feature: Add "!cmd" user shell execution

- protocol: add Op::RunUserShellCommand to model a user-initiated one-off command
- core: handle new Op by spawning a cancellable task that runs the command using the user’s default shell; stream output via ExecCommand* events; send TaskStarted/TaskComplete; track as current_task so Interrupt works
- tui: detect leading '!' in composer submission and dispatch Op::RunUserShellCommand instead of sending a user message

No changes to sandbox env var behavior; uses existing exec pipeline and event types.

codex-rs/Cargo.lock

@@ -212,6 +212,50 @@ dependencies = [
  "term",
 ]
 
+[[package]]
+name = "askama"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "b79091df18a97caea757e28cd2d5fda49c6cd4bd01ddffd7ff01ace0c0ad2c28"
+dependencies = [
+ "askama_derive",
+ "askama_escape",
+ "humansize",
+ "num-traits",
+ "percent-encoding",
+]
+
+[[package]]
+name = "askama_derive"
+version = "0.12.5"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "19fe8d6cb13c4714962c072ea496f3392015f0989b1a2847bb4b2d9effd71d83"
+dependencies = [
+ "askama_parser",
+ "basic-toml",
+ "mime",
+ "mime_guess",
+ "proc-macro2",
+ "quote",
+ "serde",
+ "syn 2.0.104",
+]
+
+[[package]]
+name = "askama_escape"
+version = "0.10.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "619743e34b5ba4e9703bba34deac3427c72507c7159f5fd030aea8cac0cfe341"
+
+[[package]]
+name = "askama_parser"
+version = "0.2.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "acb1161c6b64d1c3d83108213c2a2533a342ac225aabd0bda218278c2ddb00c0"
+dependencies = [
+ "nom",
+]
+
 [[package]]
 name = "assert-json-diff"
 version = "2.0.2"
@@ -305,6 +349,15 @@ version = "0.22.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "72b3254f16251a8381aa12e40e3c4d2f0199f8c6508fbecb9d91f575e0fbb8c6"
 
+[[package]]
+name = "basic-toml"
+version = "0.1.10"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba62675e8242a4c4e806d12f11d136e626e6c8361d6b829310732241652a178a"
+dependencies = [
+ "serde",
+]
+
 [[package]]
 name = "beef"
 version = "0.5.2"
@@ -606,6 +659,7 @@ name = "codex-core"
 version = "0.0.0"
 dependencies = [
  "anyhow",
+ "askama",
  "assert_cmd",
  "async-channel",
  "base64",
@@ -1981,6 +2035,15 @@ version = "1.0.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "df3b46402a9d5adb4c86a0cf463f42e19994e3ee891101b1841f30a545cb49a9"
 
+[[package]]
+name = "humansize"
+version = "2.1.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6cb51c9a029ddc91b07a787f1d86b53ccfa49b0e86688c946ebe8d3555685dd7"
+dependencies = [
+ "libm",
+]
+
 [[package]]
 name = "hyper"
 version = "1.7.0"
@@ -2536,6 +2599,12 @@ version = "0.2.175"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6a82ae493e598baaea5209805c49bbf2ea7de956d50d7da0da1164f9c6d28543"
 
+[[package]]
+name = "libm"
+version = "0.2.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f9fbbcab51052fe104eb5e5d351cf728d30a5be1fe14d9be8a3b097481fb97de"
+
 [[package]]
 name = "libredox"
 version = "0.1.6"

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

@@ -40,6 +40,11 @@ pub enum ApplyPatchError {
     /// Error that occurs while computing replacements when applying patch chunks
     #[error("{0}")]
     ComputeReplacements(String),
+    /// A raw patch body was provided without an explicit `apply_patch` invocation.
+    #[error(
+        "patch detected without explicit call to apply_patch. Rerun as [\"apply_patch\", \"<patch>\"]"
+    )]
+    ImplicitInvocation,
 }
 
 impl From<std::io::Error> for ApplyPatchError {
@@ -93,10 +98,12 @@ pub struct ApplyPatchArgs {
 
 pub fn maybe_parse_apply_patch(argv: &[String]) -> MaybeApplyPatch {
     match argv {
+        // Direct invocation: apply_patch <patch>
         [cmd, body] if APPLY_PATCH_COMMANDS.contains(&cmd.as_str()) => match parse_patch(body) {
             Ok(source) => MaybeApplyPatch::Body(source),
             Err(e) => MaybeApplyPatch::PatchParseError(e),
         },
+        // Bash heredoc form: (optional `cd <path> &&`) apply_patch <<'EOF' ...
         [bash, flag, script] if bash == "bash" && flag == "-lc" => {
             match extract_apply_patch_from_bash(script) {
                 Ok((body, workdir)) => match parse_patch(&body) {
@@ -207,6 +214,26 @@ impl ApplyPatchAction {
 /// cwd must be an absolute path so that we can resolve relative paths in the
 /// patch.
 pub fn maybe_parse_apply_patch_verified(argv: &[String], cwd: &Path) -> MaybeApplyPatchVerified {
+    // Detect a raw patch body passed directly as the command or as the body of a bash -lc
+    // script. In these cases, report an explicit error rather than applying the patch.
+    match argv {
+        [body] => {
+            if parse_patch(body).is_ok() {
+                return MaybeApplyPatchVerified::CorrectnessError(
+                    ApplyPatchError::ImplicitInvocation,
+                );
+            }
+        }
+        [bash, flag, script] if bash == "bash" && flag == "-lc" => {
+            if parse_patch(script).is_ok() {
+                return MaybeApplyPatchVerified::CorrectnessError(
+                    ApplyPatchError::ImplicitInvocation,
+                );
+            }
+        }
+        _ => {}
+    }
+
     match maybe_parse_apply_patch(argv) {
         MaybeApplyPatch::Body(ApplyPatchArgs {
             patch,
@@ -875,6 +902,28 @@ mod tests {
         ));
     }
 
+    #[test]
+    fn test_implicit_patch_single_arg_is_error() {
+        let patch = "*** Begin Patch\n*** Add File: foo\n+hi\n*** End Patch".to_string();
+        let args = vec![patch];
+        let dir = tempdir().unwrap();
+        assert!(matches!(
+            maybe_parse_apply_patch_verified(&args, dir.path()),
+            MaybeApplyPatchVerified::CorrectnessError(ApplyPatchError::ImplicitInvocation)
+        ));
+    }
+
+    #[test]
+    fn test_implicit_patch_bash_script_is_error() {
+        let script = "*** Begin Patch\n*** Add File: foo\n+hi\n*** End Patch";
+        let args = args_bash(script);
+        let dir = tempdir().unwrap();
+        assert!(matches!(
+            maybe_parse_apply_patch_verified(&args, dir.path()),
+            MaybeApplyPatchVerified::CorrectnessError(ApplyPatchError::ImplicitInvocation)
+        ));
+    }
+
     #[test]
     fn test_literal() {
         let args = strs_to_strings(&[

codex-rs/common/src/config_summary.rs

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

codex-rs/common/src/model_presets.rs

@@ -12,7 +12,7 @@ pub struct ModelPreset {
     /// Model slug (e.g., "gpt-5").
     pub model: &'static str,
     /// Reasoning effort to apply for this preset.
-    pub effort: ReasoningEffort,
+    pub effort: Option<ReasoningEffort>,
 }
 
 /// Built-in list of model presets that pair a model with a reasoning effort.
@@ -26,35 +26,35 @@ pub fn builtin_model_presets() -> &'static [ModelPreset] {
             label: "gpt-5 minimal",
             description: "— fastest responses with limited reasoning; ideal for coding, instructions, or lightweight tasks",
             model: "gpt-5",
-            effort: ReasoningEffort::Minimal,
+            effort: Some(ReasoningEffort::Minimal),
         },
         ModelPreset {
             id: "gpt-5-low",
             label: "gpt-5 low",
             description: "— balances speed with some reasoning; useful for straightforward queries and short explanations",
             model: "gpt-5",
-            effort: ReasoningEffort::Low,
+            effort: Some(ReasoningEffort::Low),
         },
         ModelPreset {
             id: "gpt-5-medium",
             label: "gpt-5 medium",
             description: "— default setting; provides a solid balance of reasoning depth and latency for general-purpose tasks",
             model: "gpt-5",
-            effort: ReasoningEffort::Medium,
+            effort: Some(ReasoningEffort::Medium),
         },
         ModelPreset {
             id: "gpt-5-high",
             label: "gpt-5 high",
             description: "— maximizes reasoning depth for complex or ambiguous problems",
             model: "gpt-5",
-            effort: ReasoningEffort::High,
+            effort: Some(ReasoningEffort::High),
         },
         ModelPreset {
             id: "gpt-5-high-new",
             label: "gpt-5 high new",
             description: "— our latest release tuned to rely on the model's built-in reasoning defaults",
             model: "gpt-5-high-new",
-            effort: ReasoningEffort::Medium,
+            effort: None,
         },
     ];
     PRESETS

codex-rs/core/Cargo.toml

@@ -13,6 +13,7 @@ workspace = true
 
 [dependencies]
 anyhow = "1"
+askama = "0.12"
 async-channel = "2.3.1"
 base64 = "0.22"
 bytes = "1.10.1"

codex-rs/core/review_prompt.md

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

codex-rs/core/src/client.rs

@@ -72,7 +72,7 @@ pub struct ModelClient {
     client: reqwest::Client,
     provider: ModelProviderInfo,
     conversation_id: ConversationId,
-    effort: ReasoningEffortConfig,
+    effort: Option<ReasoningEffortConfig>,
     summary: ReasoningSummaryConfig,
 }
 
@@ -81,7 +81,7 @@ impl ModelClient {
         config: Arc<Config>,
         auth_manager: Option<Arc<AuthManager>>,
         provider: ModelProviderInfo,
-        effort: ReasoningEffortConfig,
+        effort: Option<ReasoningEffortConfig>,
         summary: ReasoningSummaryConfig,
         conversation_id: ConversationId,
     ) -> Self {
@@ -104,6 +104,12 @@ impl ModelClient {
             .or_else(|| get_model_info(&self.config.model_family).map(|info| info.context_window))
     }
 
+    pub fn get_auto_compact_token_limit(&self) -> Option<i64> {
+        self.config.model_auto_compact_token_limit.or_else(|| {
+            get_model_info(&self.config.model_family).and_then(|info| info.auto_compact_token_limit)
+        })
+    }
+
     /// Dispatches to either the Responses or Chat implementation depending on
     /// the provider config.  Public callers always invoke `stream()` – the
     /// specialised helpers are private to avoid accidental misuse.
@@ -187,6 +193,15 @@ impl ModelClient {
             None
         };
 
+        // In general, we want to explicitly send `store: false` when using the Responses API,
+        // but in practice, the Azure Responses API rejects `store: false`:
+        //
+        // - If store = false and id is sent an error is thrown that ID is not found
+        // - If store = false and id is not sent an error is thrown that ID is required
+        //
+        // For Azure, we send `store: true` and preserve reasoning item IDs.
+        let azure_workaround = self.provider.is_azure_responses_endpoint();
+
         let payload = ResponsesApiRequest {
             model: &self.config.model,
             instructions: &full_instructions,
@@ -195,13 +210,19 @@ impl ModelClient {
             tool_choice: "auto",
             parallel_tool_calls: false,
             reasoning,
-            store: false,
+            store: azure_workaround,
             stream: true,
             include,
             prompt_cache_key: Some(self.conversation_id.to_string()),
             text,
         };
 
+        let mut payload_json = serde_json::to_value(&payload)?;
+        if azure_workaround {
+            attach_item_ids(&mut payload_json, &input_with_instructions);
+        }
+        let payload_body = serde_json::to_string(&payload_json)?;
+
         let mut attempt = 0;
         let max_retries = self.provider.request_max_retries();
 
@@ -214,7 +235,7 @@ impl ModelClient {
             trace!(
                 "POST to {}: {}",
                 self.provider.get_full_url(&auth),
-                serde_json::to_string(&payload)?
+                payload_body.as_str()
             );
 
             let mut req_builder = self
@@ -228,7 +249,7 @@ impl ModelClient {
                 .header("conversation_id", self.conversation_id.to_string())
                 .header("session_id", self.conversation_id.to_string())
                 .header(reqwest::header::ACCEPT, "text/event-stream")
-                .json(&payload);
+                .json(&payload_json);
 
             if let Some(auth) = auth.as_ref()
                 && auth.mode == AuthMode::ChatGPT
@@ -356,7 +377,7 @@ impl ModelClient {
     }
 
     /// Returns the current reasoning effort setting.
-    pub fn get_reasoning_effort(&self) -> ReasoningEffortConfig {
+    pub fn get_reasoning_effort(&self) -> Option<ReasoningEffortConfig> {
         self.effort
     }
 
@@ -425,6 +446,33 @@ struct ResponseCompletedOutputTokensDetails {
     reasoning_tokens: u64,
 }
 
+fn attach_item_ids(payload_json: &mut Value, original_items: &[ResponseItem]) {
+    let Some(input_value) = payload_json.get_mut("input") else {
+        return;
+    };
+    let serde_json::Value::Array(items) = input_value else {
+        return;
+    };
+
+    for (value, item) in items.iter_mut().zip(original_items.iter()) {
+        if let ResponseItem::Reasoning { id, .. }
+        | ResponseItem::Message { id: Some(id), .. }
+        | ResponseItem::WebSearchCall { id: Some(id), .. }
+        | ResponseItem::FunctionCall { id: Some(id), .. }
+        | ResponseItem::LocalShellCall { id: Some(id), .. }
+        | ResponseItem::CustomToolCall { id: Some(id), .. } = item
+        {
+            if id.is_empty() {
+                continue;
+            }
+
+            if let Some(obj) = value.as_object_mut() {
+                obj.insert("id".to_string(), Value::String(id.clone()));
+            }
+        }
+    }
+}
+
 async fn process_sse<S>(
     stream: S,
     tx_event: mpsc::Sender<Result<ResponseEvent>>,

codex-rs/core/src/client_common.rs

@@ -19,6 +19,9 @@ use tokio::sync::mpsc;
 /// with this content.
 const BASE_INSTRUCTIONS: &str = include_str!("../prompt.md");
 
+/// Review thread system prompt. Edit `core/src/review_prompt.md` to customize.
+pub const REVIEW_PROMPT: &str = include_str!("../review_prompt.md");
+
 /// API request payload for a single model turn
 #[derive(Default, Debug, Clone)]
 pub struct Prompt {
@@ -81,8 +84,10 @@ pub enum ResponseEvent {
 
 #[derive(Debug, Serialize)]
 pub(crate) struct Reasoning {
-    pub(crate) effort: ReasoningEffortConfig,
-    pub(crate) summary: ReasoningSummaryConfig,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(crate) effort: Option<ReasoningEffortConfig>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub(crate) summary: Option<ReasoningSummaryConfig>,
 }
 
 /// Controls under the `text` field in the Responses API for GPT-5.
@@ -136,14 +141,17 @@ pub(crate) struct ResponsesApiRequest<'a> {
 
 pub(crate) fn create_reasoning_param_for_request(
     model_family: &ModelFamily,
-    effort: ReasoningEffortConfig,
+    effort: Option<ReasoningEffortConfig>,
     summary: ReasoningSummaryConfig,
 ) -> Option<Reasoning> {
-    if model_family.supports_reasoning_summaries {
-        Some(Reasoning { effort, summary })
-    } else {
-        None
+    if !model_family.supports_reasoning_summaries {
+        return None;
     }
+
+    Some(Reasoning {
+        effort,
+        summary: Some(summary),
+    })
 }
 
 pub(crate) fn create_text_param_for_request(

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

@@ -0,0 +1,401 @@
+use std::sync::Arc;
+
+use super::AgentTask;
+use super::MutexExt;
+use super::Session;
+use super::TurnContext;
+use super::get_last_assistant_message_from_turn;
+use crate::Prompt;
+use crate::client_common::ResponseEvent;
+use crate::error::CodexErr;
+use crate::error::Result as CodexResult;
+use crate::protocol::AgentMessageEvent;
+use crate::protocol::CompactedItem;
+use crate::protocol::ErrorEvent;
+use crate::protocol::Event;
+use crate::protocol::EventMsg;
+use crate::protocol::InputItem;
+use crate::protocol::InputMessageKind;
+use crate::protocol::TaskCompleteEvent;
+use crate::protocol::TaskStartedEvent;
+use crate::protocol::TurnContextItem;
+use crate::util::backoff;
+use askama::Template;
+use codex_protocol::models::ContentItem;
+use codex_protocol::models::ResponseInputItem;
+use codex_protocol::models::ResponseItem;
+use codex_protocol::protocol::RolloutItem;
+use futures::prelude::*;
+
+pub(super) const COMPACT_TRIGGER_TEXT: &str = "Start Summarization";
+const SUMMARIZATION_PROMPT: &str = include_str!("../../templates/compact/prompt.md");
+
+#[derive(Template)]
+#[template(path = "compact/history_bridge.md", escape = "none")]
+struct HistoryBridgeTemplate<'a> {
+    user_messages_text: &'a str,
+    summary_text: &'a str,
+}
+
+pub(super) fn spawn_compact_task(
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+    sub_id: String,
+    input: Vec<InputItem>,
+) {
+    let task = AgentTask::compact(
+        sess.clone(),
+        turn_context,
+        sub_id,
+        input,
+        SUMMARIZATION_PROMPT.to_string(),
+    );
+    sess.set_task(task);
+}
+
+pub(super) async fn run_inline_auto_compact_task(
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+) {
+    let sub_id = sess.next_internal_sub_id();
+    let input = vec![InputItem::Text {
+        text: COMPACT_TRIGGER_TEXT.to_string(),
+    }];
+    run_compact_task_inner(
+        sess,
+        turn_context,
+        sub_id,
+        input,
+        SUMMARIZATION_PROMPT.to_string(),
+        false,
+    )
+    .await;
+}
+
+pub(super) async fn run_compact_task(
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+    sub_id: String,
+    input: Vec<InputItem>,
+    compact_instructions: String,
+) {
+    run_compact_task_inner(
+        sess,
+        turn_context,
+        sub_id,
+        input,
+        compact_instructions,
+        true,
+    )
+    .await;
+}
+
+async fn run_compact_task_inner(
+    sess: Arc<Session>,
+    turn_context: Arc<TurnContext>,
+    sub_id: String,
+    input: Vec<InputItem>,
+    compact_instructions: String,
+    remove_task_on_completion: bool,
+) {
+    let model_context_window = turn_context.client.get_model_context_window();
+    let start_event = Event {
+        id: sub_id.clone(),
+        msg: EventMsg::TaskStarted(TaskStartedEvent {
+            model_context_window,
+        }),
+    };
+    sess.send_event(start_event).await;
+
+    let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
+    let instructions_override = compact_instructions;
+    let turn_input = sess.turn_input_with_history(vec![initial_input_for_turn.clone().into()]);
+
+    let prompt = Prompt {
+        input: turn_input,
+        tools: Vec::new(),
+        base_instructions_override: Some(instructions_override),
+    };
+
+    let max_retries = turn_context.client.get_provider().stream_max_retries();
+    let mut retries = 0;
+
+    let rollout_item = RolloutItem::TurnContext(TurnContextItem {
+        cwd: turn_context.cwd.clone(),
+        approval_policy: turn_context.approval_policy,
+        sandbox_policy: turn_context.sandbox_policy.clone(),
+        model: turn_context.client.get_model(),
+        effort: turn_context.client.get_reasoning_effort(),
+        summary: turn_context.client.get_reasoning_summary(),
+    });
+    sess.persist_rollout_items(&[rollout_item]).await;
+
+    loop {
+        let attempt_result = drain_to_completed(&sess, turn_context.as_ref(), &prompt).await;
+
+        match attempt_result {
+            Ok(()) => {
+                break;
+            }
+            Err(CodexErr::Interrupted) => {
+                return;
+            }
+            Err(e) => {
+                if retries < max_retries {
+                    retries += 1;
+                    let delay = backoff(retries);
+                    sess.notify_stream_error(
+                        &sub_id,
+                        format!(
+                            "stream error: {e}; retrying {retries}/{max_retries} in {delay:?}…"
+                        ),
+                    )
+                    .await;
+                    tokio::time::sleep(delay).await;
+                    continue;
+                } else {
+                    let event = Event {
+                        id: sub_id.clone(),
+                        msg: EventMsg::Error(ErrorEvent {
+                            message: e.to_string(),
+                        }),
+                    };
+                    sess.send_event(event).await;
+                    return;
+                }
+            }
+        }
+    }
+
+    if remove_task_on_completion {
+        sess.remove_task(&sub_id);
+    }
+    let history_snapshot = {
+        let state = sess.state.lock_unchecked();
+        state.history.contents()
+    };
+    let summary_text = get_last_assistant_message_from_turn(&history_snapshot).unwrap_or_default();
+    let user_messages = collect_user_messages(&history_snapshot);
+    let new_history =
+        build_compacted_history(&sess, turn_context.as_ref(), &user_messages, &summary_text);
+    {
+        let mut state = sess.state.lock_unchecked();
+        state.history.replace(new_history);
+    }
+
+    let rollout_item = RolloutItem::Compacted(CompactedItem {
+        message: summary_text.clone(),
+    });
+    sess.persist_rollout_items(&[rollout_item]).await;
+
+    let event = Event {
+        id: sub_id.clone(),
+        msg: EventMsg::AgentMessage(AgentMessageEvent {
+            message: "Compact task completed".to_string(),
+        }),
+    };
+    sess.send_event(event).await;
+    let event = Event {
+        id: sub_id.clone(),
+        msg: EventMsg::TaskComplete(TaskCompleteEvent {
+            last_agent_message: None,
+        }),
+    };
+    sess.send_event(event).await;
+}
+
+fn content_items_to_text(content: &[ContentItem]) -> Option<String> {
+    let mut pieces = Vec::new();
+    for item in content {
+        match item {
+            ContentItem::InputText { text } | ContentItem::OutputText { text } => {
+                if !text.is_empty() {
+                    pieces.push(text.as_str());
+                }
+            }
+            ContentItem::InputImage { .. } => {}
+        }
+    }
+    if pieces.is_empty() {
+        None
+    } else {
+        Some(pieces.join("\n"))
+    }
+}
+
+fn collect_user_messages(items: &[ResponseItem]) -> Vec<String> {
+    items
+        .iter()
+        .filter_map(|item| match item {
+            ResponseItem::Message { role, content, .. } if role == "user" => {
+                content_items_to_text(content)
+            }
+            _ => None,
+        })
+        .filter(|text| !is_session_prefix_message(text))
+        .collect()
+}
+
+fn is_session_prefix_message(text: &str) -> bool {
+    matches!(
+        InputMessageKind::from(("user", text)),
+        InputMessageKind::UserInstructions | InputMessageKind::EnvironmentContext
+    )
+}
+
+fn build_compacted_history(
+    sess: &Session,
+    turn_context: &TurnContext,
+    user_messages: &[String],
+    summary_text: &str,
+) -> Vec<ResponseItem> {
+    let mut history = sess.build_initial_context(turn_context);
+    let user_messages_text = if user_messages.is_empty() {
+        "(none)".to_string()
+    } else {
+        user_messages.join("\n\n")
+    };
+    let summary_text = if summary_text.is_empty() {
+        "(no summary available)".to_string()
+    } else {
+        summary_text.to_string()
+    };
+    let Ok(bridge) = HistoryBridgeTemplate {
+        user_messages_text: &user_messages_text,
+        summary_text: &summary_text,
+    }
+    .render() else {
+        return vec![];
+    };
+    history.push(ResponseItem::Message {
+        id: None,
+        role: "user".to_string(),
+        content: vec![ContentItem::InputText { text: bridge }],
+    });
+    history
+}
+
+async fn drain_to_completed(
+    sess: &Session,
+    turn_context: &TurnContext,
+    prompt: &Prompt,
+) -> CodexResult<()> {
+    let mut stream = turn_context.client.clone().stream(prompt).await?;
+    loop {
+        let maybe_event = stream.next().await;
+        let Some(event) = maybe_event else {
+            return Err(CodexErr::Stream(
+                "stream closed before response.completed".into(),
+                None,
+            ));
+        };
+        match event {
+            Ok(ResponseEvent::OutputItemDone(item)) => {
+                let mut state = sess.state.lock_unchecked();
+                state.history.record_items(std::slice::from_ref(&item));
+            }
+            Ok(ResponseEvent::Completed { .. }) => {
+                return Ok(());
+            }
+            Ok(_) => continue,
+            Err(e) => return Err(e),
+        }
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use pretty_assertions::assert_eq;
+
+    #[test]
+    fn content_items_to_text_joins_non_empty_segments() {
+        let items = vec![
+            ContentItem::InputText {
+                text: "hello".to_string(),
+            },
+            ContentItem::OutputText {
+                text: String::new(),
+            },
+            ContentItem::OutputText {
+                text: "world".to_string(),
+            },
+        ];
+
+        let joined = content_items_to_text(&items);
+
+        assert_eq!(Some("hello\nworld".to_string()), joined);
+    }
+
+    #[test]
+    fn content_items_to_text_ignores_image_only_content() {
+        let items = vec![ContentItem::InputImage {
+            image_url: "file://image.png".to_string(),
+        }];
+
+        let joined = content_items_to_text(&items);
+
+        assert_eq!(None, joined);
+    }
+
+    #[test]
+    fn collect_user_messages_extracts_user_text_only() {
+        let items = vec![
+            ResponseItem::Message {
+                id: Some("assistant".to_string()),
+                role: "assistant".to_string(),
+                content: vec![ContentItem::OutputText {
+                    text: "ignored".to_string(),
+                }],
+            },
+            ResponseItem::Message {
+                id: Some("user".to_string()),
+                role: "user".to_string(),
+                content: vec![
+                    ContentItem::InputText {
+                        text: "first".to_string(),
+                    },
+                    ContentItem::OutputText {
+                        text: "second".to_string(),
+                    },
+                ],
+            },
+            ResponseItem::Other,
+        ];
+
+        let collected = collect_user_messages(&items);
+
+        assert_eq!(vec!["first\nsecond".to_string()], collected);
+    }
+
+    #[test]
+    fn collect_user_messages_filters_session_prefix_entries() {
+        let items = vec![
+            ResponseItem::Message {
+                id: None,
+                role: "user".to_string(),
+                content: vec![ContentItem::InputText {
+                    text: "<user_instructions>do things</user_instructions>".to_string(),
+                }],
+            },
+            ResponseItem::Message {
+                id: None,
+                role: "user".to_string(),
+                content: vec![ContentItem::InputText {
+                    text: "<ENVIRONMENT_CONTEXT>cwd=/tmp</ENVIRONMENT_CONTEXT>".to_string(),
+                }],
+            },
+            ResponseItem::Message {
+                id: None,
+                role: "user".to_string(),
+                content: vec![ContentItem::InputText {
+                    text: "real user message".to_string(),
+                }],
+            },
+        ];
+
+        let collected = collect_user_messages(&items);
+
+        assert_eq!(vec!["real user message".to_string()], collected);
+    }
+}

codex-rs/core/src/config.rs

@@ -32,7 +32,8 @@ use toml::Value as TomlValue;
 use toml_edit::DocumentMut;
 
 const OPENAI_DEFAULT_MODEL: &str = "gpt-5";
-pub const GPT5_HIGH_MODEL: &str = "gpt-5-high";
+const OPENAI_DEFAULT_REVIEW_MODEL: &str = "gpt-5";
+pub const GPT5_HIGH_MODEL: &str = "gpt-5-high-new";
 
 /// 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
@@ -47,6 +48,9 @@ pub struct Config {
     /// Optional override of model selection.
     pub model: String,
 
+    /// Model used specifically for review sessions. Defaults to "gpt-5".
+    pub review_model: String,
+
     pub model_family: ModelFamily,
 
     /// Size of the context window for the model, in tokens.
@@ -55,6 +59,9 @@ pub struct Config {
     /// Maximum number of output tokens.
     pub model_max_output_tokens: Option<u64>,
 
+    /// Token usage threshold triggering auto-compaction of conversation history.
+    pub model_auto_compact_token_limit: Option<i64>,
+
     /// Key into the model_providers map that specifies which provider to use.
     pub model_provider_id: String,
 
@@ -140,7 +147,7 @@ pub struct Config {
 
     /// Value to use for `reasoning.effort` when making a request using the
     /// Responses API.
-    pub model_reasoning_effort: ReasoningEffort,
+    pub model_reasoning_effort: Option<ReasoningEffort>,
 
     /// If not "none", the value to use for `reasoning.summary` when making a
     /// request using the Responses API.
@@ -423,14 +430,24 @@ pub async fn persist_model_selection(
     if let Some(profile_name) = active_profile {
         let profile_table = ensure_profile_table(&mut doc, profile_name)?;
         profile_table["model"] = toml_edit::value(model);
-        if let Some(effort) = effort {
-            profile_table["model_reasoning_effort"] = toml_edit::value(effort.to_string());
+        match effort {
+            Some(effort) => {
+                profile_table["model_reasoning_effort"] = toml_edit::value(effort.to_string());
+            }
+            None => {
+                profile_table.remove("model_reasoning_effort");
+            }
         }
     } else {
         let table = doc.as_table_mut();
         table["model"] = toml_edit::value(model);
-        if let Some(effort) = effort {
-            table["model_reasoning_effort"] = toml_edit::value(effort.to_string());
+        match effort {
+            Some(effort) => {
+                table["model_reasoning_effort"] = toml_edit::value(effort.to_string());
+            }
+            None => {
+                table.remove("model_reasoning_effort");
+            }
         }
     }
 
@@ -499,6 +516,8 @@ fn apply_toml_override(root: &mut TomlValue, path: &str, value: TomlValue) {
 pub struct ConfigToml {
     /// Optional override of model selection.
     pub model: Option<String>,
+    /// Review model override used by the `/review` feature.
+    pub review_model: Option<String>,
 
     /// Provider to use from the model_providers map.
     pub model_provider: Option<String>,
@@ -509,6 +528,9 @@ pub struct ConfigToml {
     /// Maximum number of output tokens.
     pub model_max_output_tokens: Option<u64>,
 
+    /// Token usage threshold triggering auto-compaction of conversation history.
+    pub model_auto_compact_token_limit: Option<i64>,
+
     /// Default approval policy for executing commands.
     pub approval_policy: Option<AskForApproval>,
 
@@ -724,6 +746,7 @@ impl ConfigToml {
 #[derive(Default, Debug, Clone)]
 pub struct ConfigOverrides {
     pub model: Option<String>,
+    pub review_model: Option<String>,
     pub cwd: Option<PathBuf>,
     pub approval_policy: Option<AskForApproval>,
     pub sandbox_mode: Option<SandboxMode>,
@@ -751,6 +774,7 @@ impl Config {
         // Destructure ConfigOverrides fully to ensure all overrides are applied.
         let ConfigOverrides {
             model,
+            review_model: override_review_model,
             cwd,
             approval_policy,
             sandbox_mode,
@@ -867,6 +891,11 @@ impl Config {
                 .as_ref()
                 .map(|info| info.max_output_tokens)
         });
+        let model_auto_compact_token_limit = cfg.model_auto_compact_token_limit.or_else(|| {
+            openai_model_info
+                .as_ref()
+                .and_then(|info| info.auto_compact_token_limit)
+        });
 
         let experimental_resume = cfg.experimental_resume;
 
@@ -881,11 +910,18 @@ impl Config {
             Self::get_base_instructions(experimental_instructions_path, &resolved_cwd)?;
         let base_instructions = base_instructions.or(file_base_instructions);
 
+        // Default review model when not set in config; allow CLI override to take precedence.
+        let review_model = override_review_model
+            .or(cfg.review_model)
+            .unwrap_or_else(default_review_model);
+
         let config = Self {
             model,
+            review_model,
             model_family,
             model_context_window,
             model_max_output_tokens,
+            model_auto_compact_token_limit,
             model_provider_id,
             model_provider,
             cwd: resolved_cwd,
@@ -913,8 +949,7 @@ impl Config {
                 .unwrap_or(false),
             model_reasoning_effort: config_profile
                 .model_reasoning_effort
-                .or(cfg.model_reasoning_effort)
-                .unwrap_or_default(),
+                .or(cfg.model_reasoning_effort),
             model_reasoning_summary: config_profile
                 .model_reasoning_summary
                 .or(cfg.model_reasoning_summary)
@@ -1006,6 +1041,10 @@ fn default_model() -> String {
     OPENAI_DEFAULT_MODEL.to_string()
 }
 
+fn default_review_model() -> String {
+    OPENAI_DEFAULT_REVIEW_MODEL.to_string()
+}
+
 /// Returns the path to the Codex configuration directory, which can be
 /// specified by the `CODEX_HOME` environment variable. If not set, defaults to
 /// `~/.codex`.
@@ -1418,9 +1457,11 @@ model_verbosity = "high"
         assert_eq!(
             Config {
                 model: "o3".to_string(),
+                review_model: "gpt-5".to_string(),
                 model_family: find_family_for_model("o3").expect("known model slug"),
                 model_context_window: Some(200_000),
                 model_max_output_tokens: Some(100_000),
+                model_auto_compact_token_limit: None,
                 model_provider_id: "openai".to_string(),
                 model_provider: fixture.openai_provider.clone(),
                 approval_policy: AskForApproval::Never,
@@ -1438,7 +1479,7 @@ model_verbosity = "high"
                 codex_linux_sandbox_exe: None,
                 hide_agent_reasoning: false,
                 show_raw_agent_reasoning: false,
-                model_reasoning_effort: ReasoningEffort::High,
+                model_reasoning_effort: Some(ReasoningEffort::High),
                 model_reasoning_summary: ReasoningSummary::Detailed,
                 model_verbosity: None,
                 chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
@@ -1474,9 +1515,11 @@ model_verbosity = "high"
         )?;
         let expected_gpt3_profile_config = Config {
             model: "gpt-3.5-turbo".to_string(),
+            review_model: "gpt-5".to_string(),
             model_family: find_family_for_model("gpt-3.5-turbo").expect("known model slug"),
             model_context_window: Some(16_385),
             model_max_output_tokens: Some(4_096),
+            model_auto_compact_token_limit: None,
             model_provider_id: "openai-chat-completions".to_string(),
             model_provider: fixture.openai_chat_completions_provider.clone(),
             approval_policy: AskForApproval::UnlessTrusted,
@@ -1494,7 +1537,7 @@ model_verbosity = "high"
             codex_linux_sandbox_exe: None,
             hide_agent_reasoning: false,
             show_raw_agent_reasoning: false,
-            model_reasoning_effort: ReasoningEffort::default(),
+            model_reasoning_effort: None,
             model_reasoning_summary: ReasoningSummary::default(),
             model_verbosity: None,
             chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
@@ -1545,9 +1588,11 @@ model_verbosity = "high"
         )?;
         let expected_zdr_profile_config = Config {
             model: "o3".to_string(),
+            review_model: "gpt-5".to_string(),
             model_family: find_family_for_model("o3").expect("known model slug"),
             model_context_window: Some(200_000),
             model_max_output_tokens: Some(100_000),
+            model_auto_compact_token_limit: None,
             model_provider_id: "openai".to_string(),
             model_provider: fixture.openai_provider.clone(),
             approval_policy: AskForApproval::OnFailure,
@@ -1565,7 +1610,7 @@ model_verbosity = "high"
             codex_linux_sandbox_exe: None,
             hide_agent_reasoning: false,
             show_raw_agent_reasoning: false,
-            model_reasoning_effort: ReasoningEffort::default(),
+            model_reasoning_effort: None,
             model_reasoning_summary: ReasoningSummary::default(),
             model_verbosity: None,
             chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),
@@ -1602,9 +1647,11 @@ model_verbosity = "high"
         )?;
         let expected_gpt5_profile_config = Config {
             model: "gpt-5".to_string(),
+            review_model: "gpt-5".to_string(),
             model_family: find_family_for_model("gpt-5").expect("known model slug"),
             model_context_window: Some(272_000),
             model_max_output_tokens: Some(128_000),
+            model_auto_compact_token_limit: None,
             model_provider_id: "openai".to_string(),
             model_provider: fixture.openai_provider.clone(),
             approval_policy: AskForApproval::OnFailure,
@@ -1622,7 +1669,7 @@ model_verbosity = "high"
             codex_linux_sandbox_exe: None,
             hide_agent_reasoning: false,
             show_raw_agent_reasoning: false,
-            model_reasoning_effort: ReasoningEffort::High,
+            model_reasoning_effort: Some(ReasoningEffort::High),
             model_reasoning_summary: ReasoningSummary::Detailed,
             model_verbosity: Some(Verbosity::High),
             chatgpt_base_url: "https://chatgpt.com/backend-api/".to_string(),

codex-rs/core/src/config_edit.rs

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

codex-rs/core/src/conversation_history.rs

@@ -1,4 +1,3 @@
-use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseItem;
 
 /// Transcript of conversation history
@@ -33,52 +32,8 @@ impl ConversationHistory {
         }
     }
 
-    pub(crate) fn keep_last_messages(&mut self, n: usize) {
-        if n == 0 {
-            self.items.clear();
-            return;
-        }
-
-        // Collect the last N message items (assistant/user), newest to oldest.
-        let mut kept: Vec<ResponseItem> = Vec::with_capacity(n);
-        for item in self.items.iter().rev() {
-            if let ResponseItem::Message { role, content, .. } = item {
-                kept.push(ResponseItem::Message {
-                    // we need to remove the id or the model will complain that messages are sent without
-                    // their reasonings
-                    id: None,
-                    role: role.clone(),
-                    content: content.clone(),
-                });
-                if kept.len() == n {
-                    break;
-                }
-            }
-        }
-
-        // Preserve chronological order (oldest to newest) within the kept slice.
-        kept.reverse();
-        self.items = kept;
-    }
-
-    pub(crate) fn last_agent_message(&self) -> String {
-        for item in self.items.iter().rev() {
-            if let ResponseItem::Message { role, content, .. } = item
-                && role == "assistant"
-            {
-                return content
-                    .iter()
-                    .find_map(|ci| {
-                        if let ContentItem::OutputText { text } = ci {
-                            Some(text.clone())
-                        } else {
-                            None
-                        }
-                    })
-                    .unwrap_or_default();
-            }
-        }
-        String::new()
+    pub(crate) fn replace(&mut self, items: Vec<ResponseItem>) {
+        self.items = items;
     }
 }
 

codex-rs/core/src/model_provider_info.rs

@@ -162,6 +162,21 @@ impl ModelProviderInfo {
         }
     }
 
+    pub(crate) fn is_azure_responses_endpoint(&self) -> bool {
+        if self.wire_api != WireApi::Responses {
+            return false;
+        }
+
+        if self.name.eq_ignore_ascii_case("azure") {
+            return true;
+        }
+
+        self.base_url
+            .as_ref()
+            .map(|base| matches_azure_responses_base_url(base))
+            .unwrap_or(false)
+    }
+
     /// Apply provider-specific HTTP headers (both static and environment-based)
     /// onto an existing `reqwest::RequestBuilder` and return the updated
     /// builder.
@@ -329,6 +344,18 @@ pub fn create_oss_provider_with_base_url(base_url: &str) -> ModelProviderInfo {
     }
 }
 
+fn matches_azure_responses_base_url(base_url: &str) -> bool {
+    let base = base_url.to_ascii_lowercase();
+    const AZURE_MARKERS: [&str; 5] = [
+        "openai.azure.",
+        "cognitiveservices.azure.",
+        "aoai.azure.",
+        "azure-api.",
+        "azurefd.",
+    ];
+    AZURE_MARKERS.iter().any(|marker| base.contains(marker))
+}
+
 #[cfg(test)]
 mod tests {
     use super::*;
@@ -419,4 +446,69 @@ env_http_headers = { "X-Example-Env-Header" = "EXAMPLE_ENV_VAR" }
         let provider: ModelProviderInfo = toml::from_str(azure_provider_toml).unwrap();
         assert_eq!(expected_provider, provider);
     }
+
+    #[test]
+    fn detects_azure_responses_base_urls() {
+        fn provider_for(base_url: &str) -> ModelProviderInfo {
+            ModelProviderInfo {
+                name: "test".into(),
+                base_url: Some(base_url.into()),
+                env_key: None,
+                env_key_instructions: None,
+                wire_api: WireApi::Responses,
+                query_params: None,
+                http_headers: None,
+                env_http_headers: None,
+                request_max_retries: None,
+                stream_max_retries: None,
+                stream_idle_timeout_ms: None,
+                requires_openai_auth: false,
+            }
+        }
+
+        let positive_cases = [
+            "https://foo.openai.azure.com/openai",
+            "https://foo.openai.azure.us/openai/deployments/bar",
+            "https://foo.cognitiveservices.azure.cn/openai",
+            "https://foo.aoai.azure.com/openai",
+            "https://foo.openai.azure-api.net/openai",
+            "https://foo.z01.azurefd.net/",
+        ];
+        for base_url in positive_cases {
+            let provider = provider_for(base_url);
+            assert!(
+                provider.is_azure_responses_endpoint(),
+                "expected {base_url} to be detected as Azure"
+            );
+        }
+
+        let named_provider = ModelProviderInfo {
+            name: "Azure".into(),
+            base_url: Some("https://example.com".into()),
+            env_key: None,
+            env_key_instructions: None,
+            wire_api: WireApi::Responses,
+            query_params: None,
+            http_headers: None,
+            env_http_headers: None,
+            request_max_retries: None,
+            stream_max_retries: None,
+            stream_idle_timeout_ms: None,
+            requires_openai_auth: false,
+        };
+        assert!(named_provider.is_azure_responses_endpoint());
+
+        let negative_cases = [
+            "https://api.openai.com/v1",
+            "https://example.com/openai",
+            "https://myproxy.azurewebsites.net/openai",
+        ];
+        for base_url in negative_cases {
+            let provider = provider_for(base_url);
+            assert!(
+                !provider.is_azure_responses_endpoint(),
+                "expected {base_url} not to be detected as Azure"
+            );
+        }
+    }
 }

codex-rs/core/src/openai_model_info.rs

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

codex-rs/core/src/openai_tools.rs

@@ -8,6 +8,7 @@ use std::collections::HashMap;
 use crate::model_family::ModelFamily;
 use crate::plan_tool::PLAN_TOOL;
 use crate::protocol::AskForApproval;
+use crate::protocol::SandboxPolicy;
 use crate::tool_apply_patch::ApplyPatchToolType;
 use crate::tool_apply_patch::create_apply_patch_freeform_tool;
 use crate::tool_apply_patch::create_apply_patch_json_tool;
@@ -57,7 +58,7 @@ pub(crate) enum OpenAiTool {
 #[derive(Debug, Clone)]
 pub enum ConfigShellToolType {
     DefaultShell,
-    ShellWithRequest,
+    ShellWithRequest { sandbox_policy: SandboxPolicy },
     LocalShell,
     StreamableShell,
 }
@@ -75,6 +76,7 @@ pub(crate) struct ToolsConfig {
 pub(crate) struct ToolsConfigParams<'a> {
     pub(crate) model_family: &'a ModelFamily,
     pub(crate) approval_policy: AskForApproval,
+    pub(crate) sandbox_policy: SandboxPolicy,
     pub(crate) include_plan_tool: bool,
     pub(crate) include_apply_patch_tool: bool,
     pub(crate) include_web_search_request: bool,
@@ -88,6 +90,7 @@ impl ToolsConfig {
         let ToolsConfigParams {
             model_family,
             approval_policy,
+            sandbox_policy,
             include_plan_tool,
             include_apply_patch_tool,
             include_web_search_request,
@@ -103,7 +106,9 @@ impl ToolsConfig {
             ConfigShellToolType::DefaultShell
         };
         if matches!(approval_policy, AskForApproval::OnRequest) && !use_streamable_shell_tool {
-            shell_type = ConfigShellToolType::ShellWithRequest;
+            shell_type = ConfigShellToolType::ShellWithRequest {
+                sandbox_policy: sandbox_policy.clone(),
+            }
         }
 
         let apply_patch_tool_type = match model_family.apply_patch_tool_type {
@@ -246,9 +251,7 @@ fn create_unified_exec_tool() -> OpenAiTool {
     })
 }
 
-const SHELL_TOOL_DESCRIPTION: &str = r#"Runs a shell command and returns its output"#;
-
-fn create_shell_tool_for_request() -> OpenAiTool {
+fn create_shell_tool_for_sandbox(sandbox_policy: &SandboxPolicy) -> OpenAiTool {
     let mut properties = BTreeMap::new();
     properties.insert(
         "command".to_string(),
@@ -260,33 +263,34 @@ fn create_shell_tool_for_request() -> OpenAiTool {
     properties.insert(
         "workdir".to_string(),
         JsonSchema::String {
-            description: Some("Working directory to execute the command in.".to_string()),
+            description: Some("The working directory to execute the command in".to_string()),
         },
     );
     properties.insert(
         "timeout_ms".to_string(),
         JsonSchema::Number {
-            description: Some("Timeout for the command in milliseconds.".to_string()),
-        },
-    );
-    properties.insert(
-        "with_escalated_permissions".to_string(),
-        JsonSchema::Boolean {
-            description: Some("Request escalated permissions, only for when a command would otherwise be blocked by the sandbox.".to_string()),
-        },
-    );
-    properties.insert(
-        "justification".to_string(),
-        JsonSchema::String {
-            description: Some("Required if and only if with_escalated_permissions == true. One sentence explaining why escalation is needed (e.g., write outside CWD, network fetch, git commit).".to_string()),
+            description: Some("The timeout for the command in milliseconds".to_string()),
         },
     );
 
-    let description = SHELL_TOOL_DESCRIPTION.to_string();
+    if matches!(sandbox_policy, SandboxPolicy::WorkspaceWrite { .. }) {
+        properties.insert(
+        "with_escalated_permissions".to_string(),
+        JsonSchema::Boolean {
+            description: Some("Whether to request escalated permissions. Set to true if command needs to be run without sandbox restrictions".to_string()),
+        },
+    );
+        properties.insert(
+        "justification".to_string(),
+        JsonSchema::String {
+            description: Some("Only set if with_escalated_permissions is true. 1-sentence explanation of why we want to run this command.".to_string()),
+        },
+    );
+    }
 
     OpenAiTool::Function(ResponsesApiTool {
         name: "shell".to_string(),
-        description,
+        description: "Runs a shell command and returns its output.".to_string(),
         strict: false,
         parameters: JsonSchema::Object {
             properties,
@@ -295,6 +299,7 @@ fn create_shell_tool_for_request() -> OpenAiTool {
         },
     })
 }
+
 fn create_view_image_tool() -> OpenAiTool {
     // Support only local filesystem path.
     let mut properties = BTreeMap::new();
@@ -535,8 +540,8 @@ pub(crate) fn get_openai_tools(
             ConfigShellToolType::DefaultShell => {
                 tools.push(create_shell_tool());
             }
-            ConfigShellToolType::ShellWithRequest => {
-                tools.push(create_shell_tool_for_request());
+            ConfigShellToolType::ShellWithRequest { sandbox_policy } => {
+                tools.push(create_shell_tool_for_sandbox(sandbox_policy));
             }
             ConfigShellToolType::LocalShell => {
                 tools.push(OpenAiTool::LocalShell {});
@@ -632,6 +637,7 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: true,
             include_apply_patch_tool: false,
             include_web_search_request: true,
@@ -653,6 +659,7 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: true,
             include_apply_patch_tool: false,
             include_web_search_request: true,
@@ -674,6 +681,7 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             include_web_search_request: true,
@@ -779,6 +787,7 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             include_web_search_request: false,
@@ -856,6 +865,7 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             include_web_search_request: true,
@@ -918,6 +928,7 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             include_web_search_request: true,
@@ -975,6 +986,7 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             include_web_search_request: true,
@@ -1035,6 +1047,7 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_family: &model_family,
             approval_policy: AskForApproval::Never,
+            sandbox_policy: SandboxPolicy::ReadOnly,
             include_plan_tool: false,
             include_apply_patch_tool: false,
             include_web_search_request: true,
@@ -1088,7 +1101,13 @@ mod tests {
 
     #[test]
     fn test_shell_tool_for_sandbox_workspace_write() {
-        let tool = super::create_shell_tool_for_request();
+        let sandbox_policy = SandboxPolicy::WorkspaceWrite {
+            writable_roots: vec!["workspace".into()],
+            network_access: false,
+            exclude_tmpdir_env_var: false,
+            exclude_slash_tmp: false,
+        };
+        let tool = super::create_shell_tool_for_sandbox(&sandbox_policy);
         let OpenAiTool::Function(ResponsesApiTool {
             description, name, ..
         }) = &tool
@@ -1097,13 +1116,13 @@ mod tests {
         };
         assert_eq!(name, "shell");
 
-        let expected = super::SHELL_TOOL_DESCRIPTION;
+        let expected = "Runs a shell command and returns its output.";
         assert_eq!(description, expected);
     }
 
     #[test]
     fn test_shell_tool_for_sandbox_readonly() {
-        let tool = super::create_shell_tool_for_request();
+        let tool = super::create_shell_tool_for_sandbox(&SandboxPolicy::ReadOnly);
         let OpenAiTool::Function(ResponsesApiTool {
             description, name, ..
         }) = &tool
@@ -1112,13 +1131,13 @@ mod tests {
         };
         assert_eq!(name, "shell");
 
-        let expected = super::SHELL_TOOL_DESCRIPTION;
+        let expected = "Runs a shell command and returns its output.";
         assert_eq!(description, expected);
     }
 
     #[test]
     fn test_shell_tool_for_sandbox_danger_full_access() {
-        let tool = super::create_shell_tool_for_request();
+        let tool = super::create_shell_tool_for_sandbox(&SandboxPolicy::DangerFullAccess);
         let OpenAiTool::Function(ResponsesApiTool {
             description, name, ..
         }) = &tool
@@ -1127,7 +1146,6 @@ mod tests {
         };
         assert_eq!(name, "shell");
 
-        let expected = super::SHELL_TOOL_DESCRIPTION;
-        assert_eq!(description, expected);
+        assert_eq!(description, "Runs a shell command and returns its output.");
     }
 }

codex-rs/core/src/project_doc.rs

@@ -26,7 +26,7 @@ const PROJECT_DOC_SEPARATOR: &str = "\n\n--- project-doc ---\n\n";
 
 /// Combines `Config::instructions` and `AGENTS.md` (if present) into a single
 /// string of instructions.
-pub async fn get_user_instructions(config: &Config) -> Option<String> {
+pub(crate) async fn get_user_instructions(config: &Config) -> Option<String> {
     match read_project_docs(config).await {
         Ok(Some(project_doc)) => match &config.user_instructions {
             Some(original_instructions) => Some(format!(

codex-rs/core/src/prompt_for_compact_command.md

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

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

@@ -38,7 +38,9 @@ pub(crate) fn should_persist_event_msg(ev: &EventMsg) -> bool {
         | EventMsg::AgentMessage(_)
         | EventMsg::AgentReasoning(_)
         | EventMsg::AgentReasoningRawContent(_)
-        | EventMsg::TokenCount(_) => true,
+        | EventMsg::TokenCount(_)
+        | EventMsg::EnteredReviewMode(_)
+        | EventMsg::ExitedReviewMode(_) => true,
         EventMsg::Error(_)
         | EventMsg::TaskStarted(_)
         | EventMsg::TaskComplete(_)

codex-rs/core/src/seatbelt_base_policy.sbpl

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

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

@@ -421,7 +421,7 @@ mod tests {
             .handle_request(UnifiedExecRequest {
                 session_id: None,
                 input_chunks: &["bash".to_string(), "-i".to_string()],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
         let session_id = open_shell.session_id.expect("expected session_id");
@@ -441,7 +441,7 @@ mod tests {
             .handle_request(UnifiedExecRequest {
                 session_id: Some(session_id),
                 input_chunks: &["echo $CODEX_INTERACTIVE_SHELL_VAR\n".to_string()],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
         assert!(out_2.output.contains("codex"));
@@ -458,7 +458,7 @@ mod tests {
             .handle_request(UnifiedExecRequest {
                 session_id: None,
                 input_chunks: &["/bin/bash".to_string(), "-i".to_string()],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
         let session_a = shell_a.session_id.expect("expected session id");
@@ -467,7 +467,7 @@ mod tests {
             .handle_request(UnifiedExecRequest {
                 session_id: Some(session_a),
                 input_chunks: &["export CODEX_INTERACTIVE_SHELL_VAR=codex\n".to_string()],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
 
@@ -478,7 +478,7 @@ mod tests {
                     "echo".to_string(),
                     "$CODEX_INTERACTIVE_SHELL_VAR\n".to_string(),
                 ],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
         assert!(!out_2.output.contains("codex"));
@@ -487,7 +487,7 @@ mod tests {
             .handle_request(UnifiedExecRequest {
                 session_id: Some(session_a),
                 input_chunks: &["echo $CODEX_INTERACTIVE_SHELL_VAR\n".to_string()],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
         assert!(out_3.output.contains("codex"));
@@ -504,7 +504,7 @@ mod tests {
             .handle_request(UnifiedExecRequest {
                 session_id: None,
                 input_chunks: &["bash".to_string(), "-i".to_string()],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
         let session_id = open_shell.session_id.expect("expected session id");
@@ -516,7 +516,7 @@ mod tests {
                     "export".to_string(),
                     "CODEX_INTERACTIVE_SHELL_VAR=codex\n".to_string(),
                 ],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
 
@@ -574,7 +574,7 @@ mod tests {
             .handle_request(UnifiedExecRequest {
                 session_id: None,
                 input_chunks: &["/bin/echo".to_string(), "codex".to_string()],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
 
@@ -595,7 +595,7 @@ mod tests {
             .handle_request(UnifiedExecRequest {
                 session_id: None,
                 input_chunks: &["/bin/bash".to_string(), "-i".to_string()],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
         let session_id = open_shell.session_id.expect("expected session id");
@@ -604,7 +604,7 @@ mod tests {
             .handle_request(UnifiedExecRequest {
                 session_id: Some(session_id),
                 input_chunks: &["exit\n".to_string()],
-                timeout_ms: Some(1_500),
+                timeout_ms: Some(2_500),
             })
             .await?;
 

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

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

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

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

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

@@ -1,20 +1,32 @@
 use codex_core::CodexAuth;
+use codex_core::ContentItem;
 use codex_core::ConversationManager;
+use codex_core::LocalShellAction;
+use codex_core::LocalShellExecAction;
+use codex_core::LocalShellStatus;
+use codex_core::ModelClient;
 use codex_core::ModelProviderInfo;
 use codex_core::NewConversation;
+use codex_core::Prompt;
+use codex_core::ReasoningItemContent;
+use codex_core::ResponseEvent;
+use codex_core::ResponseItem;
 use codex_core::WireApi;
 use codex_core::built_in_model_providers;
-use codex_core::project_doc::get_user_instructions;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::InputItem;
 use codex_core::protocol::Op;
-use codex_core::shell::default_user_shell;
 use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
+use codex_protocol::mcp_protocol::ConversationId;
+use codex_protocol::models::ReasoningItemReasoningSummary;
+use codex_protocol::models::WebSearchAction;
 use core_test_support::load_default_config_for_test;
 use core_test_support::load_sse_fixture_with_id;
 use core_test_support::wait_for_event;
+use futures::StreamExt;
 use serde_json::json;
 use std::io::Write;
+use std::sync::Arc;
 use tempfile::TempDir;
 use uuid::Uuid;
 use wiremock::Mock;
@@ -223,8 +235,6 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
     };
     let codex_home = TempDir::new().unwrap();
     let mut config = load_default_config_for_test(&codex_home);
-    let cwd = TempDir::new().unwrap();
-    config.cwd = cwd.path().to_path_buf();
     config.model_provider = model_provider;
     config.experimental_resume = Some(session_path.clone());
     // Also configure user instructions to ensure they are NOT delivered on resume.
@@ -263,29 +273,6 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
 
     let request = &server.received_requests().await.unwrap()[0];
     let request_body = request.body_json::<serde_json::Value>().unwrap();
-
-    // Build expected environment context for this turn.
-    let shell = default_user_shell().await;
-    let shell_line = match shell.name() {
-        Some(name) => format!("  <shell>{name}</shell>\n"),
-        None => String::new(),
-    };
-    let expected_env_text_turn = format!(
-        r#"<environment_context>
-  <cwd>{}</cwd>
-  <approval_policy>on-request</approval_policy>
-  <sandbox_mode>read-only</sandbox_mode>
-  <network_access>restricted</network_access>
-{}</environment_context>"#,
-        cwd.path().to_string_lossy(),
-        shell_line.as_str(),
-    );
-    let expected_env_msg_turn = json!({
-        "type": "message",
-        "role": "user",
-        "content": [ { "type": "input_text", "text": expected_env_text_turn } ]
-    });
-
     let expected_input = json!([
         {
             "type": "message",
@@ -297,14 +284,12 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
             "role": "assistant",
             "content": [{ "type": "output_text", "text": "resumed assistant message" }]
         },
-        expected_env_msg_turn,
         {
             "type": "message",
             "role": "user",
             "content": [{ "type": "input_text", "text": "hello" }]
         }
     ]);
-
     assert_eq!(request_body["input"], expected_input);
 }
 
@@ -649,6 +634,147 @@ async fn includes_user_instructions_message_in_request() {
     assert_message_ends_with(&request_body["input"][1], "</environment_context>");
 }
 
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn azure_responses_request_includes_store_and_reasoning_ids() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let server = MockServer::start().await;
+
+    let sse_body = concat!(
+        "data: {\"type\":\"response.created\",\"response\":{}}\n\n",
+        "data: {\"type\":\"response.completed\",\"response\":{\"id\":\"resp_1\"}}\n\n",
+    );
+
+    let template = ResponseTemplate::new(200)
+        .insert_header("content-type", "text/event-stream")
+        .set_body_raw(sse_body, "text/event-stream");
+
+    Mock::given(method("POST"))
+        .and(path("/openai/responses"))
+        .respond_with(template)
+        .expect(1)
+        .mount(&server)
+        .await;
+
+    let provider = ModelProviderInfo {
+        name: "azure".into(),
+        base_url: Some(format!("{}/openai", server.uri())),
+        env_key: None,
+        env_key_instructions: None,
+        wire_api: WireApi::Responses,
+        query_params: None,
+        http_headers: None,
+        env_http_headers: None,
+        request_max_retries: Some(0),
+        stream_max_retries: Some(0),
+        stream_idle_timeout_ms: Some(5_000),
+        requires_openai_auth: false,
+    };
+
+    let codex_home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&codex_home);
+    config.model_provider_id = provider.name.clone();
+    config.model_provider = provider.clone();
+    let effort = config.model_reasoning_effort;
+    let summary = config.model_reasoning_summary;
+    let config = Arc::new(config);
+
+    let client = ModelClient::new(
+        Arc::clone(&config),
+        None,
+        provider,
+        effort,
+        summary,
+        ConversationId::new(),
+    );
+
+    let mut prompt = Prompt::default();
+    prompt.input.push(ResponseItem::Reasoning {
+        id: "reasoning-id".into(),
+        summary: vec![ReasoningItemReasoningSummary::SummaryText {
+            text: "summary".into(),
+        }],
+        content: Some(vec![ReasoningItemContent::ReasoningText {
+            text: "content".into(),
+        }]),
+        encrypted_content: None,
+    });
+    prompt.input.push(ResponseItem::Message {
+        id: Some("message-id".into()),
+        role: "assistant".into(),
+        content: vec![ContentItem::OutputText {
+            text: "message".into(),
+        }],
+    });
+    prompt.input.push(ResponseItem::WebSearchCall {
+        id: Some("web-search-id".into()),
+        status: Some("completed".into()),
+        action: WebSearchAction::Search {
+            query: "weather".into(),
+        },
+    });
+    prompt.input.push(ResponseItem::FunctionCall {
+        id: Some("function-id".into()),
+        name: "do_thing".into(),
+        arguments: "{}".into(),
+        call_id: "function-call-id".into(),
+    });
+    prompt.input.push(ResponseItem::LocalShellCall {
+        id: Some("local-shell-id".into()),
+        call_id: Some("local-shell-call-id".into()),
+        status: LocalShellStatus::Completed,
+        action: LocalShellAction::Exec(LocalShellExecAction {
+            command: vec!["echo".into(), "hello".into()],
+            timeout_ms: None,
+            working_directory: None,
+            env: None,
+            user: None,
+        }),
+    });
+    prompt.input.push(ResponseItem::CustomToolCall {
+        id: Some("custom-tool-id".into()),
+        status: Some("completed".into()),
+        call_id: "custom-tool-call-id".into(),
+        name: "custom_tool".into(),
+        input: "{}".into(),
+    });
+
+    let mut stream = client
+        .stream(&prompt)
+        .await
+        .expect("responses stream to start");
+
+    while let Some(event) = stream.next().await {
+        if let Ok(ResponseEvent::Completed { .. }) = event {
+            break;
+        }
+    }
+
+    let requests = server
+        .received_requests()
+        .await
+        .expect("mock server collected requests");
+    assert_eq!(requests.len(), 1, "expected a single request");
+    let body: serde_json::Value = requests[0]
+        .body_json()
+        .expect("request body to be valid JSON");
+
+    assert_eq!(body["store"], serde_json::Value::Bool(true));
+    assert_eq!(body["stream"], serde_json::Value::Bool(true));
+    assert_eq!(body["input"].as_array().map(Vec::len), Some(6));
+    assert_eq!(body["input"][0]["id"].as_str(), Some("reasoning-id"));
+    assert_eq!(body["input"][1]["id"].as_str(), Some("message-id"));
+    assert_eq!(body["input"][2]["id"].as_str(), Some("web-search-id"));
+    assert_eq!(body["input"][3]["id"].as_str(), Some("function-id"));
+    assert_eq!(body["input"][4]["id"].as_str(), Some("local-shell-id"));
+    assert_eq!(body["input"][5]["id"].as_str(), Some("custom-tool-id"));
+}
+
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn azure_overrides_assign_properties_used_for_responses_url() {
     let existing_env_var_with_random_value = if cfg!(windows) { "USERNAME" } else { "USER" };
@@ -867,7 +993,7 @@ async fn history_dedupes_streamed_and_final_messages_across_turns() {
         conversation: codex,
         ..
     } = conversation_manager
-        .new_conversation(config.clone())
+        .new_conversation(config)
         .await
         .expect("create new conversation");
 
@@ -902,49 +1028,34 @@ async fn history_dedupes_streamed_and_final_messages_across_turns() {
     let requests = server.received_requests().await.unwrap();
     assert_eq!(requests.len(), 3, "expected 3 requests (one per turn)");
 
-    // Build expected environment context dynamically to avoid OS-dependent flakiness.
-    let user_instructions = get_user_instructions(&config).await;
-    let shell = default_user_shell().await;
-    let shell_line = match shell.name() {
-        Some(name) => format!("  <shell>{name}</shell>\n"),
-        None => String::new(),
-    };
-    let expected_env_text = format!(
-        r#"<environment_context>
-  <cwd>{}</cwd>
-  <approval_policy>on-request</approval_policy>
-  <sandbox_mode>read-only</sandbox_mode>
-  <network_access>restricted</network_access>
-{}</environment_context>"#,
-        std::env::current_dir().unwrap().to_string_lossy(),
-        shell_line.as_str(),
-    );
-    let expected_env_msg = json!({
-        "type": "message",
-        "role": "user",
-        "content": [ { "type": "input_text", "text": expected_env_text } ]
-    });
-    // Wrap user instructions in the XML container to match the raw/ingest view
-    let expected_ui_text = format!(
-        "<user_instructions>\n\n{}\n\n</user_instructions>",
-        user_instructions.clone().unwrap()
-    );
-    let expected_ui_msg = json!({
-        "type": "message",
-        "role": "user",
-        "content": [ { "type": "input_text", "text": expected_ui_text } ]
-    });
-
-    let expected_full = json!([
-        expected_ui_msg,
-        expected_env_msg.clone(),
-        {"type":"message","role":"user","content":[{"type":"input_text","text":"U1"}]},
-        {"type":"message","role":"assistant","content":[{"type":"output_text","text":"Hey there!\n"}]},
-        expected_env_msg.clone(),
-        {"type":"message","role":"user","content":[{"type":"input_text","text":"U2"}]},
-        {"type":"message","role":"assistant","content":[{"type":"output_text","text":"Hey there!\n"}]},
-        expected_env_msg,
-        {"type":"message","role":"user","content":[{"type":"input_text","text":"U3"}]}]);
+    // Replace full-array compare with tail-only raw JSON compare using a single hard-coded value.
+    let r3_tail_expected = json!([
+        {
+            "type": "message",
+            "role": "user",
+            "content": [{"type":"input_text","text":"U1"}]
+        },
+        {
+            "type": "message",
+            "role": "assistant",
+            "content": [{"type":"output_text","text":"Hey there!\n"}]
+        },
+        {
+            "type": "message",
+            "role": "user",
+            "content": [{"type":"input_text","text":"U2"}]
+        },
+        {
+            "type": "message",
+            "role": "assistant",
+            "content": [{"type":"output_text","text":"Hey there!\n"}]
+        },
+        {
+            "type": "message",
+            "role": "user",
+            "content": [{"type":"input_text","text":"U3"}]
+        }
+    ]);
 
     let r3_input_array = requests[2]
         .body_json::<serde_json::Value>()
@@ -953,6 +1064,12 @@ async fn history_dedupes_streamed_and_final_messages_across_turns() {
         .and_then(|v| v.as_array())
         .cloned()
         .expect("r3 missing input array");
-
-    assert_eq!(json!(r3_input_array), expected_full);
+    // skipping earlier context and developer messages
+    let tail_len = r3_tail_expected.as_array().unwrap().len();
+    let actual_tail = &r3_input_array[r3_input_array.len() - tail_len..];
+    assert_eq!(
+        serde_json::Value::Array(actual_tail.to_vec()),
+        r3_tail_expected,
+        "request 3 tail mismatch",
+    );
 }

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

@@ -5,6 +5,7 @@ use codex_core::ConversationManager;
 use codex_core::ModelProviderInfo;
 use codex_core::NewConversation;
 use codex_core::built_in_model_providers;
+use codex_core::protocol::ErrorEvent;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::InputItem;
 use codex_core::protocol::Op;
@@ -15,13 +16,20 @@ use core_test_support::load_default_config_for_test;
 use core_test_support::wait_for_event;
 use serde_json::Value;
 use tempfile::TempDir;
+use wiremock::BodyPrintLimit;
 use wiremock::Mock;
 use wiremock::MockServer;
+use wiremock::Request;
+use wiremock::Respond;
 use wiremock::ResponseTemplate;
 use wiremock::matchers::method;
 use wiremock::matchers::path;
 
 use pretty_assertions::assert_eq;
+use std::sync::Arc;
+use std::sync::Mutex;
+use std::sync::atomic::AtomicUsize;
+use std::sync::atomic::Ordering;
 
 // --- Test helpers -----------------------------------------------------------
 
@@ -52,6 +60,22 @@ fn ev_completed(id: &str) -> Value {
     })
 }
 
+fn ev_completed_with_tokens(id: &str, total_tokens: u64) -> Value {
+    serde_json::json!({
+        "type": "response.completed",
+        "response": {
+            "id": id,
+            "usage": {
+                "input_tokens": total_tokens,
+                "input_tokens_details": null,
+                "output_tokens": 0,
+                "output_tokens_details": null,
+                "total_tokens": total_tokens
+            }
+        }
+    })
+}
+
 /// Convenience: SSE event for a single assistant message output item.
 fn ev_assistant_message(id: &str, text: &str) -> Value {
     serde_json::json!({
@@ -65,6 +89,18 @@ fn ev_assistant_message(id: &str, text: &str) -> Value {
     })
 }
 
+fn ev_function_call(call_id: &str, name: &str, arguments: &str) -> Value {
+    serde_json::json!({
+        "type": "response.output_item.done",
+        "item": {
+            "type": "function_call",
+            "call_id": call_id,
+            "name": name,
+            "arguments": arguments
+        }
+    })
+}
+
 fn sse_response(body: String) -> ResponseTemplate {
     ResponseTemplate::new(200)
         .insert_header("content-type", "text/event-stream")
@@ -84,10 +120,28 @@ where
         .await;
 }
 
+async fn start_mock_server() -> MockServer {
+    MockServer::builder()
+        .body_print_limit(BodyPrintLimit::Limited(80_000))
+        .start()
+        .await
+}
+
 const FIRST_REPLY: &str = "FIRST_REPLY";
 const SUMMARY_TEXT: &str = "SUMMARY_ONLY_CONTEXT";
 const SUMMARIZE_TRIGGER: &str = "Start Summarization";
 const THIRD_USER_MSG: &str = "next turn";
+const AUTO_SUMMARY_TEXT: &str = "AUTO_SUMMARY";
+const FIRST_AUTO_MSG: &str = "token limit start";
+const SECOND_AUTO_MSG: &str = "token limit push";
+const STILL_TOO_BIG_REPLY: &str = "STILL_TOO_BIG";
+const MULTI_AUTO_MSG: &str = "multi auto";
+const SECOND_LARGE_REPLY: &str = "SECOND_LARGE_REPLY";
+const FIRST_AUTO_SUMMARY: &str = "FIRST_AUTO_SUMMARY";
+const SECOND_AUTO_SUMMARY: &str = "SECOND_AUTO_SUMMARY";
+const FINAL_REPLY: &str = "FINAL_REPLY";
+const DUMMY_FUNCTION_NAME: &str = "unsupported_tool";
+const DUMMY_CALL_ID: &str = "call-multi-auto";
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn summarize_context_three_requests_and_instructions() {
@@ -99,7 +153,7 @@ async fn summarize_context_three_requests_and_instructions() {
     }
 
     // Set up a mock server that we can inspect after the run.
-    let server = MockServer::start().await;
+    let server = start_mock_server().await;
 
     // SSE 1: assistant replies normally so it is recorded in history.
     let sse1 = sse(vec![
@@ -144,6 +198,7 @@ async fn summarize_context_three_requests_and_instructions() {
     let home = TempDir::new().unwrap();
     let mut config = load_default_config_for_test(&home);
     config.model_provider = model_provider;
+    config.model_auto_compact_token_limit = Some(200_000);
     let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
     let NewConversation {
         conversation: codex,
@@ -198,7 +253,7 @@ async fn summarize_context_three_requests_and_instructions() {
         "summarization should override base instructions"
     );
     assert!(
-        instr2.contains("You are a summarization assistant"),
+        instr2.contains("You have exceeded the maximum number of tokens"),
         "summarization instructions not applied"
     );
 
@@ -209,14 +264,17 @@ async fn summarize_context_three_requests_and_instructions() {
     assert_eq!(last2.get("type").unwrap().as_str().unwrap(), "message");
     assert_eq!(last2.get("role").unwrap().as_str().unwrap(), "user");
     let text2 = last2["content"][0]["text"].as_str().unwrap();
-    assert!(text2.contains(SUMMARIZE_TRIGGER));
+    assert!(
+        text2.contains(SUMMARIZE_TRIGGER),
+        "expected summarize trigger, got `{text2}`"
+    );
 
-    // Third request must contain only the summary from step 2 as prior history plus new user msg.
+    // Third request must contain the refreshed instructions, bridge summary message and new user msg.
     let input3 = body3.get("input").and_then(|v| v.as_array()).unwrap();
     println!("third request body: {body3}");
     assert!(
-        input3.len() >= 2,
-        "expected summary + new user message in third request"
+        input3.len() >= 3,
+        "expected refreshed context and new user message in third request"
     );
 
     // Collect all (role, text) message tuples.
@@ -232,24 +290,35 @@ async fn summarize_context_three_requests_and_instructions() {
         }
     }
 
-    // Exactly one assistant message should remain after compaction and the new user message is present.
+    // No previous assistant messages should remain and the new user message is present.
     let assistant_count = messages.iter().filter(|(r, _)| r == "assistant").count();
-    assert_eq!(
-        assistant_count, 1,
-        "exactly one assistant message should remain after compaction"
-    );
+    assert_eq!(assistant_count, 0, "assistant history should be cleared");
     assert!(
         messages
             .iter()
             .any(|(r, t)| r == "user" && t == THIRD_USER_MSG),
         "third request should include the new user message"
     );
+    let Some((_, bridge_text)) = messages.iter().find(|(role, text)| {
+        role == "user"
+            && (text.contains("Here were the user messages")
+                || text.contains("Here are all the user messages"))
+            && text.contains(SUMMARY_TEXT)
+    }) else {
+        panic!("expected a bridge message containing the summary");
+    };
     assert!(
-        !messages.iter().any(|(_, t)| t.contains("hello world")),
-        "third request should not include the original user input"
+        bridge_text.contains("hello world"),
+        "bridge should capture earlier user messages"
     );
     assert!(
-        !messages.iter().any(|(_, t)| t.contains(SUMMARIZE_TRIGGER)),
+        !bridge_text.contains(SUMMARIZE_TRIGGER),
+        "bridge text should not echo the summarize trigger"
+    );
+    assert!(
+        !messages
+            .iter()
+            .any(|(_, text)| text.contains(SUMMARIZE_TRIGGER)),
         "third request should not include the summarize trigger"
     );
 
@@ -258,6 +327,7 @@ async fn summarize_context_three_requests_and_instructions() {
     wait_for_event(&codex, |ev| matches!(ev, EventMsg::ShutdownComplete)).await;
 
     // Verify rollout contains APITurn entries for each API call and a Compacted entry.
+    println!("rollout path: {}", rollout_path.display());
     let text = std::fs::read_to_string(&rollout_path).unwrap_or_else(|e| {
         panic!(
             "failed to read rollout file {}: {e}",
@@ -296,3 +366,506 @@ async fn summarize_context_three_requests_and_instructions() {
         "expected a Compacted entry containing the summarizer output"
     );
 }
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn auto_compact_runs_after_token_limit_hit() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let server = start_mock_server().await;
+
+    let sse1 = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed_with_tokens("r1", 70_000),
+    ]);
+
+    let sse2 = sse(vec![
+        ev_assistant_message("m2", "SECOND_REPLY"),
+        ev_completed_with_tokens("r2", 330_000),
+    ]);
+
+    let sse3 = sse(vec![
+        ev_assistant_message("m3", AUTO_SUMMARY_TEXT),
+        ev_completed_with_tokens("r3", 200),
+    ]);
+
+    let first_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(FIRST_AUTO_MSG)
+            && !body.contains(SECOND_AUTO_MSG)
+            && !body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(first_matcher)
+        .respond_with(sse_response(sse1))
+        .mount(&server)
+        .await;
+
+    let second_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(SECOND_AUTO_MSG)
+            && body.contains(FIRST_AUTO_MSG)
+            && !body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(second_matcher)
+        .respond_with(sse_response(sse2))
+        .mount(&server)
+        .await;
+
+    let third_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(third_matcher)
+        .respond_with(sse_response(sse3))
+        .mount(&server)
+        .await;
+
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    let home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&home);
+    config.model_provider = model_provider;
+    config.model_auto_compact_token_limit = Some(200_000);
+    let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
+    let codex = conversation_manager
+        .new_conversation(config)
+        .await
+        .unwrap()
+        .conversation;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: FIRST_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: SECOND_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+    // wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    let requests = server.received_requests().await.unwrap();
+    assert_eq!(requests.len(), 3, "auto compact should add a third request");
+
+    let body3 = requests[2].body_json::<serde_json::Value>().unwrap();
+    let instructions = body3
+        .get("instructions")
+        .and_then(|v| v.as_str())
+        .unwrap_or_default();
+    assert!(
+        instructions.contains("You have exceeded the maximum number of tokens"),
+        "auto compact should reuse summarization instructions"
+    );
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn auto_compact_persists_rollout_entries() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let server = start_mock_server().await;
+
+    let sse1 = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed_with_tokens("r1", 70_000),
+    ]);
+
+    let sse2 = sse(vec![
+        ev_assistant_message("m2", "SECOND_REPLY"),
+        ev_completed_with_tokens("r2", 330_000),
+    ]);
+
+    let sse3 = sse(vec![
+        ev_assistant_message("m3", AUTO_SUMMARY_TEXT),
+        ev_completed_with_tokens("r3", 200),
+    ]);
+
+    let first_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(FIRST_AUTO_MSG)
+            && !body.contains(SECOND_AUTO_MSG)
+            && !body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(first_matcher)
+        .respond_with(sse_response(sse1))
+        .mount(&server)
+        .await;
+
+    let second_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(SECOND_AUTO_MSG)
+            && body.contains(FIRST_AUTO_MSG)
+            && !body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(second_matcher)
+        .respond_with(sse_response(sse2))
+        .mount(&server)
+        .await;
+
+    let third_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(third_matcher)
+        .respond_with(sse_response(sse3))
+        .mount(&server)
+        .await;
+
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    let home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&home);
+    config.model_provider = model_provider;
+    let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
+    let NewConversation {
+        conversation: codex,
+        session_configured,
+        ..
+    } = conversation_manager.new_conversation(config).await.unwrap();
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: FIRST_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: SECOND_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    codex.submit(Op::Shutdown).await.unwrap();
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::ShutdownComplete)).await;
+
+    let rollout_path = session_configured.rollout_path;
+    let text = std::fs::read_to_string(&rollout_path).unwrap_or_else(|e| {
+        panic!(
+            "failed to read rollout file {}: {e}",
+            rollout_path.display()
+        )
+    });
+
+    let mut turn_context_count = 0usize;
+    for line in text.lines() {
+        let trimmed = line.trim();
+        if trimmed.is_empty() {
+            continue;
+        }
+        let Ok(entry): Result<RolloutLine, _> = serde_json::from_str(trimmed) else {
+            continue;
+        };
+        match entry.item {
+            RolloutItem::TurnContext(_) => {
+                turn_context_count += 1;
+            }
+            RolloutItem::Compacted(_) => {}
+            _ => {}
+        }
+    }
+
+    assert!(
+        turn_context_count >= 2,
+        "expected at least two turn context entries, got {turn_context_count}"
+    );
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn auto_compact_stops_after_failed_attempt() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let server = start_mock_server().await;
+
+    let sse1 = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed_with_tokens("r1", 500),
+    ]);
+
+    let sse2 = sse(vec![
+        ev_assistant_message("m2", SUMMARY_TEXT),
+        ev_completed_with_tokens("r2", 50),
+    ]);
+
+    let sse3 = sse(vec![
+        ev_assistant_message("m3", STILL_TOO_BIG_REPLY),
+        ev_completed_with_tokens("r3", 500),
+    ]);
+
+    let first_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains(FIRST_AUTO_MSG)
+            && !body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(first_matcher)
+        .respond_with(sse_response(sse1.clone()))
+        .mount(&server)
+        .await;
+
+    let second_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        body.contains("You have exceeded the maximum number of tokens")
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(second_matcher)
+        .respond_with(sse_response(sse2.clone()))
+        .mount(&server)
+        .await;
+
+    let third_matcher = |req: &wiremock::Request| {
+        let body = std::str::from_utf8(&req.body).unwrap_or("");
+        !body.contains("You have exceeded the maximum number of tokens")
+            && body.contains(SUMMARY_TEXT)
+    };
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .and(third_matcher)
+        .respond_with(sse_response(sse3.clone()))
+        .mount(&server)
+        .await;
+
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    let home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&home);
+    config.model_provider = model_provider;
+    config.model_auto_compact_token_limit = Some(200);
+    let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
+    let codex = conversation_manager
+        .new_conversation(config)
+        .await
+        .unwrap()
+        .conversation;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: FIRST_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+
+    let error_event = wait_for_event(&codex, |ev| matches!(ev, EventMsg::Error(_))).await;
+    let EventMsg::Error(ErrorEvent { message }) = error_event else {
+        panic!("expected error event");
+    };
+    assert!(
+        message.contains("limit"),
+        "error message should include limit information: {message}"
+    );
+    wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    let requests = server.received_requests().await.unwrap();
+    assert_eq!(
+        requests.len(),
+        3,
+        "auto compact should attempt at most one summarization before erroring"
+    );
+
+    let last_body = requests[2].body_json::<serde_json::Value>().unwrap();
+    let instructions = last_body
+        .get("instructions")
+        .and_then(|v| v.as_str())
+        .unwrap_or_default();
+    assert!(
+        !instructions.contains("You have exceeded the maximum number of tokens"),
+        "third request should be the follow-up turn, not another summarization"
+    );
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn auto_compact_allows_multiple_attempts_when_interleaved_with_other_turn_events() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let server = start_mock_server().await;
+
+    let sse1 = sse(vec![
+        ev_assistant_message("m1", FIRST_REPLY),
+        ev_completed_with_tokens("r1", 500),
+    ]);
+    let sse2 = sse(vec![
+        ev_assistant_message("m2", FIRST_AUTO_SUMMARY),
+        ev_completed_with_tokens("r2", 50),
+    ]);
+    let sse3 = sse(vec![
+        ev_function_call(DUMMY_CALL_ID, DUMMY_FUNCTION_NAME, "{}"),
+        ev_completed_with_tokens("r3", 150),
+    ]);
+    let sse4 = sse(vec![
+        ev_assistant_message("m4", SECOND_LARGE_REPLY),
+        ev_completed_with_tokens("r4", 450),
+    ]);
+    let sse5 = sse(vec![
+        ev_assistant_message("m5", SECOND_AUTO_SUMMARY),
+        ev_completed_with_tokens("r5", 60),
+    ]);
+    let sse6 = sse(vec![
+        ev_assistant_message("m6", FINAL_REPLY),
+        ev_completed_with_tokens("r6", 120),
+    ]);
+
+    #[derive(Clone)]
+    struct SeqResponder {
+        bodies: Arc<Vec<String>>,
+        calls: Arc<AtomicUsize>,
+        requests: Arc<Mutex<Vec<Vec<u8>>>>,
+    }
+
+    impl SeqResponder {
+        fn new(bodies: Vec<String>) -> Self {
+            Self {
+                bodies: Arc::new(bodies),
+                calls: Arc::new(AtomicUsize::new(0)),
+                requests: Arc::new(Mutex::new(Vec::new())),
+            }
+        }
+
+        fn recorded_requests(&self) -> Vec<Vec<u8>> {
+            self.requests.lock().unwrap().clone()
+        }
+    }
+
+    impl Respond for SeqResponder {
+        fn respond(&self, req: &Request) -> ResponseTemplate {
+            let idx = self.calls.fetch_add(1, Ordering::SeqCst);
+            self.requests.lock().unwrap().push(req.body.clone());
+            let body = self
+                .bodies
+                .get(idx)
+                .unwrap_or_else(|| panic!("unexpected request index {idx}"))
+                .clone();
+            ResponseTemplate::new(200)
+                .insert_header("content-type", "text/event-stream")
+                .set_body_raw(body, "text/event-stream")
+        }
+    }
+
+    let responder = SeqResponder::new(vec![sse1, sse2, sse3, sse4, sse5, sse6]);
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .respond_with(responder.clone())
+        .expect(6)
+        .mount(&server)
+        .await;
+
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    let home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&home);
+    config.model_provider = model_provider;
+    config.model_auto_compact_token_limit = Some(200);
+    let conversation_manager = ConversationManager::with_auth(CodexAuth::from_api_key("dummy"));
+    let codex = conversation_manager
+        .new_conversation(config)
+        .await
+        .unwrap()
+        .conversation;
+
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: MULTI_AUTO_MSG.into(),
+            }],
+        })
+        .await
+        .unwrap();
+
+    loop {
+        let event = codex.next_event().await.unwrap();
+        if let EventMsg::TaskComplete(_) = &event.msg
+            && !event.id.starts_with("auto-compact-")
+        {
+            break;
+        }
+    }
+
+    let request_bodies: Vec<String> = responder
+        .recorded_requests()
+        .into_iter()
+        .map(|body| String::from_utf8(body).unwrap_or_default())
+        .collect();
+    assert_eq!(
+        request_bodies.len(),
+        6,
+        "expected six requests including two auto compactions"
+    );
+    assert!(
+        request_bodies[0].contains(MULTI_AUTO_MSG),
+        "first request should contain the user input"
+    );
+    assert!(
+        request_bodies[1].contains("You have exceeded the maximum number of tokens"),
+        "first auto compact request should use summarization instructions"
+    );
+    assert!(
+        request_bodies[3].contains(&format!("unsupported call: {DUMMY_FUNCTION_NAME}")),
+        "function call output should be sent before the second auto compact"
+    );
+    assert!(
+        request_bodies[4].contains("You have exceeded the maximum number of tokens"),
+        "second auto compact request should reuse summarization instructions"
+    );
+}

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

@@ -1,5 +1,6 @@
 // Aggregates all former standalone integration tests as modules.
 
+mod user_shell_cmd;
 mod cli_stream;
 mod client;
 mod compact;
@@ -9,6 +10,7 @@ mod fork_conversation;
 mod live_cli;
 mod model_overrides;
 mod prompt_caching;
+mod review;
 mod seatbelt;
 mod stream_error_allows_next_turn;
 mod stream_no_completed;

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

@@ -36,7 +36,7 @@ async fn override_turn_context_does_not_persist_when_config_exists() {
             approval_policy: None,
             sandbox_policy: None,
             model: Some("o3".to_string()),
-            effort: Some(ReasoningEffort::High),
+            effort: Some(Some(ReasoningEffort::High)),
             summary: None,
         })
         .await
@@ -76,7 +76,7 @@ async fn override_turn_context_does_not_create_config_file() {
             approval_policy: None,
             sandbox_policy: None,
             model: Some("o3".to_string()),
-            effort: Some(ReasoningEffort::Medium),
+            effort: Some(Some(ReasoningEffort::Medium)),
             summary: None,
         })
         .await

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

@@ -270,13 +270,8 @@ async fn prefixes_context_and_instructions_once_and_consistently_across_requests
     assert_eq!(requests.len(), 2, "expected two POST requests");
 
     let shell = default_user_shell().await;
-    let shell_line = match shell.name() {
-        Some(name) => format!("  <shell>{name}</shell>\n"),
-        None => String::new(),
-    };
 
-    // Per-turn environment context includes the shell tag.
-    let expected_env_text_turn = format!(
+    let expected_env_text = format!(
         r#"<environment_context>
   <cwd>{}</cwd>
   <approval_policy>on-request</approval_policy>
@@ -284,15 +279,18 @@ async fn prefixes_context_and_instructions_once_and_consistently_across_requests
   <network_access>restricted</network_access>
 {}</environment_context>"#,
         cwd.path().to_string_lossy(),
-        shell_line.as_str(),
+        match shell.name() {
+            Some(name) => format!("  <shell>{name}</shell>\n"),
+            None => String::new(),
+        }
     );
     let expected_ui_text =
         "<user_instructions>\n\nbe consistent and helpful\n\n</user_instructions>";
 
-    let expected_env_msg_turn = serde_json::json!({
+    let expected_env_msg = serde_json::json!({
         "type": "message",
         "role": "user",
-        "content": [ { "type": "input_text", "text": expected_env_text_turn } ]
+        "content": [ { "type": "input_text", "text": expected_env_text } ]
     });
     let expected_ui_msg = serde_json::json!({
         "type": "message",
@@ -306,29 +304,11 @@ async fn prefixes_context_and_instructions_once_and_consistently_across_requests
         "content": [ { "type": "input_text", "text": "hello 1" } ]
     });
     let body1 = requests[0].body_json::<serde_json::Value>().unwrap();
-    let body1_input = body1["input"].as_array().unwrap();
     assert_eq!(
         body1["input"],
-        serde_json::json!([
-            expected_ui_msg,
-            expected_env_msg_turn,
-            expected_user_message_1
-        ])
+        serde_json::json!([expected_ui_msg, expected_env_msg, expected_user_message_1])
     );
 
-    let env_texts: Vec<&str> = body1_input
-        .iter()
-        .filter_map(|msg| {
-            msg.get("content")
-                .and_then(|content| content.as_array())
-                .and_then(|content| content.first())
-                .and_then(|item| item.get("text"))
-                .and_then(|text| text.as_str())
-        })
-        .filter(|text| text.starts_with("<environment_context>"))
-        .collect();
-    assert_eq!(env_texts, vec![expected_env_text_turn.as_str()]);
-
     let expected_user_message_2 = serde_json::json!({
         "type": "message",
         "role": "user",
@@ -338,7 +318,7 @@ async fn prefixes_context_and_instructions_once_and_consistently_across_requests
     let expected_body2 = serde_json::json!(
         [
             body1["input"].as_array().unwrap().as_slice(),
-            [expected_env_msg_turn, expected_user_message_2].as_slice(),
+            [expected_user_message_2].as_slice(),
         ]
         .concat()
     );
@@ -407,7 +387,7 @@ async fn overrides_turn_context_but_keeps_cached_prefix_and_key_constant() {
                 exclude_slash_tmp: true,
             }),
             model: Some("o3".to_string()),
-            effort: Some(ReasoningEffort::High),
+            effort: Some(Some(ReasoningEffort::High)),
             summary: Some(ReasoningSummary::Detailed),
         })
         .await
@@ -443,28 +423,19 @@ async fn overrides_turn_context_but_keeps_cached_prefix_and_key_constant() {
         "role": "user",
         "content": [ { "type": "input_text", "text": "hello 2" } ]
     });
-    let shell = default_user_shell().await;
-    let shell_line = match shell.name() {
-        Some(name) => format!("  <shell>{name}</shell>\n"),
-        None => String::new(),
-    };
-
     // After overriding the turn context, the environment context should be emitted again
     // reflecting the new approval policy and sandbox settings. Omit cwd because it did
     // not change.
     let expected_env_text_2 = format!(
         r#"<environment_context>
-  <cwd>{}</cwd>
   <approval_policy>never</approval_policy>
   <sandbox_mode>workspace-write</sandbox_mode>
   <network_access>enabled</network_access>
   <writable_roots>
     <root>{}</root>
   </writable_roots>
-{}</environment_context>"#,
-        cwd.path().to_string_lossy(),
-        writable.path().to_string_lossy(),
-        shell_line.as_str()
+</environment_context>"#,
+        writable.path().to_string_lossy()
     );
     let expected_env_msg_2 = serde_json::json!({
         "type": "message",
@@ -548,7 +519,7 @@ async fn per_turn_overrides_keep_cached_prefix_and_key_constant() {
                 exclude_slash_tmp: true,
             },
             model: "o3".to_string(),
-            effort: ReasoningEffort::High,
+            effort: Some(ReasoningEffort::High),
             summary: ReasoningSummary::Detailed,
         })
         .await
@@ -575,165 +546,12 @@ async fn per_turn_overrides_keep_cached_prefix_and_key_constant() {
         "role": "user",
         "content": [ { "type": "input_text", "text": "hello 2" } ]
     });
-    let shell = default_user_shell().await;
-    let shell_line = match shell.name() {
-        Some(name) => format!("  <shell>{name}</shell>\n"),
-        None => String::new(),
-    };
-    let expected_env_text_2 = format!(
-        r#"<environment_context>
-  <cwd>{}</cwd>
-  <approval_policy>never</approval_policy>
-  <sandbox_mode>workspace-write</sandbox_mode>
-  <network_access>enabled</network_access>
-  <writable_roots>
-    <root>{}</root>
-  </writable_roots>
-{}</environment_context>"#,
-        new_cwd.path().to_string_lossy(),
-        writable.path().to_string_lossy(),
-        shell_line.as_str()
-    );
-    let expected_env_msg_2 = serde_json::json!({
-        "type": "message",
-        "role": "user",
-        "content": [ { "type": "input_text", "text": expected_env_text_2 } ]
-    });
     let expected_body2 = serde_json::json!(
         [
             body1["input"].as_array().unwrap().as_slice(),
-            [expected_env_msg_2, expected_user_message_2].as_slice(),
+            [expected_user_message_2].as_slice(),
         ]
         .concat()
     );
     assert_eq!(body2["input"], expected_body2);
 }
-
-#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
-async fn tools_stable_across_all_approval_policy_transitions() {
-    use pretty_assertions::assert_eq;
-
-    let server = MockServer::start().await;
-
-    let sse = sse_completed("resp");
-    let template = ResponseTemplate::new(200)
-        .insert_header("content-type", "text/event-stream")
-        .set_body_raw(sse, "text/event-stream");
-
-    // Build all transitions FROM each to each other (exclude self transitions)
-    let policies = vec![
-        AskForApproval::UnlessTrusted,
-        AskForApproval::OnFailure,
-        AskForApproval::OnRequest,
-        AskForApproval::Never,
-    ];
-    let mut transitions: Vec<(AskForApproval, AskForApproval)> = Vec::new();
-    for &from in &policies {
-        for &to in &policies {
-            if from != to {
-                transitions.push((from, to));
-            }
-        }
-    }
-
-    // Expect 2 POSTs per transition
-    Mock::given(method("POST"))
-        .and(path("/v1/responses"))
-        .respond_with(template)
-        .expect((transitions.len() * 2) as u64)
-        .mount(&server)
-        .await;
-
-    let model_provider = ModelProviderInfo {
-        base_url: Some(format!("{}/v1", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-
-    let cwd = TempDir::new().unwrap();
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home);
-    config.cwd = cwd.path().to_path_buf();
-    config.model_provider = model_provider;
-    config.user_instructions = Some("be consistent and helpful".to_string());
-    // Keep tools stable and minimal
-    config.include_plan_tool = false;
-    config.include_apply_patch_tool = false;
-    config.tools_web_search_request = false;
-    config.use_experimental_unified_exec_tool = true; // policy-independent tool
-
-    let conversation_manager =
-        ConversationManager::with_auth(CodexAuth::from_api_key("Test API Key"));
-    let codex = conversation_manager
-        .new_conversation(config)
-        .await
-        .expect("create new conversation")
-        .conversation;
-
-    for (i, (from, to)) in transitions.iter().enumerate() {
-        // Ensure a known starting policy for this pair
-        codex
-            .submit(Op::OverrideTurnContext {
-                cwd: None,
-                approval_policy: Some(*from),
-                sandbox_policy: None,
-                model: None,
-                effort: None,
-                summary: None,
-            })
-            .await
-            .unwrap();
-
-        codex
-            .submit(Op::UserInput {
-                items: vec![InputItem::Text {
-                    text: format!("turn {i}-a"),
-                }],
-            })
-            .await
-            .unwrap();
-        wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
-
-        // Override to the target policy and send next turn
-        codex
-            .submit(Op::OverrideTurnContext {
-                cwd: None,
-                approval_policy: Some(*to),
-                sandbox_policy: None,
-                model: None,
-                effort: None,
-                summary: None,
-            })
-            .await
-            .unwrap();
-
-        codex
-            .submit(Op::UserInput {
-                items: vec![InputItem::Text {
-                    text: format!("turn {i}-b"),
-                }],
-            })
-            .await
-            .unwrap();
-        wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
-    }
-
-    // Verify tool arrays are identical across each pair of requests
-    let requests = server.received_requests().await.unwrap();
-    assert_eq!(
-        requests.len(),
-        transitions.len() * 2,
-        "expected 2 requests per transition"
-    );
-
-    for i in 0..transitions.len() {
-        let body_a = requests[2 * i].body_json::<serde_json::Value>().unwrap();
-        let body_b = requests[2 * i + 1]
-            .body_json::<serde_json::Value>()
-            .unwrap();
-        assert_eq!(
-            body_a["tools"], body_b["tools"],
-            "tools changed between requests for transition #{i}: {:?}",
-            transitions[i]
-        );
-    }
-}

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

@@ -0,0 +1,542 @@
+use codex_core::CodexAuth;
+use codex_core::CodexConversation;
+use codex_core::ConversationManager;
+use codex_core::ModelProviderInfo;
+use codex_core::built_in_model_providers;
+use codex_core::config::Config;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::InputItem;
+use codex_core::protocol::Op;
+use codex_core::protocol::ReviewCodeLocation;
+use codex_core::protocol::ReviewFinding;
+use codex_core::protocol::ReviewLineRange;
+use codex_core::protocol::ReviewOutputEvent;
+use codex_core::protocol::ReviewRequest;
+use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
+use core_test_support::load_default_config_for_test;
+use core_test_support::load_sse_fixture_with_id_from_str;
+use core_test_support::wait_for_event;
+use pretty_assertions::assert_eq;
+use std::path::PathBuf;
+use std::sync::Arc;
+use tempfile::TempDir;
+use tokio::io::AsyncWriteExt as _;
+use uuid::Uuid;
+use wiremock::Mock;
+use wiremock::MockServer;
+use wiremock::ResponseTemplate;
+use wiremock::matchers::method;
+use wiremock::matchers::path;
+
+/// Verify that submitting `Op::Review` spawns a child task and emits
+/// EnteredReviewMode -> ExitedReviewMode(None) -> TaskComplete
+/// in that order when the model returns a structured review JSON payload.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn review_op_emits_lifecycle_and_review_output() {
+    // Skip under Codex sandbox network restrictions.
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    // Start mock Responses API server. Return a single assistant message whose
+    // text is a JSON-encoded ReviewOutputEvent.
+    let review_json = serde_json::json!({
+        "findings": [
+            {
+                "title": "Prefer Stylize helpers",
+                "body": "Use .dim()/.bold() chaining instead of manual Style where possible.",
+                "confidence_score": 0.9,
+                "priority": 1,
+                "code_location": {
+                    "absolute_file_path": "/tmp/file.rs",
+                    "line_range": {"start": 10, "end": 20}
+                }
+            }
+        ],
+        "overall_correctness": "good",
+        "overall_explanation": "All good with some improvements suggested.",
+        "overall_confidence_score": 0.8
+    })
+    .to_string();
+    let sse_template = r#"[
+            {"type":"response.output_item.done", "item":{
+                "type":"message", "role":"assistant",
+                "content":[{"type":"output_text","text":__REVIEW__}]
+            }},
+            {"type":"response.completed", "response": {"id": "__ID__"}}
+        ]"#;
+    let review_json_escaped = serde_json::to_string(&review_json).unwrap();
+    let sse_raw = sse_template.replace("__REVIEW__", &review_json_escaped);
+    let server = start_responses_server_with_sse(&sse_raw, 1).await;
+    let codex_home = TempDir::new().unwrap();
+    let codex = new_conversation_for_server(&server, &codex_home, |_| {}).await;
+
+    // Submit review request.
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: "Please review my changes".to_string(),
+                user_facing_hint: "my changes".to_string(),
+            },
+        })
+        .await
+        .unwrap();
+
+    // Verify lifecycle: Entered -> Exited(Some(review)) -> TaskComplete.
+    let _entered = wait_for_event(&codex, |ev| matches!(ev, EventMsg::EnteredReviewMode(_))).await;
+    let closed = wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExitedReviewMode(_))).await;
+    let review = match closed {
+        EventMsg::ExitedReviewMode(Some(r)) => r,
+        other => panic!("expected ExitedReviewMode(Some(..)), got {other:?}"),
+    };
+
+    // Deep compare full structure using PartialEq (floats are f32 on both sides).
+    let expected = ReviewOutputEvent {
+        findings: vec![ReviewFinding {
+            title: "Prefer Stylize helpers".to_string(),
+            body: "Use .dim()/.bold() chaining instead of manual Style where possible.".to_string(),
+            confidence_score: 0.9,
+            priority: 1,
+            code_location: ReviewCodeLocation {
+                absolute_file_path: PathBuf::from("/tmp/file.rs"),
+                line_range: ReviewLineRange { start: 10, end: 20 },
+            },
+        }],
+        overall_correctness: "good".to_string(),
+        overall_explanation: "All good with some improvements suggested.".to_string(),
+        overall_confidence_score: 0.8,
+    };
+    assert_eq!(expected, review);
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    server.verify().await;
+}
+
+/// When the model returns plain text that is not JSON, ensure the child
+/// lifecycle still occurs and the plain text is surfaced via
+/// ExitedReviewMode(Some(..)) as the overall_explanation.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn review_op_with_plain_text_emits_review_fallback() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let sse_raw = r#"[
+        {"type":"response.output_item.done", "item":{
+            "type":"message", "role":"assistant",
+            "content":[{"type":"output_text","text":"just plain text"}]
+        }},
+        {"type":"response.completed", "response": {"id": "__ID__"}}
+    ]"#;
+    let server = start_responses_server_with_sse(sse_raw, 1).await;
+    let codex_home = TempDir::new().unwrap();
+    let codex = new_conversation_for_server(&server, &codex_home, |_| {}).await;
+
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: "Plain text review".to_string(),
+                user_facing_hint: "plain text review".to_string(),
+            },
+        })
+        .await
+        .unwrap();
+
+    let _entered = wait_for_event(&codex, |ev| matches!(ev, EventMsg::EnteredReviewMode(_))).await;
+    let closed = wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExitedReviewMode(_))).await;
+    let review = match closed {
+        EventMsg::ExitedReviewMode(Some(r)) => r,
+        other => panic!("expected ExitedReviewMode(Some(..)), got {other:?}"),
+    };
+
+    // Expect a structured fallback carrying the plain text.
+    let expected = ReviewOutputEvent {
+        overall_explanation: "just plain text".to_string(),
+        ..Default::default()
+    };
+    assert_eq!(expected, review);
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    server.verify().await;
+}
+
+/// When the model returns structured JSON in a review, ensure no AgentMessage
+/// is emitted; the UI consumes the structured result via ExitedReviewMode.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn review_does_not_emit_agent_message_on_structured_output() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    let review_json = serde_json::json!({
+        "findings": [
+            {
+                "title": "Example",
+                "body": "Structured review output.",
+                "confidence_score": 0.5,
+                "priority": 1,
+                "code_location": {
+                    "absolute_file_path": "/tmp/file.rs",
+                    "line_range": {"start": 1, "end": 2}
+                }
+            }
+        ],
+        "overall_correctness": "ok",
+        "overall_explanation": "ok",
+        "overall_confidence_score": 0.5
+    })
+    .to_string();
+    let sse_template = r#"[
+            {"type":"response.output_item.done", "item":{
+                "type":"message", "role":"assistant",
+                "content":[{"type":"output_text","text":__REVIEW__}]
+            }},
+            {"type":"response.completed", "response": {"id": "__ID__"}}
+        ]"#;
+    let review_json_escaped = serde_json::to_string(&review_json).unwrap();
+    let sse_raw = sse_template.replace("__REVIEW__", &review_json_escaped);
+    let server = start_responses_server_with_sse(&sse_raw, 1).await;
+    let codex_home = TempDir::new().unwrap();
+    let codex = new_conversation_for_server(&server, &codex_home, |_| {}).await;
+
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: "check structured".to_string(),
+                user_facing_hint: "check structured".to_string(),
+            },
+        })
+        .await
+        .unwrap();
+
+    // Drain events until TaskComplete; ensure none are AgentMessage.
+    use tokio::time::Duration;
+    use tokio::time::timeout;
+    let mut saw_entered = false;
+    let mut saw_exited = false;
+    loop {
+        let ev = timeout(Duration::from_secs(5), codex.next_event())
+            .await
+            .expect("timeout waiting for event")
+            .expect("stream ended unexpectedly");
+        match ev.msg {
+            EventMsg::TaskComplete(_) => break,
+            EventMsg::AgentMessage(_) => {
+                panic!("unexpected AgentMessage during review with structured output")
+            }
+            EventMsg::EnteredReviewMode(_) => saw_entered = true,
+            EventMsg::ExitedReviewMode(_) => saw_exited = true,
+            _ => {}
+        }
+    }
+    assert!(saw_entered && saw_exited, "missing review lifecycle events");
+
+    server.verify().await;
+}
+
+/// Ensure that when a custom `review_model` is set in the config, the review
+/// request uses that model (and not the main chat model).
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn review_uses_custom_review_model_from_config() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    // Minimal stream: just a completed event
+    let sse_raw = r#"[
+        {"type":"response.completed", "response": {"id": "__ID__"}}
+    ]"#;
+    let server = start_responses_server_with_sse(sse_raw, 1).await;
+    let codex_home = TempDir::new().unwrap();
+    // Choose a review model different from the main model; ensure it is used.
+    let codex = new_conversation_for_server(&server, &codex_home, |cfg| {
+        cfg.model = "gpt-4.1".to_string();
+        cfg.review_model = "gpt-5".to_string();
+    })
+    .await;
+
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: "use custom model".to_string(),
+                user_facing_hint: "use custom model".to_string(),
+            },
+        })
+        .await
+        .unwrap();
+
+    // Wait for completion
+    let _entered = wait_for_event(&codex, |ev| matches!(ev, EventMsg::EnteredReviewMode(_))).await;
+    let _closed = wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExitedReviewMode(None))).await;
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Assert the request body model equals the configured review model
+    let request = &server.received_requests().await.unwrap()[0];
+    let body = request.body_json::<serde_json::Value>().unwrap();
+    assert_eq!(body["model"].as_str().unwrap(), "gpt-5");
+
+    server.verify().await;
+}
+
+/// When a review session begins, it must not prepend prior chat history from
+/// the parent session. The request `input` should contain only the review
+/// prompt from the user.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn review_input_isolated_from_parent_history() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    // Mock server for the single review request
+    let sse_raw = r#"[
+        {"type":"response.completed", "response": {"id": "__ID__"}}
+    ]"#;
+    let server = start_responses_server_with_sse(sse_raw, 1).await;
+
+    // Seed a parent session history via resume file with both user + assistant items.
+    let codex_home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&codex_home);
+    config.model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+
+    let session_file = codex_home.path().join("resume.jsonl");
+    {
+        let mut f = tokio::fs::File::create(&session_file).await.unwrap();
+        let convo_id = Uuid::new_v4();
+        // Proper session_meta line (enveloped) with a conversation id
+        let meta_line = serde_json::json!({
+            "timestamp": "2024-01-01T00:00:00.000Z",
+            "type": "session_meta",
+            "payload": {
+                "id": convo_id,
+                "timestamp": "2024-01-01T00:00:00Z",
+                "instructions": null,
+                "cwd": ".",
+                "originator": "test_originator",
+                "cli_version": "test_version"
+            }
+        });
+        f.write_all(format!("{meta_line}\n").as_bytes())
+            .await
+            .unwrap();
+
+        // Prior user message (enveloped response_item)
+        let user = codex_protocol::models::ResponseItem::Message {
+            id: None,
+            role: "user".to_string(),
+            content: vec![codex_protocol::models::ContentItem::InputText {
+                text: "parent: earlier user message".to_string(),
+            }],
+        };
+        let user_json = serde_json::to_value(&user).unwrap();
+        let user_line = serde_json::json!({
+            "timestamp": "2024-01-01T00:00:01.000Z",
+            "type": "response_item",
+            "payload": user_json
+        });
+        f.write_all(format!("{user_line}\n").as_bytes())
+            .await
+            .unwrap();
+
+        // Prior assistant message (enveloped response_item)
+        let assistant = codex_protocol::models::ResponseItem::Message {
+            id: None,
+            role: "assistant".to_string(),
+            content: vec![codex_protocol::models::ContentItem::OutputText {
+                text: "parent: assistant reply".to_string(),
+            }],
+        };
+        let assistant_json = serde_json::to_value(&assistant).unwrap();
+        let assistant_line = serde_json::json!({
+            "timestamp": "2024-01-01T00:00:02.000Z",
+            "type": "response_item",
+            "payload": assistant_json
+        });
+        f.write_all(format!("{assistant_line}\n").as_bytes())
+            .await
+            .unwrap();
+    }
+    config.experimental_resume = Some(session_file);
+
+    let codex = new_conversation_for_server(&server, &codex_home, |cfg| {
+        // apply resume file
+        cfg.experimental_resume = config.experimental_resume.clone();
+    })
+    .await;
+
+    // Submit review request; it must start fresh (no parent history in `input`).
+    let review_prompt = "Please review only this".to_string();
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: review_prompt.clone(),
+                user_facing_hint: review_prompt.clone(),
+            },
+        })
+        .await
+        .unwrap();
+
+    let _entered = wait_for_event(&codex, |ev| matches!(ev, EventMsg::EnteredReviewMode(_))).await;
+    let _closed = wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExitedReviewMode(None))).await;
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Assert the request `input` contains only the single review user message.
+    let request = &server.received_requests().await.unwrap()[0];
+    let body = request.body_json::<serde_json::Value>().unwrap();
+    let expected_input = serde_json::json!([
+        {
+            "type": "message",
+            "role": "user",
+            "content": [{"type": "input_text", "text": review_prompt}]
+        }
+    ]);
+    assert_eq!(body["input"], expected_input);
+
+    server.verify().await;
+}
+
+/// After a review thread finishes, its conversation should not leak into the
+/// parent session. A subsequent parent turn must not include any review
+/// messages in its request `input`.
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn review_history_does_not_leak_into_parent_session() {
+    if std::env::var(CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR).is_ok() {
+        println!(
+            "Skipping test because it cannot execute when network is disabled in a Codex sandbox."
+        );
+        return;
+    }
+
+    // Respond to both the review request and the subsequent parent request.
+    let sse_raw = r#"[
+        {"type":"response.output_item.done", "item":{
+            "type":"message", "role":"assistant",
+            "content":[{"type":"output_text","text":"review assistant output"}]
+        }},
+        {"type":"response.completed", "response": {"id": "__ID__"}}
+    ]"#;
+    let server = start_responses_server_with_sse(sse_raw, 2).await;
+    let codex_home = TempDir::new().unwrap();
+    let codex = new_conversation_for_server(&server, &codex_home, |_| {}).await;
+
+    // 1) Run a review turn that produces an assistant message (isolated in child).
+    codex
+        .submit(Op::Review {
+            review_request: ReviewRequest {
+                prompt: "Start a review".to_string(),
+                user_facing_hint: "Start a review".to_string(),
+            },
+        })
+        .await
+        .unwrap();
+    let _entered = wait_for_event(&codex, |ev| matches!(ev, EventMsg::EnteredReviewMode(_))).await;
+    let _closed = wait_for_event(&codex, |ev| {
+        matches!(ev, EventMsg::ExitedReviewMode(Some(_)))
+    })
+    .await;
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // 2) Continue in the parent session; request input must not include any review items.
+    let followup = "back to parent".to_string();
+    codex
+        .submit(Op::UserInput {
+            items: vec![InputItem::Text {
+                text: followup.clone(),
+            }],
+        })
+        .await
+        .unwrap();
+    let _complete = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TaskComplete(_))).await;
+
+    // Inspect the second request (parent turn) input contents.
+    // Parent turns include session initial messages (user_instructions, environment_context).
+    // Critically, no messages from the review thread should appear.
+    let requests = server.received_requests().await.unwrap();
+    assert_eq!(requests.len(), 2);
+    let body = requests[1].body_json::<serde_json::Value>().unwrap();
+    let input = body["input"].as_array().expect("input array");
+
+    // Must include the followup as the last item for this turn
+    let last = input.last().expect("at least one item in input");
+    assert_eq!(last["role"].as_str().unwrap(), "user");
+    let last_text = last["content"][0]["text"].as_str().unwrap();
+    assert_eq!(last_text, followup);
+
+    // Ensure no review-thread content leaked into the parent request
+    let contains_review_prompt = input
+        .iter()
+        .any(|msg| msg["content"][0]["text"].as_str().unwrap_or_default() == "Start a review");
+    let contains_review_assistant = input.iter().any(|msg| {
+        msg["content"][0]["text"].as_str().unwrap_or_default() == "review assistant output"
+    });
+    assert!(
+        !contains_review_prompt,
+        "review prompt leaked into parent turn input"
+    );
+    assert!(
+        !contains_review_assistant,
+        "review assistant output leaked into parent turn input"
+    );
+
+    server.verify().await;
+}
+
+/// Start a mock Responses API server and mount the given SSE stream body.
+async fn start_responses_server_with_sse(sse_raw: &str, expected_requests: usize) -> MockServer {
+    let server = MockServer::start().await;
+    let sse = load_sse_fixture_with_id_from_str(sse_raw, &Uuid::new_v4().to_string());
+    Mock::given(method("POST"))
+        .and(path("/v1/responses"))
+        .respond_with(
+            ResponseTemplate::new(200)
+                .insert_header("content-type", "text/event-stream")
+                .set_body_raw(sse.clone(), "text/event-stream"),
+        )
+        .expect(expected_requests as u64)
+        .mount(&server)
+        .await;
+    server
+}
+
+/// Create a conversation configured to talk to the provided mock server.
+#[expect(clippy::expect_used)]
+async fn new_conversation_for_server<F>(
+    server: &MockServer,
+    codex_home: &TempDir,
+    mutator: F,
+) -> Arc<CodexConversation>
+where
+    F: FnOnce(&mut Config),
+{
+    let model_provider = ModelProviderInfo {
+        base_url: Some(format!("{}/v1", server.uri())),
+        ..built_in_model_providers()["openai"].clone()
+    };
+    let mut config = load_default_config_for_test(codex_home);
+    config.model_provider = model_provider;
+    mutator(&mut config);
+    let conversation_manager =
+        ConversationManager::with_auth(CodexAuth::from_api_key("Test API Key"));
+    conversation_manager
+        .new_conversation(config)
+        .await
+        .expect("create conversation")
+        .conversation
+}

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

@@ -0,0 +1,112 @@
+#![cfg(unix)]
+
+use codex_core::ConversationManager;
+use codex_core::NewConversation;
+use codex_core::protocol::EventMsg;
+use codex_core::protocol::ExecCommandEndEvent;
+use codex_core::protocol::Op;
+use codex_core::protocol::TurnAbortReason;
+use core_test_support::load_default_config_for_test;
+use core_test_support::wait_for_event;
+use std::fs;
+use std::path::PathBuf;
+use tempfile::TempDir;
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn user_shell_cmd_ls_and_cat_in_temp_dir() {
+    // No env overrides needed; test build hard-codes a hermetic zsh with empty rc.
+    // Create a temporary working directory with a known file.
+    let cwd = TempDir::new().unwrap();
+    let file_name = "hello.txt";
+    let file_path: PathBuf = cwd.path().join(file_name);
+    let contents = "hello from bang test\n";
+    fs::write(&file_path, contents).expect("write temp file");
+
+    // Load config and pin cwd to the temp dir so ls/cat operate there.
+    let codex_home = TempDir::new().unwrap();
+    let mut config = load_default_config_for_test(&codex_home);
+    config.cwd = cwd.path().to_path_buf();
+
+    let conversation_manager =
+        ConversationManager::with_auth(codex_core::CodexAuth::from_api_key("dummy"));
+    let NewConversation {
+        conversation: codex,
+        ..
+    } = conversation_manager
+        .new_conversation(config)
+        .await
+        .expect("create new conversation");
+
+    // 1) ls should list the file
+    codex
+        .submit(Op::RunUserShellCommand {
+            command: "ls".to_string(),
+        })
+        .await
+        .unwrap();
+    let msg = wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExecCommandEnd(_))).await;
+    let EventMsg::ExecCommandEnd(ExecCommandEndEvent {
+        stdout, exit_code, ..
+    }) = msg
+    else {
+        unreachable!()
+    };
+    assert_eq!(exit_code, 0);
+    assert!(
+        stdout.contains(file_name),
+        "ls output should include {file_name}, got: {stdout:?}"
+    );
+
+    // 2) cat the file should return exact contents
+    codex
+        .submit(Op::RunUserShellCommand {
+            command: format!("cat {}", file_name),
+        })
+        .await
+        .unwrap();
+    let msg = wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExecCommandEnd(_))).await;
+    let EventMsg::ExecCommandEnd(ExecCommandEndEvent {
+        stdout, exit_code, ..
+    }) = msg
+    else {
+        unreachable!()
+    };
+    assert_eq!(exit_code, 0);
+    assert_eq!(stdout, contents);
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
+async fn user_shell_cmd_can_be_interrupted() {
+    // No env overrides needed; test build hard-codes a hermetic zsh with empty rc.
+    // Set up isolated config and conversation.
+    let codex_home = TempDir::new().unwrap();
+    let config = load_default_config_for_test(&codex_home);
+    let conversation_manager =
+        ConversationManager::with_auth(codex_core::CodexAuth::from_api_key("dummy"));
+    let NewConversation {
+        conversation: codex,
+        ..
+    } = conversation_manager
+        .new_conversation(config)
+        .await
+        .expect("create new conversation");
+
+    // Start a long-running command and then interrupt it.
+    codex
+        .submit(Op::RunUserShellCommand {
+            command: "sleep 5".to_string(),
+        })
+        .await
+        .unwrap();
+
+    // Wait until it has started (ExecCommandBegin), then interrupt.
+    let _ = wait_for_event(&codex, |ev| matches!(ev, EventMsg::ExecCommandBegin(_))).await;
+    codex.submit(Op::Interrupt).await.unwrap();
+
+    // Expect a TurnAborted(Interrupted) notification.
+    let msg = wait_for_event(&codex, |ev| matches!(ev, EventMsg::TurnAborted(_))).await;
+    let EventMsg::TurnAborted(ev) = msg else {
+        unreachable!()
+    };
+    assert_eq!(ev.reason, TurnAbortReason::Interrupted);
+}

codex-rs/exec/src/event_processor_with_human_output.rs

@@ -278,6 +278,7 @@ impl EventProcessor for EventProcessorWithHumanOutput {
                 command,
                 cwd,
                 parsed_cmd: _,
+                ..
             }) => {
                 self.call_id_to_command.insert(
                     call_id,
@@ -562,6 +563,8 @@ impl EventProcessor for EventProcessorWithHumanOutput {
             EventMsg::ShutdownComplete => return CodexStatus::Shutdown,
             EventMsg::ConversationPath(_) => {}
             EventMsg::UserMessage(_) => {}
+            EventMsg::EnteredReviewMode(_) => {}
+            EventMsg::ExitedReviewMode(_) => {}
         }
         CodexStatus::Running
     }

codex-rs/exec/src/lib.rs

@@ -137,6 +137,7 @@ pub async fn run_main(cli: Cli, codex_linux_sandbox_exe: Option<PathBuf>) -> any
     // Load configuration and determine approval policy
     let overrides = ConfigOverrides {
         model,
+        review_model: None,
         config_profile,
         // This CLI is intended to be headless and has no affordances for asking
         // the user for approval.

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

@@ -20,7 +20,7 @@ use codex_core::config::ConfigToml;
 use codex_core::config::load_config_as_toml;
 use codex_core::config_edit::CONFIG_KEY_EFFORT;
 use codex_core::config_edit::CONFIG_KEY_MODEL;
-use codex_core::config_edit::persist_non_null_overrides;
+use codex_core::config_edit::persist_overrides_and_clear_if_none;
 use codex_core::default_client::get_codex_user_agent;
 use codex_core::exec::ExecParams;
 use codex_core::exec_env::create_env;
@@ -519,7 +519,7 @@ impl CodexMessageProcessor {
             (&[CONFIG_KEY_EFFORT], effort_str.as_deref()),
         ];
 
-        match persist_non_null_overrides(
+        match persist_overrides_and_clear_if_none(
             &self.config.codex_home,
             self.config.active_profile.as_deref(),
             &overrides,
@@ -1248,6 +1248,7 @@ fn derive_config_from_params(
     } = params;
     let overrides = ConfigOverrides {
         model,
+        review_model: None,
         config_profile: profile,
         cwd: cwd.map(PathBuf::from),
         approval_policy,

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

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

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

@@ -279,7 +279,9 @@ async fn run_codex_tool_session_inner(
                     | EventMsg::TurnAborted(_)
                     | EventMsg::ConversationPath(_)
                     | EventMsg::UserMessage(_)
-                    | EventMsg::ShutdownComplete => {
+                    | EventMsg::ShutdownComplete
+                    | EventMsg::EnteredReviewMode(_)
+                    | EventMsg::ExitedReviewMode(_) => {
                         // For now, we do not do anything extra for these
                         // events. Note that
                         // send(codex_event_to_notification(&event)) above has

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

@@ -280,7 +280,7 @@ mod tests {
             msg: EventMsg::SessionConfigured(SessionConfiguredEvent {
                 session_id: conversation_id,
                 model: "gpt-4o".to_string(),
-                reasoning_effort: ReasoningEffort::default(),
+                reasoning_effort: Some(ReasoningEffort::default()),
                 history_log_id: 1,
                 history_entry_count: 1000,
                 initial_messages: None,
@@ -314,7 +314,7 @@ mod tests {
         let session_configured_event = SessionConfiguredEvent {
             session_id: conversation_id,
             model: "gpt-4o".to_string(),
-            reasoning_effort: ReasoningEffort::default(),
+            reasoning_effort: Some(ReasoningEffort::default()),
             history_log_id: 1,
             history_entry_count: 1000,
             initial_messages: None,

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

@@ -320,7 +320,7 @@ async fn test_send_user_turn_changes_approval_policy_behavior() {
             approval_policy: AskForApproval::Never,
             sandbox_policy: SandboxPolicy::new_read_only_policy(),
             model: "mock-model".to_string(),
-            effort: ReasoningEffort::Medium,
+            effort: Some(ReasoningEffort::Medium),
             summary: ReasoningSummary::Auto,
         })
         .await

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

@@ -1,5 +1,6 @@
+use std::path::Path;
+
 use codex_core::config::ConfigToml;
-use codex_protocol::config_types::ReasoningEffort;
 use codex_protocol::mcp_protocol::SetDefaultModelParams;
 use codex_protocol::mcp_protocol::SetDefaultModelResponse;
 use mcp_test_support::McpProcess;
@@ -14,7 +15,8 @@ const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn set_default_model_persists_overrides() {
-    let codex_home = TempDir::new().unwrap_or_else(|e| panic!("create tempdir: {e}"));
+    let codex_home = TempDir::new().expect("create tempdir");
+    create_config_toml(codex_home.path()).expect("write config.toml");
 
     let mut mcp = McpProcess::new(codex_home.path())
         .await
@@ -25,8 +27,8 @@ async fn set_default_model_persists_overrides() {
         .expect("init failed");
 
     let params = SetDefaultModelParams {
-        model: Some("o4-mini".to_string()),
-        reasoning_effort: Some(ReasoningEffort::High),
+        model: Some("gpt-4.1".to_string()),
+        reasoning_effort: None,
     };
 
     let request_id = mcp
@@ -53,10 +55,22 @@ async fn set_default_model_persists_overrides() {
 
     assert_eq!(
         ConfigToml {
-            model: Some("o4-mini".to_string()),
-            model_reasoning_effort: Some(ReasoningEffort::High),
+            model: Some("gpt-4.1".to_string()),
+            model_reasoning_effort: None,
             ..Default::default()
         },
         config_toml,
     );
 }
+
+// Helper to create a config.toml; mirrors create_conversation.rs
+fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        r#"
+model = "gpt-5"
+model_reasoning_effort = "medium"
+"#,
+    )
+}

codex-rs/protocol/src/mcp_protocol.rs

@@ -223,7 +223,8 @@ pub struct NewConversationResponse {
     pub conversation_id: ConversationId,
     pub model: String,
     /// Note this could be ignored by the model.
-    pub reasoning_effort: ReasoningEffort,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub reasoning_effort: Option<ReasoningEffort>,
     pub rollout_path: PathBuf,
 }
 
@@ -424,8 +425,11 @@ pub struct GetUserSavedConfigResponse {
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, TS)]
 #[serde(rename_all = "camelCase")]
 pub struct SetDefaultModelParams {
+    /// If set to None, this means `model` should be cleared in config.toml.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub model: Option<String>,
+    /// If set to None, this means `model_reasoning_effort` should be cleared
+    /// in config.toml.
     #[serde(skip_serializing_if = "Option::is_none")]
     pub reasoning_effort: Option<ReasoningEffort>,
 }
@@ -523,7 +527,8 @@ pub struct SendUserTurnParams {
     pub approval_policy: AskForApproval,
     pub sandbox_policy: SandboxPolicy,
     pub model: String,
-    pub effort: ReasoningEffort,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub effort: Option<ReasoningEffort>,
     pub summary: ReasoningSummary,
 }
 

codex-rs/protocol/src/protocol.rs

@@ -81,7 +81,8 @@ pub enum Op {
         model: String,
 
         /// Will only be honored if the model is configured to use reasoning.
-        effort: ReasoningEffortConfig,
+        #[serde(skip_serializing_if = "Option::is_none")]
+        effort: Option<ReasoningEffortConfig>,
 
         /// Will only be honored if the model is configured to use reasoning.
         summary: ReasoningSummaryConfig,
@@ -111,8 +112,11 @@ pub enum Op {
         model: Option<String>,
 
         /// Updated reasoning effort (honored only for reasoning-capable models).
+        ///
+        /// Use `Some(Some(_))` to set a specific effort, `Some(None)` to clear
+        /// the effort, or `None` to leave the existing value unchanged.
         #[serde(skip_serializing_if = "Option::is_none")]
-        effort: Option<ReasoningEffortConfig>,
+        effort: Option<Option<ReasoningEffortConfig>>,
 
         /// Updated reasoning summary preference (honored only for reasoning-capable models).
         #[serde(skip_serializing_if = "Option::is_none")]
@@ -162,8 +166,22 @@ pub enum Op {
     /// The agent will use its existing context (either conversation history or previous response id)
     /// to generate a summary which will be returned as an AgentMessage event.
     Compact,
+
+    /// Request a code review from the agent.
+    Review { review_request: ReviewRequest },
+
     /// Request to shut down codex instance.
     Shutdown,
+
+    /// Execute a user-initiated one-off shell command (triggered by "!cmd").
+    ///
+    /// The command string is executed using the user's default shell and may
+    /// include shell syntax (pipes, redirects, etc.). Output is streamed via
+    /// `ExecCommand*` events and the UI regains control upon `TaskComplete`.
+    RunUserShellCommand {
+        /// The raw command string after '!'
+        command: String,
+    },
 }
 
 /// Determines the conditions under which the user is consulted to approve
@@ -500,6 +518,12 @@ pub enum EventMsg {
     ShutdownComplete,
 
     ConversationPath(ConversationPathResponseEvent),
+
+    /// Entered review mode.
+    EnteredReviewMode(ReviewRequest),
+
+    /// Exited review mode with an optional final result to apply.
+    ExitedReviewMode(Option<ReviewOutputEvent>),
 }
 
 // Individual event payload types matching each `EventMsg` variant.
@@ -708,12 +732,12 @@ where
         let (_role, message) = value;
         let message = message.as_ref();
         let trimmed = message.trim();
-        if trimmed.starts_with(ENVIRONMENT_CONTEXT_OPEN_TAG)
-            && trimmed.ends_with(ENVIRONMENT_CONTEXT_CLOSE_TAG)
+        if starts_with_ignore_ascii_case(trimmed, ENVIRONMENT_CONTEXT_OPEN_TAG)
+            && ends_with_ignore_ascii_case(trimmed, ENVIRONMENT_CONTEXT_CLOSE_TAG)
         {
             InputMessageKind::EnvironmentContext
-        } else if trimmed.starts_with(USER_INSTRUCTIONS_OPEN_TAG)
-            && trimmed.ends_with(USER_INSTRUCTIONS_CLOSE_TAG)
+        } else if starts_with_ignore_ascii_case(trimmed, USER_INSTRUCTIONS_OPEN_TAG)
+            && ends_with_ignore_ascii_case(trimmed, USER_INSTRUCTIONS_CLOSE_TAG)
         {
             InputMessageKind::UserInstructions
         } else {
@@ -722,6 +746,26 @@ where
     }
 }
 
+fn starts_with_ignore_ascii_case(text: &str, prefix: &str) -> bool {
+    let text_bytes = text.as_bytes();
+    let prefix_bytes = prefix.as_bytes();
+    text_bytes.len() >= prefix_bytes.len()
+        && text_bytes
+            .iter()
+            .zip(prefix_bytes.iter())
+            .all(|(a, b)| a.eq_ignore_ascii_case(b))
+}
+
+fn ends_with_ignore_ascii_case(text: &str, suffix: &str) -> bool {
+    let text_bytes = text.as_bytes();
+    let suffix_bytes = suffix.as_bytes();
+    text_bytes.len() >= suffix_bytes.len()
+        && text_bytes[text_bytes.len() - suffix_bytes.len()..]
+            .iter()
+            .zip(suffix_bytes.iter())
+            .all(|(a, b)| a.eq_ignore_ascii_case(b))
+}
+
 #[derive(Debug, Clone, Deserialize, Serialize, TS)]
 pub struct AgentMessageDeltaEvent {
     pub delta: String,
@@ -913,7 +957,8 @@ pub struct TurnContextItem {
     pub approval_policy: AskForApproval,
     pub sandbox_policy: SandboxPolicy,
     pub model: String,
-    pub effort: ReasoningEffortConfig,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub effort: Option<ReasoningEffortConfig>,
     pub summary: ReasoningSummaryConfig,
 }
 
@@ -937,6 +982,57 @@ pub struct GitInfo {
     pub repository_url: Option<String>,
 }
 
+/// Review request sent to the review session.
+#[derive(Debug, Clone, Deserialize, Serialize, PartialEq, TS)]
+pub struct ReviewRequest {
+    pub prompt: String,
+    pub user_facing_hint: String,
+}
+
+/// Structured review result produced by a child review session.
+#[derive(Debug, Clone, Deserialize, Serialize, PartialEq, TS)]
+pub struct ReviewOutputEvent {
+    pub findings: Vec<ReviewFinding>,
+    pub overall_correctness: String,
+    pub overall_explanation: String,
+    pub overall_confidence_score: f32,
+}
+
+impl Default for ReviewOutputEvent {
+    fn default() -> Self {
+        Self {
+            findings: Vec::new(),
+            overall_correctness: String::default(),
+            overall_explanation: String::default(),
+            overall_confidence_score: 0.0,
+        }
+    }
+}
+
+/// A single review finding describing an observed issue or recommendation.
+#[derive(Debug, Clone, Deserialize, Serialize, PartialEq, TS)]
+pub struct ReviewFinding {
+    pub title: String,
+    pub body: String,
+    pub confidence_score: f32,
+    pub priority: i32,
+    pub code_location: ReviewCodeLocation,
+}
+
+/// Location of the code related to a review finding.
+#[derive(Debug, Clone, Deserialize, Serialize, PartialEq, TS)]
+pub struct ReviewCodeLocation {
+    pub absolute_file_path: PathBuf,
+    pub line_range: ReviewLineRange,
+}
+
+/// Inclusive line range in a file associated with the finding.
+#[derive(Debug, Clone, Deserialize, Serialize, PartialEq, TS)]
+pub struct ReviewLineRange {
+    pub start: u32,
+    pub end: u32,
+}
+
 #[derive(Debug, Clone, Deserialize, Serialize, TS)]
 pub struct ExecCommandBeginEvent {
     /// Identifier so this can be paired with the ExecCommandEnd event.
@@ -946,6 +1042,10 @@ pub struct ExecCommandBeginEvent {
     /// The command's working directory if not the default cwd for the agent.
     pub cwd: PathBuf,
     pub parsed_cmd: Vec<ParsedCommand>,
+    /// True when this exec was initiated directly by the user (e.g. bang command),
+    /// not by the agent/model. Defaults to false for backwards compatibility.
+    #[serde(default)]
+    pub user_initiated_shell_command: bool,
 }
 
 #[derive(Debug, Clone, Deserialize, Serialize, TS)]
@@ -1082,7 +1182,8 @@ pub struct SessionConfiguredEvent {
     pub model: String,
 
     /// The effort the model is putting into reasoning about the user's request.
-    pub reasoning_effort: ReasoningEffortConfig,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub reasoning_effort: Option<ReasoningEffortConfig>,
 
     /// Identifier of the history log file (inode on Unix, 0 otherwise).
     pub history_log_id: u64,
@@ -1172,7 +1273,7 @@ mod tests {
             msg: EventMsg::SessionConfigured(SessionConfiguredEvent {
                 session_id: conversation_id,
                 model: "codex-mini-latest".to_string(),
-                reasoning_effort: ReasoningEffortConfig::default(),
+                reasoning_effort: Some(ReasoningEffortConfig::default()),
                 history_log_id: 0,
                 history_entry_count: 0,
                 initial_messages: None,
@@ -1211,4 +1312,19 @@ mod tests {
         let deserialized: ExecCommandOutputDeltaEvent = serde_json::from_str(&serialized).unwrap();
         assert_eq!(deserialized, event);
     }
+
+    #[test]
+    fn serialize_run_user_shell_command_op() {
+        let op = Op::RunUserShellCommand {
+            command: "echo hi".to_string(),
+        };
+        let value = serde_json::to_value(op).unwrap();
+        assert_eq!(
+            value,
+            json!({
+                "type": "run_user_shell_command",
+                "command": "echo hi",
+            })
+        );
+    }
 }

codex-rs/tui/src/app.rs

@@ -327,17 +327,19 @@ impl App {
     fn show_model_save_hint(&mut self) {
         let model = self.config.model.clone();
         if self.active_profile.is_some() {
-            self.chat_widget.add_info_message(format!(
-                "Model switched to {model}. Press Ctrl+S to save it for this profile, then press Ctrl+S again to set it as your global default."
-            ));
+            self.chat_widget.add_info_message(
+                format!("Model changed to {model} for the current session"),
+                Some("(ctrl+s to set as profile default)".to_string()),
+            );
         } else {
-            self.chat_widget.add_info_message(format!(
-                "Model switched to {model}. Press Ctrl+S to save it as your global default."
-            ));
+            self.chat_widget.add_info_message(
+                format!("Model changed to {model} for the current session"),
+                Some("(ctrl+s to set as default)".to_string()),
+            );
         }
     }
 
-    fn on_update_reasoning_effort(&mut self, effort: ReasoningEffortConfig) {
+    fn on_update_reasoning_effort(&mut self, effort: Option<ReasoningEffortConfig>) {
         let changed = self.config.model_reasoning_effort != effort;
         self.chat_widget.set_reasoning_effort(effort);
         self.config.model_reasoning_effort = effort;
@@ -376,14 +378,13 @@ impl App {
 
         match scope {
             SaveScope::Profile(profile) => {
-                match persist_model_selection(&codex_home, Some(profile), &model, Some(effort))
-                    .await
-                {
+                match persist_model_selection(&codex_home, Some(profile), &model, effort).await {
                     Ok(()) => {
                         self.model_saved_to_profile = true;
-                        self.chat_widget.add_info_message(format!(
-                            "Saved model {model} ({effort}) for profile `{profile}`. Press Ctrl+S again to make this your global default."
-                        ));
+                        self.chat_widget.add_info_message(
+                            format!("Profile model changed to {model} for all sessions"),
+                            Some("(view global config in config.toml)".to_string()),
+                        );
                     }
                     Err(err) => {
                         tracing::error!(
@@ -397,12 +398,13 @@ impl App {
                 }
             }
             SaveScope::Global => {
-                match persist_model_selection(&codex_home, None, &model, Some(effort)).await {
+                match persist_model_selection(&codex_home, None, &model, effort).await {
                     Ok(()) => {
                         self.model_saved_to_global = true;
-                        self.chat_widget.add_info_message(format!(
-                            "Saved model {model} ({effort}) as your global default."
-                        ));
+                        self.chat_widget.add_info_message(
+                            format!("Default model changed to {model} for all sessions"),
+                            Some("(view global config in config.toml)".to_string()),
+                        )
                     }
                     Err(err) => {
                         tracing::error!(
@@ -419,6 +421,7 @@ impl App {
                 self.chat_widget.add_info_message(
                     "Model preference already saved globally; no further action needed."
                         .to_string(),
+                    None,
                 );
             }
         }
@@ -537,19 +540,19 @@ mod tests {
         let mut app = make_test_app();
         app.model_saved_to_profile = true;
         app.model_saved_to_global = true;
-        app.config.model_reasoning_effort = ReasoningEffortConfig::Medium;
+        app.config.model_reasoning_effort = Some(ReasoningEffortConfig::Medium);
         app.chat_widget
-            .set_reasoning_effort(ReasoningEffortConfig::Medium);
+            .set_reasoning_effort(Some(ReasoningEffortConfig::Medium));
 
-        app.on_update_reasoning_effort(ReasoningEffortConfig::High);
+        app.on_update_reasoning_effort(Some(ReasoningEffortConfig::High));
 
         assert_eq!(
             app.config.model_reasoning_effort,
-            ReasoningEffortConfig::High
+            Some(ReasoningEffortConfig::High)
         );
         assert_eq!(
             app.chat_widget.config_ref().model_reasoning_effort,
-            ReasoningEffortConfig::High
+            Some(ReasoningEffortConfig::High)
         );
         assert!(!app.model_saved_to_profile);
         assert!(!app.model_saved_to_global);

codex-rs/tui/src/app_event.rs

@@ -46,7 +46,7 @@ pub(crate) enum AppEvent {
     CommitTick,
 
     /// Update the current reasoning effort in the running app and widget.
-    UpdateReasoningEffort(ReasoningEffort),
+    UpdateReasoningEffort(Option<ReasoningEffort>),
 
     /// Update the current model slug in the running app and widget.
     UpdateModel(String),

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

@@ -95,6 +95,10 @@ enum ActivePopup {
     File(FileSearchPopup),
 }
 
+const FOOTER_HINT_HEIGHT: u16 = 1;
+const FOOTER_SPACING_HEIGHT: u16 = 1;
+const FOOTER_HEIGHT_WITH_HINT: u16 = FOOTER_HINT_HEIGHT + FOOTER_SPACING_HEIGHT;
+
 impl ChatComposer {
     pub fn new(
         has_input_focus: bool,
@@ -134,20 +138,20 @@ impl ChatComposer {
     pub fn desired_height(&self, width: u16) -> u16 {
         self.textarea.desired_height(width - 1)
             + match &self.active_popup {
-                ActivePopup::None => 1u16,
+                ActivePopup::None => FOOTER_HEIGHT_WITH_HINT,
                 ActivePopup::Command(c) => c.calculate_required_height(),
                 ActivePopup::File(c) => c.calculate_required_height(),
             }
     }
 
     pub fn cursor_pos(&self, area: Rect) -> Option<(u16, u16)> {
-        let popup_height = match &self.active_popup {
-            ActivePopup::Command(popup) => popup.calculate_required_height(),
-            ActivePopup::File(popup) => popup.calculate_required_height(),
-            ActivePopup::None => 1,
+        let popup_constraint = match &self.active_popup {
+            ActivePopup::Command(popup) => Constraint::Max(popup.calculate_required_height()),
+            ActivePopup::File(popup) => Constraint::Max(popup.calculate_required_height()),
+            ActivePopup::None => Constraint::Max(FOOTER_HEIGHT_WITH_HINT),
         };
         let [textarea_rect, _] =
-            Layout::vertical([Constraint::Min(1), Constraint::Max(popup_height)]).areas(area);
+            Layout::vertical([Constraint::Min(1), popup_constraint]).areas(area);
         let mut textarea_rect = textarea_rect;
         textarea_rect.width = textarea_rect.width.saturating_sub(1);
         textarea_rect.x += 1;
@@ -1223,13 +1227,16 @@ impl ChatComposer {
 
 impl WidgetRef for ChatComposer {
     fn render_ref(&self, area: Rect, buf: &mut Buffer) {
-        let popup_height = match &self.active_popup {
-            ActivePopup::Command(popup) => popup.calculate_required_height(),
-            ActivePopup::File(popup) => popup.calculate_required_height(),
-            ActivePopup::None => 1,
+        let (popup_constraint, hint_spacing) = match &self.active_popup {
+            ActivePopup::Command(popup) => (Constraint::Max(popup.calculate_required_height()), 0),
+            ActivePopup::File(popup) => (Constraint::Max(popup.calculate_required_height()), 0),
+            ActivePopup::None => (
+                Constraint::Length(FOOTER_HEIGHT_WITH_HINT),
+                FOOTER_SPACING_HEIGHT,
+            ),
         };
         let [textarea_rect, popup_rect] =
-            Layout::vertical([Constraint::Min(1), Constraint::Max(popup_height)]).areas(area);
+            Layout::vertical([Constraint::Min(1), popup_constraint]).areas(area);
         match &self.active_popup {
             ActivePopup::Command(popup) => {
                 popup.render_ref(popup_rect, buf);
@@ -1238,7 +1245,16 @@ impl WidgetRef for ChatComposer {
                 popup.render_ref(popup_rect, buf);
             }
             ActivePopup::None => {
-                let bottom_line_rect = popup_rect;
+                let hint_rect = if hint_spacing > 0 {
+                    let [_, hint_rect] = Layout::vertical([
+                        Constraint::Length(hint_spacing),
+                        Constraint::Length(FOOTER_HINT_HEIGHT),
+                    ])
+                    .areas(popup_rect);
+                    hint_rect
+                } else {
+                    popup_rect
+                };
                 let mut hint: Vec<Span<'static>> = if self.ctrl_c_quit_hint {
                     let ctrl_c_followup = if self.is_task_running {
                         " to interrupt"
@@ -1309,7 +1325,7 @@ impl WidgetRef for ChatComposer {
 
                 Line::from(hint)
                     .style(Style::default().dim())
-                    .render_ref(bottom_line_rect, buf);
+                    .render_ref(hint_rect, buf);
             }
         }
         let border_style = if self.has_focus {
@@ -1344,6 +1360,7 @@ mod tests {
     use super::*;
     use image::ImageBuffer;
     use image::Rgba;
+    use pretty_assertions::assert_eq;
     use std::path::PathBuf;
     use tempfile::tempdir;
 
@@ -1356,6 +1373,60 @@ mod tests {
     use crate::bottom_pane::textarea::TextArea;
     use tokio::sync::mpsc::unbounded_channel;
 
+    #[test]
+    fn footer_hint_row_is_separated_from_composer() {
+        let (tx, _rx) = unbounded_channel::<AppEvent>();
+        let sender = AppEventSender::new(tx);
+        let composer = ChatComposer::new(
+            true,
+            sender,
+            false,
+            "Ask Codex to do anything".to_string(),
+            false,
+        );
+
+        let area = Rect::new(0, 0, 40, 6);
+        let mut buf = Buffer::empty(area);
+        composer.render_ref(area, &mut buf);
+
+        let row_to_string = |y: u16| {
+            let mut row = String::new();
+            for x in 0..area.width {
+                row.push(buf[(x, y)].symbol().chars().next().unwrap_or(' '));
+            }
+            row
+        };
+
+        let mut hint_row: Option<(u16, String)> = None;
+        for y in 0..area.height {
+            let row = row_to_string(y);
+            if row.contains(" send") {
+                hint_row = Some((y, row));
+                break;
+            }
+        }
+
+        let (hint_row_idx, hint_row_contents) =
+            hint_row.expect("expected footer hint row to be rendered");
+        assert_eq!(
+            hint_row_idx,
+            area.height - 1,
+            "hint row should occupy the bottom line: {hint_row_contents:?}",
+        );
+
+        assert!(
+            hint_row_idx > 0,
+            "expected a spacing row above the footer hints",
+        );
+
+        let spacing_row = row_to_string(hint_row_idx - 1);
+        assert_eq!(
+            spacing_row.trim(),
+            "",
+            "expected blank spacing row above hints but saw: {spacing_row:?}",
+        );
+    }
+
     #[test]
     fn test_current_at_token_basic_cases() {
         let test_cases = vec![

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

@@ -137,11 +137,12 @@ impl BottomPaneView for ListSelectionView {
 
     fn desired_height(&self, _width: u16) -> u16 {
         let rows = (self.items.len()).clamp(1, MAX_POPUP_ROWS);
-        // +1 for the title row, +1 for optional subtitle, +1 for optional footer
-        let mut height = rows as u16 + 1;
+        // +1 for the title row, +1 for a spacer line beneath the header,
+        // +1 for optional subtitle, +1 for optional footer
+        let mut height = rows as u16 + 2;
         if self.subtitle.is_some() {
-            // +1 for subtitle, +1 for a blank spacer line beneath it
-            height = height.saturating_add(2);
+            // +1 for subtitle (the spacer is accounted for above)
+            height = height.saturating_add(1);
         }
         if self.footer_hint.is_some() {
             height = height.saturating_add(2);
@@ -178,17 +179,18 @@ impl BottomPaneView for ListSelectionView {
                 vec![Self::dim_prefix_span(), sub.clone().dim()];
             let subtitle_para = Paragraph::new(Line::from(subtitle_spans));
             subtitle_para.render(subtitle_area, buf);
-            // Render the extra spacer line with the dimmed prefix to align with title/subtitle
-            let spacer_area = Rect {
-                x: area.x,
-                y: next_y.saturating_add(1),
-                width: area.width,
-                height: 1,
-            };
-            Self::render_dim_prefix_line(spacer_area, buf);
-            next_y = next_y.saturating_add(2);
+            next_y = next_y.saturating_add(1);
         }
 
+        let spacer_area = Rect {
+            x: area.x,
+            y: next_y,
+            width: area.width,
+            height: 1,
+        };
+        Self::render_dim_prefix_line(spacer_area, buf);
+        next_y = next_y.saturating_add(1);
+
         let footer_reserved = if self.footer_hint.is_some() { 2 } else { 0 };
         let rows_area = Rect {
             x: area.x,
@@ -245,3 +247,78 @@ impl BottomPaneView for ListSelectionView {
         }
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::BottomPaneView;
+    use super::*;
+    use crate::app_event::AppEvent;
+    use insta::assert_snapshot;
+    use ratatui::layout::Rect;
+    use tokio::sync::mpsc::unbounded_channel;
+
+    fn make_selection_view(subtitle: Option<&str>) -> ListSelectionView {
+        let (tx_raw, _rx) = unbounded_channel::<AppEvent>();
+        let tx = AppEventSender::new(tx_raw);
+        let items = vec![
+            SelectionItem {
+                name: "Read Only".to_string(),
+                description: Some("Codex can read files".to_string()),
+                is_current: true,
+                actions: vec![],
+            },
+            SelectionItem {
+                name: "Full Access".to_string(),
+                description: Some("Codex can edit files".to_string()),
+                is_current: false,
+                actions: vec![],
+            },
+        ];
+        ListSelectionView::new(
+            "Select Approval Mode".to_string(),
+            subtitle.map(str::to_string),
+            Some("Press Enter to confirm or Esc to go back".to_string()),
+            items,
+            tx,
+        )
+    }
+
+    fn render_lines(view: &ListSelectionView) -> String {
+        let width = 48;
+        let height = BottomPaneView::desired_height(view, width);
+        let area = Rect::new(0, 0, width, height);
+        let mut buf = Buffer::empty(area);
+        view.render(area, &mut buf);
+
+        let lines: Vec<String> = (0..area.height)
+            .map(|row| {
+                let mut line = String::new();
+                for col in 0..area.width {
+                    let symbol = buf[(area.x + col, area.y + row)].symbol();
+                    if symbol.is_empty() {
+                        line.push(' ');
+                    } else {
+                        line.push_str(symbol);
+                    }
+                }
+                line
+            })
+            .collect();
+        lines.join("\n")
+    }
+
+    #[test]
+    fn renders_blank_line_between_title_and_items_without_subtitle() {
+        let view = make_selection_view(None);
+        assert_snapshot!(
+            "list_selection_spacing_without_subtitle",
+            render_lines(&view)
+        );
+    }
+
+    #[test]
+    fn renders_blank_line_between_subtitle_and_items() {
+        let view = make_selection_view(Some("Switch between Codex approval presets"));
+        assert_snapshot!("list_selection_spacing_with_subtitle", render_lines(&view));
+    }
+}

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

@@ -10,5 +10,5 @@ expression: terminal.backend()
 "▌                                                                                                   "
 "▌                                                                                                   "
 "▌                                                                                                   "
-"▌                                                                                                   "
+"                                                                                                    "
 " ⏎ send   ⌃J newline   ⌃T transcript   ⌃C quit                                                      "

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

@@ -10,5 +10,5 @@ expression: terminal.backend()
 "▌                                                                                                   "
 "▌                                                                                                   "
 "▌                                                                                                   "
-"▌                                                                                                   "
+"                                                                                                    "
 " ⏎ send   ⌃J newline   ⌃T transcript   ⌃C quit                                                      "

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

@@ -10,5 +10,5 @@ expression: terminal.backend()
 "▌                                                                                                   "
 "▌                                                                                                   "
 "▌                                                                                                   "
-"▌                                                                                                   "
+"                                                                                                    "
 " ⏎ send   ⌃J newline   ⌃T transcript   ⌃C quit                                                      "

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

@@ -10,5 +10,5 @@ expression: terminal.backend()
 "▌                                                                                                   "
 "▌                                                                                                   "
 "▌                                                                                                   "
-"▌                                                                                                   "
+"                                                                                                    "
 " ⏎ send   ⌃J newline   ⌃T transcript   ⌃C quit                                                      "

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

@@ -10,5 +10,5 @@ expression: terminal.backend()
 "▌                                                                                                   "
 "▌                                                                                                   "
 "▌                                                                                                   "
-"▌                                                                                                   "
+"                                                                                                    "
 " ⏎ send   ⌃J newline   ⌃T transcript   ⌃C quit                                                      "

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

@@ -0,0 +1,11 @@
+---
+source: tui/src/bottom_pane/list_selection_view.rs
+expression: render_lines(&view)
+---
+▌ Select Approval Mode                          
+▌ Switch between Codex approval presets         
+▌                                               
+▌> 1. Read Only (current)  Codex can read files 
+▌  2. Full Access  Codex can edit files         
+                                                
+Press Enter to confirm or Esc to go back

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

@@ -0,0 +1,10 @@
+---
+source: tui/src/bottom_pane/list_selection_view.rs
+expression: render_lines(&view)
+---
+▌ Select Approval Mode                          
+▌                                               
+▌> 1. Read Only (current)  Codex can read files 
+▌  2. Full Access  Codex can edit files         
+                                                
+Press Enter to confirm or Esc to go back

codex-rs/tui/src/chatwidget.rs

@@ -91,6 +91,7 @@ use codex_protocol::mcp_protocol::ConversationId;
 struct RunningCommand {
     command: Vec<String>,
     parsed_cmd: Vec<ParsedCommand>,
+    user_initiated_shell_command: bool,
 }
 
 /// Common initialization parameters shared by all `ChatWidget` constructors.
@@ -289,7 +290,9 @@ impl ChatWidget {
     /// separated by newlines rather than auto‑submitting the next one.
     fn on_interrupted_turn(&mut self) {
         // Finalize, log a gentle prompt, and clear running state.
-        self.finalize_turn_with_error_message("Tell the model what to do differently".to_owned());
+        self.finalize_turn_with_error_message(
+            "Conversation interrupted - tell the model what to do differently".to_owned(),
+        );
 
         // If any messages were queued during the task, restore them into the composer.
         if !self.queued_user_messages.is_empty() {
@@ -475,9 +478,9 @@ impl ChatWidget {
 
     pub(crate) fn handle_exec_end_now(&mut self, ev: ExecCommandEndEvent) {
         let running = self.running_commands.remove(&ev.call_id);
-        let (command, parsed) = match running {
-            Some(rc) => (rc.command, rc.parsed_cmd),
-            None => (vec![ev.call_id.clone()], Vec::new()),
+        let (command, parsed, user_initiated_shell_command) = match running {
+            Some(rc) => (rc.command, rc.parsed_cmd, rc.user_initiated_shell_command),
+            None => (vec![ev.call_id.clone()], Vec::new(), false),
         };
 
         if self.active_exec_cell.is_none() {
@@ -487,6 +490,7 @@ impl ChatWidget {
                 ev.call_id.clone(),
                 command,
                 parsed,
+                user_initiated_shell_command,
             ));
         }
         if let Some(cell) = self.active_exec_cell.as_mut() {
@@ -559,6 +563,7 @@ impl ChatWidget {
             RunningCommand {
                 command: ev.command.clone(),
                 parsed_cmd: ev.parsed_cmd.clone(),
+                user_initiated_shell_command: ev.user_initiated_shell_command,
             },
         );
         if let Some(exec) = &self.active_exec_cell {
@@ -566,6 +571,7 @@ impl ChatWidget {
                 ev.call_id.clone(),
                 ev.command.clone(),
                 ev.parsed_cmd.clone(),
+                ev.user_initiated_shell_command,
             ) {
                 self.active_exec_cell = Some(new_exec);
             } else {
@@ -575,6 +581,7 @@ impl ChatWidget {
                     ev.call_id.clone(),
                     ev.command.clone(),
                     ev.parsed_cmd,
+                    ev.user_initiated_shell_command,
                 ));
             }
         } else {
@@ -582,6 +589,7 @@ impl ChatWidget {
                 ev.call_id.clone(),
                 ev.command.clone(),
                 ev.parsed_cmd,
+                ev.user_initiated_shell_command,
             ));
         }
 
@@ -957,6 +965,15 @@ impl ChatWidget {
         let UserMessage { text, image_paths } = user_message;
         let mut items: Vec<InputItem> = Vec::new();
 
+        // Special-case: "!cmd" executes a local shell command instead of sending to the model.
+        if let Some(stripped) = text.strip_prefix('!') {
+            let cmd = stripped.trim().to_string();
+            if !cmd.is_empty() {
+                self.submit_op(Op::RunUserShellCommand { command: cmd });
+                return;
+            }
+        }
+
         if !text.is_empty() {
             items.push(InputItem::Text { text: text.clone() });
         }
@@ -1087,6 +1104,8 @@ impl ChatWidget {
                 self.app_event_tx
                     .send(crate::app_event::AppEvent::ConversationHistory(ev));
             }
+            EventMsg::EnteredReviewMode(_) => {}
+            EventMsg::ExitedReviewMode(_) => {}
         }
     }
 
@@ -1191,9 +1210,13 @@ impl ChatWidget {
                 tracing::info!(
                     "New model: {}, New effort: {}, Current model: {}, Current effort: {}",
                     model_slug.clone(),
-                    effort,
+                    effort
+                        .map(|effort| effort.to_string())
+                        .unwrap_or_else(|| "none".to_string()),
                     current_model,
                     current_effort
+                        .map(|effort| effort.to_string())
+                        .unwrap_or_else(|| "none".to_string())
                 );
             })];
             items.push(SelectionItem {
@@ -1264,7 +1287,7 @@ impl ChatWidget {
     }
 
     /// Set the reasoning effort in the widget's config copy.
-    pub(crate) fn set_reasoning_effort(&mut self, effort: ReasoningEffortConfig) {
+    pub(crate) fn set_reasoning_effort(&mut self, effort: Option<ReasoningEffortConfig>) {
         self.config.model_reasoning_effort = effort;
     }
 
@@ -1273,8 +1296,8 @@ impl ChatWidget {
         self.config.model = model;
     }
 
-    pub(crate) fn add_info_message(&mut self, message: String) {
-        self.add_to_history(history_cell::new_info_event(message));
+    pub(crate) fn add_info_message(&mut self, message: String, hint: Option<String>) {
+        self.add_to_history(history_cell::new_info_event(message, hint));
         self.request_redraw();
     }
 

codex-rs/tui/src/chatwidget/snapshots/codex_tui__chatwidget__tests__chat_small_idle_h2.snap

@@ -3,4 +3,4 @@ source: tui/src/chatwidget/tests.rs
 expression: terminal.backend()
 ---
 "▌ Ask Codex to do anything              "
-" ⏎ send   ⌃J newline   ⌃T transcript   ⌃"
+"                                        "

codex-rs/tui/src/chatwidget/snapshots/codex_tui__chatwidget__tests__chatwidget_exec_and_status_layout_vt100_snapshot.snap

@@ -12,4 +12,5 @@ expression: visual
  Investigating rendering code (0s • Esc to interrupt)
 
 ▌Summarize recent commits
+
  ⏎ send   ⌃J newline   ⌃T transcript   ⌃C quit

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

@@ -2,4 +2,4 @@
 source: tui/src/chatwidget/tests.rs
 expression: blob
 ---
-🖐  '/model' is disabled while a task is in progress.
+■ '/model' is disabled while a task is in progress.

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

@@ -6,5 +6,6 @@ expression: terminal.backend()
 " Analyzing (0s • Esc to interrupt)                                              "
 "                                                                                "
 "▌ Ask Codex to do anything                                                      "
+"                                                                                "
 " ⏎ send   ⌃J newline   ⌃T transcript   ⌃C quit                                  "
 "                                                                                "

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

@@ -138,7 +138,7 @@ fn resumed_initial_messages_render_history() {
     let configured = codex_core::protocol::SessionConfiguredEvent {
         session_id: conversation_id,
         model: "test-model".to_string(),
-        reasoning_effort: ReasoningEffortConfig::default(),
+        reasoning_effort: Some(ReasoningEffortConfig::default()),
         history_log_id: 0,
         history_entry_count: 0,
         initial_messages: Some(vec![
@@ -399,6 +399,7 @@ fn begin_exec(chat: &mut ChatWidget, call_id: &str, raw_cmd: &str) {
             command,
             cwd: std::env::current_dir().unwrap_or_else(|_| PathBuf::from(".")),
             parsed_cmd,
+            user_initiated_shell_command: false,
         }),
     });
 }
@@ -691,6 +692,7 @@ async fn binary_size_transcript_snapshot() {
                                     command: e.command,
                                     cwd: e.cwd,
                                     parsed_cmd: parsed_cmd.into_iter().map(|c| c.into()).collect(),
+                                    user_initiated_shell_command: false,
                                 }),
                             }
                         }
@@ -1674,6 +1676,7 @@ fn chatwidget_exec_and_status_layout_vt100_snapshot() {
                 }
                 .into(),
             ],
+            user_initiated_shell_command: false,
         }),
     });
     chat.handle_codex_event(Event {

codex-rs/tui/src/history_cell.rs

@@ -213,6 +213,7 @@ pub(crate) struct ExecCall {
     pub(crate) command: Vec<String>,
     pub(crate) parsed: Vec<ParsedCommand>,
     pub(crate) output: Option<CommandOutput>,
+    pub(crate) user_initiated_shell_command: bool,
     start_time: Option<Instant>,
     duration: Option<Duration>,
 }
@@ -426,17 +427,28 @@ impl ExecCell {
                 body_lines.extend(wrapped_borrowed.iter().map(|l| line_to_static(l)));
             }
         }
-        if let Some(output) = call.output.as_ref()
-            && output.exit_code != 0
-        {
-            let out = output_lines(Some(output), false, false, false)
-                .into_iter()
-                .join("\n");
-            if !out.trim().is_empty() {
-                // Wrap the output.
-                for line in out.lines() {
-                    let wrapped = textwrap::wrap(line, TwOptions::new(width as usize - 4));
-                    body_lines.extend(wrapped.into_iter().map(|l| Line::from(l.to_string().dim())));
+        if let Some(output) = call.output.as_ref() {
+            // ExecCell-level classification: if any call in the cell represents
+            // a user shell command (ParsedCommand::Unknown), show output
+            // unconditionally. Otherwise, only show on errors.
+            let is_shell_exec_cell = self.calls.iter().any(|c| c.user_initiated_shell_command);
+            let should_show_output = is_shell_exec_cell || output.exit_code != 0;
+            if should_show_output {
+                let line_limit = if is_shell_exec_cell {
+                    USER_SHELL_TOOL_CALL_MAX_LINES
+                } else {
+                    TOOL_CALL_MAX_LINES
+                };
+                let out = output_lines(Some(output), false, false, false, line_limit)
+                    .into_iter()
+                    .join("\n");
+                if !out.trim().is_empty() {
+                    // Wrap the output.
+                    for line in out.lines() {
+                        let wrapped = textwrap::wrap(line, TwOptions::new(width as usize - 4));
+                        body_lines
+                            .extend(wrapped.into_iter().map(|l| Line::from(l.to_string().dim())));
+                    }
                 }
             }
         }
@@ -519,11 +531,13 @@ impl ExecCell {
         call_id: String,
         command: Vec<String>,
         parsed: Vec<ParsedCommand>,
+        user_initiated_shell_command: bool,
     ) -> Option<Self> {
         let call = ExecCall {
             call_id,
             command,
             parsed,
+            user_initiated_shell_command,
             output: None,
             start_time: Some(Instant::now()),
             duration: None,
@@ -566,6 +580,7 @@ impl HistoryCell for CompletedMcpToolCallWithImageOutput {
 }
 
 const TOOL_CALL_MAX_LINES: usize = 5;
+const USER_SHELL_TOOL_CALL_MAX_LINES: usize = 50;
 
 fn title_case(s: &str) -> String {
     if s.is_empty() {
@@ -676,11 +691,13 @@ pub(crate) fn new_active_exec_command(
     call_id: String,
     command: Vec<String>,
     parsed: Vec<ParsedCommand>,
+    user_initiated_shell_command: bool,
 ) -> ExecCell {
     ExecCell::new(ExecCall {
         call_id,
         command,
         parsed,
+        user_initiated_shell_command,
         output: None,
         start_time: Some(Instant::now()),
         duration: None,
@@ -1053,9 +1070,13 @@ pub(crate) fn new_mcp_tools_output(
     PlainHistoryCell { lines }
 }
 
-pub(crate) fn new_info_event(message: String) -> PlainHistoryCell {
-    let lines: Vec<Line<'static>> =
-        vec![vec![padded_emoji("💾").green(), " ".into(), message.into()].into()];
+pub(crate) fn new_info_event(message: String, hint: Option<String>) -> PlainHistoryCell {
+    let mut line = vec!["> ".into(), message.into()];
+    if let Some(hint) = hint {
+        line.push(" ".into());
+        line.push(hint.dark_gray());
+    }
+    let lines: Vec<Line<'static>> = vec![line.into()];
     PlainHistoryCell { lines }
 }
 
@@ -1063,8 +1084,7 @@ pub(crate) fn new_error_event(message: String) -> PlainHistoryCell {
     // Use a hair space (U+200A) to create a subtle, near-invisible separation
     // before the text. VS16 is intentionally omitted to keep spacing tighter
     // in terminals like Ghostty.
-    let lines: Vec<Line<'static>> =
-        vec![vec![padded_emoji("🖐").red().bold(), " ".into(), message.into()].into()];
+    let lines: Vec<Line<'static>> = vec![vec![format!("■ {message}").red()].into()];
     PlainHistoryCell { lines }
 }
 
@@ -1171,6 +1191,7 @@ pub(crate) fn new_patch_apply_failure(stderr: String) -> PlainHistoryCell {
             true,
             true,
             true,
+            TOOL_CALL_MAX_LINES,
         ));
     }
 
@@ -1256,6 +1277,7 @@ fn output_lines(
     only_err: bool,
     include_angle_pipe: bool,
     include_prefix: bool,
+    line_limit: usize,
 ) -> Vec<Line<'static>> {
     let CommandOutput {
         exit_code,
@@ -1271,7 +1293,7 @@ fn output_lines(
     let src = if *exit_code == 0 { stdout } else { stderr };
     let lines: Vec<&str> = src.lines().collect();
     let total = lines.len();
-    let limit = TOOL_CALL_MAX_LINES;
+    let limit = line_limit;
 
     let mut out = Vec::new();
 
@@ -1394,6 +1416,7 @@ mod tests {
                 },
             ],
             output: None,
+            user_initiated_shell_command: false,
             start_time: Some(Instant::now()),
             duration: None,
         });
@@ -1425,6 +1448,7 @@ mod tests {
                 cmd: "rg shimmer_spans".into(),
             }],
             output: None,
+            user_initiated_shell_command: false,
             start_time: Some(Instant::now()),
             duration: None,
         });
@@ -1448,6 +1472,7 @@ mod tests {
                     name: "shimmer.rs".into(),
                     cmd: "cat shimmer.rs".into(),
                 }],
+                false,
             )
             .unwrap();
         cell.complete_call(
@@ -1469,6 +1494,7 @@ mod tests {
                     name: "status_indicator_widget.rs".into(),
                     cmd: "cat status_indicator_widget.rs".into(),
                 }],
+                false,
             )
             .unwrap();
         cell.complete_call(
@@ -1507,6 +1533,7 @@ mod tests {
                 },
             ],
             output: None,
+            user_initiated_shell_command: false,
             start_time: Some(Instant::now()),
             duration: None,
         });
@@ -1535,6 +1562,7 @@ mod tests {
             command: vec!["bash".into(), "-lc".into(), cmd],
             parsed: Vec::new(),
             output: None,
+            user_initiated_shell_command: false,
             start_time: Some(Instant::now()),
             duration: None,
         });
@@ -1565,6 +1593,7 @@ mod tests {
             command: vec!["echo".into(), "ok".into()],
             parsed: Vec::new(),
             output: None,
+            user_initiated_shell_command: false,
             start_time: Some(Instant::now()),
             duration: None,
         });
@@ -1593,6 +1622,7 @@ mod tests {
             command: vec!["bash".into(), "-lc".into(), long],
             parsed: Vec::new(),
             output: None,
+            user_initiated_shell_command: false,
             start_time: Some(Instant::now()),
             duration: None,
         });
@@ -1620,6 +1650,7 @@ mod tests {
             command: vec!["bash".into(), "-lc".into(), cmd],
             parsed: Vec::new(),
             output: None,
+            user_initiated_shell_command: false,
             start_time: Some(Instant::now()),
             duration: None,
         });
@@ -1648,6 +1679,7 @@ mod tests {
             command: vec!["bash".into(), "-lc".into(), cmd],
             parsed: Vec::new(),
             output: None,
+            user_initiated_shell_command: false,
             start_time: Some(Instant::now()),
             duration: None,
         });
@@ -1676,6 +1708,7 @@ mod tests {
             command: vec!["bash".into(), "-lc".into(), "seq 1 10 1>&2 && false".into()],
             parsed: Vec::new(),
             output: None,
+            user_initiated_shell_command: false,
             start_time: Some(Instant::now()),
             duration: None,
         });
@@ -1722,6 +1755,7 @@ mod tests {
             command: vec!["bash".into(), "-lc".into(), long_cmd.to_string()],
             parsed: Vec::new(),
             output: None,
+            user_initiated_shell_command: false,
             start_time: Some(Instant::now()),
             duration: None,
         });

codex-rs/tui/src/lib.rs

@@ -122,6 +122,7 @@ pub async fn run_main(
 
     let overrides = ConfigOverrides {
         model,
+        review_model: None,
         approval_policy,
         sandbox_mode,
         cwd,
@@ -380,11 +381,12 @@ async fn run_ratatui_app(
 
         if switch_to_new_model {
             config.model = GPT5_HIGH_MODEL.to_owned();
+            config.model_reasoning_effort = None;
             if let Err(e) = persist_model_selection(
                 &config.codex_home,
                 active_profile.as_deref(),
                 &config.model,
-                None,
+                config.model_reasoning_effort,
             )
             .await
             {

codex-rs/tui/src/new_model_popup.rs

@@ -82,12 +82,14 @@ impl WidgetRef for &ModelUpgradePopup {
         Clear.render(area, buf);
 
         let mut lines: Vec<Line> = vec![
+            String::new().into(),
+            format!("   Codex is now powered by {GPT5_HIGH_MODEL}, a new model that is").into(),
             Line::from(vec![
-                "> ".into(),
-                format!("Try {GPT5_HIGH_MODEL} as your default model").bold(),
+                "   ".into(),
+                "faster, a better collaborator, ".bold(),
+                "and ".into(),
+                "more steerable.".bold(),
             ]),
-            format!("  {GPT5_HIGH_MODEL} is our latest model tuned for coding workflows.").into(),
-            "  Switch now or keep your current default – you can change models any time.".into(),
             "".into(),
         ];