
-
-            let mut argv: Vec<*const libc::c_char> =
-                argv_cstrs.iter().map(|s| s.as_ptr()).collect();
-            argv.push(std::ptr::null());
-
-            let err = unsafe {
-                libc::execv(file.as_ptr(), argv.as_ptr());
-                std::io::Error::last_os_error()
-            };
-
-            Err(err.into())
-        }
-    }
-}

codex-rs/exec-server/src/posix/escalate_protocol.rs

@@ -1,49 +0,0 @@
-use std::collections::HashMap;
-use std::os::fd::RawFd;
-use std::path::PathBuf;
-
-use serde::Deserialize;
-use serde::Serialize;
-
-/// 'exec-server escalate' reads this to find the inherited FD for the escalate socket.
-pub(super) const ESCALATE_SOCKET_ENV_VAR: &str = "CODEX_ESCALATE_SOCKET";
-
-/// The patched bash uses this to wrap exec() calls.
-pub(super) const BASH_EXEC_WRAPPER_ENV_VAR: &str = "BASH_EXEC_WRAPPER";
-
-/// The client sends this to the server to request an exec() call.
-#[derive(Clone, Serialize, Deserialize, Debug, PartialEq, Eq)]
-pub(super) struct EscalateRequest {
-    /// The absolute path to the executable to run, i.e. the first arg to exec.
-    pub(super) file: PathBuf,
-    /// The argv, including the program name (argv[0]).
-    pub(super) argv: Vec<String>,
-    pub(super) workdir: PathBuf,
-    pub(super) env: HashMap<String, String>,
-}
-
-/// The server sends this to the client to respond to an exec() request.
-#[derive(Clone, Serialize, Deserialize, Debug, PartialEq, Eq)]
-pub(super) struct EscalateResponse {
-    pub(super) action: EscalateAction,
-}
-
-#[derive(Clone, Serialize, Deserialize, Debug, PartialEq, Eq)]
-pub(super) enum EscalateAction {
-    /// The command should be run directly by the client.
-    Run,
-    /// The command should be escalated to the server for execution.
-    Escalate,
-}
-
-/// The client sends this to the server to forward its open FDs.
-#[derive(Clone, Serialize, Deserialize, Debug)]
-pub(super) struct SuperExecMessage {
-    pub(super) fds: Vec<RawFd>,
-}
-
-/// The server responds when the exec()'d command has exited.
-#[derive(Clone, Serialize, Deserialize, Debug)]
-pub(super) struct SuperExecResult {
-    pub(super) exit_code: i32,
-}

codex-rs/exec-server/src/posix/escalate_server.rs

@@ -1,274 +0,0 @@
-use std::collections::HashMap;
-use std::os::fd::AsRawFd;
-use std::path::Path;
-use std::path::PathBuf;
-use std::process::Stdio;
-use std::time::Duration;
-
-use anyhow::Context as _;
-use path_absolutize::Absolutize as _;
-
-use codex_core::exec::SandboxType;
-use codex_core::exec::process_exec_tool_call;
-use codex_core::get_platform_sandbox;
-use codex_core::protocol::SandboxPolicy;
-use tokio::process::Command;
-
-use crate::posix::escalate_protocol::BASH_EXEC_WRAPPER_ENV_VAR;
-use crate::posix::escalate_protocol::ESCALATE_SOCKET_ENV_VAR;
-use crate::posix::escalate_protocol::EscalateAction;
-use crate::posix::escalate_protocol::EscalateRequest;
-use crate::posix::escalate_protocol::EscalateResponse;
-use crate::posix::escalate_protocol::SuperExecMessage;
-use crate::posix::escalate_protocol::SuperExecResult;
-use crate::posix::socket::AsyncDatagramSocket;
-use crate::posix::socket::AsyncSocket;
-
-/// This is the policy which decides how to handle an exec() call.
-///
-/// `file` is the absolute, canonical path to the executable to run, i.e. the first arg to exec.
-/// `argv` is the argv, including the program name (`argv[0]`).
-/// `workdir` is the absolute, canonical path to the working directory in which to execute the
-/// command.
-pub(crate) type ExecPolicy = fn(file: &Path, argv: &[String], workdir: &Path) -> EscalateAction;
-
-pub(crate) struct EscalateServer {
-    bash_path: PathBuf,
-    policy: ExecPolicy,
-}
-
-impl EscalateServer {
-    pub fn new(bash_path: PathBuf, policy: ExecPolicy) -> Self {
-        Self { bash_path, policy }
-    }
-
-    pub async fn exec(
-        &self,
-        command: String,
-        env: HashMap<String, String>,
-        workdir: PathBuf,
-        timeout_ms: Option<u64>,
-    ) -> anyhow::Result<ExecResult> {
-        let (escalate_server, escalate_client) = AsyncDatagramSocket::pair()?;
-        let client_socket = escalate_client.into_inner();
-        client_socket.set_cloexec(false)?;
-
-        let escalate_task = tokio::spawn(escalate_task(escalate_server, self.policy));
-        let mut env = env.clone();
-        env.insert(
-            ESCALATE_SOCKET_ENV_VAR.to_string(),
-            client_socket.as_raw_fd().to_string(),
-        );
-        env.insert(
-            BASH_EXEC_WRAPPER_ENV_VAR.to_string(),
-            format!("{} escalate", std::env::current_exe()?.to_string_lossy()),
-        );
-        let result = process_exec_tool_call(
-            codex_core::exec::ExecParams {
-                command: vec![
-                    self.bash_path.to_string_lossy().to_string(),
-                    "-c".to_string(),
-                    command,
-                ],
-                cwd: PathBuf::from(&workdir),
-                timeout_ms,
-                env,
-                with_escalated_permissions: None,
-                justification: None,
-                arg0: None,
-            },
-            get_platform_sandbox().unwrap_or(SandboxType::None),
-            // TODO: use the sandbox policy and cwd from the calling client
-            &SandboxPolicy::ReadOnly,
-            &PathBuf::from("/__NONEXISTENT__"), // This is ignored for ReadOnly
-            &None,
-            None,
-        )
-        .await?;
-        escalate_task.abort();
-        let result = ExecResult {
-            exit_code: result.exit_code,
-            output: result.aggregated_output.text,
-            duration: result.duration,
-            timed_out: result.timed_out,
-        };
-        Ok(result)
-    }
-}
-
-async fn escalate_task(socket: AsyncDatagramSocket, policy: ExecPolicy) -> anyhow::Result<()> {
-    loop {
-        let (_, mut fds) = socket.receive_with_fds().await?;
-        if fds.len() != 1 {
-            tracing::error!("expected 1 fd in datagram handshake, got {}", fds.len());
-            continue;
-        }
-        let stream_socket = AsyncSocket::from_fd(fds.remove(0))?;
-        tokio::spawn(async move {
-            if let Err(err) = handle_escalate_session_with_policy(stream_socket, policy).await {
-                tracing::error!("escalate session failed: {err:?}");
-            }
-        });
-    }
-}
-
-#[derive(Debug)]
-pub(crate) struct ExecResult {
-    pub(crate) exit_code: i32,
-    pub(crate) output: String,
-    pub(crate) duration: Duration,
-    pub(crate) timed_out: bool,
-}
-
-async fn handle_escalate_session_with_policy(
-    socket: AsyncSocket,
-    policy: ExecPolicy,
-) -> anyhow::Result<()> {
-    let EscalateRequest {
-        file,
-        argv,
-        workdir,
-        env,
-    } = socket.receive::<EscalateRequest>().await?;
-    let file = PathBuf::from(&file).absolutize()?.into_owned();
-    let workdir = PathBuf::from(&workdir).absolutize()?.into_owned();
-    let action = policy(file.as_path(), &argv, &workdir);
-    tracing::debug!("decided {action:?} for {file:?} {argv:?} {workdir:?}");
-    match action {
-        EscalateAction::Run => {
-            socket
-                .send(EscalateResponse {
-                    action: EscalateAction::Run,
-                })
-                .await?;
-        }
-        EscalateAction::Escalate => {
-            socket
-                .send(EscalateResponse {
-                    action: EscalateAction::Escalate,
-                })
-                .await?;
-            let (msg, fds) = socket
-                .receive_with_fds::<SuperExecMessage>()
-                .await
-                .context("failed to receive SuperExecMessage")?;
-            if fds.len() != msg.fds.len() {
-                return Err(anyhow::anyhow!(
-                    "mismatched number of fds in SuperExecMessage: {} in the message, {} from the control message",
-                    msg.fds.len(),
-                    fds.len()
-                ));
-            }
-
-            if msg
-                .fds
-                .iter()
-                .any(|src_fd| fds.iter().any(|dst_fd| dst_fd.as_raw_fd() == *src_fd))
-            {
-                return Err(anyhow::anyhow!(
-                    "overlapping fds not yet supported in SuperExecMessage"
-                ));
-            }
-
-            let mut command = Command::new(file);
-            command
-                .args(&argv[1..])
-                .arg0(argv[0].clone())
-                .envs(&env)
-                .current_dir(&workdir)
-                .stdin(Stdio::null())
-                .stdout(Stdio::null())
-                .stderr(Stdio::null());
-            unsafe {
-                command.pre_exec(move || {
-                    for (dst_fd, src_fd) in msg.fds.iter().zip(&fds) {
-                        libc::dup2(src_fd.as_raw_fd(), *dst_fd);
-                    }
-                    Ok(())
-                });
-            }
-            let mut child = command.spawn()?;
-            let exit_status = child.wait().await?;
-            socket
-                .send(SuperExecResult {
-                    exit_code: exit_status.code().unwrap_or(127),
-                })
-                .await?;
-        }
-    }
-    Ok(())
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use pretty_assertions::assert_eq;
-    use std::collections::HashMap;
-    use std::path::PathBuf;
-
-    #[tokio::test]
-    async fn handle_escalate_session_respects_run_in_sandbox_decision() -> anyhow::Result<()> {
-        let (server, client) = AsyncSocket::pair()?;
-        let server_task = tokio::spawn(handle_escalate_session_with_policy(
-            server,
-            |_file, _argv, _workdir| EscalateAction::Run,
-        ));
-
-        client
-            .send(EscalateRequest {
-                file: PathBuf::from("/bin/echo"),
-                argv: vec!["echo".to_string()],
-                workdir: PathBuf::from("/tmp"),
-                env: HashMap::new(),
-            })
-            .await?;
-
-        let response = client.receive::<EscalateResponse>().await?;
-        assert_eq!(
-            EscalateResponse {
-                action: EscalateAction::Run,
-            },
-            response
-        );
-        server_task.await?
-    }
-
-    #[tokio::test]
-    async fn handle_escalate_session_executes_escalated_command() -> anyhow::Result<()> {
-        let (server, client) = AsyncSocket::pair()?;
-        let server_task = tokio::spawn(handle_escalate_session_with_policy(
-            server,
-            |_file, _argv, _workdir| EscalateAction::Escalate,
-        ));
-
-        client
-            .send(EscalateRequest {
-                file: PathBuf::from("/bin/sh"),
-                argv: vec![
-                    "sh".to_string(),
-                    "-c".to_string(),
-                    r#"if [ "$KEY" = VALUE ]; then exit 42; else exit 1; fi"#.to_string(),
-                ],
-                workdir: std::env::current_dir()?,
-                env: HashMap::from([("KEY".to_string(), "VALUE".to_string())]),
-            })
-            .await?;
-
-        let response = client.receive::<EscalateResponse>().await?;
-        assert_eq!(
-            EscalateResponse {
-                action: EscalateAction::Escalate,
-            },
-            response
-        );
-
-        client
-            .send_with_fds(SuperExecMessage { fds: Vec::new() }, &[])
-            .await?;
-
-        let result = client.receive::<SuperExecResult>().await?;
-        assert_eq!(42, result.exit_code);
-
-        server_task.await?
-    }
-}

codex-rs/exec-server/src/posix/mcp.rs

@@ -1,154 +0,0 @@
-use std::path::PathBuf;
-use std::time::Duration;
-
-use anyhow::Context as _;
-use anyhow::Result;
-use rmcp::ErrorData as McpError;
-use rmcp::RoleServer;
-use rmcp::ServerHandler;
-use rmcp::ServiceExt;
-use rmcp::handler::server::router::tool::ToolRouter;
-use rmcp::handler::server::wrapper::Parameters;
-use rmcp::model::*;
-use rmcp::schemars;
-use rmcp::service::RequestContext;
-use rmcp::service::RunningService;
-use rmcp::tool;
-use rmcp::tool_handler;
-use rmcp::tool_router;
-use rmcp::transport::stdio;
-
-use crate::posix::escalate_server;
-use crate::posix::escalate_server::EscalateServer;
-use crate::posix::escalate_server::ExecPolicy;
-
-/// Path to our patched bash.
-const CODEX_BASH_PATH_ENV_VAR: &str = "CODEX_BASH_PATH";
-
-pub(crate) fn get_bash_path() -> Result<PathBuf> {
-    std::env::var(CODEX_BASH_PATH_ENV_VAR)
-        .map(PathBuf::from)
-        .context(format!("{CODEX_BASH_PATH_ENV_VAR} must be set"))
-}
-
-#[derive(Debug, serde::Deserialize, schemars::JsonSchema)]
-pub struct ExecParams {
-    /// The bash string to execute.
-    pub command: String,
-    /// The working directory to execute the command in. Must be an absolute path.
-    pub workdir: String,
-    /// The timeout for the command in milliseconds.
-    pub timeout_ms: Option<u64>,
-}
-
-#[derive(Debug, serde::Serialize, schemars::JsonSchema)]
-pub struct ExecResult {
-    pub exit_code: i32,
-    pub output: String,
-    pub duration: Duration,
-    pub timed_out: bool,
-}
-
-impl From<escalate_server::ExecResult> for ExecResult {
-    fn from(result: escalate_server::ExecResult) -> Self {
-        Self {
-            exit_code: result.exit_code,
-            output: result.output,
-            duration: result.duration,
-            timed_out: result.timed_out,
-        }
-    }
-}
-
-#[derive(Clone)]
-pub struct ExecTool {
-    tool_router: ToolRouter<ExecTool>,
-    bash_path: PathBuf,
-    policy: ExecPolicy,
-}
-
-#[tool_router]
-impl ExecTool {
-    pub fn new(bash_path: PathBuf, policy: ExecPolicy) -> Self {
-        Self {
-            tool_router: Self::tool_router(),
-            bash_path,
-            policy,
-        }
-    }
-
-    /// Runs a shell command and returns its output. You MUST provide the workdir as an absolute path.
-    #[tool]
-    async fn shell(
-        &self,
-        _context: RequestContext<RoleServer>,
-        Parameters(params): Parameters<ExecParams>,
-    ) -> Result<CallToolResult, McpError> {
-        let escalate_server = EscalateServer::new(self.bash_path.clone(), self.policy);
-        let result = escalate_server
-            .exec(
-                params.command,
-                // TODO: use ShellEnvironmentPolicy
-                std::env::vars().collect(),
-                PathBuf::from(&params.workdir),
-                params.timeout_ms,
-            )
-            .await
-            .map_err(|e| McpError::internal_error(e.to_string(), None))?;
-        Ok(CallToolResult::success(vec![Content::json(
-            ExecResult::from(result),
-        )?]))
-    }
-
-    #[allow(dead_code)]
-    async fn prompt(
-        &self,
-        command: String,
-        workdir: String,
-        context: RequestContext<RoleServer>,
-    ) -> Result<CreateElicitationResult, McpError> {
-        context
-            .peer
-            .create_elicitation(CreateElicitationRequestParam {
-                message: format!("Allow Codex to run `{command:?}` in `{workdir:?}`?"),
-                #[allow(clippy::expect_used)]
-                requested_schema: ElicitationSchema::builder()
-                    .property("dummy", PrimitiveSchema::String(StringSchema::new()))
-                    .build()
-                    .expect("failed to build elicitation schema"),
-            })
-            .await
-            .map_err(|e| McpError::internal_error(e.to_string(), None))
-    }
-}
-
-#[tool_handler]
-impl ServerHandler for ExecTool {
-    fn get_info(&self) -> ServerInfo {
-        ServerInfo {
-            protocol_version: ProtocolVersion::V_2025_06_18,
-            capabilities: ServerCapabilities::builder().enable_tools().build(),
-            server_info: Implementation::from_build_env(),
-            instructions: Some(
-                "This server provides a tool to execute shell commands and return their output."
-                    .to_string(),
-            ),
-        }
-    }
-
-    async fn initialize(
-        &self,
-        _request: InitializeRequestParam,
-        _context: RequestContext<RoleServer>,
-    ) -> Result<InitializeResult, McpError> {
-        Ok(self.get_info())
-    }
-}
-
-pub(crate) async fn serve(
-    bash_path: PathBuf,
-    policy: ExecPolicy,
-) -> Result<RunningService<RoleServer, ExecTool>, rmcp::service::ServerInitializeError> {
-    let tool = ExecTool::new(bash_path, policy);
-    tool.serve(stdio()).await
-}

codex-rs/exec-server/src/posix/socket.rs

@@ -1,486 +0,0 @@
-use libc::c_uint;
-use serde::Deserialize;
-use serde::Serialize;
-use socket2::Domain;
-use socket2::MaybeUninitSlice;
-use socket2::MsgHdr;
-use socket2::MsgHdrMut;
-use socket2::Socket;
-use socket2::Type;
-use std::io::IoSlice;
-use std::mem::MaybeUninit;
-use std::os::fd::AsRawFd;
-use std::os::fd::FromRawFd;
-use std::os::fd::OwnedFd;
-use std::os::fd::RawFd;
-use tokio::io::Interest;
-use tokio::io::unix::AsyncFd;
-
-const MAX_FDS_PER_MESSAGE: usize = 16;
-const LENGTH_PREFIX_SIZE: usize = size_of::<u32>();
-const MAX_DATAGRAM_SIZE: usize = 8192;
-
-/// Converts a slice of MaybeUninit<T> to a slice of T.
-///
-/// The caller guarantees that every element of `buf` is initialized.
-fn assume_init<T>(buf: &[MaybeUninit<T>]) -> &[T] {
-    unsafe { std::slice::from_raw_parts(buf.as_ptr().cast(), buf.len()) }
-}
-
-fn assume_init_slice<T, const N: usize>(buf: &[MaybeUninit<T>; N]) -> &[T; N] {
-    unsafe { &*(buf as *const [MaybeUninit<T>; N] as *const [T; N]) }
-}
-
-fn assume_init_vec<T>(mut buf: Vec<MaybeUninit<T>>) -> Vec<T> {
-    unsafe {
-        let ptr = buf.as_mut_ptr() as *mut T;
-        let len = buf.len();
-        let cap = buf.capacity();
-        std::mem::forget(buf);
-        Vec::from_raw_parts(ptr, len, cap)
-    }
-}
-
-fn control_space_for_fds(count: usize) -> usize {
-    unsafe { libc::CMSG_SPACE((count * size_of::<RawFd>()) as _) as usize }
-}
-
-/// Extracts the FDs from a SCM_RIGHTS control message.
-fn extract_fds(control: &[u8]) -> Vec<OwnedFd> {
-    let mut fds = Vec::new();
-    let mut hdr: libc::msghdr = unsafe { std::mem::zeroed() };
-    hdr.msg_control = control.as_ptr() as *mut libc::c_void;
-    hdr.msg_controllen = control.len() as _;
-    let hdr = hdr; // drop mut
-
-    let mut cmsg = unsafe { libc::CMSG_FIRSTHDR(&hdr) as *const libc::cmsghdr };
-    while !cmsg.is_null() {
-        let level = unsafe { (*cmsg).cmsg_level };
-        let ty = unsafe { (*cmsg).cmsg_type };
-        if level == libc::SOL_SOCKET && ty == libc::SCM_RIGHTS {
-            let data_ptr = unsafe { libc::CMSG_DATA(cmsg).cast::<RawFd>() };
-            let fd_count: usize = {
-                let cmsg_data_len =
-                    unsafe { (*cmsg).cmsg_len as usize } - unsafe { libc::CMSG_LEN(0) as usize };
-                cmsg_data_len / size_of::<RawFd>()
-            };
-            for i in 0..fd_count {
-                let fd = unsafe { data_ptr.add(i).read() };
-                fds.push(unsafe { OwnedFd::from_raw_fd(fd) });
-            }
-        }
-        cmsg = unsafe { libc::CMSG_NXTHDR(&hdr, cmsg) };
-    }
-    fds
-}
-
-/// Read a frame from a SOCK_STREAM socket.
-///
-/// A frame is a message length prefix followed by a payload. FDs may be included in the control
-/// message when receiving the frame header.
-async fn read_frame(async_socket: &AsyncFd<Socket>) -> std::io::Result<(Vec<u8>, Vec<OwnedFd>)> {
-    let (message_len, fds) = read_frame_header(async_socket).await?;
-    let payload = read_frame_payload(async_socket, message_len).await?;
-    Ok((payload, fds))
-}
-
-/// Read the frame header (i.e. length) and any FDs from a SOCK_STREAM socket.
-async fn read_frame_header(
-    async_socket: &AsyncFd<Socket>,
-) -> std::io::Result<(usize, Vec<OwnedFd>)> {
-    let mut header = [MaybeUninit::<u8>::uninit(); LENGTH_PREFIX_SIZE];
-    let mut filled = 0;
-    let mut control = vec![MaybeUninit::<u8>::uninit(); control_space_for_fds(MAX_FDS_PER_MESSAGE)];
-    let mut captured_control = false;
-
-    while filled < LENGTH_PREFIX_SIZE {
-        let mut guard = async_socket.readable().await?;
-        // The first read should come with a control message containing any FDs.
-        let result = if !captured_control {
-            guard.try_io(|inner| {
-                let mut bufs = [MaybeUninitSlice::new(&mut header[filled..])];
-                let (read, control_len) = {
-                    let mut msg = MsgHdrMut::new()
-                        .with_buffers(&mut bufs)
-                        .with_control(&mut control);
-                    let read = inner.get_ref().recvmsg(&mut msg, 0)?;
-                    (read, msg.control_len())
-                };
-                control.truncate(control_len);
-                captured_control = true;
-                Ok(read)
-            })
-        } else {
-            guard.try_io(|inner| inner.get_ref().recv(&mut header[filled..]))
-        };
-        let Ok(result) = result else {
-            // Would block, try again.
-            continue;
-        };
-
-        let read = result?;
-        if read == 0 {
-            return Err(std::io::Error::new(
-                std::io::ErrorKind::UnexpectedEof,
-                "socket closed while receiving frame header",
-            ));
-        }
-
-        filled += read;
-        assert!(filled <= LENGTH_PREFIX_SIZE);
-        if filled == LENGTH_PREFIX_SIZE {
-            let len_bytes = assume_init_slice(&header);
-            let payload_len = u32::from_le_bytes(*len_bytes) as usize;
-            let fds = extract_fds(assume_init(&control));
-            return Ok((payload_len, fds));
-        }
-    }
-    unreachable!("header loop always returns")
-}
-
-/// Read `message_len` bytes from a SOCK_STREAM socket.
-async fn read_frame_payload(
-    async_socket: &AsyncFd<Socket>,
-    message_len: usize,
-) -> std::io::Result<Vec<u8>> {
-    if message_len == 0 {
-        return Ok(Vec::new());
-    }
-    let mut payload = vec![MaybeUninit::<u8>::uninit(); message_len];
-    let mut filled = 0;
-    while filled < message_len {
-        let mut guard = async_socket.readable().await?;
-        let result = guard.try_io(|inner| inner.get_ref().recv(&mut payload[filled..]));
-        let Ok(result) = result else {
-            // Would block, try again.
-            continue;
-        };
-        let read = result?;
-        if read == 0 {
-            return Err(std::io::Error::new(
-                std::io::ErrorKind::UnexpectedEof,
-                "socket closed while receiving frame payload",
-            ));
-        }
-        filled += read;
-        assert!(filled <= message_len);
-        if filled == message_len {
-            return Ok(assume_init_vec(payload));
-        }
-    }
-    unreachable!("loop exits only after returning payload")
-}
-
-fn send_message_bytes(socket: &Socket, data: &[u8], fds: &[OwnedFd]) -> std::io::Result<()> {
-    if fds.len() > MAX_FDS_PER_MESSAGE {
-        return Err(std::io::Error::new(
-            std::io::ErrorKind::InvalidInput,
-            format!("too many fds: {}", fds.len()),
-        ));
-    }
-    let mut frame = Vec::with_capacity(LENGTH_PREFIX_SIZE + data.len());
-    frame.extend_from_slice(&encode_length(data.len())?);
-    frame.extend_from_slice(data);
-
-    let mut control = vec![0u8; control_space_for_fds(fds.len())];
-    unsafe {
-        let cmsg = control.as_mut_ptr().cast::<libc::cmsghdr>();
-        (*cmsg).cmsg_len = libc::CMSG_LEN(size_of::<RawFd>() as c_uint * fds.len() as c_uint) as _;
-        (*cmsg).cmsg_level = libc::SOL_SOCKET;
-        (*cmsg).cmsg_type = libc::SCM_RIGHTS;
-        let data_ptr = libc::CMSG_DATA(cmsg).cast::<RawFd>();
-        for (i, fd) in fds.iter().enumerate() {
-            data_ptr.add(i).write(fd.as_raw_fd());
-        }
-    }
-
-    let payload = [IoSlice::new(&frame)];
-    let msg = MsgHdr::new().with_buffers(&payload).with_control(&control);
-    let mut sent = socket.sendmsg(&msg, 0)?;
-    while sent < frame.len() {
-        let bytes = socket.send(&frame[sent..])?;
-        if bytes == 0 {
-            return Err(std::io::Error::new(
-                std::io::ErrorKind::WriteZero,
-                "socket closed while sending frame payload",
-            ));
-        }
-        sent += bytes;
-    }
-    Ok(())
-}
-
-fn encode_length(len: usize) -> std::io::Result<[u8; LENGTH_PREFIX_SIZE]> {
-    let len_u32 = u32::try_from(len).map_err(|_| {
-        std::io::Error::new(
-            std::io::ErrorKind::InvalidInput,
-            format!("message too large: {len}"),
-        )
-    })?;
-    Ok(len_u32.to_le_bytes())
-}
-
-pub(crate) fn send_json_message<T: Serialize>(
-    socket: &Socket,
-    msg: T,
-    fds: &[OwnedFd],
-) -> std::io::Result<()> {
-    let data = serde_json::to_vec(&msg)?;
-    send_message_bytes(socket, &data, fds)
-}
-
-fn send_datagram_bytes(socket: &Socket, data: &[u8], fds: &[OwnedFd]) -> std::io::Result<()> {
-    if fds.len() > MAX_FDS_PER_MESSAGE {
-        return Err(std::io::Error::new(
-            std::io::ErrorKind::InvalidInput,
-            format!("too many fds: {}", fds.len()),
-        ));
-    }
-    let mut control = vec![0u8; control_space_for_fds(fds.len())];
-    if !fds.is_empty() {
-        unsafe {
-            let cmsg = control.as_mut_ptr().cast::<libc::cmsghdr>();
-            (*cmsg).cmsg_len =
-                libc::CMSG_LEN(size_of::<RawFd>() as c_uint * fds.len() as c_uint) as _;
-            (*cmsg).cmsg_level = libc::SOL_SOCKET;
-            (*cmsg).cmsg_type = libc::SCM_RIGHTS;
-            let data_ptr = libc::CMSG_DATA(cmsg).cast::<RawFd>();
-            for (i, fd) in fds.iter().enumerate() {
-                data_ptr.add(i).write(fd.as_raw_fd());
-            }
-        }
-    }
-    let payload = [IoSlice::new(data)];
-    let msg = MsgHdr::new().with_buffers(&payload).with_control(&control);
-    let written = socket.sendmsg(&msg, 0)?;
-    if written != data.len() {
-        return Err(std::io::Error::new(
-            std::io::ErrorKind::WriteZero,
-            format!(
-                "short datagram write: wrote {written} bytes out of {}",
-                data.len()
-            ),
-        ));
-    }
-    Ok(())
-}
-
-fn receive_datagram_bytes(socket: &Socket) -> std::io::Result<(Vec<u8>, Vec<OwnedFd>)> {
-    let mut buffer = vec![MaybeUninit::<u8>::uninit(); MAX_DATAGRAM_SIZE];
-    let mut control = vec![MaybeUninit::<u8>::uninit(); control_space_for_fds(MAX_FDS_PER_MESSAGE)];
-    let (read, control_len) = {
-        let mut bufs = [MaybeUninitSlice::new(&mut buffer)];
-        let mut msg = MsgHdrMut::new()
-            .with_buffers(&mut bufs)
-            .with_control(&mut control);
-        let read = socket.recvmsg(&mut msg, 0)?;
-        (read, msg.control_len())
-    };
-    let data = assume_init(&buffer[..read]).to_vec();
-    let fds = extract_fds(assume_init(&control[..control_len]));
-    Ok((data, fds))
-}
-
-pub(crate) struct AsyncSocket {
-    inner: AsyncFd<Socket>,
-}
-
-impl AsyncSocket {
-    fn new(socket: Socket) -> std::io::Result<AsyncSocket> {
-        socket.set_nonblocking(true)?;
-        let async_socket = AsyncFd::new(socket)?;
-        Ok(AsyncSocket {
-            inner: async_socket,
-        })
-    }
-
-    pub fn from_fd(fd: OwnedFd) -> std::io::Result<AsyncSocket> {
-        AsyncSocket::new(Socket::from(fd))
-    }
-
-    pub fn pair() -> std::io::Result<(AsyncSocket, AsyncSocket)> {
-        let (server, client) = Socket::pair(Domain::UNIX, Type::STREAM, None)?;
-        Ok((AsyncSocket::new(server)?, AsyncSocket::new(client)?))
-    }
-
-    pub async fn send_with_fds<T: Serialize>(
-        &self,
-        msg: T,
-        fds: &[OwnedFd],
-    ) -> std::io::Result<()> {
-        self.inner
-            .async_io(Interest::WRITABLE, |socket| {
-                send_json_message(socket, &msg, fds)
-            })
-            .await
-    }
-
-    pub async fn receive_with_fds<T: for<'de> Deserialize<'de>>(
-        &self,
-    ) -> std::io::Result<(T, Vec<OwnedFd>)> {
-        let (payload, fds) = read_frame(&self.inner).await?;
-        let message: T = serde_json::from_slice(&payload)?;
-        Ok((message, fds))
-    }
-
-    pub async fn send<T>(&self, msg: T) -> std::io::Result<()>
-    where
-        T: Serialize,
-    {
-        self.send_with_fds(&msg, &[]).await
-    }
-
-    pub async fn receive<T: for<'de> Deserialize<'de>>(&self) -> std::io::Result<T> {
-        let (msg, fds) = self.receive_with_fds().await?;
-        if !fds.is_empty() {
-            tracing::warn!("unexpected fds in receive: {}", fds.len());
-        }
-        Ok(msg)
-    }
-
-    pub fn into_inner(self) -> Socket {
-        self.inner.into_inner()
-    }
-}
-
-pub(crate) struct AsyncDatagramSocket {
-    inner: AsyncFd<Socket>,
-}
-
-impl AsyncDatagramSocket {
-    fn new(socket: Socket) -> std::io::Result<Self> {
-        socket.set_nonblocking(true)?;
-        Ok(Self {
-            inner: AsyncFd::new(socket)?,
-        })
-    }
-
-    pub unsafe fn from_raw_fd(fd: RawFd) -> std::io::Result<Self> {
-        Self::new(unsafe { Socket::from_raw_fd(fd) })
-    }
-
-    pub fn pair() -> std::io::Result<(Self, Self)> {
-        let (server, client) = Socket::pair(Domain::UNIX, Type::DGRAM, None)?;
-        Ok((Self::new(server)?, Self::new(client)?))
-    }
-
-    pub async fn send_with_fds(&self, data: &[u8], fds: &[OwnedFd]) -> std::io::Result<()> {
-        self.inner
-            .async_io(Interest::WRITABLE, |socket| {
-                send_datagram_bytes(socket, data, fds)
-            })
-            .await
-    }
-
-    pub async fn receive_with_fds(&self) -> std::io::Result<(Vec<u8>, Vec<OwnedFd>)> {
-        self.inner
-            .async_io(Interest::READABLE, receive_datagram_bytes)
-            .await
-    }
-
-    pub fn into_inner(self) -> Socket {
-        self.inner.into_inner()
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-    use pretty_assertions::assert_eq;
-    use serde::Deserialize;
-    use serde::Serialize;
-    use std::os::fd::AsFd;
-    use std::os::fd::AsRawFd;
-    use tempfile::NamedTempFile;
-
-    #[derive(Debug, Serialize, Deserialize, PartialEq, Eq, Clone)]
-    struct TestPayload {
-        id: i32,
-        label: String,
-    }
-
-    fn fd_list(count: usize) -> std::io::Result<Vec<OwnedFd>> {
-        let file = NamedTempFile::new()?;
-        let mut fds = Vec::new();
-        for _ in 0..count {
-            fds.push(file.as_fd().try_clone_to_owned()?);
-        }
-        Ok(fds)
-    }
-
-    #[tokio::test]
-    async fn async_socket_round_trips_payload_and_fds() -> std::io::Result<()> {
-        let (server, client) = AsyncSocket::pair()?;
-        let payload = TestPayload {
-            id: 7,
-            label: "round-trip".to_string(),
-        };
-        let send_fds = fd_list(1)?;
-
-        let receive_task =
-            tokio::spawn(async move { server.receive_with_fds::<TestPayload>().await });
-        client.send_with_fds(payload.clone(), &send_fds).await?;
-        drop(send_fds);
-
-        let (received_payload, received_fds) = receive_task.await.unwrap()?;
-        assert_eq!(payload, received_payload);
-        assert_eq!(1, received_fds.len());
-        let fd_status = unsafe { libc::fcntl(received_fds[0].as_raw_fd(), libc::F_GETFD) };
-        assert!(
-            fd_status >= 0,
-            "expected received file descriptor to be valid, but got {fd_status}",
-        );
-        Ok(())
-    }
-
-    #[tokio::test]
-    async fn async_datagram_sockets_round_trip_messages() -> std::io::Result<()> {
-        let (server, client) = AsyncDatagramSocket::pair()?;
-        let data = b"datagram payload".to_vec();
-        let send_fds = fd_list(1)?;
-        let receive_task = tokio::spawn(async move { server.receive_with_fds().await });
-
-        client.send_with_fds(&data, &send_fds).await?;
-        drop(send_fds);
-
-        let (received_bytes, received_fds) = receive_task.await.unwrap()?;
-        assert_eq!(data, received_bytes);
-        assert_eq!(1, received_fds.len());
-        Ok(())
-    }
-
-    #[test]
-    fn send_message_bytes_rejects_excessive_fd_counts() -> std::io::Result<()> {
-        let (socket, _peer) = Socket::pair(Domain::UNIX, Type::STREAM, None)?;
-        let fds = fd_list(MAX_FDS_PER_MESSAGE + 1)?;
-        let err = send_message_bytes(&socket, b"hello", &fds).unwrap_err();
-        assert_eq!(std::io::ErrorKind::InvalidInput, err.kind());
-        Ok(())
-    }
-
-    #[test]
-    fn send_datagram_bytes_rejects_excessive_fd_counts() -> std::io::Result<()> {
-        let (socket, _peer) = Socket::pair(Domain::UNIX, Type::DGRAM, None)?;
-        let fds = fd_list(MAX_FDS_PER_MESSAGE + 1)?;
-        let err = send_datagram_bytes(&socket, b"hi", &fds).unwrap_err();
-        assert_eq!(std::io::ErrorKind::InvalidInput, err.kind());
-        Ok(())
-    }
-
-    #[test]
-    fn encode_length_errors_for_oversized_messages() {
-        let err = encode_length(usize::MAX).unwrap_err();
-        assert_eq!(std::io::ErrorKind::InvalidInput, err.kind());
-    }
-
-    #[tokio::test]
-    async fn receive_fails_when_peer_closes_before_header() {
-        let (server, client) = AsyncSocket::pair().expect("failed to create socket pair");
-        drop(client);
-        let err = server
-            .receive::<serde_json::Value>()
-            .await
-            .expect_err("expected read failure");
-        assert_eq!(std::io::ErrorKind::UnexpectedEof, err.kind());
-    }
-}

codex-rs/exec/src/event_processor_with_human_output.rs

@@ -480,7 +480,11 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                 let SessionConfiguredEvent {
                     session_id: conversation_id,
                     model,
-                    ..
+                    reasoning_effort: _,
+                    history_log_id: _,
+                    history_entry_count: _,
+                    initial_messages: _,
+                    rollout_path: _,
                 } = session_configured_event;
 
                 ts_msg!(

codex-rs/exec/tests/event_processor_with_json_output.rs

@@ -1,6 +1,5 @@
 use codex_core::protocol::AgentMessageEvent;
 use codex_core::protocol::AgentReasoningEvent;
-use codex_core::protocol::AskForApproval;
 use codex_core::protocol::ErrorEvent;
 use codex_core::protocol::Event;
 use codex_core::protocol::EventMsg;
@@ -13,7 +12,6 @@ use codex_core::protocol::McpToolCallBeginEvent;
 use codex_core::protocol::McpToolCallEndEvent;
 use codex_core::protocol::PatchApplyBeginEvent;
 use codex_core::protocol::PatchApplyEndEvent;
-use codex_core::protocol::SandboxPolicy;
 use codex_core::protocol::SessionConfiguredEvent;
 use codex_core::protocol::WarningEvent;
 use codex_core::protocol::WebSearchEndEvent;
@@ -74,10 +72,6 @@ fn session_configured_produces_thread_started_event() {
         EventMsg::SessionConfigured(SessionConfiguredEvent {
             session_id,
             model: "codex-mini-latest".to_string(),
-            model_provider_id: "test-provider".to_string(),
-            approval_policy: AskForApproval::Never,
-            sandbox_policy: SandboxPolicy::ReadOnly,
-            cwd: PathBuf::from("/home/user/project"),
             reasoning_effort: None,
             history_log_id: 0,
             history_entry_count: 0,

codex-rs/execpolicy2/README.md

@@ -5,7 +5,7 @@
 - This release covers only the prefix-rule subset of the planned execpolicy v2 language; a richer language will follow.
 - Tokens are matched in order; any `pattern` element may be a list to denote alternatives. `decision` defaults to `allow`; valid values: `allow`, `prompt`, `forbidden`.
 - `match` / `not_match` supply example invocations that are validated at load time (think of them as unit tests); examples can be token arrays or strings (strings are tokenized with `shlex`).
-- The CLI always prints the JSON serialization of the evaluation result (whether a match or not).
+- The CLI always prints the JSON serialization of the evaluation result.
 
 ## Policy shapes
 - Prefix rules use Starlark syntax:
@@ -18,6 +18,24 @@ prefix_rule(
 )
 ```
 
+## CLI
+- From the Codex CLI, run `codex execpolicycheck` with one or more policy files (for example `src/default.codexpolicy`) to check a command:
+```bash
+codex execpolicycheck --policy path/to/policy.codexpolicy git status
+```
+- Pass multiple `--policy` flags to merge rules, evaluated in the order provided:
+```bash
+codex execpolicycheck --policy base.codexpolicy --policy overrides.codexpolicy git status
+```
+- Output is JSON by default; pass `--pretty` for pretty-printed JSON
+- You can also run the standalone dev binary directly during development:
+```bash
+cargo run -p codex-execpolicy2 -- check --policy path/to/policy.codexpolicy git status
+```
+- Example outcomes:
+  - Match: `{"match": { ... "decision": "allow" ... }}`
+  - No match: `{"noMatch": {}}`
+
 ## Response shapes
 - Match:
 ```json
@@ -38,22 +56,8 @@ prefix_rule(
 
 - No match:
 ```json
-"noMatch"
+{"noMatch": {}}
 ```
 
 - `matchedRules` lists every rule whose prefix matched the command; `matchedPrefix` is the exact prefix that matched.
 - The effective `decision` is the strictest severity across all matches (`forbidden` > `prompt` > `allow`).
-
-## CLI
-- Provide one or more policy files (for example `src/default.codexpolicy`) to check a command:
-```bash
-cargo run -p codex-execpolicy2 -- check --policy path/to/policy.codexpolicy git status
-```
-- Pass multiple `--policy` flags to merge rules, evaluated in the order provided:
-```bash
-cargo run -p codex-execpolicy2 -- check --policy base.codexpolicy --policy overrides.codexpolicy git status
-```
-- Output is newline-delimited JSON by default; pass `--pretty` for pretty-printed JSON if desired.
-- Example outcomes:
-  - Match: `{"match": { ... "decision": "allow" ... }}`
-  - No match: `"noMatch"`

codex-rs/execpolicy2/src/execpolicycheck.rs

@@ -0,0 +1,64 @@
+use std::fs;
+use std::path::PathBuf;
+
+use anyhow::Context;
+use anyhow::Result;
+use clap::Parser;
+
+use crate::Evaluation;
+use crate::Policy;
+use crate::PolicyParser;
+
+/// Arguments for evaluating a command against one or more execpolicy files.
+#[derive(Debug, Parser, Clone)]
+pub struct ExecPolicyCheckCommand {
+    /// Paths to execpolicy files to evaluate (repeatable).
+    #[arg(short, long = "policy", value_name = "PATH", required = true)]
+    pub policies: Vec<PathBuf>,
+
+    /// Pretty-print the JSON output.
+    #[arg(long)]
+    pub pretty: bool,
+
+    /// Command tokens to check against the policy.
+    #[arg(
+        value_name = "COMMAND",
+        required = true,
+        trailing_var_arg = true,
+        allow_hyphen_values = true
+    )]
+    pub command: Vec<String>,
+}
+
+impl ExecPolicyCheckCommand {
+    /// Load the policies for this command, evaluate the command, and render JSON output.
+    pub fn run(&self) -> Result<String> {
+        let policy = load_policies(&self.policies)?;
+        let evaluation = policy.check(&self.command);
+
+        format_evaluation_json(&evaluation, self.pretty)
+    }
+}
+
+pub fn format_evaluation_json(evaluation: &Evaluation, pretty: bool) -> Result<String> {
+    if pretty {
+        serde_json::to_string_pretty(evaluation).map_err(Into::into)
+    } else {
+        serde_json::to_string(evaluation).map_err(Into::into)
+    }
+}
+
+pub fn load_policies(policy_paths: &[PathBuf]) -> Result<Policy> {
+    let mut parser = PolicyParser::new();
+
+    for policy_path in policy_paths {
+        let policy_file_contents = fs::read_to_string(policy_path)
+            .with_context(|| format!("failed to read policy at {}", policy_path.display()))?;
+        let policy_identifier = policy_path.to_string_lossy().to_string();
+        parser
+            .parse(&policy_identifier, &policy_file_contents)
+            .with_context(|| format!("failed to parse policy at {}", policy_path.display()))?;
+    }
+
+    Ok(parser.build())
+}

codex-rs/execpolicy2/src/lib.rs

@@ -1,5 +1,6 @@
 pub mod decision;
 pub mod error;
+pub mod execpolicycheck;
 pub mod parser;
 pub mod policy;
 pub mod rule;
@@ -7,6 +8,7 @@ pub mod rule;
 pub use decision::Decision;
 pub use error::Error;
 pub use error::Result;
+pub use execpolicycheck::ExecPolicyCheckCommand;
 pub use parser::PolicyParser;
 pub use policy::Evaluation;
 pub use policy::Policy;

codex-rs/execpolicy2/src/main.rs

@@ -1,66 +1,24 @@
-use std::fs;
-use std::path::PathBuf;
-
-use anyhow::Context;
 use anyhow::Result;
 use clap::Parser;
-use codex_execpolicy2::PolicyParser;
+use codex_execpolicy2::ExecPolicyCheckCommand;
 
 /// CLI for evaluating exec policies
 #[derive(Parser)]
 #[command(name = "codex-execpolicy2")]
 enum Cli {
     /// Evaluate a command against a policy.
-    Check {
-        #[arg(short, long = "policy", value_name = "PATH", required = true)]
-        policies: Vec<PathBuf>,
-
-        /// Pretty-print the JSON output.
-        #[arg(long)]
-        pretty: bool,
-
-        /// Command tokens to check.
-        #[arg(
-            value_name = "COMMAND",
-            required = true,
-            trailing_var_arg = true,
-            allow_hyphen_values = true
-        )]
-        command: Vec<String>,
-    },
+    Check(ExecPolicyCheckCommand),
 }
 
 fn main() -> Result<()> {
     let cli = Cli::parse();
     match cli {
-        Cli::Check {
-            policies,
-            command,
-            pretty,
-        } => cmd_check(policies, command, pretty),
+        Cli::Check(cmd) => cmd_check(cmd),
     }
 }
 
-fn cmd_check(policy_paths: Vec<PathBuf>, args: Vec<String>, pretty: bool) -> Result<()> {
-    let policy = load_policies(&policy_paths)?;
-
-    let eval = policy.check(&args);
-    let json = if pretty {
-        serde_json::to_string_pretty(&eval)?
-    } else {
-        serde_json::to_string(&eval)?
-    };
+fn cmd_check(cmd: ExecPolicyCheckCommand) -> Result<()> {
+    let json = cmd.run()?;
     println!("{json}");
     Ok(())
 }
-
-fn load_policies(policy_paths: &[PathBuf]) -> Result<codex_execpolicy2::Policy> {
-    let mut parser = PolicyParser::new();
-    for policy_path in policy_paths {
-        let policy_file_contents = fs::read_to_string(policy_path)
-            .with_context(|| format!("failed to read policy at {}", policy_path.display()))?;
-        let policy_identifier = policy_path.to_string_lossy().to_string();
-        parser.parse(&policy_identifier, &policy_file_contents)?;
-    }
-    Ok(parser.build())
-}

codex-rs/execpolicy2/src/policy.rs

@@ -15,6 +15,10 @@ impl Policy {
         Self { rules_by_program }
     }
 
+    pub fn empty() -> Self {
+        Self::new(MultiMap::new())
+    }
+
     pub fn rules(&self) -> &MultiMap<String, RuleRef> {
         &self.rules_by_program
     }
@@ -23,9 +27,9 @@ impl Policy {
         let rules = match cmd.first() {
             Some(first) => match self.rules_by_program.get_vec(first) {
                 Some(rules) => rules,
-                None => return Evaluation::NoMatch,
+                None => return Evaluation::NoMatch {},
             },
-            None => return Evaluation::NoMatch,
+            None => return Evaluation::NoMatch {},
         };
 
         let matched_rules: Vec<RuleMatch> =
@@ -35,7 +39,7 @@ impl Policy {
                 decision,
                 matched_rules,
             },
-            None => Evaluation::NoMatch,
+            None => Evaluation::NoMatch {},
         }
     }
 
@@ -48,7 +52,7 @@ impl Policy {
             .into_iter()
             .flat_map(|command| match self.check(command.as_ref()) {
                 Evaluation::Match { matched_rules, .. } => matched_rules,
-                Evaluation::NoMatch => Vec::new(),
+                Evaluation::NoMatch { .. } => Vec::new(),
             })
             .collect();
 
@@ -57,7 +61,7 @@ impl Policy {
                 decision,
                 matched_rules,
             },
-            None => Evaluation::NoMatch,
+            None => Evaluation::NoMatch {},
         }
     }
 }
@@ -65,7 +69,7 @@ impl Policy {
 #[derive(Clone, Debug, Eq, PartialEq, Serialize, Deserialize)]
 #[serde(rename_all = "camelCase")]
 pub enum Evaluation {
-    NoMatch,
+    NoMatch {},
     Match {
         decision: Decision,
         #[serde(rename = "matchedRules")]

codex-rs/execpolicy2/tests/basic.rs

@@ -10,6 +10,7 @@ use codex_execpolicy2::rule::PatternToken;
 use codex_execpolicy2::rule::PrefixPattern;
 use codex_execpolicy2::rule::PrefixRule;
 use pretty_assertions::assert_eq;
+use serde_json::json;
 
 fn tokens(cmd: &[&str]) -> Vec<String> {
     cmd.iter().map(std::string::ToString::to_string).collect()
@@ -60,6 +61,14 @@ prefix_rule(
     );
 }
 
+#[test]
+fn serializes_no_match_as_object() {
+    let serialized =
+        serde_json::to_value(&Evaluation::NoMatch {}).expect("should serialize evaluation");
+
+    assert_eq!(json!({"noMatch": {}}), serialized);
+}
+
 #[test]
 fn parses_multiple_policy_files() {
     let first_policy = r#"
@@ -288,7 +297,7 @@ prefix_rule(
         "color.status=always",
         "status",
     ]));
-    assert_eq!(Evaluation::NoMatch, no_match_eval);
+    assert_eq!(Evaluation::NoMatch {}, no_match_eval);
 }
 
 #[test]

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

@@ -231,12 +231,8 @@ pub(crate) struct OutgoingError {
 
 #[cfg(test)]
 mod tests {
-    use std::path::PathBuf;
-
     use anyhow::Result;
-    use codex_core::protocol::AskForApproval;
     use codex_core::protocol::EventMsg;
-    use codex_core::protocol::SandboxPolicy;
     use codex_core::protocol::SessionConfiguredEvent;
     use codex_protocol::ConversationId;
     use codex_protocol::config_types::ReasoningEffort;
@@ -258,10 +254,6 @@ mod tests {
             msg: EventMsg::SessionConfigured(SessionConfiguredEvent {
                 session_id: conversation_id,
                 model: "gpt-4o".to_string(),
-                model_provider_id: "test-provider".to_string(),
-                approval_policy: AskForApproval::Never,
-                sandbox_policy: SandboxPolicy::ReadOnly,
-                cwd: PathBuf::from("/home/user/project"),
                 reasoning_effort: Some(ReasoningEffort::default()),
                 history_log_id: 1,
                 history_entry_count: 1000,
@@ -297,10 +289,6 @@ mod tests {
         let session_configured_event = SessionConfiguredEvent {
             session_id: conversation_id,
             model: "gpt-4o".to_string(),
-            model_provider_id: "test-provider".to_string(),
-            approval_policy: AskForApproval::Never,
-            sandbox_policy: SandboxPolicy::ReadOnly,
-            cwd: PathBuf::from("/home/user/project"),
             reasoning_effort: Some(ReasoningEffort::default()),
             history_log_id: 1,
             history_entry_count: 1000,
@@ -330,18 +318,12 @@ mod tests {
             },
             "id": "1",
             "msg": {
-                "type": "session_configured",
                 "session_id": session_configured_event.session_id,
-                "model": "gpt-4o",
-                "model_provider_id": "test-provider",
-                "approval_policy": "never",
-                "sandbox_policy": {
-                    "type": "read-only"
-                },
-                "cwd": "/home/user/project",
+                "model": session_configured_event.model,
                 "reasoning_effort": session_configured_event.reasoning_effort,
                 "history_log_id": session_configured_event.history_log_id,
                 "history_entry_count": session_configured_event.history_entry_count,
+                "type": "session_configured",
                 "rollout_path": rollout_file.path().to_path_buf(),
             }
         });