compaction

### Motivation
- Improve UX by making it explicit that `VISUAL`/`EDITOR` must be set
before launching Codex, not during a running session.

### Description
- Update the external editor error text in `codex-rs/tui/src/app.rs` to:
`"Cannot open external editor: set $VISUAL or $EDITOR before starting
Codex."` and run `just fmt` to apply formatting.

### Testing
- Ran `just fmt` successfully; attempted `cargo test -p codex-tui` but
it failed due to network errors when fetching git dependencies (tests
did not complete).

------
[Codex
Task](https://chatgpt.com/codex/tasks/task_i_6972c2c984948329b1a37d5c5839aff3)

.codespellrc

@@ -1,6 +1,6 @@
 [codespell]
 # Ref: https://github.com/codespell-project/codespell#using-a-config-file
-skip = .git*,vendor,*-lock.yaml,*.lock,.codespellrc,*test.ts,*.jsonl,frame*.txt
+skip = .git*,vendor,*-lock.yaml,*.lock,.codespellrc,*test.ts,*.jsonl,frame*.txt,*.snap,*.snap.new
 check-hidden = true
 ignore-regex = ^\s*"image/\S+": ".*|\b(afterAll)\b
 ignore-words-list = ratatui,ser,iTerm,iterm2,iterm

.github/workflows/rust-release.yml

@@ -252,6 +252,7 @@ jobs:
           # Path that contains the uncompressed binaries for the current
           # ${{ matrix.target }}
           dest="dist/${{ matrix.target }}"
+          repo_root=$PWD
 
           # We want to ship the raw Windows executables in the GitHub Release
           # in addition to the compressed archives. Keep the originals for
@@ -303,7 +304,9 @@ jobs:
                   cp "$dest/$base" "$bundle_dir/$base"
                   cp "$runner_src" "$bundle_dir/codex-command-runner.exe"
                   cp "$setup_src" "$bundle_dir/codex-windows-sandbox-setup.exe"
-                  (cd "$bundle_dir" && 7z a "$dest/${base}.zip" .)
+                  # Use an absolute path so bundle zips land in the real dist
+                  # dir even when 7z runs from a temp directory.
+                  (cd "$bundle_dir" && 7z a "$repo_root/$dest/${base}.zip" .)
                 else
                   echo "warning: missing sandbox binaries; falling back to single-binary zip"
                   echo "warning: expected $runner_src and $setup_src"

AGENTS.md

@@ -11,6 +11,7 @@ In the codex-rs folder where the rust code lives:
 - Always collapse if statements per https://rust-lang.github.io/rust-clippy/master/index.html#collapsible_if
 - Always inline format! args when possible per https://rust-lang.github.io/rust-clippy/master/index.html#uninlined_format_args
 - Use method references over closures when possible per https://rust-lang.github.io/rust-clippy/master/index.html#redundant_closure_for_method_calls
+- When possible, make `match` statements exhaustive and avoid wildcard arms.
 - When writing tests, prefer comparing the equality of entire objects over fields one by one.
 - When making a change that adds or changes an API, ensure that the documentation in the `docs/` folder is up to date if applicable.
 - If you change `ConfigToml` or nested config types, run `just write-config-schema` to update `codex-rs/core/config.schema.json`.

PNPM.md

@@ -15,7 +15,7 @@ This project has been migrated from npm to pnpm to improve dependency management
 
 ```bash
 # Global installation of pnpm
-npm install -g pnpm@10.8.1
+npm install -g pnpm@10.28.2
 
 # Or with corepack (available with Node.js 22+)
 corepack enable
@@ -59,12 +59,12 @@ codex/
 
 ## CI/CD
 
-CI/CD workflows have been updated to use pnpm instead of npm. Make sure your CI environments use pnpm 10.8.1 or higher.
+CI/CD workflows have been updated to use pnpm instead of npm. Make sure your CI environments use pnpm 10.28.2 or higher.
 
 ## Known issues
 
 If you encounter issues with pnpm, try the following solutions:
 
 1. Remove the `node_modules` folder and `pnpm-lock.yaml` file, then run `pnpm install`
-2. Make sure you're using pnpm 10.8.1 or higher
+2. Make sure you're using pnpm 10.28.2 or higher
 3. Verify that Node.js 22 or higher is installed

codex-rs/Cargo.lock

@@ -361,7 +361,7 @@ dependencies = [
  "objc2-foundation",
  "parking_lot",
  "percent-encoding",
- "windows-sys 0.60.2",
+ "windows-sys 0.52.0",
  "wl-clipboard-rs",
  "x11rb",
 ]
@@ -616,9 +616,9 @@ checksum = "c08606f8c3cbf4ce6ec8e28fb0014a2c086708fe954eaa885384a6165172e7e8"
 
 [[package]]
 name = "axum"
-version = "0.8.4"
+version = "0.8.8"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "021e862c184ae977658b36c4500f7feac3221ca5da43e3f25bd04ab6c79a29b5"
+checksum = "8b52af3cb4058c895d37317bb27508dccc8e5f2d39454016b297bf4a400597b8"
 dependencies = [
  "axum-core",
  "bytes",
@@ -634,8 +634,7 @@ dependencies = [
  "mime",
  "percent-encoding",
  "pin-project-lite",
- "rustversion",
- "serde",
+ "serde_core",
  "serde_json",
  "serde_path_to_error",
  "sync_wrapper",
@@ -647,9 +646,9 @@ dependencies = [
 
 [[package]]
 name = "axum-core"
-version = "0.5.2"
+version = "0.5.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "68464cd0412f486726fb3373129ef5d2993f90c34bc2bc1c1e9943b2f4fc7ca6"
+checksum = "08c78f31d7b1291f7ee735c1c6780ccde7785daae9a9206026862dab7d8792d1"
 dependencies = [
  "bytes",
  "futures-core",
@@ -658,7 +657,6 @@ dependencies = [
  "http-body-util",
  "mime",
  "pin-project-lite",
- "rustversion",
  "sync_wrapper",
  "tower-layer",
  "tower-service",
@@ -1685,6 +1683,7 @@ dependencies = [
  "rama-http",
  "rama-http-backend",
  "rama-net",
+ "rama-socks5",
  "rama-tcp",
  "rama-tls-boring",
  "rama-unix",
@@ -2006,6 +2005,7 @@ dependencies = [
  "chrono",
  "codex-protocol",
  "codex-utils-absolute-path",
+ "codex-utils-string",
  "dirs-next",
  "dunce",
  "pretty_assertions",
@@ -2905,7 +2905,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "778e2ac28f6c47af28e4907f13ffd1e1ddbd400980a9abd7c8df189bf578a5ad"
 dependencies = [
  "libc",
- "windows-sys 0.60.2",
+ "windows-sys 0.52.0",
 ]
 
 [[package]]
@@ -3005,7 +3005,7 @@ checksum = "0ce92ff622d6dadf7349484f42c93271a0d49b7cc4d466a936405bacbe10aa78"
 dependencies = [
  "cfg-if",
  "rustix 1.0.8",
- "windows-sys 0.59.0",
+ "windows-sys 0.52.0",
 ]
 
 [[package]]
@@ -3310,7 +3310,7 @@ dependencies = [
  "libc",
  "log",
  "rustversion",
- "windows-link 0.2.0",
+ "windows-link 0.1.3",
  "windows-result 0.3.4",
 ]
 
@@ -3384,9 +3384,9 @@ checksum = "0cc23270f6e1808e30a928bdc84dea0b9b4136a8bc82338574f23baf47bbd280"
 
 [[package]]
 name = "globset"
-version = "0.4.16"
+version = "0.4.18"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "54a1028dfc5f5df5da8a56a73e6c153c9a9708ec57232470703592a3f18e49f5"
+checksum = "52dfc19153a48bde0cbd630453615c8151bce3a5adfac7a0aebfbf0a1e1f57e3"
 dependencies = [
  "aho-corasick",
  "bstr",
@@ -3715,7 +3715,7 @@ dependencies = [
  "libc",
  "percent-encoding",
  "pin-project-lite",
- "socket2 0.5.10",
+ "socket2 0.6.1",
  "system-configuration",
  "tokio",
  "tower-service",
@@ -4092,7 +4092,7 @@ checksum = "e04d7f318608d35d4b61ddd75cbdaee86b023ebe2bd5a66ee0915f0bf93095a9"
 dependencies = [
  "hermit-abi",
  "libc",
- "windows-sys 0.59.0",
+ "windows-sys 0.52.0",
 ]
 
 [[package]]
@@ -5631,7 +5631,7 @@ dependencies = [
  "quinn-udp",
  "rustc-hash",
  "rustls",
- "socket2 0.5.10",
+ "socket2 0.6.1",
  "thiserror 2.0.17",
  "tokio",
  "tracing",
@@ -5668,9 +5668,9 @@ dependencies = [
  "cfg_aliases 0.2.1",
  "libc",
  "once_cell",
- "socket2 0.5.10",
+ "socket2 0.6.1",
  "tracing",
- "windows-sys 0.60.2",
+ "windows-sys 0.52.0",
 ]
 
 [[package]]
@@ -5960,6 +5960,21 @@ dependencies = [
  "tokio",
 ]
 
+[[package]]
+name = "rama-socks5"
+version = "0.3.0-alpha.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "5468b263516daaf258de32542c1974b7cbe962363ad913dcb669f5d46db0ef3e"
+dependencies = [
+ "byteorder",
+ "rama-core",
+ "rama-net",
+ "rama-tcp",
+ "rama-udp",
+ "rama-utils",
+ "tokio",
+]
+
 [[package]]
 name = "rama-tcp"
 version = "0.3.0-alpha.4"
@@ -5998,6 +6013,18 @@ dependencies = [
  "tokio",
 ]
 
+[[package]]
+name = "rama-udp"
+version = "0.3.0-alpha.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "36ed05e0ecac73e084e92a3a8b1fbf16fdae8958c506f0f0eada180a2d99eef4"
+dependencies = [
+ "rama-core",
+ "rama-net",
+ "tokio",
+ "tokio-util",
+]
+
 [[package]]
 name = "rama-unix"
 version = "0.3.0-alpha.4"
@@ -6365,7 +6392,7 @@ dependencies = [
  "errno",
  "libc",
  "linux-raw-sys 0.4.15",
- "windows-sys 0.59.0",
+ "windows-sys 0.52.0",
 ]
 
 [[package]]
@@ -6378,7 +6405,7 @@ dependencies = [
  "errno",
  "libc",
  "linux-raw-sys 0.9.4",
- "windows-sys 0.60.2",
+ "windows-sys 0.52.0",
 ]
 
 [[package]]
@@ -7465,7 +7492,7 @@ dependencies = [
  "getrandom 0.3.3",
  "once_cell",
  "rustix 1.0.8",
- "windows-sys 0.61.1",
+ "windows-sys 0.52.0",
 ]
 
 [[package]]
@@ -7796,12 +7823,10 @@ dependencies = [
 
 [[package]]
 name = "tokio-test"
-version = "0.4.4"
+version = "0.4.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2468baabc3311435b55dd935f702f42cd1b8abb7e754fb7dfb16bd36aa88f9f7"
+checksum = "3f6d24790a10a7af737693a3e8f1d03faef7e6ca0cc99aae5066f533766de545"
 dependencies = [
- "async-stream",
- "bytes",
  "futures-core",
  "tokio",
  "tokio-stream",
@@ -8000,9 +8025,9 @@ checksum = "8df9b6e13f2d32c91b9bd719c00d1958837bc7dec474d94952798cc8e69eeec3"
 
 [[package]]
 name = "tracing"
-version = "0.1.43"
+version = "0.1.44"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "2d15d90a0b5c19378952d479dc858407149d7bb45a14de0142f6c534b16fc647"
+checksum = "63e71662fa4b2a2c3a26f570f037eb95bb1f85397f3cd8076caed2f026a6d100"
 dependencies = [
  "log",
  "pin-project-lite",
@@ -8035,9 +8060,9 @@ dependencies = [
 
 [[package]]
 name = "tracing-core"
-version = "0.1.35"
+version = "0.1.36"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7a04e24fab5c89c6a36eb8558c9656f30d81de51dfa4d3b45f26b21d61fa0a6c"
+checksum = "db97caf9d906fbde555dd62fa95ddba9eecfd14cb388e4f491a66d74cd5fb79a"
 dependencies = [
  "once_cell",
  "valuable",
@@ -8749,7 +8774,7 @@ version = "0.1.9"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "cf221c93e13a30d793f7645a0e7762c55d169dbb0a49671918a2319d289b10bb"
 dependencies = [
- "windows-sys 0.59.0",
+ "windows-sys 0.48.0",
 ]
 
 [[package]]

codex-rs/Cargo.toml

@@ -216,7 +216,7 @@ tokio-tungstenite = { version = "0.28.0", features = ["proxy", "rustls-tls-nativ
 tokio-util = "0.7.18"
 toml = "0.9.5"
 toml_edit = "0.24.0"
-tracing = "0.1.43"
+tracing = "0.1.44"
 tracing-appender = "0.2.3"
 tracing-subscriber = "0.3.22"
 tracing-test = "0.2.5"

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

@@ -117,6 +117,10 @@ client_request_definitions! {
         params: v2::ThreadArchiveParams,
         response: v2::ThreadArchiveResponse,
     },
+    ThreadUnarchive => "thread/unarchive" {
+        params: v2::ThreadUnarchiveParams,
+        response: v2::ThreadUnarchiveResponse,
+    },
     ThreadRollback => "thread/rollback" {
         params: v2::ThreadRollbackParams,
         response: v2::ThreadRollbackResponse,
@@ -524,6 +528,12 @@ server_request_definitions! {
         response: v2::ToolRequestUserInputResponse,
     },
 
+    /// Execute a dynamic tool call on the client.
+    DynamicToolCall => "item/tool/call" {
+        params: v2::DynamicToolCallParams,
+        response: v2::DynamicToolCallResponse,
+    },
+
     /// DEPRECATED APIs below
     /// Request to approve a patch.
     /// This request is used for Turns started via the legacy APIs (i.e. SendUserTurn, SendUserMessage).
@@ -588,6 +598,7 @@ server_notification_definitions! {
     ReasoningSummaryTextDelta => "item/reasoning/summaryTextDelta" (v2::ReasoningSummaryTextDeltaNotification),
     ReasoningSummaryPartAdded => "item/reasoning/summaryPartAdded" (v2::ReasoningSummaryPartAddedNotification),
     ReasoningTextDelta => "item/reasoning/textDelta" (v2::ReasoningTextDeltaNotification),
+    ContextCompactionStarted => "thread/compaction/started" (v2::ContextCompactionStartedNotification),
     ContextCompacted => "thread/compacted" (v2::ContextCompactedNotification),
     DeprecationNotice => "deprecationNotice" (v2::DeprecationNoticeNotification),
     ConfigWarning => "configWarning" (v2::ConfigWarningNotification),

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

@@ -56,6 +56,8 @@ impl ThreadHistoryBuilder {
                 self.handle_agent_reasoning_raw_content(payload)
             }
             EventMsg::TokenCount(_) => {}
+            EventMsg::ContextCompactionStarted(_) => {}
+            EventMsg::ContextCompactionEnded(_) => {}
             EventMsg::EnteredReviewMode(_) => {}
             EventMsg::ExitedReviewMode(_) => {}
             EventMsg::ThreadRolledBack(payload) => self.handle_thread_rollback(payload),

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

@@ -325,6 +325,15 @@ pub struct ToolsV2 {
     pub view_image: Option<bool>,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct DynamicToolSpec {
+    pub name: String,
+    pub description: String,
+    pub input_schema: JsonValue,
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "snake_case")]
 #[ts(export_to = "v2/")]
@@ -905,6 +914,8 @@ pub struct Model {
     pub description: String,
     pub supported_reasoning_efforts: Vec<ReasoningEffortOption>,
     pub default_reasoning_effort: ReasoningEffort,
+    #[serde(default)]
+    pub supports_personality: bool,
     // Only one model should be marked as default.
     pub is_default: bool,
 }
@@ -1088,6 +1099,7 @@ pub struct ThreadStartParams {
     pub developer_instructions: Option<String>,
     pub personality: Option<Personality>,
     pub ephemeral: Option<bool>,
+    pub dynamic_tools: Option<Vec<DynamicToolSpec>>,
     /// If true, opt into emitting raw response items on the event stream.
     ///
     /// This is for internal use only (e.g. Codex Cloud).
@@ -1211,6 +1223,20 @@ pub struct ThreadArchiveParams {
 #[ts(export_to = "v2/")]
 pub struct ThreadArchiveResponse {}
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadUnarchiveParams {
+    pub thread_id: String,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ThreadUnarchiveResponse {
+    pub thread: Thread,
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
@@ -1248,11 +1274,32 @@ pub struct ThreadListParams {
     /// Optional provider filter; when set, only sessions recorded under these
     /// providers are returned. When present but empty, includes all providers.
     pub model_providers: Option<Vec<String>>,
+    /// Optional source filter; when set, only sessions from these source kinds
+    /// are returned. When omitted or empty, defaults to interactive sources.
+    pub source_kinds: Option<Vec<ThreadSourceKind>>,
     /// Optional archived filter; when set to true, only archived threads are returned.
     /// If false or null, only non-archived threads are returned.
     pub archived: Option<bool>,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, Copy, PartialEq, Eq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(rename_all = "camelCase", export_to = "v2/")]
+pub enum ThreadSourceKind {
+    Cli,
+    #[serde(rename = "vscode")]
+    #[ts(rename = "vscode")]
+    VsCode,
+    Exec,
+    AppServer,
+    SubAgent,
+    SubAgentReview,
+    SubAgentCompact,
+    SubAgentThreadSpawn,
+    SubAgentOther,
+    Unknown,
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, Copy, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "snake_case")]
 #[ts(export_to = "v2/")]
@@ -1922,6 +1969,9 @@ pub enum ThreadItem {
     #[serde(rename_all = "camelCase")]
     #[ts(rename_all = "camelCase")]
     ExitedReviewMode { id: String, review: String },
+    #[serde(rename_all = "camelCase")]
+    #[ts(rename_all = "camelCase")]
+    ContextCompaction { id: String },
 }
 
 impl From<CoreTurnItem> for ThreadItem {
@@ -1950,6 +2000,9 @@ impl From<CoreTurnItem> for ThreadItem {
                 id: search.id,
                 query: search.query,
             },
+            CoreTurnItem::ContextCompaction(compaction) => {
+                ThreadItem::ContextCompaction { id: compaction.id }
+            }
         }
     }
 }
@@ -2320,6 +2373,14 @@ pub struct ContextCompactedNotification {
     pub turn_id: String,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct ContextCompactionStartedNotification {
+    pub thread_id: String,
+    pub turn_id: String,
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
@@ -2372,6 +2433,25 @@ pub struct FileChangeRequestApprovalResponse {
     pub decision: FileChangeApprovalDecision,
 }
 
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct DynamicToolCallParams {
+    pub thread_id: String,
+    pub turn_id: String,
+    pub call_id: String,
+    pub tool: String,
+    pub arguments: JsonValue,
+}
+
+#[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
+#[serde(rename_all = "camelCase")]
+#[ts(export_to = "v2/")]
+pub struct DynamicToolCallResponse {
+    pub output: String,
+    pub success: bool,
+}
+
 #[derive(Serialize, Deserialize, Debug, Clone, PartialEq, JsonSchema, TS)]
 #[serde(rename_all = "camelCase")]
 #[ts(export_to = "v2/")]
@@ -2389,6 +2469,8 @@ pub struct ToolRequestUserInputQuestion {
     pub id: String,
     pub header: String,
     pub question: String,
+    #[serde(default)]
+    pub is_other: bool,
     pub options: Option<Vec<ToolRequestUserInputOption>>,
 }
 
@@ -2549,10 +2631,12 @@ mod tests {
     use super::*;
     use codex_protocol::items::AgentMessageContent;
     use codex_protocol::items::AgentMessageItem;
+    use codex_protocol::items::ContextCompactionItem;
     use codex_protocol::items::ReasoningItem;
     use codex_protocol::items::TurnItem;
     use codex_protocol::items::UserMessageItem;
     use codex_protocol::items::WebSearchItem;
+    use codex_protocol::models::WebSearchAction;
     use codex_protocol::protocol::NetworkAccess as CoreNetworkAccess;
     use codex_protocol::user_input::UserInput as CoreUserInput;
     use pretty_assertions::assert_eq;
@@ -2660,6 +2744,9 @@ mod tests {
         let search_item = TurnItem::WebSearch(WebSearchItem {
             id: "search-1".to_string(),
             query: "docs".to_string(),
+            action: WebSearchAction::Search {
+                query: Some("docs".to_string()),
+            },
         });
 
         assert_eq!(
@@ -2669,6 +2756,17 @@ mod tests {
                 query: "docs".to_string(),
             }
         );
+
+        let compaction_item = TurnItem::ContextCompaction(ContextCompactionItem {
+            id: "compact-1".to_string(),
+        });
+
+        assert_eq!(
+            ThreadItem::from(compaction_item),
+            ThreadItem::ContextCompaction {
+                id: "compact-1".to_string(),
+            }
+        );
     }
 
     #[test]

codex-rs/app-server/README.md

@@ -81,6 +81,7 @@ Example (from OpenAI's official VSCode extension):
 - `thread/loaded/list` — list the thread ids currently loaded in memory.
 - `thread/read` — read a stored thread by id without resuming it; optionally include turns via `includeTurns`.
 - `thread/archive` — move a thread’s rollout file into the archived directory; returns `{}` on success.
+- `thread/unarchive` — move an archived rollout file back into the sessions directory; returns the restored `thread` on success.
 - `thread/rollback` — drop the last N turns from the agent’s in-memory context and persist a rollback marker in the rollout so future resumes see the pruned history; returns the updated `thread` (with `turns` populated) on success.
 - `turn/start` — add user input to a thread and begin Codex generation; responds with the initial `turn` object and streams `turn/started`, `item/*`, and `turn/completed` notifications.
 - `turn/interrupt` — request cancellation of an in-flight turn by `(thread_id, turn_id)`; success is an empty `{}` response and the turn finishes with `status: "interrupted"`.
@@ -114,7 +115,20 @@ Start a fresh thread when you need a new Codex conversation.
     "cwd": "/Users/me/project",
     "approvalPolicy": "never",
     "sandbox": "workspaceWrite",
-    "personality": "friendly"
+    "personality": "friendly",
+    "dynamicTools": [
+        {
+            "name": "lookup_ticket",
+            "description": "Fetch a ticket by id",
+            "inputSchema": {
+                "type": "object",
+                "properties": {
+                    "id": { "type": "string" }
+                },
+                "required": ["id"]
+            }
+        }
+    ],
 } }
 { "id": 10, "result": {
     "thread": {
@@ -153,6 +167,7 @@ To branch from a stored session, call `thread/fork` with the `thread.id`. This c
 - `limit` — server defaults to a reasonable page size if unset.
 - `sortKey` — `created_at` (default) or `updated_at`.
 - `modelProviders` — restrict results to specific providers; unset, null, or an empty array will include all providers.
+- `sourceKinds` — restrict results to specific sources; omit or pass `[]` for interactive sessions only (`cli`, `vscode`).
 - `archived` — when `true`, list archived threads only. When `false` or `null`, list non-archived threads (default).
 
 Example:
@@ -210,6 +225,15 @@ Use `thread/archive` to move the persisted rollout (stored as a JSONL file on di
 
 An archived thread will not appear in `thread/list` unless `archived` is set to `true`.
 
+### Example: Unarchive a thread
+
+Use `thread/unarchive` to move an archived rollout back into the sessions directory.
+
+```json
+{ "method": "thread/unarchive", "id": 24, "params": { "threadId": "thr_b" } }
+{ "id": 24, "result": { "thread": { "id": "thr_b" } } }
+```
+
 ### Example: Start a turn (send user input)
 
 Turns attach user input (text or images) to a thread and trigger Codex generation. The `input` field is a list of discriminated unions:

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

@@ -24,7 +24,9 @@ use codex_app_server_protocol::CommandExecutionRequestApprovalParams;
 use codex_app_server_protocol::CommandExecutionRequestApprovalResponse;
 use codex_app_server_protocol::CommandExecutionStatus;
 use codex_app_server_protocol::ContextCompactedNotification;
+use codex_app_server_protocol::ContextCompactionStartedNotification;
 use codex_app_server_protocol::DeprecationNoticeNotification;
+use codex_app_server_protocol::DynamicToolCallParams;
 use codex_app_server_protocol::ErrorNotification;
 use codex_app_server_protocol::ExecCommandApprovalParams;
 use codex_app_server_protocol::ExecCommandApprovalResponse;
@@ -85,6 +87,7 @@ use codex_core::protocol::TurnDiffEvent;
 use codex_core::review_format::format_review_findings_block;
 use codex_core::review_prompts;
 use codex_protocol::ThreadId;
+use codex_protocol::dynamic_tools::DynamicToolResponse as CoreDynamicToolResponse;
 use codex_protocol::plan_tool::UpdatePlanArgs;
 use codex_protocol::protocol::ReviewOutputEvent;
 use codex_protocol::request_user_input::RequestUserInputAnswer as CoreRequestUserInputAnswer;
@@ -276,6 +279,7 @@ pub(crate) async fn apply_bespoke_event_handling(
                         id: question.id,
                         header: question.header,
                         question: question.question,
+                        is_other: question.is_other,
                         options: question.options.map(|options| {
                             options
                                 .into_iter()
@@ -318,6 +322,40 @@ pub(crate) async fn apply_bespoke_event_handling(
                 }
             }
         }
+        EventMsg::DynamicToolCallRequest(request) => {
+            if matches!(api_version, ApiVersion::V2) {
+                let call_id = request.call_id;
+                let params = DynamicToolCallParams {
+                    thread_id: conversation_id.to_string(),
+                    turn_id: request.turn_id,
+                    call_id: call_id.clone(),
+                    tool: request.tool,
+                    arguments: request.arguments,
+                };
+                let rx = outgoing
+                    .send_request(ServerRequestPayload::DynamicToolCall(params))
+                    .await;
+                tokio::spawn(async move {
+                    crate::dynamic_tools::on_call_response(call_id, rx, conversation).await;
+                });
+            } else {
+                error!(
+                    "dynamic tool calls are only supported on api v2 (call_id: {})",
+                    request.call_id
+                );
+                let call_id = request.call_id;
+                let _ = conversation
+                    .submit(Op::DynamicToolResponse {
+                        id: call_id.clone(),
+                        response: CoreDynamicToolResponse {
+                            call_id,
+                            output: "dynamic tool calls require api v2".to_string(),
+                            success: false,
+                        },
+                    })
+                    .await;
+            }
+        }
         // TODO(celia): properly construct McpToolCall TurnItem in core.
         EventMsg::McpToolCallBegin(begin_event) => {
             let notification = construct_mcp_tool_call_notification(
@@ -564,7 +602,18 @@ pub(crate) async fn apply_bespoke_event_handling(
                 .send_server_notification(ServerNotification::AgentMessageDelta(notification))
                 .await;
         }
-        EventMsg::ContextCompacted(..) => {
+        EventMsg::ContextCompactionStarted(..) => {
+            let notification = ContextCompactionStartedNotification {
+                thread_id: conversation_id.to_string(),
+                turn_id: event_turn_id.clone(),
+            };
+            outgoing
+                .send_server_notification(ServerNotification::ContextCompactionStarted(
+                    notification,
+                ))
+                .await;
+        }
+        EventMsg::ContextCompactionEnded(..) => {
             let notification = ContextCompactedNotification {
                 thread_id: conversation_id.to_string(),
                 turn_id: event_turn_id.clone(),

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

@@ -31,6 +31,7 @@ use codex_app_server_protocol::CollaborationModeListResponse;
 use codex_app_server_protocol::CommandExecParams;
 use codex_app_server_protocol::ConversationGitInfo;
 use codex_app_server_protocol::ConversationSummary;
+use codex_app_server_protocol::DynamicToolSpec as ApiDynamicToolSpec;
 use codex_app_server_protocol::ExecOneOffCommandResponse;
 use codex_app_server_protocol::FeedbackUploadParams;
 use codex_app_server_protocol::FeedbackUploadResponse;
@@ -110,9 +111,12 @@ use codex_app_server_protocol::ThreadResumeParams;
 use codex_app_server_protocol::ThreadResumeResponse;
 use codex_app_server_protocol::ThreadRollbackParams;
 use codex_app_server_protocol::ThreadSortKey;
+use codex_app_server_protocol::ThreadSourceKind;
 use codex_app_server_protocol::ThreadStartParams;
 use codex_app_server_protocol::ThreadStartResponse;
 use codex_app_server_protocol::ThreadStartedNotification;
+use codex_app_server_protocol::ThreadUnarchiveParams;
+use codex_app_server_protocol::ThreadUnarchiveResponse;
 use codex_app_server_protocol::Turn;
 use codex_app_server_protocol::TurnError;
 use codex_app_server_protocol::TurnInterruptParams;
@@ -129,7 +133,6 @@ use codex_chatgpt::connectors;
 use codex_core::AuthManager;
 use codex_core::CodexThread;
 use codex_core::Cursor as RolloutCursor;
-use codex_core::INTERACTIVE_SESSION_SOURCES;
 use codex_core::InitialHistory;
 use codex_core::NewThread;
 use codex_core::RolloutRecorder;
@@ -150,6 +153,7 @@ use codex_core::error::CodexErr;
 use codex_core::exec::ExecParams;
 use codex_core::exec_env::create_env;
 use codex_core::features::Feature;
+use codex_core::find_archived_thread_path_by_id_str;
 use codex_core::find_thread_path_by_id_str;
 use codex_core::git_info::git_diff_to_remote;
 use codex_core::mcp::collect_mcp_snapshot;
@@ -163,7 +167,9 @@ use codex_core::protocol::ReviewTarget as CoreReviewTarget;
 use codex_core::protocol::SessionConfiguredEvent;
 use codex_core::read_head_for_summary;
 use codex_core::read_session_meta_line;
+use codex_core::rollout_date_parts;
 use codex_core::sandboxing::SandboxPermissions;
+use codex_core::windows_sandbox::WindowsSandboxLevelExt;
 use codex_feedback::CodexFeedback;
 use codex_login::ServerOptions as LoginServerOptions;
 use codex_login::ShutdownHandle;
@@ -171,6 +177,8 @@ use codex_login::run_login_server;
 use codex_protocol::ThreadId;
 use codex_protocol::config_types::ForcedLoginMethod;
 use codex_protocol::config_types::Personality;
+use codex_protocol::config_types::WindowsSandboxLevel;
+use codex_protocol::dynamic_tools::DynamicToolSpec as CoreDynamicToolSpec;
 use codex_protocol::items::TurnItem;
 use codex_protocol::models::ResponseItem;
 use codex_protocol::protocol::AgentStatus;
@@ -203,6 +211,9 @@ use tracing::info;
 use tracing::warn;
 use uuid::Uuid;
 
+use crate::filters::compute_source_filters;
+use crate::filters::source_kind_matches;
+
 type PendingInterruptQueue = Vec<(RequestId, ApiVersion)>;
 pub(crate) type PendingInterrupts = Arc<Mutex<HashMap<ThreadId, PendingInterruptQueue>>>;
 
@@ -402,6 +413,9 @@ impl CodexMessageProcessor {
             ClientRequest::ThreadArchive { request_id, params } => {
                 self.thread_archive(request_id, params).await;
             }
+            ClientRequest::ThreadUnarchive { request_id, params } => {
+                self.thread_unarchive(request_id, params).await;
+            }
             ClientRequest::ThreadRollback { request_id, params } => {
                 self.thread_rollback(request_id, params).await;
             }
@@ -1247,12 +1261,14 @@ impl CodexMessageProcessor {
         let timeout_ms = params
             .timeout_ms
             .and_then(|timeout_ms| u64::try_from(timeout_ms).ok());
+        let windows_sandbox_level = WindowsSandboxLevel::from_config(&self.config);
         let exec_params = ExecParams {
             command: params.command,
             cwd,
             expiration: timeout_ms.into(),
             env,
             sandbox_permissions: SandboxPermissions::UseDefault,
+            windows_sandbox_level,
             justification: None,
             arg0: None,
         };
@@ -1411,35 +1427,81 @@ impl CodexMessageProcessor {
     }
 
     async fn thread_start(&mut self, request_id: RequestId, params: ThreadStartParams) {
+        let ThreadStartParams {
+            model,
+            model_provider,
+            cwd,
+            approval_policy,
+            sandbox,
+            config,
+            base_instructions,
+            developer_instructions,
+            dynamic_tools,
+            experimental_raw_events,
+            personality,
+            ephemeral,
+        } = params;
         let mut typesafe_overrides = self.build_thread_config_overrides(
-            params.model,
-            params.model_provider,
-            params.cwd,
-            params.approval_policy,
-            params.sandbox,
-            params.base_instructions,
-            params.developer_instructions,
-            params.personality,
+            model,
+            model_provider,
+            cwd,
+            approval_policy,
+            sandbox,
+            base_instructions,
+            developer_instructions,
+            personality,
         );
-        typesafe_overrides.ephemeral = Some(params.ephemeral.unwrap_or_default());
+        typesafe_overrides.ephemeral = ephemeral;
 
-        let config =
-            match derive_config_from_params(&self.cli_overrides, params.config, typesafe_overrides)
-                .await
-            {
-                Ok(config) => config,
-                Err(err) => {
-                    let error = JSONRPCErrorError {
-                        code: INVALID_REQUEST_ERROR_CODE,
-                        message: format!("error deriving config: {err}"),
-                        data: None,
-                    };
-                    self.outgoing.send_error(request_id, error).await;
-                    return;
-                }
-            };
+        let config = match derive_config_from_params(
+            &self.cli_overrides,
+            config,
+            typesafe_overrides,
+        )
+        .await
+        {
+            Ok(config) => config,
+            Err(err) => {
+                let error = JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message: format!("error deriving config: {err}"),
+                    data: None,
+                };
+                self.outgoing.send_error(request_id, error).await;
+                return;
+            }
+        };
 
-        match self.thread_manager.start_thread(config).await {
+        let dynamic_tools = dynamic_tools.unwrap_or_default();
+        let core_dynamic_tools = if dynamic_tools.is_empty() {
+            Vec::new()
+        } else {
+            let snapshot = collect_mcp_snapshot(&config).await;
+            let mcp_tool_names = snapshot.tools.keys().cloned().collect::<HashSet<_>>();
+            if let Err(message) = validate_dynamic_tools(&dynamic_tools, &mcp_tool_names) {
+                let error = JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message,
+                    data: None,
+                };
+                self.outgoing.send_error(request_id, error).await;
+                return;
+            }
+            dynamic_tools
+                .into_iter()
+                .map(|tool| CoreDynamicToolSpec {
+                    name: tool.name,
+                    description: tool.description,
+                    input_schema: tool.input_schema,
+                })
+                .collect()
+        };
+
+        match self
+            .thread_manager
+            .start_thread_with_tools(config, core_dynamic_tools)
+            .await
+        {
             Ok(new_conv) => {
                 let NewThread {
                     thread_id,
@@ -1489,7 +1551,7 @@ impl CodexMessageProcessor {
                 if let Err(err) = self
                     .attach_conversation_listener(
                         thread_id,
-                        params.experimental_raw_events,
+                        experimental_raw_events,
                         ApiVersion::V2,
                     )
                     .await
@@ -1595,6 +1657,150 @@ impl CodexMessageProcessor {
         }
     }
 
+    async fn thread_unarchive(&mut self, request_id: RequestId, params: ThreadUnarchiveParams) {
+        let thread_id = match ThreadId::from_string(&params.thread_id) {
+            Ok(id) => id,
+            Err(err) => {
+                let error = JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message: format!("invalid thread id: {err}"),
+                    data: None,
+                };
+                self.outgoing.send_error(request_id, error).await;
+                return;
+            }
+        };
+
+        let archived_path = match find_archived_thread_path_by_id_str(
+            &self.config.codex_home,
+            &thread_id.to_string(),
+        )
+        .await
+        {
+            Ok(Some(path)) => path,
+            Ok(None) => {
+                let error = JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message: format!("no archived rollout found for thread id {thread_id}"),
+                    data: None,
+                };
+                self.outgoing.send_error(request_id, error).await;
+                return;
+            }
+            Err(err) => {
+                let error = JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message: format!("failed to locate archived thread id {thread_id}: {err}"),
+                    data: None,
+                };
+                self.outgoing.send_error(request_id, error).await;
+                return;
+            }
+        };
+
+        let rollout_path_display = archived_path.display().to_string();
+        let fallback_provider = self.config.model_provider_id.clone();
+        let archived_folder = self
+            .config
+            .codex_home
+            .join(codex_core::ARCHIVED_SESSIONS_SUBDIR);
+
+        let result: Result<Thread, JSONRPCErrorError> = async {
+            let canonical_archived_dir = tokio::fs::canonicalize(&archived_folder).await.map_err(
+                |err| JSONRPCErrorError {
+                    code: INTERNAL_ERROR_CODE,
+                    message: format!(
+                        "failed to unarchive thread: unable to resolve archived directory: {err}"
+                    ),
+                    data: None,
+                },
+            )?;
+            let canonical_rollout_path = tokio::fs::canonicalize(&archived_path).await;
+            let canonical_rollout_path = if let Ok(path) = canonical_rollout_path
+                && path.starts_with(&canonical_archived_dir)
+            {
+                path
+            } else {
+                return Err(JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message: format!(
+                        "rollout path `{rollout_path_display}` must be in archived directory"
+                    ),
+                    data: None,
+                });
+            };
+
+            let required_suffix = format!("{thread_id}.jsonl");
+            let Some(file_name) = canonical_rollout_path.file_name().map(OsStr::to_owned) else {
+                return Err(JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message: format!("rollout path `{rollout_path_display}` missing file name"),
+                    data: None,
+                });
+            };
+            if !file_name
+                .to_string_lossy()
+                .ends_with(required_suffix.as_str())
+            {
+                return Err(JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message: format!(
+                        "rollout path `{rollout_path_display}` does not match thread id {thread_id}"
+                    ),
+                    data: None,
+                });
+            }
+
+            let Some((year, month, day)) = rollout_date_parts(&file_name) else {
+                return Err(JSONRPCErrorError {
+                    code: INVALID_REQUEST_ERROR_CODE,
+                    message: format!(
+                        "rollout path `{rollout_path_display}` missing filename timestamp"
+                    ),
+                    data: None,
+                });
+            };
+
+            let sessions_folder = self.config.codex_home.join(codex_core::SESSIONS_SUBDIR);
+            let dest_dir = sessions_folder.join(year).join(month).join(day);
+            let restored_path = dest_dir.join(&file_name);
+            tokio::fs::create_dir_all(&dest_dir)
+                .await
+                .map_err(|err| JSONRPCErrorError {
+                    code: INTERNAL_ERROR_CODE,
+                    message: format!("failed to unarchive thread: {err}"),
+                    data: None,
+                })?;
+            tokio::fs::rename(&canonical_rollout_path, &restored_path)
+                .await
+                .map_err(|err| JSONRPCErrorError {
+                    code: INTERNAL_ERROR_CODE,
+                    message: format!("failed to unarchive thread: {err}"),
+                    data: None,
+                })?;
+            let summary =
+                read_summary_from_rollout(restored_path.as_path(), fallback_provider.as_str())
+                    .await
+                    .map_err(|err| JSONRPCErrorError {
+                        code: INTERNAL_ERROR_CODE,
+                        message: format!("failed to read unarchived thread: {err}"),
+                        data: None,
+                    })?;
+            Ok(summary_to_thread(summary))
+        }
+        .await;
+
+        match result {
+            Ok(thread) => {
+                let response = ThreadUnarchiveResponse { thread };
+                self.outgoing.send_response(request_id, response).await;
+            }
+            Err(err) => {
+                self.outgoing.send_error(request_id, err).await;
+            }
+        }
+    }
+
     async fn thread_rollback(&mut self, request_id: RequestId, params: ThreadRollbackParams) {
         let ThreadRollbackParams {
             thread_id,
@@ -1646,6 +1852,7 @@ impl CodexMessageProcessor {
             limit,
             sort_key,
             model_providers,
+            source_kinds,
             archived,
         } = params;
 
@@ -1662,6 +1869,7 @@ impl CodexMessageProcessor {
                 requested_page_size,
                 cursor,
                 model_providers,
+                source_kinds,
                 core_sort_key,
                 archived.unwrap_or(false),
             )
@@ -2339,6 +2547,7 @@ impl CodexMessageProcessor {
                 requested_page_size,
                 cursor,
                 model_providers,
+                None,
                 CoreThreadSortKey::UpdatedAt,
                 false,
             )
@@ -2359,6 +2568,7 @@ impl CodexMessageProcessor {
         requested_page_size: usize,
         cursor: Option<String>,
         model_providers: Option<Vec<String>>,
+        source_kinds: Option<Vec<ThreadSourceKind>>,
         sort_key: CoreThreadSortKey,
         archived: bool,
     ) -> Result<(Vec<ConversationSummary>, Option<String>), JSONRPCErrorError> {
@@ -2388,6 +2598,8 @@ impl CodexMessageProcessor {
             None => Some(vec![self.config.model_provider_id.clone()]),
         };
         let fallback_provider = self.config.model_provider_id.clone();
+        let (allowed_sources_vec, source_kind_filter) = compute_source_filters(source_kinds);
+        let allowed_sources = allowed_sources_vec.as_slice();
 
         while remaining > 0 {
             let page_size = remaining.min(THREAD_LIST_MAX_LIMIT);
@@ -2397,7 +2609,7 @@ impl CodexMessageProcessor {
                     page_size,
                     cursor_obj.as_ref(),
                     sort_key,
-                    INTERACTIVE_SESSION_SOURCES,
+                    allowed_sources,
                     model_provider_filter.as_deref(),
                     fallback_provider.as_str(),
                 )
@@ -2413,7 +2625,7 @@ impl CodexMessageProcessor {
                     page_size,
                     cursor_obj.as_ref(),
                     sort_key,
-                    INTERACTIVE_SESSION_SOURCES,
+                    allowed_sources,
                     model_provider_filter.as_deref(),
                     fallback_provider.as_str(),
                 )
@@ -2442,6 +2654,11 @@ impl CodexMessageProcessor {
                         updated_at,
                     )
                 })
+                .filter(|summary| {
+                    source_kind_filter
+                        .as_ref()
+                        .is_none_or(|filter| source_kind_matches(&summary.source, filter))
+                })
                 .collect::<Vec<_>>();
             if filtered.len() > remaining {
                 filtered.truncate(remaining);
@@ -2652,6 +2869,8 @@ impl CodexMessageProcessor {
             }
         };
 
+        let scopes = scopes.or_else(|| server.scopes.clone());
+
         match perform_oauth_login_return_url(
             &name,
             &url,
@@ -3672,6 +3891,7 @@ impl CodexMessageProcessor {
                     cwd: params.cwd,
                     approval_policy: params.approval_policy.map(AskForApproval::to_core),
                     sandbox_policy: params.sandbox_policy.map(|p| p.to_core()),
+                    windows_sandbox_level: None,
                     model: params.model,
                     effort: params.effort.map(Some),
                     summary: params.summary,
@@ -4322,6 +4542,41 @@ fn errors_to_info(
         .collect()
 }
 
+fn validate_dynamic_tools(
+    tools: &[ApiDynamicToolSpec],
+    mcp_tool_names: &HashSet<String>,
+) -> Result<(), String> {
+    let mut seen = HashSet::new();
+    for tool in tools {
+        let name = tool.name.trim();
+        if name.is_empty() {
+            return Err("dynamic tool name must not be empty".to_string());
+        }
+        if name != tool.name {
+            return Err(format!(
+                "dynamic tool name has leading/trailing whitespace: {}",
+                tool.name
+            ));
+        }
+        if name == "mcp" || name.starts_with("mcp__") {
+            return Err(format!("dynamic tool name is reserved: {name}"));
+        }
+        if mcp_tool_names.contains(name) {
+            return Err(format!("dynamic tool name conflicts with MCP tool: {name}"));
+        }
+        if !seen.insert(name.to_string()) {
+            return Err(format!("duplicate dynamic tool name: {name}"));
+        }
+
+        if let Err(err) = codex_core::parse_tool_input_schema(&tool.input_schema) {
+            return Err(format!(
+                "dynamic tool input schema is not supported for {name}: {err}"
+            ));
+        }
+    }
+    Ok(())
+}
+
 /// Derive the effective [`Config`] by layering three override sources.
 ///
 /// Precedence (lowest to highest):
@@ -4602,6 +4857,28 @@ mod tests {
     use serde_json::json;
     use tempfile::TempDir;
 
+    #[test]
+    fn validate_dynamic_tools_rejects_unsupported_input_schema() {
+        let tools = vec![ApiDynamicToolSpec {
+            name: "my_tool".to_string(),
+            description: "test".to_string(),
+            input_schema: json!({"type": "null"}),
+        }];
+        let err = validate_dynamic_tools(&tools, &HashSet::new()).expect_err("invalid schema");
+        assert!(err.contains("my_tool"), "unexpected error: {err}");
+    }
+
+    #[test]
+    fn validate_dynamic_tools_accepts_sanitizable_input_schema() {
+        let tools = vec![ApiDynamicToolSpec {
+            name: "my_tool".to_string(),
+            description: "test".to_string(),
+            // Missing `type` is common; core sanitizes these to a supported schema.
+            input_schema: json!({"properties": {}}),
+        }];
+        validate_dynamic_tools(&tools, &HashSet::new()).expect("valid schema");
+    }
+
     #[test]
     fn extract_conversation_summary_prefers_plain_user_messages() -> Result<()> {
         let conversation_id = ThreadId::from_string("3f941c35-29b3-493b-b0a4-e25800d9aeb0")?;

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

@@ -0,0 +1,58 @@
+use codex_app_server_protocol::DynamicToolCallResponse;
+use codex_core::CodexThread;
+use codex_protocol::dynamic_tools::DynamicToolResponse as CoreDynamicToolResponse;
+use codex_protocol::protocol::Op;
+use std::sync::Arc;
+use tokio::sync::oneshot;
+use tracing::error;
+
+pub(crate) async fn on_call_response(
+    call_id: String,
+    receiver: oneshot::Receiver<serde_json::Value>,
+    conversation: Arc<CodexThread>,
+) {
+    let response = receiver.await;
+    let value = match response {
+        Ok(value) => value,
+        Err(err) => {
+            error!("request failed: {err:?}");
+            let fallback = CoreDynamicToolResponse {
+                call_id: call_id.clone(),
+                output: "dynamic tool request failed".to_string(),
+                success: false,
+            };
+            if let Err(err) = conversation
+                .submit(Op::DynamicToolResponse {
+                    id: call_id.clone(),
+                    response: fallback,
+                })
+                .await
+            {
+                error!("failed to submit DynamicToolResponse: {err}");
+            }
+            return;
+        }
+    };
+
+    let response = serde_json::from_value::<DynamicToolCallResponse>(value).unwrap_or_else(|err| {
+        error!("failed to deserialize DynamicToolCallResponse: {err}");
+        DynamicToolCallResponse {
+            output: "dynamic tool response was invalid".to_string(),
+            success: false,
+        }
+    });
+    let response = CoreDynamicToolResponse {
+        call_id: call_id.clone(),
+        output: response.output,
+        success: response.success,
+    };
+    if let Err(err) = conversation
+        .submit(Op::DynamicToolResponse {
+            id: call_id,
+            response,
+        })
+        .await
+    {
+        error!("failed to submit DynamicToolResponse: {err}");
+    }
+}

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

@@ -0,0 +1,155 @@
+use codex_app_server_protocol::ThreadSourceKind;
+use codex_core::INTERACTIVE_SESSION_SOURCES;
+use codex_protocol::protocol::SessionSource as CoreSessionSource;
+use codex_protocol::protocol::SubAgentSource as CoreSubAgentSource;
+
+pub(crate) fn compute_source_filters(
+    source_kinds: Option<Vec<ThreadSourceKind>>,
+) -> (Vec<CoreSessionSource>, Option<Vec<ThreadSourceKind>>) {
+    let Some(source_kinds) = source_kinds else {
+        return (INTERACTIVE_SESSION_SOURCES.to_vec(), None);
+    };
+
+    if source_kinds.is_empty() {
+        return (INTERACTIVE_SESSION_SOURCES.to_vec(), None);
+    }
+
+    let requires_post_filter = source_kinds.iter().any(|kind| {
+        matches!(
+            kind,
+            ThreadSourceKind::Exec
+                | ThreadSourceKind::AppServer
+                | ThreadSourceKind::SubAgent
+                | ThreadSourceKind::SubAgentReview
+                | ThreadSourceKind::SubAgentCompact
+                | ThreadSourceKind::SubAgentThreadSpawn
+                | ThreadSourceKind::SubAgentOther
+                | ThreadSourceKind::Unknown
+        )
+    });
+
+    if requires_post_filter {
+        (Vec::new(), Some(source_kinds))
+    } else {
+        let interactive_sources = source_kinds
+            .iter()
+            .filter_map(|kind| match kind {
+                ThreadSourceKind::Cli => Some(CoreSessionSource::Cli),
+                ThreadSourceKind::VsCode => Some(CoreSessionSource::VSCode),
+                ThreadSourceKind::Exec
+                | ThreadSourceKind::AppServer
+                | ThreadSourceKind::SubAgent
+                | ThreadSourceKind::SubAgentReview
+                | ThreadSourceKind::SubAgentCompact
+                | ThreadSourceKind::SubAgentThreadSpawn
+                | ThreadSourceKind::SubAgentOther
+                | ThreadSourceKind::Unknown => None,
+            })
+            .collect::<Vec<_>>();
+        (interactive_sources, Some(source_kinds))
+    }
+}
+
+pub(crate) fn source_kind_matches(source: &CoreSessionSource, filter: &[ThreadSourceKind]) -> bool {
+    filter.iter().any(|kind| match kind {
+        ThreadSourceKind::Cli => matches!(source, CoreSessionSource::Cli),
+        ThreadSourceKind::VsCode => matches!(source, CoreSessionSource::VSCode),
+        ThreadSourceKind::Exec => matches!(source, CoreSessionSource::Exec),
+        ThreadSourceKind::AppServer => matches!(source, CoreSessionSource::Mcp),
+        ThreadSourceKind::SubAgent => matches!(source, CoreSessionSource::SubAgent(_)),
+        ThreadSourceKind::SubAgentReview => {
+            matches!(
+                source,
+                CoreSessionSource::SubAgent(CoreSubAgentSource::Review)
+            )
+        }
+        ThreadSourceKind::SubAgentCompact => {
+            matches!(
+                source,
+                CoreSessionSource::SubAgent(CoreSubAgentSource::Compact)
+            )
+        }
+        ThreadSourceKind::SubAgentThreadSpawn => matches!(
+            source,
+            CoreSessionSource::SubAgent(CoreSubAgentSource::ThreadSpawn { .. })
+        ),
+        ThreadSourceKind::SubAgentOther => matches!(
+            source,
+            CoreSessionSource::SubAgent(CoreSubAgentSource::Other(_))
+        ),
+        ThreadSourceKind::Unknown => matches!(source, CoreSessionSource::Unknown),
+    })
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use codex_protocol::ThreadId;
+    use pretty_assertions::assert_eq;
+    use uuid::Uuid;
+
+    #[test]
+    fn compute_source_filters_defaults_to_interactive_sources() {
+        let (allowed_sources, filter) = compute_source_filters(None);
+
+        assert_eq!(allowed_sources, INTERACTIVE_SESSION_SOURCES.to_vec());
+        assert_eq!(filter, None);
+    }
+
+    #[test]
+    fn compute_source_filters_empty_means_interactive_sources() {
+        let (allowed_sources, filter) = compute_source_filters(Some(Vec::new()));
+
+        assert_eq!(allowed_sources, INTERACTIVE_SESSION_SOURCES.to_vec());
+        assert_eq!(filter, None);
+    }
+
+    #[test]
+    fn compute_source_filters_interactive_only_skips_post_filtering() {
+        let source_kinds = vec![ThreadSourceKind::Cli, ThreadSourceKind::VsCode];
+        let (allowed_sources, filter) = compute_source_filters(Some(source_kinds.clone()));
+
+        assert_eq!(
+            allowed_sources,
+            vec![CoreSessionSource::Cli, CoreSessionSource::VSCode]
+        );
+        assert_eq!(filter, Some(source_kinds));
+    }
+
+    #[test]
+    fn compute_source_filters_subagent_variant_requires_post_filtering() {
+        let source_kinds = vec![ThreadSourceKind::SubAgentReview];
+        let (allowed_sources, filter) = compute_source_filters(Some(source_kinds.clone()));
+
+        assert_eq!(allowed_sources, Vec::new());
+        assert_eq!(filter, Some(source_kinds));
+    }
+
+    #[test]
+    fn source_kind_matches_distinguishes_subagent_variants() {
+        let parent_thread_id =
+            ThreadId::from_string(&Uuid::new_v4().to_string()).expect("valid thread id");
+        let review = CoreSessionSource::SubAgent(CoreSubAgentSource::Review);
+        let spawn = CoreSessionSource::SubAgent(CoreSubAgentSource::ThreadSpawn {
+            parent_thread_id,
+            depth: 1,
+        });
+
+        assert!(source_kind_matches(
+            &review,
+            &[ThreadSourceKind::SubAgentReview]
+        ));
+        assert!(!source_kind_matches(
+            &review,
+            &[ThreadSourceKind::SubAgentThreadSpawn]
+        ));
+        assert!(source_kind_matches(
+            &spawn,
+            &[ThreadSourceKind::SubAgentThreadSpawn]
+        ));
+        assert!(!source_kind_matches(
+            &spawn,
+            &[ThreadSourceKind::SubAgentReview]
+        ));
+    }
+}

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

@@ -40,7 +40,9 @@ use tracing_subscriber::util::SubscriberInitExt;
 mod bespoke_event_handling;
 mod codex_message_processor;
 mod config_api;
+mod dynamic_tools;
 mod error_code;
+mod filters;
 mod fuzzy_file_search;
 mod message_processor;
 mod models;
@@ -133,7 +135,7 @@ fn project_config_warning(config: &Config) -> Option<ConfigWarningNotification>
                     .disabled_reason
                     .as_ref()
                     .map(ToString::to_string)
-                    .unwrap_or_else(|| "Config folder disabled.".to_string()),
+                    .unwrap_or_else(|| "config.toml is disabled.".to_string()),
             ));
         }
     }
@@ -142,7 +144,11 @@ fn project_config_warning(config: &Config) -> Option<ConfigWarningNotification>
         return None;
     }
 
-    let mut message = "The following config folders are disabled:\n".to_string();
+    let mut message = concat!(
+        "Project config.toml files are disabled in the following folders. ",
+        "Settings in those files are ignored, but skills and exec policies still load.\n",
+    )
+    .to_string();
     for (index, (folder, reason)) in disabled_folders.iter().enumerate() {
         let display_index = index + 1;
         message.push_str(&format!("    {display_index}. {folder}\n"));

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

@@ -28,6 +28,7 @@ fn model_from_preset(preset: ModelPreset) -> Model {
             preset.supported_reasoning_efforts,
         ),
         default_reasoning_effort: preset.default_reasoning_effort,
+        supports_personality: preset.supports_personality,
         is_default: preset.is_default,
     }
 }

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

@@ -30,6 +30,7 @@ pub use responses::create_final_assistant_message_sse_response;
 pub use responses::create_request_user_input_sse_response;
 pub use responses::create_shell_command_sse_response;
 pub use rollout::create_fake_rollout;
+pub use rollout::create_fake_rollout_with_source;
 pub use rollout::create_fake_rollout_with_text_elements;
 pub use rollout::rollout_path;
 use serde::de::DeserializeOwned;

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

@@ -53,6 +53,7 @@ use codex_app_server_protocol::ThreadReadParams;
 use codex_app_server_protocol::ThreadResumeParams;
 use codex_app_server_protocol::ThreadRollbackParams;
 use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadUnarchiveParams;
 use codex_app_server_protocol::TurnInterruptParams;
 use codex_app_server_protocol::TurnStartParams;
 use codex_core::default_client::CODEX_INTERNAL_ORIGINATOR_OVERRIDE_ENV_VAR;
@@ -365,6 +366,15 @@ impl McpProcess {
         self.send_request("thread/archive", params).await
     }
 
+    /// Send a `thread/unarchive` JSON-RPC request.
+    pub async fn send_thread_unarchive_request(
+        &mut self,
+        params: ThreadUnarchiveParams,
+    ) -> anyhow::Result<i64> {
+        let params = Some(serde_json::to_value(params)?);
+        self.send_request("thread/unarchive", params).await
+    }
+
     /// Send a `thread/rollback` JSON-RPC request.
     pub async fn send_thread_rollback_request(
         &mut self,

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

@@ -67,6 +67,7 @@ pub fn create_request_user_input_sse_response(call_id: &str) -> anyhow::Result<S
             "id": "confirm_path",
             "header": "Confirm",
             "question": "Proceed with the plan?",
+            "isOther": false,
             "options": [{
                 "label": "Yes (Recommended)",
                 "description": "Continue the current plan."

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

@@ -38,6 +38,27 @@ pub fn create_fake_rollout(
     preview: &str,
     model_provider: Option<&str>,
     git_info: Option<GitInfo>,
+) -> Result<String> {
+    create_fake_rollout_with_source(
+        codex_home,
+        filename_ts,
+        meta_rfc3339,
+        preview,
+        model_provider,
+        git_info,
+        SessionSource::Cli,
+    )
+}
+
+/// Create a minimal rollout file with an explicit session source.
+pub fn create_fake_rollout_with_source(
+    codex_home: &Path,
+    filename_ts: &str,
+    meta_rfc3339: &str,
+    preview: &str,
+    model_provider: Option<&str>,
+    git_info: Option<GitInfo>,
+    source: SessionSource,
 ) -> Result<String> {
     let uuid = Uuid::new_v4();
     let uuid_str = uuid.to_string();
@@ -57,7 +78,7 @@ pub fn create_fake_rollout(
         cwd: PathBuf::from("/"),
         originator: "codex".to_string(),
         cli_version: "0.0.0".to_string(),
-        source: SessionSource::Cli,
+        source,
         model_provider: model_provider.map(str::to_string),
         base_instructions: None,
     };

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

@@ -108,6 +108,10 @@ async fn test_codex_jsonrpc_conversation_flow() -> Result<()> {
     let AddConversationSubscriptionResponse { subscription_id } =
         to_response::<AddConversationSubscriptionResponse>(add_listener_resp)?;
 
+    // Drop any buffered events from conversation setup to avoid
+    // matching an earlier task_complete.
+    mcp.clear_message_buffer();
+
     // 3) sendUserMessage (should trigger notifications; we only validate an OK response)
     let send_user_id = mcp
         .send_send_user_message_request(SendUserMessageParams {
@@ -125,13 +129,38 @@ async fn test_codex_jsonrpc_conversation_flow() -> Result<()> {
     .await??;
     let SendUserMessageResponse {} = to_response::<SendUserMessageResponse>(send_user_resp)?;
 
-    // Verify the task_finished notification is received.
-    // Note this also ensures that the final request to the server was made.
-    let task_finished_notification: JSONRPCNotification = timeout(
+    let task_started_notification: JSONRPCNotification = timeout(
         DEFAULT_READ_TIMEOUT,
-        mcp.read_stream_until_notification_message("codex/event/task_complete"),
+        mcp.read_stream_until_notification_message("codex/event/task_started"),
     )
     .await??;
+    let task_started_event: Event = serde_json::from_value(
+        task_started_notification
+            .params
+            .clone()
+            .expect("task_started should have params"),
+    )
+    .expect("task_started should deserialize to Event");
+
+    // Verify the task_finished notification for this turn is received.
+    // Note this also ensures that the final request to the server was made.
+    let task_finished_notification: JSONRPCNotification = loop {
+        let notification: JSONRPCNotification = timeout(
+            DEFAULT_READ_TIMEOUT,
+            mcp.read_stream_until_notification_message("codex/event/task_complete"),
+        )
+        .await??;
+        let event: Event = serde_json::from_value(
+            notification
+                .params
+                .clone()
+                .expect("task_complete should have params"),
+        )
+        .expect("task_complete should deserialize to Event");
+        if event.id == task_started_event.id {
+            break notification;
+        }
+    };
     let serde_json::Value::Object(map) = task_finished_notification
         .params
         .expect("notification should have params")

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

@@ -0,0 +1,286 @@
+use anyhow::Context;
+use anyhow::Result;
+use app_test_support::McpProcess;
+use app_test_support::create_final_assistant_message_sse_response;
+use app_test_support::create_mock_responses_server_sequence_unchecked;
+use app_test_support::to_response;
+use codex_app_server_protocol::DynamicToolCallParams;
+use codex_app_server_protocol::DynamicToolCallResponse;
+use codex_app_server_protocol::DynamicToolSpec;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ServerRequest;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_app_server_protocol::TurnStartParams;
+use codex_app_server_protocol::TurnStartResponse;
+use codex_app_server_protocol::UserInput as V2UserInput;
+use core_test_support::responses;
+use pretty_assertions::assert_eq;
+use serde_json::Value;
+use serde_json::json;
+use std::path::Path;
+use std::time::Duration;
+use tempfile::TempDir;
+use tokio::time::timeout;
+use wiremock::MockServer;
+
+const DEFAULT_READ_TIMEOUT: Duration = Duration::from_secs(10);
+
+/// Ensures dynamic tool specs are serialized into the model request payload.
+#[tokio::test]
+async fn thread_start_injects_dynamic_tools_into_model_requests() -> Result<()> {
+    let responses = vec![create_final_assistant_message_sse_response("Done")?];
+    let server = create_mock_responses_server_sequence_unchecked(responses).await;
+
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    // Use a minimal JSON schema so we can assert the tool payload round-trips.
+    let input_schema = json!({
+        "type": "object",
+        "properties": {
+            "city": { "type": "string" }
+        },
+        "required": ["city"],
+        "additionalProperties": false,
+    });
+    let dynamic_tool = DynamicToolSpec {
+        name: "demo_tool".to_string(),
+        description: "Demo dynamic tool".to_string(),
+        input_schema: input_schema.clone(),
+    };
+
+    // Thread start injects dynamic tools into the thread's tool registry.
+    let thread_req = mcp
+        .send_thread_start_request(ThreadStartParams {
+            dynamic_tools: Some(vec![dynamic_tool.clone()]),
+            ..Default::default()
+        })
+        .await?;
+    let thread_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
+    )
+    .await??;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
+
+    // Start a turn so a model request is issued.
+    let turn_req = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "Hello".to_string(),
+                text_elements: Vec::new(),
+            }],
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_req)),
+    )
+    .await??;
+    let _turn: TurnStartResponse = to_response::<TurnStartResponse>(turn_resp)?;
+
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/completed"),
+    )
+    .await??;
+
+    // Inspect the captured model request to assert the tool spec made it through.
+    let bodies = responses_bodies(&server).await?;
+    let body = bodies
+        .first()
+        .context("expected at least one responses request")?;
+    let tool = find_tool(body, &dynamic_tool.name)
+        .context("expected dynamic tool to be injected into request")?;
+
+    assert_eq!(
+        tool.get("description"),
+        Some(&Value::String(dynamic_tool.description.clone()))
+    );
+    assert_eq!(tool.get("parameters"), Some(&input_schema));
+
+    Ok(())
+}
+
+/// Exercises the full dynamic tool call path (server request, client response, model output).
+#[tokio::test]
+async fn dynamic_tool_call_round_trip_sends_output_to_model() -> Result<()> {
+    let call_id = "dyn-call-1";
+    let tool_name = "demo_tool";
+    let tool_args = json!({ "city": "Paris" });
+    let tool_call_arguments = serde_json::to_string(&tool_args)?;
+
+    // First response triggers a dynamic tool call, second closes the turn.
+    let responses = vec![
+        responses::sse(vec![
+            responses::ev_response_created("resp-1"),
+            responses::ev_function_call(call_id, tool_name, &tool_call_arguments),
+            responses::ev_completed("resp-1"),
+        ]),
+        create_final_assistant_message_sse_response("Done")?,
+    ];
+    let server = create_mock_responses_server_sequence_unchecked(responses).await;
+
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path(), &server.uri())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let dynamic_tool = DynamicToolSpec {
+        name: tool_name.to_string(),
+        description: "Demo dynamic tool".to_string(),
+        input_schema: json!({
+            "type": "object",
+            "properties": {
+                "city": { "type": "string" }
+            },
+            "required": ["city"],
+            "additionalProperties": false,
+        }),
+    };
+
+    let thread_req = mcp
+        .send_thread_start_request(ThreadStartParams {
+            dynamic_tools: Some(vec![dynamic_tool]),
+            ..Default::default()
+        })
+        .await?;
+    let thread_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(thread_req)),
+    )
+    .await??;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(thread_resp)?;
+
+    // Start a turn so the tool call is emitted.
+    let turn_req = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: thread.id.clone(),
+            input: vec![V2UserInput::Text {
+                text: "Run the tool".to_string(),
+                text_elements: Vec::new(),
+            }],
+            ..Default::default()
+        })
+        .await?;
+    let turn_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_req)),
+    )
+    .await??;
+    let TurnStartResponse { turn } = to_response::<TurnStartResponse>(turn_resp)?;
+
+    // Read the tool call request from the app server.
+    let request = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_request_message(),
+    )
+    .await??;
+    let (request_id, params) = match request {
+        ServerRequest::DynamicToolCall { request_id, params } => (request_id, params),
+        other => panic!("expected DynamicToolCall request, got {other:?}"),
+    };
+
+    let expected = DynamicToolCallParams {
+        thread_id: thread.id,
+        turn_id: turn.id,
+        call_id: call_id.to_string(),
+        tool: tool_name.to_string(),
+        arguments: tool_args.clone(),
+    };
+    assert_eq!(params, expected);
+
+    // Respond to the tool call so the model receives a function_call_output.
+    let response = DynamicToolCallResponse {
+        output: "dynamic-ok".to_string(),
+        success: true,
+    };
+    mcp.send_response(request_id, serde_json::to_value(response)?)
+        .await?;
+
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/completed"),
+    )
+    .await??;
+
+    let bodies = responses_bodies(&server).await?;
+    let output = bodies
+        .iter()
+        .find_map(|body| function_call_output_text(body, call_id))
+        .context("expected function_call_output in follow-up request")?;
+    assert_eq!(output, "dynamic-ok");
+
+    Ok(())
+}
+
+async fn responses_bodies(server: &MockServer) -> Result<Vec<Value>> {
+    let requests = server
+        .received_requests()
+        .await
+        .context("failed to fetch received requests")?;
+
+    requests
+        .into_iter()
+        .filter(|req| req.url.path().ends_with("/responses"))
+        .map(|req| {
+            req.body_json::<Value>()
+                .context("request body should be JSON")
+        })
+        .collect()
+}
+
+fn find_tool<'a>(body: &'a Value, name: &str) -> Option<&'a Value> {
+    body.get("tools")
+        .and_then(Value::as_array)
+        .and_then(|tools| {
+            tools
+                .iter()
+                .find(|tool| tool.get("name").and_then(Value::as_str) == Some(name))
+        })
+}
+
+fn function_call_output_text(body: &Value, call_id: &str) -> Option<String> {
+    body.get("input")
+        .and_then(Value::as_array)
+        .and_then(|items| {
+            items.iter().find(|item| {
+                item.get("type").and_then(Value::as_str) == Some("function_call_output")
+                    && item.get("call_id").and_then(Value::as_str) == Some(call_id)
+            })
+        })
+        .and_then(|item| item.get("output"))
+        .and_then(Value::as_str)
+        .map(str::to_string)
+}
+
+fn create_config_toml(codex_home: &Path, server_uri: &str) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(
+        config_toml,
+        format!(
+            r#"
+model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "read-only"
+
+model_provider = "mock_provider"
+
+[model_providers.mock_provider]
+name = "Mock provider for test"
+base_url = "{server_uri}/v1"
+wire_api = "responses"
+request_max_retries = 0
+stream_max_retries = 0
+"#
+        ),
+    )
+}

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

@@ -3,6 +3,7 @@ mod analytics;
 mod app_list;
 mod collaboration_mode_list;
 mod config_rpc;
+mod dynamic_tools;
 mod initialize;
 mod model_list;
 mod output_schema;
@@ -17,5 +18,6 @@ mod thread_read;
 mod thread_resume;
 mod thread_rollback;
 mod thread_start;
+mod thread_unarchive;
 mod turn_interrupt;
 mod turn_start;

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

@@ -72,6 +72,7 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
                 },
             ],
             default_reasoning_effort: ReasoningEffort::Medium,
+            supports_personality: false,
             is_default: true,
         },
         Model {
@@ -99,6 +100,7 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
                 },
             ],
             default_reasoning_effort: ReasoningEffort::Medium,
+            supports_personality: false,
             is_default: false,
         },
         Model {
@@ -118,6 +120,7 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
                 },
             ],
             default_reasoning_effort: ReasoningEffort::Medium,
+            supports_personality: false,
             is_default: false,
         },
         Model {
@@ -151,6 +154,7 @@ async fn list_models_returns_all_models_with_large_limit() -> Result<()> {
                 },
             ],
             default_reasoning_effort: ReasoningEffort::Medium,
+            supports_personality: false,
             is_default: false,
         },
     ];

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

@@ -1,6 +1,7 @@
 use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::create_fake_rollout;
+use app_test_support::create_fake_rollout_with_source;
 use app_test_support::rollout_path;
 use app_test_support::to_response;
 use chrono::DateTime;
@@ -12,8 +13,12 @@ use codex_app_server_protocol::RequestId;
 use codex_app_server_protocol::SessionSource;
 use codex_app_server_protocol::ThreadListResponse;
 use codex_app_server_protocol::ThreadSortKey;
+use codex_app_server_protocol::ThreadSourceKind;
 use codex_core::ARCHIVED_SESSIONS_SUBDIR;
+use codex_protocol::ThreadId;
 use codex_protocol::protocol::GitInfo as CoreGitInfo;
+use codex_protocol::protocol::SessionSource as CoreSessionSource;
+use codex_protocol::protocol::SubAgentSource;
 use pretty_assertions::assert_eq;
 use std::cmp::Reverse;
 use std::fs;
@@ -38,9 +43,10 @@ async fn list_threads(
     cursor: Option<String>,
     limit: Option<u32>,
     providers: Option<Vec<String>>,
+    source_kinds: Option<Vec<ThreadSourceKind>>,
     archived: Option<bool>,
 ) -> Result<ThreadListResponse> {
-    list_threads_with_sort(mcp, cursor, limit, providers, None, archived).await
+    list_threads_with_sort(mcp, cursor, limit, providers, source_kinds, None, archived).await
 }
 
 async fn list_threads_with_sort(
@@ -48,6 +54,7 @@ async fn list_threads_with_sort(
     cursor: Option<String>,
     limit: Option<u32>,
     providers: Option<Vec<String>>,
+    source_kinds: Option<Vec<ThreadSourceKind>>,
     sort_key: Option<ThreadSortKey>,
     archived: Option<bool>,
 ) -> Result<ThreadListResponse> {
@@ -57,6 +64,7 @@ async fn list_threads_with_sort(
             limit,
             sort_key,
             model_providers: providers,
+            source_kinds,
             archived,
         })
         .await?;
@@ -131,6 +139,7 @@ async fn thread_list_basic_empty() -> Result<()> {
         Some(10),
         Some(vec!["mock_provider".to_string()]),
         None,
+        None,
     )
     .await?;
     assert!(data.is_empty());
@@ -194,6 +203,7 @@ async fn thread_list_pagination_next_cursor_none_on_last_page() -> Result<()> {
         Some(2),
         Some(vec!["mock_provider".to_string()]),
         None,
+        None,
     )
     .await?;
     assert_eq!(data1.len(), 2);
@@ -219,6 +229,7 @@ async fn thread_list_pagination_next_cursor_none_on_last_page() -> Result<()> {
         Some(2),
         Some(vec!["mock_provider".to_string()]),
         None,
+        None,
     )
     .await?;
     assert!(data2.len() <= 2);
@@ -269,6 +280,7 @@ async fn thread_list_respects_provider_filter() -> Result<()> {
         Some(10),
         Some(vec!["other_provider".to_string()]),
         None,
+        None,
     )
     .await?;
     assert_eq!(data.len(), 1);
@@ -287,6 +299,207 @@ async fn thread_list_respects_provider_filter() -> Result<()> {
     Ok(())
 }
 
+#[tokio::test]
+async fn thread_list_empty_source_kinds_defaults_to_interactive_only() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_minimal_config(codex_home.path())?;
+
+    let cli_id = create_fake_rollout(
+        codex_home.path(),
+        "2025-02-01T10-00-00",
+        "2025-02-01T10:00:00Z",
+        "CLI",
+        Some("mock_provider"),
+        None,
+    )?;
+    let exec_id = create_fake_rollout_with_source(
+        codex_home.path(),
+        "2025-02-01T11-00-00",
+        "2025-02-01T11:00:00Z",
+        "Exec",
+        Some("mock_provider"),
+        None,
+        CoreSessionSource::Exec,
+    )?;
+
+    let mut mcp = init_mcp(codex_home.path()).await?;
+
+    let ThreadListResponse { data, next_cursor } = list_threads(
+        &mut mcp,
+        None,
+        Some(10),
+        Some(vec!["mock_provider".to_string()]),
+        Some(Vec::new()),
+        None,
+    )
+    .await?;
+
+    assert_eq!(next_cursor, None);
+    let ids: Vec<_> = data.iter().map(|thread| thread.id.as_str()).collect();
+    assert_eq!(ids, vec![cli_id.as_str()]);
+    assert_ne!(cli_id, exec_id);
+    assert_eq!(data[0].source, SessionSource::Cli);
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn thread_list_filters_by_source_kind_subagent_thread_spawn() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_minimal_config(codex_home.path())?;
+
+    let cli_id = create_fake_rollout(
+        codex_home.path(),
+        "2025-02-01T10-00-00",
+        "2025-02-01T10:00:00Z",
+        "CLI",
+        Some("mock_provider"),
+        None,
+    )?;
+
+    let parent_thread_id = ThreadId::from_string(&Uuid::new_v4().to_string())?;
+    let subagent_id = create_fake_rollout_with_source(
+        codex_home.path(),
+        "2025-02-01T11-00-00",
+        "2025-02-01T11:00:00Z",
+        "SubAgent",
+        Some("mock_provider"),
+        None,
+        CoreSessionSource::SubAgent(SubAgentSource::ThreadSpawn {
+            parent_thread_id,
+            depth: 1,
+        }),
+    )?;
+
+    let mut mcp = init_mcp(codex_home.path()).await?;
+
+    let ThreadListResponse { data, next_cursor } = list_threads(
+        &mut mcp,
+        None,
+        Some(10),
+        Some(vec!["mock_provider".to_string()]),
+        Some(vec![ThreadSourceKind::SubAgentThreadSpawn]),
+        None,
+    )
+    .await?;
+
+    assert_eq!(next_cursor, None);
+    let ids: Vec<_> = data.iter().map(|thread| thread.id.as_str()).collect();
+    assert_eq!(ids, vec![subagent_id.as_str()]);
+    assert_ne!(cli_id, subagent_id);
+    assert!(matches!(data[0].source, SessionSource::SubAgent(_)));
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn thread_list_filters_by_subagent_variant() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_minimal_config(codex_home.path())?;
+
+    let parent_thread_id = ThreadId::from_string(&Uuid::new_v4().to_string())?;
+
+    let review_id = create_fake_rollout_with_source(
+        codex_home.path(),
+        "2025-02-02T09-00-00",
+        "2025-02-02T09:00:00Z",
+        "Review",
+        Some("mock_provider"),
+        None,
+        CoreSessionSource::SubAgent(SubAgentSource::Review),
+    )?;
+    let compact_id = create_fake_rollout_with_source(
+        codex_home.path(),
+        "2025-02-02T10-00-00",
+        "2025-02-02T10:00:00Z",
+        "Compact",
+        Some("mock_provider"),
+        None,
+        CoreSessionSource::SubAgent(SubAgentSource::Compact),
+    )?;
+    let spawn_id = create_fake_rollout_with_source(
+        codex_home.path(),
+        "2025-02-02T11-00-00",
+        "2025-02-02T11:00:00Z",
+        "Spawn",
+        Some("mock_provider"),
+        None,
+        CoreSessionSource::SubAgent(SubAgentSource::ThreadSpawn {
+            parent_thread_id,
+            depth: 1,
+        }),
+    )?;
+    let other_id = create_fake_rollout_with_source(
+        codex_home.path(),
+        "2025-02-02T12-00-00",
+        "2025-02-02T12:00:00Z",
+        "Other",
+        Some("mock_provider"),
+        None,
+        CoreSessionSource::SubAgent(SubAgentSource::Other("custom".to_string())),
+    )?;
+
+    let mut mcp = init_mcp(codex_home.path()).await?;
+
+    let review = list_threads(
+        &mut mcp,
+        None,
+        Some(10),
+        Some(vec!["mock_provider".to_string()]),
+        Some(vec![ThreadSourceKind::SubAgentReview]),
+        None,
+    )
+    .await?;
+    let review_ids: Vec<_> = review
+        .data
+        .iter()
+        .map(|thread| thread.id.as_str())
+        .collect();
+    assert_eq!(review_ids, vec![review_id.as_str()]);
+
+    let compact = list_threads(
+        &mut mcp,
+        None,
+        Some(10),
+        Some(vec!["mock_provider".to_string()]),
+        Some(vec![ThreadSourceKind::SubAgentCompact]),
+        None,
+    )
+    .await?;
+    let compact_ids: Vec<_> = compact
+        .data
+        .iter()
+        .map(|thread| thread.id.as_str())
+        .collect();
+    assert_eq!(compact_ids, vec![compact_id.as_str()]);
+
+    let spawn = list_threads(
+        &mut mcp,
+        None,
+        Some(10),
+        Some(vec!["mock_provider".to_string()]),
+        Some(vec![ThreadSourceKind::SubAgentThreadSpawn]),
+        None,
+    )
+    .await?;
+    let spawn_ids: Vec<_> = spawn.data.iter().map(|thread| thread.id.as_str()).collect();
+    assert_eq!(spawn_ids, vec![spawn_id.as_str()]);
+
+    let other = list_threads(
+        &mut mcp,
+        None,
+        Some(10),
+        Some(vec!["mock_provider".to_string()]),
+        Some(vec![ThreadSourceKind::SubAgentOther]),
+        None,
+    )
+    .await?;
+    let other_ids: Vec<_> = other.data.iter().map(|thread| thread.id.as_str()).collect();
+    assert_eq!(other_ids, vec![other_id.as_str()]);
+
+    Ok(())
+}
+
 #[tokio::test]
 async fn thread_list_fetches_until_limit_or_exhausted() -> Result<()> {
     let codex_home = TempDir::new()?;
@@ -319,6 +532,7 @@ async fn thread_list_fetches_until_limit_or_exhausted() -> Result<()> {
         Some(8),
         Some(vec!["target_provider".to_string()]),
         None,
+        None,
     )
     .await?;
     assert_eq!(
@@ -364,6 +578,7 @@ async fn thread_list_enforces_max_limit() -> Result<()> {
         Some(200),
         Some(vec!["mock_provider".to_string()]),
         None,
+        None,
     )
     .await?;
     assert_eq!(
@@ -410,6 +625,7 @@ async fn thread_list_stops_when_not_enough_filtered_results_exist() -> Result<()
         Some(10),
         Some(vec!["target_provider".to_string()]),
         None,
+        None,
     )
     .await?;
     assert_eq!(
@@ -457,6 +673,7 @@ async fn thread_list_includes_git_info() -> Result<()> {
         Some(10),
         Some(vec!["mock_provider".to_string()]),
         None,
+        None,
     )
     .await?;
     let thread = data
@@ -516,6 +733,7 @@ async fn thread_list_default_sorts_by_created_at() -> Result<()> {
         Some(vec!["mock_provider".to_string()]),
         None,
         None,
+        None,
     )
     .await?;
 
@@ -575,6 +793,7 @@ async fn thread_list_sort_updated_at_orders_by_mtime() -> Result<()> {
         None,
         Some(10),
         Some(vec!["mock_provider".to_string()]),
+        None,
         Some(ThreadSortKey::UpdatedAt),
         None,
     )
@@ -639,6 +858,7 @@ async fn thread_list_updated_at_paginates_with_cursor() -> Result<()> {
         None,
         Some(2),
         Some(vec!["mock_provider".to_string()]),
+        None,
         Some(ThreadSortKey::UpdatedAt),
         None,
     )
@@ -655,6 +875,7 @@ async fn thread_list_updated_at_paginates_with_cursor() -> Result<()> {
         Some(cursor1),
         Some(2),
         Some(vec!["mock_provider".to_string()]),
+        None,
         Some(ThreadSortKey::UpdatedAt),
         None,
     )
@@ -696,6 +917,7 @@ async fn thread_list_created_at_tie_breaks_by_uuid() -> Result<()> {
         Some(10),
         Some(vec!["mock_provider".to_string()]),
         None,
+        None,
     )
     .await?;
 
@@ -747,6 +969,7 @@ async fn thread_list_updated_at_tie_breaks_by_uuid() -> Result<()> {
         None,
         Some(10),
         Some(vec!["mock_provider".to_string()]),
+        None,
         Some(ThreadSortKey::UpdatedAt),
         None,
     )
@@ -787,6 +1010,7 @@ async fn thread_list_updated_at_uses_mtime() -> Result<()> {
         None,
         Some(10),
         Some(vec!["mock_provider".to_string()]),
+        None,
         Some(ThreadSortKey::UpdatedAt),
         None,
     )
@@ -846,6 +1070,7 @@ async fn thread_list_archived_filter() -> Result<()> {
         Some(10),
         Some(vec!["mock_provider".to_string()]),
         None,
+        None,
     )
     .await?;
     assert_eq!(data.len(), 1);
@@ -856,6 +1081,7 @@ async fn thread_list_archived_filter() -> Result<()> {
         None,
         Some(10),
         Some(vec!["mock_provider".to_string()]),
+        None,
         Some(true),
     )
     .await?;
@@ -878,6 +1104,7 @@ async fn thread_list_invalid_cursor_returns_error() -> Result<()> {
             limit: Some(2),
             sort_key: None,
             model_providers: Some(vec!["mock_provider".to_string()]),
+            source_kinds: None,
             archived: None,
         })
         .await?;

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

@@ -2,7 +2,9 @@ use anyhow::Result;
 use app_test_support::McpProcess;
 use app_test_support::create_fake_rollout_with_text_elements;
 use app_test_support::create_mock_responses_server_repeating_assistant;
+use app_test_support::rollout_path;
 use app_test_support::to_response;
+use chrono::Utc;
 use codex_app_server_protocol::JSONRPCResponse;
 use codex_app_server_protocol::RequestId;
 use codex_app_server_protocol::SessionSource;
@@ -22,6 +24,8 @@ use codex_protocol::user_input::TextElement;
 use core_test_support::responses;
 use core_test_support::skip_if_no_network;
 use pretty_assertions::assert_eq;
+use std::fs::FileTimes;
+use std::path::Path;
 use std::path::PathBuf;
 use tempfile::TempDir;
 use tokio::time::timeout;
@@ -147,6 +151,116 @@ async fn thread_resume_returns_rollout_history() -> Result<()> {
     Ok(())
 }
 
+#[tokio::test]
+async fn thread_resume_without_overrides_does_not_change_updated_at_or_mtime() -> Result<()> {
+    let server = create_mock_responses_server_repeating_assistant("Done").await;
+    let codex_home = TempDir::new()?;
+    let rollout = setup_rollout_fixture(codex_home.path(), &server.uri())?;
+    let thread_id = rollout.conversation_id.clone();
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let resume_id = mcp
+        .send_thread_resume_request(ThreadResumeParams {
+            thread_id: thread_id.clone(),
+            ..Default::default()
+        })
+        .await?;
+    let resume_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
+    )
+    .await??;
+    let ThreadResumeResponse { thread, .. } = to_response::<ThreadResumeResponse>(resume_resp)?;
+
+    assert_eq!(thread.updated_at, rollout.expected_updated_at);
+
+    let after_modified = std::fs::metadata(&rollout.rollout_file_path)?.modified()?;
+    assert_eq!(after_modified, rollout.before_modified);
+
+    let turn_id = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id,
+            input: vec![UserInput::Text {
+                text: "Hello".to_string(),
+                text_elements: Vec::new(),
+            }],
+            ..Default::default()
+        })
+        .await?;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_id)),
+    )
+    .await??;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/completed"),
+    )
+    .await??;
+
+    let after_turn_modified = std::fs::metadata(&rollout.rollout_file_path)?.modified()?;
+    assert!(after_turn_modified > rollout.before_modified);
+
+    Ok(())
+}
+
+#[tokio::test]
+async fn thread_resume_with_overrides_defers_updated_at_until_turn_start() -> Result<()> {
+    let server = create_mock_responses_server_repeating_assistant("Done").await;
+    let codex_home = TempDir::new()?;
+    let rollout = setup_rollout_fixture(codex_home.path(), &server.uri())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let resume_id = mcp
+        .send_thread_resume_request(ThreadResumeParams {
+            thread_id: rollout.conversation_id.clone(),
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let resume_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(resume_id)),
+    )
+    .await??;
+    let ThreadResumeResponse { thread, .. } = to_response::<ThreadResumeResponse>(resume_resp)?;
+
+    assert_eq!(thread.updated_at, rollout.expected_updated_at);
+
+    let after_resume_modified = std::fs::metadata(&rollout.rollout_file_path)?.modified()?;
+    assert_eq!(after_resume_modified, rollout.before_modified);
+
+    let turn_id = mcp
+        .send_turn_start_request(TurnStartParams {
+            thread_id: rollout.conversation_id,
+            input: vec![UserInput::Text {
+                text: "Hello".to_string(),
+                text_elements: Vec::new(),
+            }],
+            ..Default::default()
+        })
+        .await?;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(turn_id)),
+    )
+    .await??;
+    timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_notification_message("turn/completed"),
+    )
+    .await??;
+
+    let after_turn_modified = std::fs::metadata(&rollout.rollout_file_path)?.modified()?;
+    assert!(after_turn_modified > rollout.before_modified);
+
+    Ok(())
+}
+
 #[tokio::test]
 async fn thread_resume_prefers_path_over_thread_id() -> Result<()> {
     let server = create_mock_responses_server_repeating_assistant("Done").await;
@@ -364,3 +478,51 @@ stream_max_retries = 0
         ),
     )
 }
+
+fn set_rollout_mtime(path: &Path, updated_at_rfc3339: &str) -> Result<()> {
+    let parsed = chrono::DateTime::parse_from_rfc3339(updated_at_rfc3339)?.with_timezone(&Utc);
+    let times = FileTimes::new().set_modified(parsed.into());
+    std::fs::OpenOptions::new()
+        .append(true)
+        .open(path)?
+        .set_times(times)?;
+    Ok(())
+}
+
+struct RolloutFixture {
+    conversation_id: String,
+    rollout_file_path: PathBuf,
+    before_modified: std::time::SystemTime,
+    expected_updated_at: i64,
+}
+
+fn setup_rollout_fixture(codex_home: &Path, server_uri: &str) -> Result<RolloutFixture> {
+    create_config_toml(codex_home, server_uri)?;
+
+    let preview = "Saved user message";
+    let filename_ts = "2025-01-05T12-00-00";
+    let meta_rfc3339 = "2025-01-05T12:00:00Z";
+    let expected_updated_at_rfc3339 = "2025-01-07T00:00:00Z";
+    let conversation_id = create_fake_rollout_with_text_elements(
+        codex_home,
+        filename_ts,
+        meta_rfc3339,
+        preview,
+        Vec::new(),
+        Some("mock_provider"),
+        None,
+    )?;
+    let rollout_file_path = rollout_path(codex_home, filename_ts, &conversation_id);
+    set_rollout_mtime(rollout_file_path.as_path(), expected_updated_at_rfc3339)?;
+    let before_modified = std::fs::metadata(&rollout_file_path)?.modified()?;
+    let expected_updated_at = chrono::DateTime::parse_from_rfc3339(expected_updated_at_rfc3339)?
+        .with_timezone(&Utc)
+        .timestamp();
+
+    Ok(RolloutFixture {
+        conversation_id,
+        rollout_file_path,
+        before_modified,
+        expected_updated_at,
+    })
+}

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

@@ -0,0 +1,101 @@
+use anyhow::Result;
+use app_test_support::McpProcess;
+use app_test_support::to_response;
+use codex_app_server_protocol::JSONRPCResponse;
+use codex_app_server_protocol::RequestId;
+use codex_app_server_protocol::ThreadArchiveParams;
+use codex_app_server_protocol::ThreadArchiveResponse;
+use codex_app_server_protocol::ThreadStartParams;
+use codex_app_server_protocol::ThreadStartResponse;
+use codex_app_server_protocol::ThreadUnarchiveParams;
+use codex_app_server_protocol::ThreadUnarchiveResponse;
+use codex_core::find_archived_thread_path_by_id_str;
+use codex_core::find_thread_path_by_id_str;
+use std::path::Path;
+use tempfile::TempDir;
+use tokio::time::timeout;
+
+const DEFAULT_READ_TIMEOUT: std::time::Duration = std::time::Duration::from_secs(30);
+
+#[tokio::test]
+async fn thread_unarchive_moves_rollout_back_into_sessions_directory() -> Result<()> {
+    let codex_home = TempDir::new()?;
+    create_config_toml(codex_home.path())?;
+
+    let mut mcp = McpProcess::new(codex_home.path()).await?;
+    timeout(DEFAULT_READ_TIMEOUT, mcp.initialize()).await??;
+
+    let start_id = mcp
+        .send_thread_start_request(ThreadStartParams {
+            model: Some("mock-model".to_string()),
+            ..Default::default()
+        })
+        .await?;
+    let start_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(start_id)),
+    )
+    .await??;
+    let ThreadStartResponse { thread, .. } = to_response::<ThreadStartResponse>(start_resp)?;
+
+    let rollout_path = find_thread_path_by_id_str(codex_home.path(), &thread.id)
+        .await?
+        .expect("expected rollout path for thread id to exist");
+
+    let archive_id = mcp
+        .send_thread_archive_request(ThreadArchiveParams {
+            thread_id: thread.id.clone(),
+        })
+        .await?;
+    let archive_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(archive_id)),
+    )
+    .await??;
+    let _: ThreadArchiveResponse = to_response::<ThreadArchiveResponse>(archive_resp)?;
+
+    let archived_path = find_archived_thread_path_by_id_str(codex_home.path(), &thread.id)
+        .await?
+        .expect("expected archived rollout path for thread id to exist");
+    let archived_path_display = archived_path.display();
+    assert!(
+        archived_path.exists(),
+        "expected {archived_path_display} to exist"
+    );
+
+    let unarchive_id = mcp
+        .send_thread_unarchive_request(ThreadUnarchiveParams {
+            thread_id: thread.id.clone(),
+        })
+        .await?;
+    let unarchive_resp: JSONRPCResponse = timeout(
+        DEFAULT_READ_TIMEOUT,
+        mcp.read_stream_until_response_message(RequestId::Integer(unarchive_id)),
+    )
+    .await??;
+    let _: ThreadUnarchiveResponse = to_response::<ThreadUnarchiveResponse>(unarchive_resp)?;
+
+    let rollout_path_display = rollout_path.display();
+    assert!(
+        rollout_path.exists(),
+        "expected rollout path {rollout_path_display} to be restored"
+    );
+    assert!(
+        !archived_path.exists(),
+        "expected archived rollout path {archived_path_display} to be moved"
+    );
+
+    Ok(())
+}
+
+fn create_config_toml(codex_home: &Path) -> std::io::Result<()> {
+    let config_toml = codex_home.join("config.toml");
+    std::fs::write(config_toml, config_contents())
+}
+
+fn config_contents() -> &'static str {
+    r#"model = "mock-model"
+approval_policy = "never"
+sandbox_mode = "read-only"
+"#
+}

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

@@ -433,7 +433,7 @@ async fn turn_start_accepts_personality_override_v2() -> Result<()> {
 
     let thread_req = mcp
         .send_thread_start_request(ThreadStartParams {
-            model: Some("gpt-5.2-codex".to_string()),
+            model: Some("exp-codex-personality".to_string()),
             ..Default::default()
         })
         .await?;

codex-rs/backend-client/src/client.rs

@@ -1,4 +1,5 @@
 use crate::types::CodeTaskDetailsResponse;
+use crate::types::ConfigFileResponse;
 use crate::types::CreditStatusDetails;
 use crate::types::PaginatedListTaskListItem;
 use crate::types::RateLimitStatusPayload;
@@ -244,6 +245,20 @@ impl Client {
         self.decode_json::<TurnAttemptsSiblingTurnsResponse>(&url, &ct, &body)
     }
 
+    /// Fetch the managed requirements file from codex-backend.
+    ///
+    /// `GET /api/codex/config/requirements` (Codex API style) or
+    /// `GET /wham/config/requirements` (ChatGPT backend-api style).
+    pub async fn get_config_requirements_file(&self) -> Result<ConfigFileResponse> {
+        let url = match self.path_style {
+            PathStyle::CodexApi => format!("{}/api/codex/config/requirements", self.base_url),
+            PathStyle::ChatGptApi => format!("{}/wham/config/requirements", self.base_url),
+        };
+        let req = self.http.get(&url).headers(self.headers());
+        let (body, ct) = self.exec_request(req, "GET", &url).await?;
+        self.decode_json::<ConfigFileResponse>(&url, &ct, &body)
+    }
+
     /// Create a new task (user turn) by POSTing to the appropriate backend path
     /// based on `path_style`. Returns the created task id.
     pub async fn create_task(&self, request_body: serde_json::Value) -> Result<String> {

codex-rs/backend-client/src/lib.rs

@@ -4,6 +4,7 @@ pub mod types;
 pub use client::Client;
 pub use types::CodeTaskDetailsResponse;
 pub use types::CodeTaskDetailsResponseExt;
+pub use types::ConfigFileResponse;
 pub use types::PaginatedListTaskListItem;
 pub use types::TaskListItem;
 pub use types::TurnAttemptsSiblingTurnsResponse;

codex-rs/backend-client/src/types.rs

@@ -1,3 +1,4 @@
+pub use codex_backend_openapi_models::models::ConfigFileResponse;
 pub use codex_backend_openapi_models::models::CreditStatusDetails;
 pub use codex_backend_openapi_models::models::PaginatedListTaskListItem;
 pub use codex_backend_openapi_models::models::PlanType;

codex-rs/cli/src/main.rs

@@ -147,7 +147,7 @@ struct ResumeCommand {
     session_id: Option<String>,
 
     /// Continue the most recent session without showing the picker.
-    #[arg(long = "last", default_value_t = false, conflicts_with = "session_id")]
+    #[arg(long = "last", default_value_t = false)]
     last: bool,
 
     /// Show all sessions (disables cwd filtering and shows CWD column).
@@ -453,8 +453,8 @@ enum FeaturesSubcommand {
 fn stage_str(stage: codex_core::features::Stage) -> &'static str {
     use codex_core::features::Stage;
     match stage {
-        Stage::Beta => "experimental",
-        Stage::Experimental { .. } => "beta",
+        Stage::UnderDevelopment => "under development",
+        Stage::Experimental { .. } => "experimental",
         Stage::Stable => "stable",
         Stage::Deprecated => "deprecated",
         Stage::Removed => "removed",
@@ -932,6 +932,24 @@ mod tests {
         finalize_fork_interactive(interactive, root_overrides, session_id, last, all, fork_cli)
     }
 
+    #[test]
+    fn exec_resume_last_accepts_prompt_positional() {
+        let cli =
+            MultitoolCli::try_parse_from(["codex", "exec", "--json", "resume", "--last", "2+2"])
+                .expect("parse should succeed");
+
+        let Some(Subcommand::Exec(exec)) = cli.subcommand else {
+            panic!("expected exec subcommand");
+        };
+        let Some(codex_exec::Command::Resume(args)) = exec.command else {
+            panic!("expected exec resume");
+        };
+
+        assert!(args.last);
+        assert_eq!(args.session_id, None);
+        assert_eq!(args.prompt.as_deref(), Some("2+2"));
+    }
+
     fn app_server_from_args(args: &[&str]) -> AppServerCommand {
         let cli = MultitoolCli::try_parse_from(args).expect("parse");
         let Subcommand::AppServer(app_server) = cli.subcommand.expect("app-server present") else {

codex-rs/cli/src/mcp_cmd.rs

@@ -247,6 +247,7 @@ async fn run_add(config_overrides: &CliConfigOverrides, add_args: AddArgs) -> Re
         tool_timeout_sec: None,
         enabled_tools: None,
         disabled_tools: None,
+        scopes: None,
     };
 
     servers.insert(name.clone(), new_entry);
@@ -348,6 +349,11 @@ async fn run_login(config_overrides: &CliConfigOverrides, login_args: LoginArgs)
         _ => bail!("OAuth login is only supported for streamable HTTP servers."),
     };
 
+    let mut scopes = scopes;
+    if scopes.is_empty() {
+        scopes = server.scopes.clone().unwrap_or_default();
+    }
+
     perform_oauth_login(
         &name,
         &url,

codex-rs/codex-api/src/sse/responses.rs

@@ -291,7 +291,7 @@ pub fn process_responses_event(
                 if let Ok(item) = serde_json::from_value::<ResponseItem>(item_val) {
                     return Ok(Some(ResponseEvent::OutputItemAdded(item)));
                 }
-                debug!("failed to parse ResponseItem from output_item.done");
+                debug!("failed to parse ResponseItem from output_item.added");
             }
         }
         "response.reasoning_summary_part.added" => {

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

@@ -0,0 +1,40 @@
+/*
+ * codex-backend
+ *
+ * codex-backend
+ *
+ * The version of the OpenAPI document: 0.0.1
+ *
+ * Generated by: https://openapi-generator.tech
+ */
+
+use serde::Deserialize;
+use serde::Serialize;
+
+#[derive(Clone, Default, Debug, PartialEq, Serialize, Deserialize)]
+pub struct ConfigFileResponse {
+    #[serde(rename = "contents", skip_serializing_if = "Option::is_none")]
+    pub contents: Option<String>,
+    #[serde(rename = "sha256", skip_serializing_if = "Option::is_none")]
+    pub sha256: Option<String>,
+    #[serde(rename = "updated_at", skip_serializing_if = "Option::is_none")]
+    pub updated_at: Option<String>,
+    #[serde(rename = "updated_by_user_id", skip_serializing_if = "Option::is_none")]
+    pub updated_by_user_id: Option<String>,
+}
+
+impl ConfigFileResponse {
+    pub fn new(
+        contents: Option<String>,
+        sha256: Option<String>,
+        updated_at: Option<String>,
+        updated_by_user_id: Option<String>,
+    ) -> ConfigFileResponse {
+        ConfigFileResponse {
+            contents,
+            sha256,
+            updated_at,
+            updated_by_user_id,
+        }
+    }
+}

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

@@ -3,6 +3,10 @@
 // Currently export only the types referenced by the workspace
 // The process for this will change
 
+// Config
+pub mod config_file_response;
+pub use self::config_file_response::ConfigFileResponse;
+
 // Cloud Tasks
 pub mod code_task_details_response;
 pub use self::code_task_details_response::CodeTaskDetailsResponse;

codex-rs/core/config.schema.json

@@ -750,6 +750,13 @@
           },
           "type": "object"
         },
+        "scopes": {
+          "default": null,
+          "items": {
+            "type": "string"
+          },
+          "type": "array"
+        },
         "startup_timeout_ms": {
           "default": null,
           "format": "uint64",
@@ -1458,6 +1465,10 @@
       ],
       "description": "User-level skill config entries keyed by SKILL.md path."
     },
+    "suppress_unstable_features_warning": {
+      "description": "Suppress warnings about unstable (under development) features.",
+      "type": "boolean"
+    },
     "tool_output_token_limit": {
       "description": "Token budget applied when storing tool/function outputs in the context manager.",
       "format": "uint",

codex-rs/core/src/agent/role.rs

@@ -1,20 +1,22 @@
 use crate::config::Config;
 use crate::protocol::SandboxPolicy;
+use codex_protocol::openai_models::ReasoningEffort;
 use serde::Deserialize;
 use serde::Serialize;
 
 /// Base instructions for the orchestrator role.
 const ORCHESTRATOR_PROMPT: &str = include_str!("../../templates/agents/orchestrator.md");
-/// Base instructions for the worker role.
-const WORKER_PROMPT: &str = include_str!("../../gpt-5.2-codex_prompt.md");
-/// Default worker model override used by the worker role.
-const WORKER_MODEL: &str = "gpt-5.2-codex";
+/// Default model override used.
+// TODO(jif) update when we have something smarter.
+const EXPLORER_MODEL: &str = "gpt-5.2-codex";
 
 /// Enumerated list of all supported agent roles.
 const ALL_ROLES: [AgentRole; 3] = [
     AgentRole::Default,
-    AgentRole::Orchestrator,
+    AgentRole::Explorer,
     AgentRole::Worker,
+    // TODO(jif) add when we have stable prompts + models
+    // AgentRole::Orchestrator,
 ];
 
 /// Hard-coded agent role selection used when spawning sub-agents.
@@ -27,6 +29,8 @@ pub enum AgentRole {
     Orchestrator,
     /// Task-executing agent with a fixed model override.
     Worker,
+    /// Task-executing agent with a fixed model override.
+    Explorer,
 }
 
 /// Immutable profile data that drives per-agent configuration overrides.
@@ -36,8 +40,12 @@ pub struct AgentProfile {
     pub base_instructions: Option<&'static str>,
     /// Optional model override.
     pub model: Option<&'static str>,
+    /// Optional reasoning effort override.
+    pub reasoning_effort: Option<ReasoningEffort>,
     /// Whether to force a read-only sandbox policy.
     pub read_only: bool,
+    /// Description to include in the tool specs.
+    pub description: &'static str,
 }
 
 impl AgentRole {
@@ -45,7 +53,19 @@ impl AgentRole {
     pub fn enum_values() -> Vec<String> {
         ALL_ROLES
             .iter()
-            .filter_map(|role| serde_json::to_string(role).ok())
+            .filter_map(|role| {
+                let description = role.profile().description;
+                serde_json::to_string(role)
+                    .map(|role| {
+                        let description = if !description.is_empty() {
+                            format!(r#", "description": {description}"#)
+                        } else {
+                            String::new()
+                        };
+                        format!(r#"{{ "name": {role}{description}}}"#)
+                    })
+                    .ok()
+            })
             .collect()
     }
 
@@ -58,8 +78,35 @@ impl AgentRole {
                 ..Default::default()
             },
             AgentRole::Worker => AgentProfile {
-                base_instructions: Some(WORKER_PROMPT),
-                model: Some(WORKER_MODEL),
+                // base_instructions: Some(WORKER_PROMPT),
+                // model: Some(WORKER_MODEL),
+                description: r#"Use for execution and production work.
+Typical tasks:
+- Implement part of a feature
+- Fix tests or bugs
+- Split large refactors into independent chunks
+Rules:
+- Explicitly assign **ownership** of the task (files / responsibility).
+- Always tell workers they are **not alone in the codebase**, and they should ignore edits made by others without touching them"#,
+                ..Default::default()
+            },
+            AgentRole::Explorer => AgentProfile {
+                model: Some(EXPLORER_MODEL),
+                reasoning_effort: Some(ReasoningEffort::Low),
+                description: r#"Use for fast codebase understanding and information gathering.
+`explorer` are extremely fast agents so use them as much as you can to speed up the resolution of the global task.
+Typical tasks:
+- Locate usages of a symbol or concept
+- Understand how X is handled in Y
+- Review a section of code for issues
+- Assess impact of a potential change
+Rules:
+- Be explicit in what you are looking for. A good usage of `explorer` would mean that don't need to read the same code after the explorer send you the result.
+- **Always** prefer asking explorers rather than exploring the codebase yourself.
+- Spawn multiple explorers in parallel when useful and wait for all results.
+- You can ask the `explorer` to return file name, lines, entire code snippets, ...
+- Reuse the same explorer when it is relevant. If later in your process you have more questions on some code an explorer already covered, reuse this same explorer to be more efficient.
+                "#,
                 ..Default::default()
             },
         }
@@ -74,6 +121,9 @@ impl AgentRole {
         if let Some(model) = profile.model {
             config.model = Some(model.to_string());
         }
+        if let Some(reasoning_effort) = profile.reasoning_effort {
+            config.model_reasoning_effort = Some(reasoning_effort)
+        }
         if profile.read_only {
             config
                 .sandbox_policy

codex-rs/core/src/apply_patch.rs

@@ -42,6 +42,7 @@ pub(crate) async fn apply_patch(
         turn_context.approval_policy,
         &turn_context.sandbox_policy,
         &turn_context.cwd,
+        turn_context.windows_sandbox_level,
     ) {
         SafetyCheck::AutoApprove {
             user_explicitly_approved,

codex-rs/core/src/client.rs

@@ -625,11 +625,13 @@ fn build_api_prompt(prompt: &Prompt, instructions: String, tools_json: Vec<Value
     }
 }
 
-fn beta_feature_headers(config: &Config) -> ApiHeaderMap {
+fn experimental_feature_headers(config: &Config) -> ApiHeaderMap {
     let enabled = FEATURES
         .iter()
         .filter_map(|spec| {
-            if spec.stage.beta_menu_description().is_some() && config.features.enabled(spec.id) {
+            if spec.stage.experimental_menu_description().is_some()
+                && config.features.enabled(spec.id)
+            {
                 Some(spec.key)
             } else {
                 None
@@ -650,16 +652,14 @@ fn build_responses_headers(
     config: &Config,
     turn_state: Option<&Arc<OnceLock<String>>>,
 ) -> ApiHeaderMap {
-    let mut headers = beta_feature_headers(config);
+    let mut headers = experimental_feature_headers(config);
     headers.insert(
         WEB_SEARCH_ELIGIBLE_HEADER,
-        HeaderValue::from_static(
-            if matches!(config.web_search_mode, Some(WebSearchMode::Disabled)) {
-                "false"
-            } else {
-                "true"
-            },
-        ),
+        HeaderValue::from_static(if config.web_search_mode == WebSearchMode::Disabled {
+            "false"
+        } else {
+            "true"
+        }),
     );
     if let Some(turn_state) = turn_state
         && let Some(state) = turn_state.get()

codex-rs/core/src/codex.rs

@@ -12,6 +12,7 @@ use crate::CodexAuth;
 use crate::SandboxState;
 use crate::agent::AgentControl;
 use crate::agent::AgentStatus;
+use crate::agent::MAX_THREAD_SPAWN_DEPTH;
 use crate::agent::agent_status_from_event;
 use crate::compact;
 use crate::compact::run_inline_auto_compact_task;
@@ -21,6 +22,7 @@ use crate::connectors;
 use crate::exec_policy::ExecPolicyManager;
 use crate::features::Feature;
 use crate::features::Features;
+use crate::features::maybe_push_unstable_features_warning;
 use crate::models_manager::manager::ModelsManager;
 use crate::parse_command::parse_command;
 use crate::parse_turn_item;
@@ -38,6 +40,8 @@ use codex_protocol::approvals::ExecPolicyAmendment;
 use codex_protocol::config_types::ModeKind;
 use codex_protocol::config_types::Settings;
 use codex_protocol::config_types::WebSearchMode;
+use codex_protocol::dynamic_tools::DynamicToolResponse;
+use codex_protocol::dynamic_tools::DynamicToolSpec;
 use codex_protocol::items::TurnItem;
 use codex_protocol::items::UserMessageItem;
 use codex_protocol::models::BaseInstructions;
@@ -50,6 +54,7 @@ use codex_protocol::protocol::RawResponseItemEvent;
 use codex_protocol::protocol::ReviewRequest;
 use codex_protocol::protocol::RolloutItem;
 use codex_protocol::protocol::SessionSource;
+use codex_protocol::protocol::SubAgentSource;
 use codex_protocol::protocol::TurnAbortReason;
 use codex_protocol::protocol::TurnContextItem;
 use codex_protocol::protocol::TurnStartedEvent;
@@ -169,11 +174,13 @@ use crate::turn_diff_tracker::TurnDiffTracker;
 use crate::unified_exec::UnifiedExecProcessManager;
 use crate::user_notification::UserNotification;
 use crate::util::backoff;
+use crate::windows_sandbox::WindowsSandboxLevelExt;
 use codex_async_utils::OrCancelExt;
 use codex_otel::OtelManager;
 use codex_protocol::config_types::CollaborationMode;
 use codex_protocol::config_types::Personality;
 use codex_protocol::config_types::ReasoningSummary as ReasoningSummaryConfig;
+use codex_protocol::config_types::WindowsSandboxLevel;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::DeveloperInstructions;
 use codex_protocol::models::ResponseInputItem;
@@ -236,24 +243,21 @@ fn maybe_push_chat_wire_api_deprecation(
 
 impl Codex {
     /// Spawn a new [`Codex`] and initialize the session.
+    #[allow(clippy::too_many_arguments)]
     pub(crate) async fn spawn(
-        config: Config,
+        mut config: Config,
         auth_manager: Arc<AuthManager>,
         models_manager: Arc<ModelsManager>,
         skills_manager: Arc<SkillsManager>,
         conversation_history: InitialHistory,
         session_source: SessionSource,
         agent_control: AgentControl,
+        dynamic_tools: Vec<DynamicToolSpec>,
     ) -> CodexResult<CodexSpawnOk> {
         let (tx_sub, rx_sub) = async_channel::bounded(SUBMISSION_CHANNEL_CAPACITY);
         let (tx_event, rx_event) = async_channel::unbounded();
 
         let loaded_skills = skills_manager.skills_for_config(&config);
-        // let loaded_skills = if config.features.enabled(Feature::Skills) {
-        //     Some(skills_manager.skills_for_config(&config))
-        // } else {
-        //     None
-        // };
 
         for err in &loaded_skills.errors {
             error!(
@@ -263,6 +267,12 @@ impl Codex {
             );
         }
 
+        if let SessionSource::SubAgent(SubAgentSource::ThreadSpawn { depth, .. }) = session_source
+            && depth >= MAX_THREAD_SPAWN_DEPTH
+        {
+            config.features.disable(Feature::Collab);
+        }
+
         let enabled_skills = loaded_skills.enabled_skills();
         let user_instructions = get_user_instructions(&config, Some(&enabled_skills)).await;
 
@@ -317,9 +327,11 @@ impl Codex {
             compact_prompt: config.compact_prompt.clone(),
             approval_policy: config.approval_policy.clone(),
             sandbox_policy: config.sandbox_policy.clone(),
+            windows_sandbox_level: WindowsSandboxLevel::from_config(&config),
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             session_source,
+            dynamic_tools,
         };
 
         // Generate a unique ID for the lifetime of this Codex session.
@@ -436,6 +448,7 @@ pub(crate) struct TurnContext {
     pub(crate) personality: Option<Personality>,
     pub(crate) approval_policy: AskForApproval,
     pub(crate) sandbox_policy: SandboxPolicy,
+    pub(crate) windows_sandbox_level: WindowsSandboxLevel,
     pub(crate) shell_environment_policy: ShellEnvironmentPolicy,
     pub(crate) tools_config: ToolsConfig,
     pub(crate) ghost_snapshot: GhostSnapshotConfig,
@@ -443,6 +456,7 @@ pub(crate) struct TurnContext {
     pub(crate) codex_linux_sandbox_exe: Option<PathBuf>,
     pub(crate) tool_call_gate: Arc<ReadinessFlag>,
     pub(crate) truncation_policy: TruncationPolicy,
+    pub(crate) dynamic_tools: Vec<DynamicToolSpec>,
 }
 
 impl TurnContext {
@@ -486,6 +500,7 @@ pub(crate) struct SessionConfiguration {
     approval_policy: Constrained<AskForApproval>,
     /// How to sandbox commands executed in the system
     sandbox_policy: Constrained<SandboxPolicy>,
+    windows_sandbox_level: WindowsSandboxLevel,
 
     /// Working directory that should be treated as the *root* of the
     /// session. All relative paths supplied by the model as well as the
@@ -500,6 +515,7 @@ pub(crate) struct SessionConfiguration {
     original_config_do_not_use: Arc<Config>,
     /// Source of the session (cli, vscode, exec, mcp, ...)
     session_source: SessionSource,
+    dynamic_tools: Vec<DynamicToolSpec>,
 }
 
 impl SessionConfiguration {
@@ -533,6 +549,9 @@ impl SessionConfiguration {
         if let Some(sandbox_policy) = updates.sandbox_policy.clone() {
             next_configuration.sandbox_policy.set(sandbox_policy)?;
         }
+        if let Some(windows_sandbox_level) = updates.windows_sandbox_level {
+            next_configuration.windows_sandbox_level = windows_sandbox_level;
+        }
         if let Some(cwd) = updates.cwd.clone() {
             next_configuration.cwd = cwd;
         }
@@ -545,6 +564,7 @@ pub(crate) struct SessionSettingsUpdate {
     pub(crate) cwd: Option<PathBuf>,
     pub(crate) approval_policy: Option<AskForApproval>,
     pub(crate) sandbox_policy: Option<SandboxPolicy>,
+    pub(crate) windows_sandbox_level: Option<WindowsSandboxLevel>,
     pub(crate) collaboration_mode: Option<CollaborationMode>,
     pub(crate) reasoning_summary: Option<ReasoningSummaryConfig>,
     pub(crate) final_output_json_schema: Option<Option<Value>>,
@@ -609,6 +629,7 @@ impl Session {
             personality: session_configuration.personality,
             approval_policy: session_configuration.approval_policy.value(),
             sandbox_policy: session_configuration.sandbox_policy.get().clone(),
+            windows_sandbox_level: session_configuration.windows_sandbox_level,
             shell_environment_policy: per_turn_config.shell_environment_policy.clone(),
             tools_config,
             ghost_snapshot: per_turn_config.ghost_snapshot.clone(),
@@ -616,6 +637,7 @@ impl Session {
             codex_linux_sandbox_exe: per_turn_config.codex_linux_sandbox_exe.clone(),
             tool_call_gate: Arc::new(ReadinessFlag::new()),
             truncation_policy: model_info.truncation_policy.into(),
+            dynamic_tools: session_configuration.dynamic_tools.clone(),
         }
     }
 
@@ -743,6 +765,7 @@ impl Session {
             });
         }
         maybe_push_chat_wire_api_deprecation(&config, &mut post_session_configured_events);
+        maybe_push_unstable_features_warning(&config, &mut post_session_configured_events);
 
         let auth = auth.as_ref();
         let otel_manager = OtelManager::new(
@@ -920,23 +943,28 @@ impl Session {
                 // Build and record initial items (user instructions + environment context)
                 let items = self.build_initial_context(&turn_context).await;
                 self.record_conversation_items(&turn_context, &items).await;
+                {
+                    let mut state = self.state.lock().await;
+                    state.initial_context_seeded = true;
+                }
                 // Ensure initial items are visible to immediate readers (e.g., tests, forks).
                 self.flush_rollout().await;
             }
-            InitialHistory::Resumed(_) | InitialHistory::Forked(_) => {
-                let rollout_items = conversation_history.get_rollout_items();
-                let persist = matches!(conversation_history, InitialHistory::Forked(_));
+            InitialHistory::Resumed(resumed_history) => {
+                let rollout_items = resumed_history.history;
+                {
+                    let mut state = self.state.lock().await;
+                    state.initial_context_seeded = false;
+                }
 
                 // If resuming, warn when the last recorded model differs from the current one.
-                if let InitialHistory::Resumed(_) = conversation_history
-                    && let Some(prev) = rollout_items.iter().rev().find_map(|it| {
-                        if let RolloutItem::TurnContext(ctx) = it {
-                            Some(ctx.model.as_str())
-                        } else {
-                            None
-                        }
-                    })
-                {
+                if let Some(prev) = rollout_items.iter().rev().find_map(|it| {
+                    if let RolloutItem::TurnContext(ctx) = it {
+                        Some(ctx.model.as_str())
+                    } else {
+                        None
+                    }
+                }) {
                     let curr = turn_context.client.get_model();
                     if prev != curr {
                         warn!(
@@ -971,8 +999,29 @@ impl Session {
                     state.set_token_info(Some(info));
                 }
 
+                // Defer seeding the session's initial context until the first turn starts so
+                // turn/start overrides can be merged before we write to the rollout.
+                self.flush_rollout().await;
+            }
+            InitialHistory::Forked(rollout_items) => {
+                // Always add response items to conversation history
+                let reconstructed_history = self
+                    .reconstruct_history_from_rollout(&turn_context, &rollout_items)
+                    .await;
+                if !reconstructed_history.is_empty() {
+                    self.record_into_history(&reconstructed_history, &turn_context)
+                        .await;
+                }
+
+                // Seed usage info from the recorded rollout so UIs can show token counts
+                // immediately on resume/fork.
+                if let Some(info) = Self::last_token_info_from_rollout(&rollout_items) {
+                    let mut state = self.state.lock().await;
+                    state.set_token_info(Some(info));
+                }
+
                 // If persisting, persist all rollout items as-is (recorder filters)
-                if persist && !rollout_items.is_empty() {
+                if !rollout_items.is_empty() {
                     self.persist_rollout_items(&rollout_items).await;
                 }
 
@@ -980,6 +1029,10 @@ impl Session {
                 let initial_context = self.build_initial_context(&turn_context).await;
                 self.record_conversation_items(&turn_context, &initial_context)
                     .await;
+                {
+                    let mut state = self.state.lock().await;
+                    state.initial_context_seeded = true;
+                }
                 // Flush after seeding history and any persisted rollout copy.
                 self.flush_rollout().await;
             }
@@ -1495,6 +1548,27 @@ impl Session {
         }
     }
 
+    pub async fn notify_dynamic_tool_response(&self, call_id: &str, response: DynamicToolResponse) {
+        let entry = {
+            let mut active = self.active_turn.lock().await;
+            match active.as_mut() {
+                Some(at) => {
+                    let mut ts = at.turn_state.lock().await;
+                    ts.remove_pending_dynamic_tool(call_id)
+                }
+                None => None,
+            }
+        };
+        match entry {
+            Some(tx_response) => {
+                tx_response.send(response).ok();
+            }
+            None => {
+                warn!("No pending dynamic tool call found for call_id: {call_id}");
+            }
+        }
+    }
+
     pub async fn notify_approval(&self, sub_id: &str, decision: ReviewDecision) {
         let entry = {
             let mut active = self.active_turn.lock().await;
@@ -1609,6 +1683,21 @@ impl Session {
         state.replace_history(items);
     }
 
+    pub(crate) async fn seed_initial_context_if_needed(&self, turn_context: &TurnContext) {
+        {
+            let mut state = self.state.lock().await;
+            if state.initial_context_seeded {
+                return;
+            }
+            state.initial_context_seeded = true;
+        }
+
+        let initial_context = self.build_initial_context(turn_context).await;
+        self.record_conversation_items(turn_context, &initial_context)
+            .await;
+        self.flush_rollout().await;
+    }
+
     async fn persist_rollout_response_items(&self, items: &[ResponseItem]) {
         let rollout_items: Vec<RolloutItem> = items
             .iter()
@@ -2112,6 +2201,7 @@ async fn submission_loop(sess: Arc<Session>, config: Arc<Config>, rx_sub: Receiv
                 cwd,
                 approval_policy,
                 sandbox_policy,
+                windows_sandbox_level,
                 model,
                 effort,
                 summary,
@@ -2135,6 +2225,7 @@ async fn submission_loop(sess: Arc<Session>, config: Arc<Config>, rx_sub: Receiv
                         cwd,
                         approval_policy,
                         sandbox_policy,
+                        windows_sandbox_level,
                         collaboration_mode: Some(collaboration_mode),
                         reasoning_summary: summary,
                         personality,
@@ -2156,6 +2247,9 @@ async fn submission_loop(sess: Arc<Session>, config: Arc<Config>, rx_sub: Receiv
             Op::UserInputAnswer { id, response } => {
                 handlers::request_user_input_response(&sess, id, response).await;
             }
+            Op::DynamicToolResponse { id, response } => {
+                handlers::dynamic_tool_response(&sess, id, response).await;
+            }
             Op::AddToHistory { text } => {
                 handlers::add_to_history(&sess, &config, text).await;
             }
@@ -2252,6 +2346,7 @@ mod handlers {
     use codex_protocol::config_types::CollaborationMode;
     use codex_protocol::config_types::ModeKind;
     use codex_protocol::config_types::Settings;
+    use codex_protocol::dynamic_tools::DynamicToolResponse;
     use codex_protocol::user_input::UserInput;
     use codex_rmcp_client::ElicitationAction;
     use codex_rmcp_client::ElicitationResponse;
@@ -2294,6 +2389,11 @@ mod handlers {
             return;
         }
 
+        let initial_context_seeded = sess.state.lock().await.initial_context_seeded;
+        if !initial_context_seeded {
+            return;
+        }
+
         let current_context = sess.new_default_turn_with_sub_id(sub_id).await;
         let update_items = sess.build_settings_update_items(
             Some(&previous_context),
@@ -2342,6 +2442,7 @@ mod handlers {
                         cwd: Some(cwd),
                         approval_policy: Some(approval_policy),
                         sandbox_policy: Some(sandbox_policy),
+                        windows_sandbox_level: None,
                         collaboration_mode,
                         reasoning_summary: Some(summary),
                         final_output_json_schema: Some(final_output_json_schema),
@@ -2381,6 +2482,7 @@ mod handlers {
 
         // Attempt to inject input into current task
         if let Err(items) = sess.inject_input(items).await {
+            sess.seed_initial_context_if_needed(&current_context).await;
             let update_items = sess.build_settings_update_items(
                 previous_context.as_ref(),
                 &current_context,
@@ -2489,6 +2591,14 @@ mod handlers {
         sess.notify_user_input_response(&id, response).await;
     }
 
+    pub async fn dynamic_tool_response(
+        sess: &Arc<Session>,
+        id: String,
+        response: DynamicToolResponse,
+    ) {
+        sess.notify_dynamic_tool_response(&id, response).await;
+    }
+
     pub async fn add_to_history(sess: &Arc<Session>, config: &Arc<Config>, text: String) {
         let id = sess.conversation_id;
         let config = Arc::clone(config);
@@ -2778,7 +2888,7 @@ async fn spawn_review_thread(
     let tools_config = ToolsConfig::new(&ToolsConfigParams {
         model_info: &review_model_info,
         features: &review_features,
-        web_search_mode: Some(review_web_search_mode),
+        web_search_mode: review_web_search_mode,
     });
 
     let review_prompt = resolved.prompt.clone();
@@ -2790,7 +2900,7 @@ async fn spawn_review_thread(
     let mut per_turn_config = (*config).clone();
     per_turn_config.model = Some(model.clone());
     per_turn_config.features = review_features.clone();
-    per_turn_config.web_search_mode = Some(review_web_search_mode);
+    per_turn_config.web_search_mode = review_web_search_mode;
 
     let otel_manager = parent_turn_context
         .client
@@ -2821,11 +2931,13 @@ async fn spawn_review_thread(
         personality: parent_turn_context.personality,
         approval_policy: parent_turn_context.approval_policy,
         sandbox_policy: parent_turn_context.sandbox_policy.clone(),
+        windows_sandbox_level: parent_turn_context.windows_sandbox_level,
         shell_environment_policy: parent_turn_context.shell_environment_policy.clone(),
         cwd: parent_turn_context.cwd.clone(),
         final_output_json_schema: None,
         codex_linux_sandbox_exe: parent_turn_context.codex_linux_sandbox_exe.clone(),
         tool_call_gate: Arc::new(ReadinessFlag::new()),
+        dynamic_tools: parent_turn_context.dynamic_tools.clone(),
         truncation_policy: model_info.truncation_policy.into(),
     };
 
@@ -3162,6 +3274,7 @@ async fn run_sampling_request(
                 .map(|(name, tool)| (name, tool.tool))
                 .collect(),
         ),
+        turn_context.dynamic_tools.as_slice(),
     ));
 
     let model_supports_parallel = turn_context
@@ -3391,10 +3504,8 @@ async fn try_run_sampling_request(
             }
             ResponseEvent::OutputItemAdded(item) => {
                 if let Some(turn_item) = handle_non_tool_response_item(&item).await {
-                    let tracked_item = turn_item.clone();
                     sess.emit_turn_item_started(&turn_context, &turn_item).await;
-
-                    active_item = Some(tracked_item);
+                    active_item = Some(turn_item);
                 }
             }
             ResponseEvent::ServerReasoningIncluded(included) => {
@@ -3549,6 +3660,7 @@ mod tests {
     use crate::shell::default_user_shell;
     use crate::tools::format_exec_output_str;
 
+    use codex_protocol::ThreadId;
     use codex_protocol::models::FunctionCallOutputPayload;
 
     use crate::protocol::CompactedItem;
@@ -3672,6 +3784,23 @@ mod tests {
 
     #[tokio::test]
     async fn record_initial_history_reconstructs_resumed_transcript() {
+        let (session, turn_context) = make_session_and_context().await;
+        let (rollout_items, expected) = sample_rollout(&session, &turn_context).await;
+
+        session
+            .record_initial_history(InitialHistory::Resumed(ResumedHistory {
+                conversation_id: ThreadId::default(),
+                history: rollout_items,
+                rollout_path: PathBuf::from("/tmp/resume.jsonl"),
+            }))
+            .await;
+
+        let history = session.state.lock().await.clone_history();
+        assert_eq!(expected, history.raw_items());
+    }
+
+    #[tokio::test]
+    async fn resumed_history_seeds_initial_context_on_first_turn_only() {
         let (session, turn_context) = make_session_and_context().await;
         let (rollout_items, mut expected) = sample_rollout(&session, &turn_context).await;
 
@@ -3683,9 +3812,17 @@ mod tests {
             }))
             .await;
 
+        let history_before_seed = session.state.lock().await.clone_history();
+        assert_eq!(expected, history_before_seed.raw_items());
+
+        session.seed_initial_context_if_needed(&turn_context).await;
         expected.extend(session.build_initial_context(&turn_context).await);
-        let history = session.state.lock().await.clone_history();
-        assert_eq!(expected, history.raw_items());
+        let history_after_seed = session.clone_history().await;
+        assert_eq!(expected, history_after_seed.raw_items());
+
+        session.seed_initial_context_if_needed(&turn_context).await;
+        let history_after_second_seed = session.clone_history().await;
+        assert_eq!(expected, history_after_second_seed.raw_items());
     }
 
     #[tokio::test]
@@ -3939,9 +4076,11 @@ mod tests {
             compact_prompt: config.compact_prompt.clone(),
             approval_policy: config.approval_policy.clone(),
             sandbox_policy: config.sandbox_policy.clone(),
+            windows_sandbox_level: WindowsSandboxLevel::from_config(&config),
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             session_source: SessionSource::Exec,
+            dynamic_tools: Vec::new(),
         };
 
         let mut state = SessionState::new(session_configuration);
@@ -4018,9 +4157,11 @@ mod tests {
             compact_prompt: config.compact_prompt.clone(),
             approval_policy: config.approval_policy.clone(),
             sandbox_policy: config.sandbox_policy.clone(),
+            windows_sandbox_level: WindowsSandboxLevel::from_config(&config),
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             session_source: SessionSource::Exec,
+            dynamic_tools: Vec::new(),
         };
 
         let mut state = SessionState::new(session_configuration);
@@ -4281,9 +4422,11 @@ mod tests {
             compact_prompt: config.compact_prompt.clone(),
             approval_policy: config.approval_policy.clone(),
             sandbox_policy: config.sandbox_policy.clone(),
+            windows_sandbox_level: WindowsSandboxLevel::from_config(&config),
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             session_source: SessionSource::Exec,
+            dynamic_tools: Vec::new(),
         };
         let per_turn_config = Session::build_per_turn_config(&session_configuration);
         let model_info = ModelsManager::construct_model_info_offline(
@@ -4297,7 +4440,8 @@ mod tests {
             session_configuration.session_source.clone(),
         );
 
-        let state = SessionState::new(session_configuration.clone());
+        let mut state = SessionState::new(session_configuration.clone());
+        mark_state_initial_context_seeded(&mut state);
         let skills_manager = Arc::new(SkillsManager::new(config.codex_home.clone()));
 
         let services = SessionServices {
@@ -4389,9 +4533,11 @@ mod tests {
             compact_prompt: config.compact_prompt.clone(),
             approval_policy: config.approval_policy.clone(),
             sandbox_policy: config.sandbox_policy.clone(),
+            windows_sandbox_level: WindowsSandboxLevel::from_config(&config),
             cwd: config.cwd.clone(),
             original_config_do_not_use: Arc::clone(&config),
             session_source: SessionSource::Exec,
+            dynamic_tools: Vec::new(),
         };
         let per_turn_config = Session::build_per_turn_config(&session_configuration);
         let model_info = ModelsManager::construct_model_info_offline(
@@ -4405,7 +4551,8 @@ mod tests {
             session_configuration.session_source.clone(),
         );
 
-        let state = SessionState::new(session_configuration.clone());
+        let mut state = SessionState::new(session_configuration.clone());
+        mark_state_initial_context_seeded(&mut state);
         let skills_manager = Arc::new(SkillsManager::new(config.codex_home.clone()));
 
         let services = SessionServices {
@@ -4451,6 +4598,10 @@ mod tests {
         (session, turn_context, rx_event)
     }
 
+    fn mark_state_initial_context_seeded(state: &mut SessionState) {
+        state.initial_context_seeded = true;
+    }
+
     #[tokio::test]
     async fn refresh_mcp_servers_is_deferred_until_next_turn() {
         let (session, turn_context) = make_session_and_context().await;
@@ -4712,6 +4863,7 @@ mod tests {
                     .map(|(name, tool)| (name, tool.tool))
                     .collect(),
             ),
+            turn_context.dynamic_tools.as_slice(),
         );
         let item = ResponseItem::CustomToolCall {
             id: None,
@@ -4889,6 +5041,7 @@ mod tests {
             expiration: timeout_ms.into(),
             env: HashMap::new(),
             sandbox_permissions,
+            windows_sandbox_level: turn_context.windows_sandbox_level,
             justification: Some("test".to_string()),
             arg0: None,
         };
@@ -4899,6 +5052,7 @@ mod tests {
             cwd: params.cwd.clone(),
             expiration: timeout_ms.into(),
             env: HashMap::new(),
+            windows_sandbox_level: turn_context.windows_sandbox_level,
             justification: params.justification.clone(),
             arg0: None,
         };

codex-rs/core/src/codex_delegate.rs

@@ -57,6 +57,7 @@ pub(crate) async fn run_codex_thread_interactive(
         initial_history.unwrap_or(InitialHistory::New),
         SessionSource::SubAgent(SubAgentSource::Review),
         parent_session.services.agent_control.clone(),
+        Vec::new(),
     )
     .await?;
     let codex = Arc::new(codex);

codex-rs/core/src/compact.rs

@@ -10,7 +10,8 @@ use crate::error::CodexErr;
 use crate::error::Result as CodexResult;
 use crate::features::Feature;
 use crate::protocol::CompactedItem;
-use crate::protocol::ContextCompactedEvent;
+use crate::protocol::ContextCompactionEndedEvent;
+use crate::protocol::ContextCompactionStartedEvent;
 use crate::protocol::EventMsg;
 use crate::protocol::TurnContextItem;
 use crate::protocol::TurnStartedEvent;
@@ -20,6 +21,7 @@ use crate::truncate::TruncationPolicy;
 use crate::truncate::approx_token_count;
 use crate::truncate::truncate_text;
 use crate::util::backoff;
+use codex_protocol::items::ContextCompactionItem;
 use codex_protocol::items::TurnItem;
 use codex_protocol::models::ContentItem;
 use codex_protocol::models::ResponseInputItem;
@@ -28,6 +30,7 @@ use codex_protocol::protocol::RolloutItem;
 use codex_protocol::user_input::UserInput;
 use futures::prelude::*;
 use tracing::error;
+use uuid::Uuid;
 
 pub const SUMMARIZATION_PROMPT: &str = include_str!("../templates/compact/prompt.md");
 pub const SUMMARY_PREFIX: &str = include_str!("../templates/compact/summary_prefix.md");
@@ -71,6 +74,9 @@ async fn run_compact_task_inner(
     turn_context: Arc<TurnContext>,
     input: Vec<UserInput>,
 ) {
+    let compaction_item = compaction_turn_item();
+    emit_compaction_started(&sess, &turn_context, &compaction_item).await;
+
     let initial_input_for_turn: ResponseInputItem = ResponseInputItem::from(input);
 
     let mut history = sess.clone_history().await;
@@ -131,6 +137,7 @@ async fn run_compact_task_inner(
                 break;
             }
             Err(CodexErr::Interrupted) => {
+                emit_compaction_ended(&sess, &turn_context, compaction_item.clone()).await;
                 return;
             }
             Err(e @ CodexErr::ContextWindowExceeded) => {
@@ -147,6 +154,7 @@ async fn run_compact_task_inner(
                 sess.set_total_tokens_full(turn_context.as_ref()).await;
                 let event = EventMsg::Error(e.to_error_event(None));
                 sess.send_event(&turn_context, event).await;
+                emit_compaction_ended(&sess, &turn_context, compaction_item.clone()).await;
                 return;
             }
             Err(e) => {
@@ -164,6 +172,7 @@ async fn run_compact_task_inner(
                 } else {
                     let event = EventMsg::Error(e.to_error_event(None));
                     sess.send_event(&turn_context, event).await;
+                    emit_compaction_ended(&sess, &turn_context, compaction_item.clone()).await;
                     return;
                 }
             }
@@ -193,8 +202,7 @@ async fn run_compact_task_inner(
     });
     sess.persist_rollout_items(&[rollout_item]).await;
 
-    let event = EventMsg::ContextCompacted(ContextCompactedEvent {});
-    sess.send_event(&turn_context, event).await;
+    emit_compaction_ended(&sess, &turn_context, compaction_item).await;
 
     let warning = EventMsg::Warning(WarningEvent {
         message: "Heads up: Long threads and multiple compactions can cause the model to be less accurate. Start a new thread when possible to keep threads small and targeted.".to_string(),
@@ -202,6 +210,38 @@ async fn run_compact_task_inner(
     sess.send_event(&turn_context, warning).await;
 }
 
+fn compaction_turn_item() -> TurnItem {
+    TurnItem::ContextCompaction(ContextCompactionItem {
+        id: Uuid::new_v4().to_string(),
+    })
+}
+
+pub(crate) async fn emit_compaction_started(
+    sess: &Session,
+    turn_context: &TurnContext,
+    item: &TurnItem,
+) {
+    sess.send_event(
+        turn_context,
+        EventMsg::ContextCompactionStarted(ContextCompactionStartedEvent {}),
+    )
+    .await;
+    sess.emit_turn_item_started(turn_context, item).await;
+}
+
+pub(crate) async fn emit_compaction_ended(
+    sess: &Session,
+    turn_context: &TurnContext,
+    item: TurnItem,
+) {
+    sess.emit_turn_item_completed(turn_context, item).await;
+    sess.send_event(
+        turn_context,
+        EventMsg::ContextCompactionEnded(ContextCompactionEndedEvent {}),
+    )
+    .await;
+}
+
 pub fn content_items_to_text(content: &[ContentItem]) -> Option<String> {
     let mut pieces = Vec::new();
     for item in content {

codex-rs/core/src/compact_remote.rs

@@ -3,13 +3,17 @@ use std::sync::Arc;
 use crate::Prompt;
 use crate::codex::Session;
 use crate::codex::TurnContext;
+use crate::compact::emit_compaction_ended;
+use crate::compact::emit_compaction_started;
 use crate::error::Result as CodexResult;
 use crate::protocol::CompactedItem;
-use crate::protocol::ContextCompactedEvent;
 use crate::protocol::EventMsg;
 use crate::protocol::RolloutItem;
 use crate::protocol::TurnStartedEvent;
+use codex_protocol::items::ContextCompactionItem;
+use codex_protocol::items::TurnItem;
 use codex_protocol::models::ResponseItem;
+use uuid::Uuid;
 
 pub(crate) async fn run_inline_remote_auto_compact_task(
     sess: Arc<Session>,
@@ -28,12 +32,19 @@ pub(crate) async fn run_remote_compact_task(sess: Arc<Session>, turn_context: Ar
 }
 
 async fn run_remote_compact_task_inner(sess: &Arc<Session>, turn_context: &Arc<TurnContext>) {
+    let compaction_item = TurnItem::ContextCompaction(ContextCompactionItem {
+        id: Uuid::new_v4().to_string(),
+    });
+    emit_compaction_started(sess, turn_context, &compaction_item).await;
+
     if let Err(err) = run_remote_compact_task_inner_impl(sess, turn_context).await {
         let event = EventMsg::Error(
             err.to_error_event(Some("Error running remote compact task".to_string())),
         );
         sess.send_event(turn_context, event).await;
     }
+
+    emit_compaction_ended(sess, turn_context, compaction_item).await;
 }
 
 async fn run_remote_compact_task_inner_impl(
@@ -77,8 +88,5 @@ async fn run_remote_compact_task_inner_impl(
     sess.persist_rollout_items(&[RolloutItem::Compacted(compacted_item)])
         .await;
 
-    let event = EventMsg::ContextCompacted(ContextCompactedEvent {});
-    sess.send_event(turn_context, event).await;
-
     Ok(())
 }

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

@@ -4,6 +4,7 @@ use crate::config::types::Notice;
 use crate::path_utils::resolve_symlink_write_paths;
 use crate::path_utils::write_atomically;
 use anyhow::Context;
+use codex_protocol::config_types::Personality;
 use codex_protocol::config_types::TrustLevel;
 use codex_protocol::openai_models::ReasoningEffort;
 use std::collections::BTreeMap;
@@ -24,6 +25,8 @@ pub enum ConfigEdit {
         model: Option<String>,
         effort: Option<ReasoningEffort>,
     },
+    /// Update the active (or default) model personality.
+    SetModelPersonality { personality: Option<Personality> },
     /// Toggle the acknowledgement flag under `[notice]`.
     SetNoticeHideFullAccessWarning(bool),
     /// Toggle the Windows world-writable directories warning acknowledgement flag.
@@ -164,6 +167,11 @@ mod document_helpers {
         {
             entry["disabled_tools"] = array_from_iter(disabled_tools.iter().cloned());
         }
+        if let Some(scopes) = &config.scopes
+            && !scopes.is_empty()
+        {
+            entry["scopes"] = array_from_iter(scopes.iter().cloned());
+        }
 
         entry
     }
@@ -269,6 +277,10 @@ impl ConfigDocument {
                 );
                 mutated
             }),
+            ConfigEdit::SetModelPersonality { personality } => Ok(self.write_profile_value(
+                &["model_personality"],
+                personality.map(|personality| value(personality.to_string())),
+            )),
             ConfigEdit::SetNoticeHideFullAccessWarning(acknowledged) => Ok(self.write_value(
                 Scope::Global,
                 &[Notice::TABLE_KEY, "hide_full_access_warning"],
@@ -712,6 +724,12 @@ impl ConfigEditsBuilder {
         self
     }
 
+    pub fn set_model_personality(mut self, personality: Option<Personality>) -> Self {
+        self.edits
+            .push(ConfigEdit::SetModelPersonality { personality });
+        self
+    }
+
     pub fn set_hide_full_access_warning(mut self, acknowledged: bool) -> Self {
         self.edits
             .push(ConfigEdit::SetNoticeHideFullAccessWarning(acknowledged));
@@ -1360,6 +1378,7 @@ gpt-5 = "gpt-5.1"
                 tool_timeout_sec: None,
                 enabled_tools: Some(vec!["one".to_string(), "two".to_string()]),
                 disabled_tools: None,
+                scopes: None,
             },
         );
 
@@ -1382,6 +1401,7 @@ gpt-5 = "gpt-5.1"
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: Some(vec!["forbidden".to_string()]),
+                scopes: None,
             },
         );
 
@@ -1447,6 +1467,7 @@ foo = { command = "cmd" }
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         );
 
@@ -1491,6 +1512,7 @@ foo = { command = "cmd" } # keep me
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         );
 
@@ -1534,6 +1556,7 @@ foo = { command = "cmd", args = ["--flag"] } # keep me
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         );
 
@@ -1578,6 +1601,7 @@ foo = { command = "cmd" }
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         );
 

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

@@ -38,6 +38,7 @@ use crate::project_doc::DEFAULT_PROJECT_DOC_FILENAME;
 use crate::project_doc::LOCAL_PROJECT_DOC_FILENAME;
 use crate::protocol::AskForApproval;
 use crate::protocol::SandboxPolicy;
+use crate::windows_sandbox::WindowsSandboxLevelExt;
 use codex_app_server_protocol::Tools;
 use codex_app_server_protocol::UserSavedConfig;
 use codex_protocol::config_types::AltScreenMode;
@@ -49,6 +50,7 @@ use codex_protocol::config_types::SandboxMode;
 use codex_protocol::config_types::TrustLevel;
 use codex_protocol::config_types::Verbosity;
 use codex_protocol::config_types::WebSearchMode;
+use codex_protocol::config_types::WindowsSandboxLevel;
 use codex_protocol::openai_models::ReasoningEffort;
 use codex_rmcp_client::OAuthCredentialsStoreMode;
 use codex_utils_absolute_path::AbsolutePathBuf;
@@ -304,8 +306,8 @@ pub struct Config {
     /// model info's default preference.
     pub include_apply_patch_tool: bool,
 
-    /// Explicit or feature-derived web search mode.
-    pub web_search_mode: Option<WebSearchMode>,
+    /// Explicit or feature-derived web search mode. Defaults to cached.
+    pub web_search_mode: WebSearchMode,
 
     /// If set to `true`, used only the experimental unified exec tool.
     pub use_experimental_unified_exec_tool: bool,
@@ -316,6 +318,9 @@ pub struct Config {
     /// Centralized feature flags; source of truth for feature gating.
     pub features: Features,
 
+    /// When `true`, suppress warnings about unstable (under development) features.
+    pub suppress_unstable_features_warning: bool,
+
     /// The active profile name used to derive this `Config` (if any).
     pub active_profile: Option<String>,
 
@@ -906,6 +911,9 @@ pub struct ConfigToml {
     #[schemars(schema_with = "crate::config::schema::features_schema")]
     pub features: Option<FeaturesToml>,
 
+    /// Suppress warnings about unstable (under development) features.
+    pub suppress_unstable_features_warning: Option<bool>,
+
     /// Settings for ghost snapshots (used for undo).
     #[serde(default)]
     pub ghost_snapshot: Option<GhostSnapshotToml>,
@@ -1050,6 +1058,7 @@ impl ConfigToml {
         &self,
         sandbox_mode_override: Option<SandboxMode>,
         profile_sandbox_mode: Option<SandboxMode>,
+        windows_sandbox_level: WindowsSandboxLevel,
         resolved_cwd: &Path,
     ) -> SandboxPolicyResolution {
         let resolved_sandbox_mode = sandbox_mode_override
@@ -1088,7 +1097,7 @@ impl ConfigToml {
         if cfg!(target_os = "windows")
             && matches!(resolved_sandbox_mode, SandboxMode::WorkspaceWrite)
             // If the experimental Windows sandbox is enabled, do not force a downgrade.
-            && crate::safety::get_platform_sandbox().is_none()
+            && windows_sandbox_level == codex_protocol::config_types::WindowsSandboxLevel::Disabled
         {
             sandbox_policy = SandboxPolicy::new_read_only_policy();
             forced_auto_mode_downgraded_on_windows = true;
@@ -1194,22 +1203,27 @@ pub fn resolve_oss_provider(
     }
 }
 
-/// Resolve the web search mode from explicit config and feature flags.
+/// Resolve the web search mode from explicit config, feature flags, and sandbox policy.
+/// Live search is auto-enabled when sandbox policy is `DangerFullAccess`
 fn resolve_web_search_mode(
     config_toml: &ConfigToml,
     config_profile: &ConfigProfile,
     features: &Features,
-) -> Option<WebSearchMode> {
+    sandbox_policy: &SandboxPolicy,
+) -> WebSearchMode {
     if let Some(mode) = config_profile.web_search.or(config_toml.web_search) {
-        return Some(mode);
+        return mode;
     }
     if features.enabled(Feature::WebSearchCached) {
-        return Some(WebSearchMode::Cached);
+        return WebSearchMode::Cached;
     }
     if features.enabled(Feature::WebSearchRequest) {
-        return Some(WebSearchMode::Live);
+        return WebSearchMode::Live;
     }
-    None
+    if matches!(sandbox_policy, SandboxPolicy::DangerFullAccess) {
+        return WebSearchMode::Live;
+    }
+    WebSearchMode::Cached
 }
 
 impl Config {
@@ -1278,17 +1292,6 @@ impl Config {
         };
 
         let features = Features::from_config(&cfg, &config_profile, feature_overrides);
-        let web_search_mode = resolve_web_search_mode(&cfg, &config_profile, &features);
-        #[cfg(target_os = "windows")]
-        {
-            // Base flag controls sandbox on/off; elevated only applies when base is enabled.
-            let sandbox_enabled = features.enabled(Feature::WindowsSandbox);
-            crate::safety::set_windows_sandbox_enabled(sandbox_enabled);
-            let elevated_enabled =
-                sandbox_enabled && features.enabled(Feature::WindowsSandboxElevated);
-            crate::safety::set_windows_elevated_sandbox_enabled(elevated_enabled);
-        }
-
         let resolved_cwd = {
             use std::env;
 
@@ -1315,10 +1318,16 @@ impl Config {
             .get_active_project(&resolved_cwd)
             .unwrap_or(ProjectConfig { trust_level: None });
 
+        let windows_sandbox_level = WindowsSandboxLevel::from_features(&features);
         let SandboxPolicyResolution {
             policy: mut sandbox_policy,
             forced_auto_mode_downgraded_on_windows,
-        } = cfg.derive_sandbox_policy(sandbox_mode, config_profile.sandbox_mode, &resolved_cwd);
+        } = cfg.derive_sandbox_policy(
+            sandbox_mode,
+            config_profile.sandbox_mode,
+            windows_sandbox_level,
+            &resolved_cwd,
+        );
         if let SandboxPolicy::WorkspaceWrite { writable_roots, .. } = &mut sandbox_policy {
             for path in additional_writable_roots {
                 if !writable_roots.iter().any(|existing| existing == &path) {
@@ -1338,6 +1347,8 @@ impl Config {
                     AskForApproval::default()
                 }
             });
+        let web_search_mode =
+            resolve_web_search_mode(&cfg, &config_profile, &features, &sandbox_policy);
         // TODO(dylan): We should be able to leverage ConfigLayerStack so that
         // we can reliably check this at every config level.
         let did_user_set_custom_approval_policy_or_sandbox_mode = approval_policy_override
@@ -1564,6 +1575,9 @@ impl Config {
             use_experimental_unified_exec_tool,
             ghost_snapshot,
             features,
+            suppress_unstable_features_warning: cfg
+                .suppress_unstable_features_warning
+                .unwrap_or(false),
             active_profile: active_profile_name,
             active_project,
             windows_wsl_setup_acknowledged: cfg.windows_wsl_setup_acknowledged.unwrap_or(false),
@@ -1658,7 +1672,6 @@ impl Config {
     }
 
     pub fn set_windows_sandbox_globally(&mut self, value: bool) {
-        crate::safety::set_windows_sandbox_enabled(value);
         if value {
             self.features.enable(Feature::WindowsSandbox);
         } else {
@@ -1668,7 +1681,6 @@ impl Config {
     }
 
     pub fn set_windows_elevated_sandbox_globally(&mut self, value: bool) {
-        crate::safety::set_windows_elevated_sandbox_enabled(value);
         if value {
             self.features.enable(Feature::WindowsSandboxElevated);
         } else {
@@ -1772,6 +1784,7 @@ mod tests {
             tool_timeout_sec: None,
             enabled_tools: None,
             disabled_tools: None,
+            scopes: None,
         }
     }
 
@@ -1789,6 +1802,7 @@ mod tests {
             tool_timeout_sec: None,
             enabled_tools: None,
             disabled_tools: None,
+            scopes: None,
         }
     }
 
@@ -1860,6 +1874,7 @@ network_access = false  # This should be ignored.
         let resolution = sandbox_full_access_cfg.derive_sandbox_policy(
             sandbox_mode_override,
             None,
+            WindowsSandboxLevel::Disabled,
             &PathBuf::from("/tmp/test"),
         );
         assert_eq!(
@@ -1883,6 +1898,7 @@ network_access = true  # This should be ignored.
         let resolution = sandbox_read_only_cfg.derive_sandbox_policy(
             sandbox_mode_override,
             None,
+            WindowsSandboxLevel::Disabled,
             &PathBuf::from("/tmp/test"),
         );
         assert_eq!(
@@ -1914,6 +1930,7 @@ exclude_slash_tmp = true
         let resolution = sandbox_workspace_write_cfg.derive_sandbox_policy(
             sandbox_mode_override,
             None,
+            WindowsSandboxLevel::Disabled,
             &PathBuf::from("/tmp/test"),
         );
         if cfg!(target_os = "windows") {
@@ -1962,6 +1979,7 @@ trust_level = "trusted"
         let resolution = sandbox_workspace_write_cfg.derive_sandbox_policy(
             sandbox_mode_override,
             None,
+            WindowsSandboxLevel::Disabled,
             &PathBuf::from("/tmp/test"),
         );
         if cfg!(target_os = "windows") {
@@ -2253,12 +2271,15 @@ trust_level = "trusted"
     }
 
     #[test]
-    fn web_search_mode_uses_none_if_unset() {
+    fn web_search_mode_defaults_to_cached_if_unset() {
         let cfg = ConfigToml::default();
         let profile = ConfigProfile::default();
         let features = Features::with_defaults();
 
-        assert_eq!(resolve_web_search_mode(&cfg, &profile, &features), None);
+        assert_eq!(
+            resolve_web_search_mode(&cfg, &profile, &features, &SandboxPolicy::ReadOnly),
+            WebSearchMode::Cached
+        );
     }
 
     #[test]
@@ -2272,8 +2293,8 @@ trust_level = "trusted"
         features.enable(Feature::WebSearchCached);
 
         assert_eq!(
-            resolve_web_search_mode(&cfg, &profile, &features),
-            Some(WebSearchMode::Live)
+            resolve_web_search_mode(&cfg, &profile, &features, &SandboxPolicy::ReadOnly),
+            WebSearchMode::Live
         );
     }
 
@@ -2288,11 +2309,50 @@ trust_level = "trusted"
         features.enable(Feature::WebSearchRequest);
 
         assert_eq!(
-            resolve_web_search_mode(&cfg, &profile, &features),
-            Some(WebSearchMode::Disabled)
+            resolve_web_search_mode(&cfg, &profile, &features, &SandboxPolicy::ReadOnly),
+            WebSearchMode::Disabled
         );
     }
 
+    #[test]
+    fn danger_full_access_defaults_web_search_live_when_unset() -> std::io::Result<()> {
+        let codex_home = TempDir::new()?;
+        let cfg = ConfigToml {
+            sandbox_mode: Some(SandboxMode::DangerFullAccess),
+            ..Default::default()
+        };
+
+        let config = Config::load_from_base_config_with_overrides(
+            cfg,
+            ConfigOverrides::default(),
+            codex_home.path().to_path_buf(),
+        )?;
+
+        assert_eq!(config.web_search_mode, WebSearchMode::Live);
+
+        Ok(())
+    }
+
+    #[test]
+    fn explicit_web_search_mode_wins_in_danger_full_access() -> std::io::Result<()> {
+        let codex_home = TempDir::new()?;
+        let cfg = ConfigToml {
+            sandbox_mode: Some(SandboxMode::DangerFullAccess),
+            web_search: Some(WebSearchMode::Cached),
+            ..Default::default()
+        };
+
+        let config = Config::load_from_base_config_with_overrides(
+            cfg,
+            ConfigOverrides::default(),
+            codex_home.path().to_path_buf(),
+        )?;
+
+        assert_eq!(config.web_search_mode, WebSearchMode::Cached);
+
+        Ok(())
+    }
+
     #[test]
     fn profile_legacy_toggles_override_base() -> std::io::Result<()> {
         let codex_home = TempDir::new()?;
@@ -2614,6 +2674,7 @@ profile = "project"
                 tool_timeout_sec: Some(Duration::from_secs(5)),
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         );
 
@@ -2768,6 +2829,7 @@ bearer_token = "secret"
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         )]);
 
@@ -2837,6 +2899,7 @@ ZIG_VAR = "3"
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         )]);
 
@@ -2886,6 +2949,7 @@ ZIG_VAR = "3"
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         )]);
 
@@ -2933,6 +2997,7 @@ ZIG_VAR = "3"
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         )]);
 
@@ -2996,6 +3061,7 @@ startup_timeout_sec = 2.0
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         )]);
         apply_blocking(
@@ -3071,6 +3137,7 @@ X-Auth = "DOCS_AUTH"
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         )]);
 
@@ -3099,6 +3166,7 @@ X-Auth = "DOCS_AUTH"
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         );
         apply_blocking(
@@ -3165,6 +3233,7 @@ url = "https://example.com/mcp"
                     tool_timeout_sec: None,
                     enabled_tools: None,
                     disabled_tools: None,
+                    scopes: None,
                 },
             ),
             (
@@ -3183,6 +3252,7 @@ url = "https://example.com/mcp"
                     tool_timeout_sec: None,
                     enabled_tools: None,
                     disabled_tools: None,
+                    scopes: None,
                 },
             ),
         ]);
@@ -3264,6 +3334,7 @@ url = "https://example.com/mcp"
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
         )]);
 
@@ -3307,6 +3378,7 @@ url = "https://example.com/mcp"
                 tool_timeout_sec: None,
                 enabled_tools: Some(vec!["allowed".to_string()]),
                 disabled_tools: Some(vec!["blocked".to_string()]),
+                scopes: None,
             },
         )]);
 
@@ -3714,10 +3786,11 @@ model_verbosity = "high"
                 forced_chatgpt_workspace_id: None,
                 forced_login_method: None,
                 include_apply_patch_tool: false,
-                web_search_mode: None,
+                web_search_mode: WebSearchMode::Cached,
                 use_experimental_unified_exec_tool: false,
                 ghost_snapshot: GhostSnapshotConfig::default(),
                 features: Features::with_defaults(),
+                suppress_unstable_features_warning: false,
                 active_profile: Some("o3".to_string()),
                 active_project: ProjectConfig { trust_level: None },
                 windows_wsl_setup_acknowledged: false,
@@ -3796,10 +3869,11 @@ model_verbosity = "high"
             forced_chatgpt_workspace_id: None,
             forced_login_method: None,
             include_apply_patch_tool: false,
-            web_search_mode: None,
+            web_search_mode: WebSearchMode::Cached,
             use_experimental_unified_exec_tool: false,
             ghost_snapshot: GhostSnapshotConfig::default(),
             features: Features::with_defaults(),
+            suppress_unstable_features_warning: false,
             active_profile: Some("gpt3".to_string()),
             active_project: ProjectConfig { trust_level: None },
             windows_wsl_setup_acknowledged: false,
@@ -3893,10 +3967,11 @@ model_verbosity = "high"
             forced_chatgpt_workspace_id: None,
             forced_login_method: None,
             include_apply_patch_tool: false,
-            web_search_mode: None,
+            web_search_mode: WebSearchMode::Cached,
             use_experimental_unified_exec_tool: false,
             ghost_snapshot: GhostSnapshotConfig::default(),
             features: Features::with_defaults(),
+            suppress_unstable_features_warning: false,
             active_profile: Some("zdr".to_string()),
             active_project: ProjectConfig { trust_level: None },
             windows_wsl_setup_acknowledged: false,
@@ -3976,10 +4051,11 @@ model_verbosity = "high"
             forced_chatgpt_workspace_id: None,
             forced_login_method: None,
             include_apply_patch_tool: false,
-            web_search_mode: None,
+            web_search_mode: WebSearchMode::Cached,
             use_experimental_unified_exec_tool: false,
             ghost_snapshot: GhostSnapshotConfig::default(),
             features: Features::with_defaults(),
+            suppress_unstable_features_warning: false,
             active_profile: Some("gpt5".to_string()),
             active_project: ProjectConfig { trust_level: None },
             windows_wsl_setup_acknowledged: false,
@@ -4160,7 +4236,12 @@ trust_level = "untrusted"
         let cfg = toml::from_str::<ConfigToml>(config_with_untrusted)
             .expect("TOML deserialization should succeed");
 
-        let resolution = cfg.derive_sandbox_policy(None, None, &PathBuf::from("/tmp/test"));
+        let resolution = cfg.derive_sandbox_policy(
+            None,
+            None,
+            WindowsSandboxLevel::Disabled,
+            &PathBuf::from("/tmp/test"),
+        );
 
         // Verify that untrusted projects get WorkspaceWrite (or ReadOnly on Windows due to downgrade)
         if cfg!(target_os = "windows") {

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

@@ -73,6 +73,10 @@ pub struct McpServerConfig {
     /// Explicit deny-list of tools. These tools will be removed after applying `enabled_tools`.
     #[serde(default, skip_serializing_if = "Option::is_none")]
     pub disabled_tools: Option<Vec<String>>,
+
+    /// Optional OAuth scopes to request during MCP login.
+    #[serde(default, skip_serializing_if = "Option::is_none")]
+    pub scopes: Option<Vec<String>>,
 }
 
 // Raw MCP config shape used for deserialization and JSON Schema generation.
@@ -113,6 +117,8 @@ pub(crate) struct RawMcpServerConfig {
     pub enabled_tools: Option<Vec<String>>,
     #[serde(default)]
     pub disabled_tools: Option<Vec<String>>,
+    #[serde(default)]
+    pub scopes: Option<Vec<String>>,
 }
 
 impl<'de> Deserialize<'de> for McpServerConfig {
@@ -134,6 +140,7 @@ impl<'de> Deserialize<'de> for McpServerConfig {
         let enabled = raw.enabled.unwrap_or_else(default_enabled);
         let enabled_tools = raw.enabled_tools.clone();
         let disabled_tools = raw.disabled_tools.clone();
+        let scopes = raw.scopes.clone();
 
         fn throw_if_set<E, T>(transport: &str, field: &str, value: Option<&T>) -> Result<(), E>
         where
@@ -188,6 +195,7 @@ impl<'de> Deserialize<'de> for McpServerConfig {
             disabled_reason: None,
             enabled_tools,
             disabled_tools,
+            scopes,
         })
     }
 }

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

@@ -512,10 +512,10 @@ impl ProjectTrustContext {
         let user_config_file = self.user_config_file.as_path().display();
         match decision.trust_level {
             Some(TrustLevel::Untrusted) => Some(format!(
-                "{trust_key} is marked as untrusted in {user_config_file}. Mark it trusted to enable project config folders."
+                "{trust_key} is marked as untrusted in {user_config_file}. To load config.toml, mark it trusted."
             )),
             _ => Some(format!(
-                "Add {trust_key} as a trusted project in {user_config_file}."
+                "To load config.toml, add {trust_key} as a trusted project in {user_config_file}."
             )),
         }
     }
@@ -526,21 +526,16 @@ fn project_layer_entry(
     dot_codex_folder: &AbsolutePathBuf,
     layer_dir: &AbsolutePathBuf,
     config: TomlValue,
+    config_toml_exists: bool,
 ) -> ConfigLayerEntry {
-    match trust_context.disabled_reason_for_dir(layer_dir) {
-        Some(reason) => ConfigLayerEntry::new_disabled(
-            ConfigLayerSource::Project {
-                dot_codex_folder: dot_codex_folder.clone(),
-            },
-            config,
-            reason,
-        ),
-        None => ConfigLayerEntry::new(
-            ConfigLayerSource::Project {
-                dot_codex_folder: dot_codex_folder.clone(),
-            },
-            config,
-        ),
+    let source = ConfigLayerSource::Project {
+        dot_codex_folder: dot_codex_folder.clone(),
+    };
+
+    if config_toml_exists && let Some(reason) = trust_context.disabled_reason_for_dir(layer_dir) {
+        ConfigLayerEntry::new_disabled(source, config, reason)
+    } else {
+        ConfigLayerEntry::new(source, config)
     }
 }
 
@@ -715,13 +710,15 @@ async fn load_project_layers(
                             &dot_codex_abs,
                             &layer_dir,
                             TomlValue::Table(toml::map::Map::new()),
+                            true,
                         ));
                         continue;
                     }
                 };
                 let config =
                     resolve_relative_paths_in_config_toml(config, dot_codex_abs.as_path())?;
-                let entry = project_layer_entry(trust_context, &dot_codex_abs, &layer_dir, config);
+                let entry =
+                    project_layer_entry(trust_context, &dot_codex_abs, &layer_dir, config, true);
                 layers.push(entry);
             }
             Err(err) => {
@@ -734,6 +731,7 @@ async fn load_project_layers(
                         &dot_codex_abs,
                         &layer_dir,
                         TomlValue::Table(toml::map::Map::new()),
+                        false,
                     ));
                 } else {
                     let config_file_display = config_file.as_path().display();

codex-rs/core/src/event_mapping.rs

@@ -21,6 +21,7 @@ use crate::instructions::SkillInstructions;
 use crate::instructions::UserInstructions;
 use crate::session_prefix::is_session_prefix;
 use crate::user_shell_command::is_user_shell_command_text;
+use crate::web_search::web_search_action_detail;
 
 fn parse_user_message(message: &[ContentItem]) -> Option<UserMessageItem> {
     if UserInstructions::is_user_instructions(message)
@@ -127,14 +128,17 @@ pub fn parse_turn_item(item: &ResponseItem) -> Option<TurnItem> {
                 raw_content,
             }))
         }
-        ResponseItem::WebSearchCall {
-            id,
-            action: WebSearchAction::Search { query },
-            ..
-        } => Some(TurnItem::WebSearch(WebSearchItem {
-            id: id.clone().unwrap_or_default(),
-            query: query.clone().unwrap_or_default(),
-        })),
+        ResponseItem::WebSearchCall { id, action, .. } => {
+            let (action, query) = match action {
+                Some(action) => (action.clone(), web_search_action_detail(action)),
+                None => (WebSearchAction::Other, String::new()),
+            };
+            Some(TurnItem::WebSearch(WebSearchItem {
+                id: id.clone().unwrap_or_default(),
+                query,
+                action,
+            }))
+        }
         _ => None,
     }
 }
@@ -144,6 +148,7 @@ mod tests {
     use super::parse_turn_item;
     use codex_protocol::items::AgentMessageContent;
     use codex_protocol::items::TurnItem;
+    use codex_protocol::items::WebSearchItem;
     use codex_protocol::models::ContentItem;
     use codex_protocol::models::ReasoningItemContent;
     use codex_protocol::models::ReasoningItemReasoningSummary;
@@ -419,18 +424,102 @@ mod tests {
         let item = ResponseItem::WebSearchCall {
             id: Some("ws_1".to_string()),
             status: Some("completed".to_string()),
-            action: WebSearchAction::Search {
+            action: Some(WebSearchAction::Search {
                 query: Some("weather".to_string()),
-            },
+            }),
         };
 
         let turn_item = parse_turn_item(&item).expect("expected web search turn item");
 
         match turn_item {
-            TurnItem::WebSearch(search) => {
-                assert_eq!(search.id, "ws_1");
-                assert_eq!(search.query, "weather");
-            }
+            TurnItem::WebSearch(search) => assert_eq!(
+                search,
+                WebSearchItem {
+                    id: "ws_1".to_string(),
+                    query: "weather".to_string(),
+                    action: WebSearchAction::Search {
+                        query: Some("weather".to_string()),
+                    },
+                }
+            ),
+            other => panic!("expected TurnItem::WebSearch, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn parses_web_search_open_page_call() {
+        let item = ResponseItem::WebSearchCall {
+            id: Some("ws_open".to_string()),
+            status: Some("completed".to_string()),
+            action: Some(WebSearchAction::OpenPage {
+                url: Some("https://example.com".to_string()),
+            }),
+        };
+
+        let turn_item = parse_turn_item(&item).expect("expected web search turn item");
+
+        match turn_item {
+            TurnItem::WebSearch(search) => assert_eq!(
+                search,
+                WebSearchItem {
+                    id: "ws_open".to_string(),
+                    query: "https://example.com".to_string(),
+                    action: WebSearchAction::OpenPage {
+                        url: Some("https://example.com".to_string()),
+                    },
+                }
+            ),
+            other => panic!("expected TurnItem::WebSearch, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn parses_web_search_find_in_page_call() {
+        let item = ResponseItem::WebSearchCall {
+            id: Some("ws_find".to_string()),
+            status: Some("completed".to_string()),
+            action: Some(WebSearchAction::FindInPage {
+                url: Some("https://example.com".to_string()),
+                pattern: Some("needle".to_string()),
+            }),
+        };
+
+        let turn_item = parse_turn_item(&item).expect("expected web search turn item");
+
+        match turn_item {
+            TurnItem::WebSearch(search) => assert_eq!(
+                search,
+                WebSearchItem {
+                    id: "ws_find".to_string(),
+                    query: "'needle' in https://example.com".to_string(),
+                    action: WebSearchAction::FindInPage {
+                        url: Some("https://example.com".to_string()),
+                        pattern: Some("needle".to_string()),
+                    },
+                }
+            ),
+            other => panic!("expected TurnItem::WebSearch, got {other:?}"),
+        }
+    }
+
+    #[test]
+    fn parses_partial_web_search_call_without_action_as_other() {
+        let item = ResponseItem::WebSearchCall {
+            id: Some("ws_partial".to_string()),
+            status: Some("in_progress".to_string()),
+            action: None,
+        };
+
+        let turn_item = parse_turn_item(&item).expect("expected web search turn item");
+        match turn_item {
+            TurnItem::WebSearch(search) => assert_eq!(
+                search,
+                WebSearchItem {
+                    id: "ws_partial".to_string(),
+                    query: String::new(),
+                    action: WebSearchAction::Other,
+                }
+            ),
             other => panic!("expected TurnItem::WebSearch, got {other:?}"),
         }
     }

codex-rs/core/src/exec.rs

@@ -64,6 +64,7 @@ pub struct ExecParams {
     pub expiration: ExecExpiration,
     pub env: HashMap<String, String>,
     pub sandbox_permissions: SandboxPermissions,
+    pub windows_sandbox_level: codex_protocol::config_types::WindowsSandboxLevel,
     pub justification: Option<String>,
     pub arg0: Option<String>,
 }
@@ -141,11 +142,15 @@ pub async fn process_exec_tool_call(
     codex_linux_sandbox_exe: &Option<PathBuf>,
     stdout_stream: Option<StdoutStream>,
 ) -> Result<ExecToolCallOutput> {
+    let windows_sandbox_level = params.windows_sandbox_level;
     let sandbox_type = match &sandbox_policy {
         SandboxPolicy::DangerFullAccess | SandboxPolicy::ExternalSandbox { .. } => {
             SandboxType::None
         }
-        _ => get_platform_sandbox().unwrap_or(SandboxType::None),
+        _ => get_platform_sandbox(
+            windows_sandbox_level != codex_protocol::config_types::WindowsSandboxLevel::Disabled,
+        )
+        .unwrap_or(SandboxType::None),
     };
     tracing::debug!("Sandbox type: {sandbox_type:?}");
 
@@ -155,6 +160,7 @@ pub async fn process_exec_tool_call(
         expiration,
         env,
         sandbox_permissions,
+        windows_sandbox_level,
         justification,
         arg0: _,
     } = params;
@@ -184,6 +190,7 @@ pub async fn process_exec_tool_call(
             sandbox_type,
             sandbox_cwd,
             codex_linux_sandbox_exe.as_ref(),
+            windows_sandbox_level,
         )
         .map_err(CodexErr::from)?;
 
@@ -202,6 +209,7 @@ pub(crate) async fn execute_exec_env(
         env,
         expiration,
         sandbox,
+        windows_sandbox_level,
         sandbox_permissions,
         justification,
         arg0,
@@ -213,6 +221,7 @@ pub(crate) async fn execute_exec_env(
         expiration,
         env,
         sandbox_permissions,
+        windows_sandbox_level,
         justification,
         arg0,
     };
@@ -229,7 +238,7 @@ async fn exec_windows_sandbox(
     sandbox_policy: &SandboxPolicy,
 ) -> Result<RawExecToolCallOutput> {
     use crate::config::find_codex_home;
-    use crate::safety::is_windows_elevated_sandbox_enabled;
+    use codex_protocol::config_types::WindowsSandboxLevel;
     use codex_windows_sandbox::run_windows_sandbox_capture;
     use codex_windows_sandbox::run_windows_sandbox_capture_elevated;
 
@@ -238,6 +247,7 @@ async fn exec_windows_sandbox(
         cwd,
         env,
         expiration,
+        windows_sandbox_level,
         ..
     } = params;
     // TODO(iceweasel-oai): run_windows_sandbox_capture should support all
@@ -255,7 +265,7 @@ async fn exec_windows_sandbox(
             "windows sandbox: failed to resolve codex_home: {err}"
         )))
     })?;
-    let use_elevated = is_windows_elevated_sandbox_enabled();
+    let use_elevated = matches!(windows_sandbox_level, WindowsSandboxLevel::Elevated);
     let spawn_res = tokio::task::spawn_blocking(move || {
         if use_elevated {
             run_windows_sandbox_capture_elevated(
@@ -312,20 +322,7 @@ async fn exec_windows_sandbox(
         text: stderr_text,
         truncated_after_lines: None,
     };
-    // Best-effort aggregate: stdout then stderr (capped).
-    let mut aggregated = Vec::with_capacity(
-        stdout
-            .text
-            .len()
-            .saturating_add(stderr.text.len())
-            .min(EXEC_OUTPUT_MAX_BYTES),
-    );
-    append_capped(&mut aggregated, &stdout.text, EXEC_OUTPUT_MAX_BYTES);
-    append_capped(&mut aggregated, &stderr.text, EXEC_OUTPUT_MAX_BYTES);
-    let aggregated_output = StreamOutput {
-        text: aggregated,
-        truncated_after_lines: None,
-    };
+    let aggregated_output = aggregate_output(&stdout, &stderr);
 
     Ok(RawExecToolCallOutput {
         exit_status,
@@ -519,6 +516,39 @@ fn append_capped(dst: &mut Vec<u8>, src: &[u8], max_bytes: usize) {
     dst.extend_from_slice(&src[..take]);
 }
 
+fn aggregate_output(
+    stdout: &StreamOutput<Vec<u8>>,
+    stderr: &StreamOutput<Vec<u8>>,
+) -> StreamOutput<Vec<u8>> {
+    let total_len = stdout.text.len().saturating_add(stderr.text.len());
+    let max_bytes = EXEC_OUTPUT_MAX_BYTES;
+    let mut aggregated = Vec::with_capacity(total_len.min(max_bytes));
+
+    if total_len <= max_bytes {
+        aggregated.extend_from_slice(&stdout.text);
+        aggregated.extend_from_slice(&stderr.text);
+        return StreamOutput {
+            text: aggregated,
+            truncated_after_lines: None,
+        };
+    }
+
+    // Under contention, reserve 1/3 for stdout and 2/3 for stderr; rebalance unused stderr to stdout.
+    let want_stdout = stdout.text.len().min(max_bytes / 3);
+    let want_stderr = stderr.text.len();
+    let stderr_take = want_stderr.min(max_bytes.saturating_sub(want_stdout));
+    let remaining = max_bytes.saturating_sub(want_stdout + stderr_take);
+    let stdout_take = want_stdout + remaining.min(stdout.text.len().saturating_sub(want_stdout));
+
+    aggregated.extend_from_slice(&stdout.text[..stdout_take]);
+    aggregated.extend_from_slice(&stderr.text[..stderr_take]);
+
+    StreamOutput {
+        text: aggregated,
+        truncated_after_lines: None,
+    }
+}
+
 #[derive(Clone, Debug)]
 pub struct ExecToolCallOutput {
     pub exit_code: i32,
@@ -564,6 +594,7 @@ async fn exec(
         env,
         arg0,
         expiration,
+        windows_sandbox_level: _,
         ..
     } = params;
 
@@ -683,20 +714,7 @@ async fn consume_truncated_output(
         Duration::from_millis(IO_DRAIN_TIMEOUT_MS),
     )
     .await?;
-    // Best-effort aggregate: stdout then stderr (capped).
-    let mut aggregated = Vec::with_capacity(
-        stdout
-            .text
-            .len()
-            .saturating_add(stderr.text.len())
-            .min(EXEC_OUTPUT_MAX_BYTES),
-    );
-    append_capped(&mut aggregated, &stdout.text, EXEC_OUTPUT_MAX_BYTES);
-    append_capped(&mut aggregated, &stderr.text, EXEC_OUTPUT_MAX_BYTES * 2);
-    let aggregated_output = StreamOutput {
-        text: aggregated,
-        truncated_after_lines: None,
-    };
+    let aggregated_output = aggregate_output(&stdout, &stderr);
 
     Ok(RawExecToolCallOutput {
         exit_status,
@@ -771,6 +789,7 @@ fn synthetic_exit_status(code: i32) -> ExitStatus {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use pretty_assertions::assert_eq;
     use std::time::Duration;
     use tokio::io::AsyncWriteExt;
 
@@ -846,6 +865,85 @@ mod tests {
         assert_eq!(out.text.len(), EXEC_OUTPUT_MAX_BYTES);
     }
 
+    #[test]
+    fn aggregate_output_prefers_stderr_on_contention() {
+        let stdout = StreamOutput {
+            text: vec![b'a'; EXEC_OUTPUT_MAX_BYTES],
+            truncated_after_lines: None,
+        };
+        let stderr = StreamOutput {
+            text: vec![b'b'; EXEC_OUTPUT_MAX_BYTES],
+            truncated_after_lines: None,
+        };
+
+        let aggregated = aggregate_output(&stdout, &stderr);
+        let stdout_cap = EXEC_OUTPUT_MAX_BYTES / 3;
+        let stderr_cap = EXEC_OUTPUT_MAX_BYTES.saturating_sub(stdout_cap);
+
+        assert_eq!(aggregated.text.len(), EXEC_OUTPUT_MAX_BYTES);
+        assert_eq!(aggregated.text[..stdout_cap], vec![b'a'; stdout_cap]);
+        assert_eq!(aggregated.text[stdout_cap..], vec![b'b'; stderr_cap]);
+    }
+
+    #[test]
+    fn aggregate_output_fills_remaining_capacity_with_stderr() {
+        let stdout_len = EXEC_OUTPUT_MAX_BYTES / 10;
+        let stdout = StreamOutput {
+            text: vec![b'a'; stdout_len],
+            truncated_after_lines: None,
+        };
+        let stderr = StreamOutput {
+            text: vec![b'b'; EXEC_OUTPUT_MAX_BYTES],
+            truncated_after_lines: None,
+        };
+
+        let aggregated = aggregate_output(&stdout, &stderr);
+        let stderr_cap = EXEC_OUTPUT_MAX_BYTES.saturating_sub(stdout_len);
+
+        assert_eq!(aggregated.text.len(), EXEC_OUTPUT_MAX_BYTES);
+        assert_eq!(aggregated.text[..stdout_len], vec![b'a'; stdout_len]);
+        assert_eq!(aggregated.text[stdout_len..], vec![b'b'; stderr_cap]);
+    }
+
+    #[test]
+    fn aggregate_output_rebalances_when_stderr_is_small() {
+        let stdout = StreamOutput {
+            text: vec![b'a'; EXEC_OUTPUT_MAX_BYTES],
+            truncated_after_lines: None,
+        };
+        let stderr = StreamOutput {
+            text: vec![b'b'; 1],
+            truncated_after_lines: None,
+        };
+
+        let aggregated = aggregate_output(&stdout, &stderr);
+        let stdout_len = EXEC_OUTPUT_MAX_BYTES.saturating_sub(1);
+
+        assert_eq!(aggregated.text.len(), EXEC_OUTPUT_MAX_BYTES);
+        assert_eq!(aggregated.text[..stdout_len], vec![b'a'; stdout_len]);
+        assert_eq!(aggregated.text[stdout_len..], vec![b'b'; 1]);
+    }
+
+    #[test]
+    fn aggregate_output_keeps_stdout_then_stderr_when_under_cap() {
+        let stdout = StreamOutput {
+            text: vec![b'a'; 4],
+            truncated_after_lines: None,
+        };
+        let stderr = StreamOutput {
+            text: vec![b'b'; 3],
+            truncated_after_lines: None,
+        };
+
+        let aggregated = aggregate_output(&stdout, &stderr);
+        let mut expected = Vec::new();
+        expected.extend_from_slice(&stdout.text);
+        expected.extend_from_slice(&stderr.text);
+
+        assert_eq!(aggregated.text, expected);
+        assert_eq!(aggregated.truncated_after_lines, None);
+    }
+
     #[cfg(unix)]
     #[test]
     fn sandbox_detection_flags_sigsys_exit_code() {
@@ -878,6 +976,7 @@ mod tests {
             expiration: 500.into(),
             env,
             sandbox_permissions: SandboxPermissions::UseDefault,
+            windows_sandbox_level: codex_protocol::config_types::WindowsSandboxLevel::Disabled,
             justification: None,
             arg0: None,
         };
@@ -923,6 +1022,7 @@ mod tests {
             expiration: ExecExpiration::Cancellation(cancel_token),
             env,
             sandbox_permissions: SandboxPermissions::UseDefault,
+            windows_sandbox_level: codex_protocol::config_types::WindowsSandboxLevel::Disabled,
             justification: None,
             arg0: None,
         };

codex-rs/core/src/features.rs

@@ -5,14 +5,20 @@
 //! booleans through multiple types, call sites consult a single `Features`
 //! container attached to `Config`.
 
+use crate::config::CONFIG_TOML_FILE;
+use crate::config::Config;
 use crate::config::ConfigToml;
 use crate::config::profile::ConfigProfile;
+use crate::protocol::Event;
+use crate::protocol::EventMsg;
+use crate::protocol::WarningEvent;
 use codex_otel::OtelManager;
 use schemars::JsonSchema;
 use serde::Deserialize;
 use serde::Serialize;
 use std::collections::BTreeMap;
 use std::collections::BTreeSet;
+use toml::Value as TomlValue;
 
 mod legacy;
 pub(crate) use legacy::LegacyFeatureToggles;
@@ -21,8 +27,8 @@ pub(crate) use legacy::legacy_feature_keys;
 /// High-level lifecycle stage for a feature.
 #[derive(Debug, Clone, Copy, PartialEq, Eq)]
 pub enum Stage {
-    /// Closed beta features to be used while developing or within the company.
-    Beta,
+    /// Features that are still under development, not ready for external use
+    UnderDevelopment,
     /// Experimental features made available to users through the `/experimental` menu
     Experimental {
         name: &'static str,
@@ -38,14 +44,14 @@ pub enum Stage {
 }
 
 impl Stage {
-    pub fn beta_menu_name(self) -> Option<&'static str> {
+    pub fn experimental_menu_name(self) -> Option<&'static str> {
         match self {
             Stage::Experimental { name, .. } => Some(name),
             _ => None,
         }
     }
 
-    pub fn beta_menu_description(self) -> Option<&'static str> {
+    pub fn experimental_menu_description(self) -> Option<&'static str> {
         match self {
             Stage::Experimental {
                 menu_description, ..
@@ -54,7 +60,7 @@ impl Stage {
         }
     }
 
-    pub fn beta_announcement(self) -> Option<&'static str> {
+    pub fn experimental_announcement(self) -> Option<&'static str> {
         match self {
             Stage::Experimental { announcement, .. } => Some(announcement),
             _ => None,
@@ -343,10 +349,10 @@ pub const FEATURES: &[FeatureSpec] = &[
     FeatureSpec {
         id: Feature::WebSearchCached,
         key: "web_search_cached",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: false,
     },
-    // Beta program. Rendered in the `/experimental` menu for users.
+    // Experimental program. Rendered in the `/experimental` menu for users.
     FeatureSpec {
         id: Feature::UnifiedExec,
         key: "unified_exec",
@@ -370,43 +376,43 @@ pub const FEATURES: &[FeatureSpec] = &[
     FeatureSpec {
         id: Feature::ChildAgentsMd,
         key: "child_agents_md",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: false,
     },
     FeatureSpec {
         id: Feature::ApplyPatchFreeform,
         key: "apply_patch_freeform",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: false,
     },
     FeatureSpec {
         id: Feature::ExecPolicy,
         key: "exec_policy",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: true,
     },
     FeatureSpec {
         id: Feature::WindowsSandbox,
         key: "experimental_windows_sandbox",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: false,
     },
     FeatureSpec {
         id: Feature::WindowsSandboxElevated,
         key: "elevated_windows_sandbox",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: false,
     },
     FeatureSpec {
         id: Feature::RemoteCompaction,
         key: "remote_compaction",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: true,
     },
     FeatureSpec {
         id: Feature::RemoteModels,
         key: "remote_models",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: true,
     },
     FeatureSpec {
@@ -421,26 +427,26 @@ pub const FEATURES: &[FeatureSpec] = &[
         #[cfg(windows)]
         default_enabled: true,
         #[cfg(not(windows))]
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         #[cfg(not(windows))]
         default_enabled: false,
     },
     FeatureSpec {
         id: Feature::EnableRequestCompression,
         key: "enable_request_compression",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: false,
     },
     FeatureSpec {
         id: Feature::Collab,
         key: "collab",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: false,
     },
     FeatureSpec {
         id: Feature::Connectors,
         key: "connectors",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: false,
     },
     FeatureSpec {
@@ -456,13 +462,64 @@ pub const FEATURES: &[FeatureSpec] = &[
     FeatureSpec {
         id: Feature::CollaborationModes,
         key: "collaboration_modes",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: false,
     },
     FeatureSpec {
         id: Feature::ResponsesWebsockets,
         key: "responses_websockets",
-        stage: Stage::Beta,
+        stage: Stage::UnderDevelopment,
         default_enabled: false,
     },
 ];
+
+/// Push a warning event if any under-development features are enabled.
+pub fn maybe_push_unstable_features_warning(
+    config: &Config,
+    post_session_configured_events: &mut Vec<Event>,
+) {
+    if config.suppress_unstable_features_warning {
+        return;
+    }
+
+    let mut under_development_feature_keys = Vec::new();
+    if let Some(table) = config
+        .config_layer_stack
+        .effective_config()
+        .get("features")
+        .and_then(TomlValue::as_table)
+    {
+        for (key, value) in table {
+            if value.as_bool() != Some(true) {
+                continue;
+            }
+            let Some(spec) = FEATURES.iter().find(|spec| spec.key == key.as_str()) else {
+                continue;
+            };
+            if !config.features.enabled(spec.id) {
+                continue;
+            }
+            if matches!(spec.stage, Stage::UnderDevelopment) {
+                under_development_feature_keys.push(spec.key.to_string());
+            }
+        }
+    }
+
+    if under_development_feature_keys.is_empty() {
+        return;
+    }
+
+    let under_development_feature_keys = under_development_feature_keys.join(", ");
+    let config_path = config
+        .codex_home
+        .join(CONFIG_TOML_FILE)
+        .display()
+        .to_string();
+    let message = format!(
+        "Under-development features enabled: {under_development_feature_keys}. Under-development features are incomplete and may behave unpredictably. To suppress this warning, set `suppress_unstable_features_warning = true` in {config_path}."
+    );
+    post_session_configured_events.push(Event {
+        id: "".to_owned(),
+        msg: EventMsg::Warning(WarningEvent { message }),
+    });
+}

codex-rs/core/src/lib.rs

@@ -69,6 +69,7 @@ mod event_mapping;
 pub mod review_format;
 pub mod review_prompts;
 mod thread_manager;
+pub mod web_search;
 pub use codex_protocol::protocol::InitialHistory;
 pub use thread_manager::NewThread;
 pub use thread_manager::ThreadManager;
@@ -98,6 +99,7 @@ pub use rollout::INTERACTIVE_SESSION_SOURCES;
 pub use rollout::RolloutRecorder;
 pub use rollout::SESSIONS_SUBDIR;
 pub use rollout::SessionMeta;
+pub use rollout::find_archived_thread_path_by_id_str;
 #[deprecated(note = "use find_thread_path_by_id_str")]
 pub use rollout::find_conversation_path_by_id_str;
 pub use rollout::find_thread_path_by_id_str;
@@ -108,6 +110,7 @@ pub use rollout::list::ThreadsPage;
 pub use rollout::list::parse_cursor;
 pub use rollout::list::read_head_for_summary;
 pub use rollout::list::read_session_meta_line;
+pub use rollout::rollout_date_parts;
 mod function_tool;
 mod state;
 mod tasks;
@@ -123,9 +126,7 @@ pub use exec_policy::ExecPolicyError;
 pub use exec_policy::check_execpolicy_for_warnings;
 pub use exec_policy::load_exec_policy;
 pub use safety::get_platform_sandbox;
-pub use safety::is_windows_elevated_sandbox_enabled;
-pub use safety::set_windows_elevated_sandbox_enabled;
-pub use safety::set_windows_sandbox_enabled;
+pub use tools::spec::parse_tool_input_schema;
 // Re-export the protocol types from the standalone `codex-protocol` crate so existing
 // `codex_core::protocol::...` references continue to work across the workspace.
 pub use codex_protocol::protocol;

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

@@ -97,6 +97,7 @@ fn codex_apps_mcp_server_config(config: &Config, auth: Option<&CodexAuth>) -> Mc
         tool_timeout_sec: None,
         enabled_tools: None,
         disabled_tools: None,
+        scopes: None,
     }
 }
 

codex-rs/core/src/mcp_connection_manager.rs

@@ -1182,6 +1182,7 @@ mod tests {
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
             auth_status: McpAuthStatus::Unsupported,
         };
@@ -1227,6 +1228,7 @@ mod tests {
                 tool_timeout_sec: None,
                 enabled_tools: None,
                 disabled_tools: None,
+                scopes: None,
             },
             auth_status: McpAuthStatus::Unsupported,
         };

codex-rs/core/src/models_manager/collaboration_mode_presets.rs

@@ -28,7 +28,7 @@ fn plan_preset() -> CollaborationModeMask {
         name: "Plan".to_string(),
         mode: Some(ModeKind::Plan),
         model: None,
-        reasoning_effort: Some(Some(ReasoningEffort::High)),
+        reasoning_effort: Some(Some(ReasoningEffort::Medium)),
         developer_instructions: Some(Some(COLLABORATION_MODE_PLAN.to_string())),
     }
 }

codex-rs/core/src/models_manager/model_info.rs

@@ -169,6 +169,16 @@ pub(crate) fn find_model_info_for_slug(slug: &str) -> ModelInfo {
         model_info!(
             slug,
             base_instructions: GPT_5_2_CODEX_INSTRUCTIONS.to_string(),
+            model_instructions_template: Some(ModelInstructionsTemplate {
+                template: GPT_5_2_CODEX_INSTRUCTIONS_TEMPLATE.to_string(),
+                personality_messages: Some(PersonalityMessages(BTreeMap::from([(
+                    Personality::Friendly,
+                    PERSONALITY_FRIENDLY.to_string(),
+                ), (
+                    Personality::Pragmatic,
+                    PERSONALITY_PRAGMATIC.to_string(),
+                )]))),
+            }),
             apply_patch_tool_type: Some(ApplyPatchToolType::Freeform),
             shell_type: ConfigShellToolType::ShellCommand,
             supports_parallel_tool_calls: true,
@@ -203,16 +213,6 @@ pub(crate) fn find_model_info_for_slug(slug: &str) -> ModelInfo {
             truncation_policy: TruncationPolicyConfig::tokens(10_000),
             context_window: Some(CONTEXT_WINDOW_272K),
             supported_reasoning_levels: supported_reasoning_level_low_medium_high_xhigh(),
-            model_instructions_template: Some(ModelInstructionsTemplate {
-                template: GPT_5_2_CODEX_INSTRUCTIONS_TEMPLATE.to_string(),
-                personality_messages: Some(PersonalityMessages(BTreeMap::from([(
-                    Personality::Friendly,
-                    PERSONALITY_FRIENDLY.to_string(),
-                ), (
-                    Personality::Pragmatic,
-                    PERSONALITY_PRAGMATIC.to_string(),
-                )]))),
-            }),
         )
     } else if slug.starts_with("gpt-5.1-codex-max") {
         model_info!(

codex-rs/core/src/models_manager/model_presets.rs

@@ -36,6 +36,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     description: "Extra high reasoning depth for complex problems".to_string(),
                 },
             ],
+            supports_personality: true,
             is_default: true,
             upgrade: None,
             show_in_picker: true,
@@ -65,6 +66,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     description: "Extra high reasoning depth for complex problems".to_string(),
                 },
             ],
+            supports_personality: false,
             is_default: false,
             upgrade: Some(gpt_52_codex_upgrade()),
             show_in_picker: true,
@@ -87,6 +89,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                         .to_string(),
                 },
             ],
+            supports_personality: false,
             is_default: false,
             upgrade: Some(gpt_52_codex_upgrade()),
             show_in_picker: true,
@@ -116,6 +119,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     description: "Extra high reasoning depth for complex problems".to_string(),
                 },
             ],
+            supports_personality: false,
             is_default: false,
             upgrade: Some(gpt_52_codex_upgrade()),
             show_in_picker: true,
@@ -145,6 +149,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     description: "Extra high reasoning depth for complex problems".to_string(),
                 },
             ],
+            supports_personality: true,
             is_default: false,
             upgrade: None,
             show_in_picker: false,
@@ -174,6 +179,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     description: "Extra high reasoning depth for complex problems".to_string(),
                 },
             ],
+            supports_personality: false,
             is_default: false,
             upgrade: None,
             show_in_picker: false,
@@ -200,6 +206,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     description: "Maximizes reasoning depth for complex or ambiguous problems".to_string(),
                 },
             ],
+            supports_personality: false,
             is_default: false,
             upgrade: Some(gpt_52_codex_upgrade()),
             show_in_picker: false,
@@ -221,6 +228,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     description: "Maximizes reasoning depth for complex or ambiguous problems".to_string(),
                 },
             ],
+            supports_personality: false,
             is_default: false,
             upgrade: Some(gpt_52_codex_upgrade()),
             show_in_picker: false,
@@ -247,6 +255,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                         .to_string(),
                 },
             ],
+            supports_personality: false,
             is_default: false,
             upgrade: Some(gpt_52_codex_upgrade()),
             show_in_picker: false,
@@ -276,6 +285,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     description: "Maximizes reasoning depth for complex or ambiguous problems".to_string(),
                 },
             ],
+            supports_personality: false,
             is_default: false,
             upgrade: Some(gpt_52_codex_upgrade()),
             show_in_picker: false,
@@ -301,6 +311,7 @@ static PRESETS: Lazy<Vec<ModelPreset>> = Lazy::new(|| {
                     description: "Maximizes reasoning depth for complex or ambiguous problems".to_string(),
                 },
             ],
+            supports_personality: false,
             is_default: false,
             upgrade: Some(gpt_52_codex_upgrade()),
             show_in_picker: false,

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

@@ -1,4 +1,5 @@
 use std::cmp::Reverse;
+use std::ffi::OsStr;
 use std::io::{self};
 use std::num::NonZero;
 use std::ops::ControlFlow;
@@ -15,6 +16,7 @@ use time::format_description::well_known::Rfc3339;
 use time::macros::format_description;
 use uuid::Uuid;
 
+use super::ARCHIVED_SESSIONS_SUBDIR;
 use super::SESSIONS_SUBDIR;
 use crate::protocol::EventMsg;
 use codex_file_search as file_search;
@@ -1054,11 +1056,9 @@ fn truncate_to_seconds(dt: OffsetDateTime) -> Option<OffsetDateTime> {
     dt.replace_nanosecond(0).ok()
 }
 
-/// Locate a recorded thread rollout file by its UUID string using the existing
-/// paginated listing implementation. Returns `Ok(Some(path))` if found, `Ok(None)` if not present
-/// or the id is invalid.
-pub async fn find_thread_path_by_id_str(
+async fn find_thread_path_by_id_str_in_subdir(
     codex_home: &Path,
+    subdir: &str,
     id_str: &str,
 ) -> io::Result<Option<PathBuf>> {
     // Validate UUID format early.
@@ -1067,7 +1067,7 @@ pub async fn find_thread_path_by_id_str(
     }
 
     let mut root = codex_home.to_path_buf();
-    root.push(SESSIONS_SUBDIR);
+    root.push(subdir);
     if !root.exists() {
         return Ok(None);
     }
@@ -1099,3 +1099,31 @@ pub async fn find_thread_path_by_id_str(
         .next()
         .map(|m| root.join(m.path)))
 }
+
+/// Locate a recorded thread rollout file by its UUID string using the existing
+/// paginated listing implementation. Returns `Ok(Some(path))` if found, `Ok(None)` if not present
+/// or the id is invalid.
+pub async fn find_thread_path_by_id_str(
+    codex_home: &Path,
+    id_str: &str,
+) -> io::Result<Option<PathBuf>> {
+    find_thread_path_by_id_str_in_subdir(codex_home, SESSIONS_SUBDIR, id_str).await
+}
+
+/// Locate an archived thread rollout file by its UUID string.
+pub async fn find_archived_thread_path_by_id_str(
+    codex_home: &Path,
+    id_str: &str,
+) -> io::Result<Option<PathBuf>> {
+    find_thread_path_by_id_str_in_subdir(codex_home, ARCHIVED_SESSIONS_SUBDIR, id_str).await
+}
+
+/// Extract the `YYYY/MM/DD` directory components from a rollout filename.
+pub fn rollout_date_parts(file_name: &OsStr) -> Option<(String, String, String)> {
+    let name = file_name.to_string_lossy();
+    let date = name.strip_prefix("rollout-")?.get(..10)?;
+    let year = date.get(..4)?.to_string();
+    let month = date.get(5..7)?.to_string();
+    let day = date.get(8..10)?.to_string();
+    Some((year, month, day))
+}

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

@@ -15,9 +15,11 @@ pub(crate) mod truncation;
 
 pub use codex_protocol::protocol::SessionMeta;
 pub(crate) use error::map_session_init_error;
+pub use list::find_archived_thread_path_by_id_str;
 pub use list::find_thread_path_by_id_str;
 #[deprecated(note = "use find_thread_path_by_id_str")]
 pub use list::find_thread_path_by_id_str as find_conversation_path_by_id_str;
+pub use list::rollout_date_parts;
 pub use recorder::RolloutRecorder;
 pub use recorder::RolloutRecorderParams;
 

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

@@ -42,7 +42,8 @@ pub(crate) fn should_persist_event_msg(ev: &EventMsg) -> bool {
         | EventMsg::AgentReasoning(_)
         | EventMsg::AgentReasoningRawContent(_)
         | EventMsg::TokenCount(_)
-        | EventMsg::ContextCompacted(_)
+        | EventMsg::ContextCompactionStarted(_)
+        | EventMsg::ContextCompactionEnded(_)
         | EventMsg::EnteredReviewMode(_)
         | EventMsg::ExitedReviewMode(_)
         | EventMsg::ThreadRolledBack(_)
@@ -68,6 +69,7 @@ pub(crate) fn should_persist_event_msg(ev: &EventMsg) -> bool {
         | EventMsg::ExecCommandEnd(_)
         | EventMsg::ExecApprovalRequest(_)
         | EventMsg::RequestUserInput(_)
+        | EventMsg::DynamicToolCallRequest(_)
         | EventMsg::ElicitationRequest(_)
         | EventMsg::ApplyPatchApprovalRequest(_)
         | EventMsg::BackgroundEvent(_)

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

@@ -1,5 +1,6 @@
 #![allow(clippy::unwrap_used, clippy::expect_used)]
 
+use std::ffi::OsStr;
 use std::fs::File;
 use std::fs::FileTimes;
 use std::fs::{self};
@@ -21,6 +22,7 @@ use crate::rollout::list::ThreadItem;
 use crate::rollout::list::ThreadSortKey;
 use crate::rollout::list::ThreadsPage;
 use crate::rollout::list::get_threads;
+use crate::rollout::rollout_date_parts;
 use anyhow::Result;
 use codex_protocol::ThreadId;
 use codex_protocol::models::ContentItem;
@@ -43,6 +45,16 @@ fn provider_vec(providers: &[&str]) -> Vec<String> {
         .collect()
 }
 
+#[test]
+fn rollout_date_parts_extracts_directory_components() {
+    let file_name = OsStr::new("rollout-2025-03-01T09-00-00-123.jsonl");
+    let parts = rollout_date_parts(file_name);
+    assert_eq!(
+        parts,
+        Some(("2025".to_string(), "03".to_string(), "01".to_string()))
+    );
+}
+
 fn write_session_file(
     root: &Path,
     ts_str: &str,

codex-rs/core/src/safety.rs

@@ -10,45 +10,7 @@ use crate::util::resolve_path;
 
 use crate::protocol::AskForApproval;
 use crate::protocol::SandboxPolicy;
-
-#[cfg(target_os = "windows")]
-use std::sync::atomic::AtomicBool;
-#[cfg(target_os = "windows")]
-use std::sync::atomic::Ordering;
-
-#[cfg(target_os = "windows")]
-static WINDOWS_SANDBOX_ENABLED: AtomicBool = AtomicBool::new(false);
-#[cfg(target_os = "windows")]
-static WINDOWS_ELEVATED_SANDBOX_ENABLED: AtomicBool = AtomicBool::new(false);
-
-#[cfg(target_os = "windows")]
-pub fn set_windows_sandbox_enabled(enabled: bool) {
-    WINDOWS_SANDBOX_ENABLED.store(enabled, Ordering::Relaxed);
-}
-
-#[cfg(not(target_os = "windows"))]
-#[allow(dead_code)]
-pub fn set_windows_sandbox_enabled(_enabled: bool) {}
-
-#[cfg(target_os = "windows")]
-pub fn set_windows_elevated_sandbox_enabled(enabled: bool) {
-    WINDOWS_ELEVATED_SANDBOX_ENABLED.store(enabled, Ordering::Relaxed);
-}
-
-#[cfg(not(target_os = "windows"))]
-#[allow(dead_code)]
-pub fn set_windows_elevated_sandbox_enabled(_enabled: bool) {}
-
-#[cfg(target_os = "windows")]
-pub fn is_windows_elevated_sandbox_enabled() -> bool {
-    WINDOWS_ELEVATED_SANDBOX_ENABLED.load(Ordering::Relaxed)
-}
-
-#[cfg(not(target_os = "windows"))]
-#[allow(dead_code)]
-pub fn is_windows_elevated_sandbox_enabled() -> bool {
-    false
-}
+use codex_protocol::config_types::WindowsSandboxLevel;
 
 #[derive(Debug, PartialEq)]
 pub enum SafetyCheck {
@@ -67,6 +29,7 @@ pub fn assess_patch_safety(
     policy: AskForApproval,
     sandbox_policy: &SandboxPolicy,
     cwd: &Path,
+    windows_sandbox_level: WindowsSandboxLevel,
 ) -> SafetyCheck {
     if action.is_empty() {
         return SafetyCheck::Reject {
@@ -104,7 +67,7 @@ pub fn assess_patch_safety(
             // Only auto‑approve when we can actually enforce a sandbox. Otherwise
             // fall back to asking the user because the patch may touch arbitrary
             // paths outside the project.
-            match get_platform_sandbox() {
+            match get_platform_sandbox(windows_sandbox_level != WindowsSandboxLevel::Disabled) {
                 Some(sandbox_type) => SafetyCheck::AutoApprove {
                     sandbox_type,
                     user_explicitly_approved: false,
@@ -122,19 +85,17 @@ pub fn assess_patch_safety(
     }
 }
 
-pub fn get_platform_sandbox() -> Option<SandboxType> {
+pub fn get_platform_sandbox(windows_sandbox_enabled: bool) -> Option<SandboxType> {
     if cfg!(target_os = "macos") {
         Some(SandboxType::MacosSeatbelt)
     } else if cfg!(target_os = "linux") {
         Some(SandboxType::LinuxSeccomp)
     } else if cfg!(target_os = "windows") {
-        #[cfg(target_os = "windows")]
-        {
-            if WINDOWS_SANDBOX_ENABLED.load(Ordering::Relaxed) {
-                return Some(SandboxType::WindowsRestrictedToken);
-            }
+        if windows_sandbox_enabled {
+            Some(SandboxType::WindowsRestrictedToken)
+        } else {
+            None
         }
-        None
     } else {
         None
     }
@@ -277,7 +238,13 @@ mod tests {
         };
 
         assert_eq!(
-            assess_patch_safety(&add_inside, AskForApproval::OnRequest, &policy, &cwd),
+            assess_patch_safety(
+                &add_inside,
+                AskForApproval::OnRequest,
+                &policy,
+                &cwd,
+                WindowsSandboxLevel::Disabled
+            ),
             SafetyCheck::AutoApprove {
                 sandbox_type: SandboxType::None,
                 user_explicitly_approved: false,

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

@@ -21,6 +21,7 @@ use crate::seatbelt::create_seatbelt_command_args;
 use crate::spawn::CODEX_SANDBOX_ENV_VAR;
 use crate::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use crate::tools::sandboxing::SandboxablePreference;
+use codex_protocol::config_types::WindowsSandboxLevel;
 pub use codex_protocol::models::SandboxPermissions;
 use std::collections::HashMap;
 use std::path::Path;
@@ -44,6 +45,7 @@ pub struct ExecEnv {
     pub env: HashMap<String, String>,
     pub expiration: ExecExpiration,
     pub sandbox: SandboxType,
+    pub windows_sandbox_level: WindowsSandboxLevel,
     pub sandbox_permissions: SandboxPermissions,
     pub justification: Option<String>,
     pub arg0: Option<String>,
@@ -76,19 +78,26 @@ impl SandboxManager {
         &self,
         policy: &SandboxPolicy,
         pref: SandboxablePreference,
+        windows_sandbox_level: WindowsSandboxLevel,
     ) -> SandboxType {
         match pref {
             SandboxablePreference::Forbid => SandboxType::None,
             SandboxablePreference::Require => {
                 // Require a platform sandbox when available; on Windows this
                 // respects the experimental_windows_sandbox feature.
-                crate::safety::get_platform_sandbox().unwrap_or(SandboxType::None)
+                crate::safety::get_platform_sandbox(
+                    windows_sandbox_level != WindowsSandboxLevel::Disabled,
+                )
+                .unwrap_or(SandboxType::None)
             }
             SandboxablePreference::Auto => match policy {
                 SandboxPolicy::DangerFullAccess | SandboxPolicy::ExternalSandbox { .. } => {
                     SandboxType::None
                 }
-                _ => crate::safety::get_platform_sandbox().unwrap_or(SandboxType::None),
+                _ => crate::safety::get_platform_sandbox(
+                    windows_sandbox_level != WindowsSandboxLevel::Disabled,
+                )
+                .unwrap_or(SandboxType::None),
             },
         }
     }
@@ -100,6 +109,7 @@ impl SandboxManager {
         sandbox: SandboxType,
         sandbox_policy_cwd: &Path,
         codex_linux_sandbox_exe: Option<&PathBuf>,
+        windows_sandbox_level: WindowsSandboxLevel,
     ) -> Result<ExecEnv, SandboxTransformError> {
         let mut env = spec.env;
         if !policy.has_full_network_access() {
@@ -160,6 +170,7 @@ impl SandboxManager {
             env,
             expiration: spec.expiration,
             sandbox,
+            windows_sandbox_level,
             sandbox_permissions: spec.sandbox_permissions,
             justification: spec.justification,
             arg0: arg0_override,

codex-rs/core/src/shell_snapshot.rs

@@ -464,8 +464,6 @@ mod tests {
     use pretty_assertions::assert_eq;
     #[cfg(unix)]
     use std::os::unix::ffi::OsStrExt;
-    #[cfg(target_os = "linux")]
-    use std::os::unix::fs::PermissionsExt;
     #[cfg(unix)]
     use std::process::Command;
     #[cfg(target_os = "linux")]
@@ -562,27 +560,16 @@ mod tests {
         use tokio::time::sleep;
 
         let dir = tempdir()?;
-        let shell_path = dir.path().join("hanging-shell.sh");
         let pid_path = dir.path().join("pid");
-
-        let script = format!(
-            "#!/bin/sh\n\
-             echo $$ > {}\n\
-             sleep 30\n",
-            pid_path.display()
-        );
-        fs::write(&shell_path, script).await?;
-        let mut permissions = std::fs::metadata(&shell_path)?.permissions();
-        permissions.set_mode(0o755);
-        std::fs::set_permissions(&shell_path, permissions)?;
+        let script = format!("echo $$ > \"{}\"; sleep 30", pid_path.display());
 
         let shell = Shell {
             shell_type: ShellType::Sh,
-            shell_path,
+            shell_path: PathBuf::from("/bin/sh"),
             shell_snapshot: crate::shell::empty_shell_snapshot_receiver(),
         };
 
-        let err = run_script_with_timeout(&shell, "ignored", Duration::from_millis(500), true)
+        let err = run_script_with_timeout(&shell, &script, Duration::from_secs(1), true)
             .await
             .expect_err("snapshot shell should time out");
         assert!(

codex-rs/core/src/skills/loader.rs

@@ -35,7 +35,7 @@ struct SkillFrontmatterMetadata {
 }
 
 #[derive(Debug, Default, Deserialize)]
-struct SkillToml {
+struct SkillMetadataFile {
     #[serde(default)]
     interface: Option<Interface>,
 }
@@ -51,7 +51,7 @@ struct Interface {
 }
 
 const SKILLS_FILENAME: &str = "SKILL.md";
-const SKILLS_TOML_FILENAME: &str = "SKILL.toml";
+const SKILLS_JSON_FILENAME: &str = "SKILL.json";
 const SKILLS_DIR_NAME: &str = "skills";
 const MAX_NAME_LEN: usize = 64;
 const MAX_DESCRIPTION_LEN: usize = 1024;
@@ -370,9 +370,9 @@ fn parse_skill_file(path: &Path, scope: SkillScope) -> Result<SkillMetadata, Ski
 }
 
 fn load_skill_interface(skill_path: &Path) -> Option<SkillInterface> {
-    // Fail open: optional SKILL.toml metadata should not block loading SKILL.md.
+    // Fail open: optional interface metadata should not block loading SKILL.md.
     let skill_dir = skill_path.parent()?;
-    let interface_path = skill_dir.join(SKILLS_TOML_FILENAME);
+    let interface_path = skill_dir.join(SKILLS_JSON_FILENAME);
     if !interface_path.exists() {
         return None;
     }
@@ -381,17 +381,17 @@ fn load_skill_interface(skill_path: &Path) -> Option<SkillInterface> {
         Ok(contents) => contents,
         Err(error) => {
             tracing::warn!(
-                "ignoring {path}: failed to read SKILL.toml: {error}",
+                "ignoring {path}: failed to read SKILL.json: {error}",
                 path = interface_path.display()
             );
             return None;
         }
     };
-    let parsed: SkillToml = match toml::from_str(&contents) {
+    let parsed: SkillMetadataFile = match serde_json::from_str(&contents) {
         Ok(parsed) => parsed,
         Err(error) => {
             tracing::warn!(
-                "ignoring {path}: invalid TOML: {error}",
+                "ignoring {path}: invalid JSON: {error}",
                 path = interface_path.display()
             );
             return None;
@@ -756,7 +756,7 @@ mod tests {
     }
 
     fn write_skill_interface_at(skill_dir: &Path, contents: &str) -> PathBuf {
-        let path = skill_dir.join(SKILLS_TOML_FILENAME);
+        let path = skill_dir.join(SKILLS_JSON_FILENAME);
         fs::write(&path, contents).unwrap();
         path
     }
@@ -764,20 +764,23 @@ mod tests {
     #[tokio::test]
     async fn loads_skill_interface_metadata_happy_path() {
         let codex_home = tempfile::tempdir().expect("tempdir");
-        let skill_path = write_skill(&codex_home, "demo", "ui-skill", "from toml");
+        let skill_path = write_skill(&codex_home, "demo", "ui-skill", "from json");
         let skill_dir = skill_path.parent().expect("skill dir");
         let normalized_skill_dir = normalized(skill_dir);
 
         write_skill_interface_at(
             skill_dir,
             r##"
-[interface]
-display_name = "UI Skill"
-short_description = "  short    desc   "
-icon_small = "./assets/small-400px.png"
-icon_large = "./assets/large-logo.svg"
-brand_color = "#3B82F6"
-default_prompt = "  default   prompt   "
+{
+  "interface": {
+    "display_name": "UI Skill",
+    "short_description": "  short    desc   ",
+    "icon_small": "./assets/small-400px.png",
+    "icon_large": "./assets/large-logo.svg",
+    "brand_color": "#3B82F6",
+    "default_prompt": "  default   prompt   "
+  }
+}
 "##,
         );
 
@@ -793,7 +796,7 @@ default_prompt = "  default   prompt   "
             outcome.skills,
             vec![SkillMetadata {
                 name: "ui-skill".to_string(),
-                description: "from toml".to_string(),
+                description: "from json".to_string(),
                 short_description: None,
                 interface: Some(SkillInterface {
                     display_name: Some("UI Skill".to_string()),
@@ -803,7 +806,7 @@ default_prompt = "  default   prompt   "
                     brand_color: Some("#3B82F6".to_string()),
                     default_prompt: Some("default prompt".to_string()),
                 }),
-                path: normalized(&skill_path),
+                path: normalized(skill_path.as_path()),
                 scope: SkillScope::User,
             }]
         );
@@ -812,17 +815,20 @@ default_prompt = "  default   prompt   "
     #[tokio::test]
     async fn accepts_icon_paths_under_assets_dir() {
         let codex_home = tempfile::tempdir().expect("tempdir");
-        let skill_path = write_skill(&codex_home, "demo", "ui-skill", "from toml");
+        let skill_path = write_skill(&codex_home, "demo", "ui-skill", "from json");
         let skill_dir = skill_path.parent().expect("skill dir");
         let normalized_skill_dir = normalized(skill_dir);
 
         write_skill_interface_at(
             skill_dir,
             r#"
-[interface]
-display_name = "UI Skill"
-icon_small = "assets/icon.png"
-icon_large = "./assets/logo.svg"
+{
+  "interface": {
+    "display_name": "UI Skill",
+    "icon_small": "assets/icon.png",
+    "icon_large": "./assets/logo.svg"
+  }
+}
 "#,
         );
 
@@ -838,7 +844,7 @@ icon_large = "./assets/logo.svg"
             outcome.skills,
             vec![SkillMetadata {
                 name: "ui-skill".to_string(),
-                description: "from toml".to_string(),
+                description: "from json".to_string(),
                 short_description: None,
                 interface: Some(SkillInterface {
                     display_name: Some("UI Skill".to_string()),
@@ -857,14 +863,17 @@ icon_large = "./assets/logo.svg"
     #[tokio::test]
     async fn ignores_invalid_brand_color() {
         let codex_home = tempfile::tempdir().expect("tempdir");
-        let skill_path = write_skill(&codex_home, "demo", "ui-skill", "from toml");
+        let skill_path = write_skill(&codex_home, "demo", "ui-skill", "from json");
         let skill_dir = skill_path.parent().expect("skill dir");
 
         write_skill_interface_at(
             skill_dir,
             r#"
-[interface]
-brand_color = "blue"
+{
+  "interface": {
+    "brand_color": "blue"
+  }
+}
 "#,
         );
 
@@ -880,7 +889,7 @@ brand_color = "blue"
             outcome.skills,
             vec![SkillMetadata {
                 name: "ui-skill".to_string(),
-                description: "from toml".to_string(),
+                description: "from json".to_string(),
                 short_description: None,
                 interface: None,
                 path: normalized(&skill_path),
@@ -892,7 +901,7 @@ brand_color = "blue"
     #[tokio::test]
     async fn ignores_default_prompt_over_max_length() {
         let codex_home = tempfile::tempdir().expect("tempdir");
-        let skill_path = write_skill(&codex_home, "demo", "ui-skill", "from toml");
+        let skill_path = write_skill(&codex_home, "demo", "ui-skill", "from json");
         let skill_dir = skill_path.parent().expect("skill dir");
         let normalized_skill_dir = normalized(skill_dir);
         let too_long = "x".repeat(MAX_DEFAULT_PROMPT_LEN + 1);
@@ -901,10 +910,13 @@ brand_color = "blue"
             skill_dir,
             &format!(
                 r##"
-[interface]
-display_name = "UI Skill"
-icon_small = "./assets/small-400px.png"
-default_prompt = "{too_long}"
+{{
+  "interface": {{
+    "display_name": "UI Skill",
+    "icon_small": "./assets/small-400px.png",
+    "default_prompt": "{too_long}"
+  }}
+}}
 "##
             ),
         );
@@ -921,7 +933,7 @@ default_prompt = "{too_long}"
             outcome.skills,
             vec![SkillMetadata {
                 name: "ui-skill".to_string(),
-                description: "from toml".to_string(),
+                description: "from json".to_string(),
                 short_description: None,
                 interface: Some(SkillInterface {
                     display_name: Some("UI Skill".to_string()),
@@ -940,15 +952,18 @@ default_prompt = "{too_long}"
     #[tokio::test]
     async fn drops_interface_when_icons_are_invalid() {
         let codex_home = tempfile::tempdir().expect("tempdir");
-        let skill_path = write_skill(&codex_home, "demo", "ui-skill", "from toml");
+        let skill_path = write_skill(&codex_home, "demo", "ui-skill", "from json");
         let skill_dir = skill_path.parent().expect("skill dir");
 
         write_skill_interface_at(
             skill_dir,
             r#"
-[interface]
-icon_small = "icon.png"
-icon_large = "./assets/../logo.svg"
+{
+  "interface": {
+    "icon_small": "icon.png",
+    "icon_large": "./assets/../logo.svg"
+  }
+}
 "#,
         );
 
@@ -964,7 +979,7 @@ icon_large = "./assets/../logo.svg"
             outcome.skills,
             vec![SkillMetadata {
                 name: "ui-skill".to_string(),
-                description: "from toml".to_string(),
+                description: "from json".to_string(),
                 short_description: None,
                 interface: None,
                 path: normalized(&skill_path),

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

@@ -15,6 +15,11 @@ pub(crate) struct SessionState {
     pub(crate) history: ContextManager,
     pub(crate) latest_rate_limits: Option<RateLimitSnapshot>,
     pub(crate) server_reasoning_included: bool,
+    /// Whether the session's initial context has been seeded into history.
+    ///
+    /// TODO(owen): This is a temporary solution to avoid updating a thread's updated_at
+    /// timestamp when resuming a session. Remove this once SQLite is in place.
+    pub(crate) initial_context_seeded: bool,
 }
 
 impl SessionState {
@@ -26,6 +31,7 @@ impl SessionState {
             history,
             latest_rate_limits: None,
             server_reasoning_included: false,
+            initial_context_seeded: false,
         }
     }
 

codex-rs/core/src/state/turn.rs

@@ -8,6 +8,7 @@ use tokio::sync::Notify;
 use tokio_util::sync::CancellationToken;
 use tokio_util::task::AbortOnDropHandle;
 
+use codex_protocol::dynamic_tools::DynamicToolResponse;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::request_user_input::RequestUserInputResponse;
 use tokio::sync::oneshot;
@@ -70,6 +71,7 @@ impl ActiveTurn {
 pub(crate) struct TurnState {
     pending_approvals: HashMap<String, oneshot::Sender<ReviewDecision>>,
     pending_user_input: HashMap<String, oneshot::Sender<RequestUserInputResponse>>,
+    pending_dynamic_tools: HashMap<String, oneshot::Sender<DynamicToolResponse>>,
     pending_input: Vec<ResponseInputItem>,
 }
 
@@ -92,6 +94,7 @@ impl TurnState {
     pub(crate) fn clear_pending(&mut self) {
         self.pending_approvals.clear();
         self.pending_user_input.clear();
+        self.pending_dynamic_tools.clear();
         self.pending_input.clear();
     }
 
@@ -110,6 +113,21 @@ impl TurnState {
         self.pending_user_input.remove(key)
     }
 
+    pub(crate) fn insert_pending_dynamic_tool(
+        &mut self,
+        key: String,
+        tx: oneshot::Sender<DynamicToolResponse>,
+    ) -> Option<oneshot::Sender<DynamicToolResponse>> {
+        self.pending_dynamic_tools.insert(key, tx)
+    }
+
+    pub(crate) fn remove_pending_dynamic_tool(
+        &mut self,
+        key: &str,
+    ) -> Option<oneshot::Sender<DynamicToolResponse>> {
+        self.pending_dynamic_tools.remove(key)
+    }
+
     pub(crate) fn push_pending_input(&mut self, input: ResponseInputItem) {
         self.pending_input.push(input);
     }

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

@@ -41,7 +41,7 @@ pub(crate) use undo::UndoTask;
 pub(crate) use user_shell::UserShellCommandTask;
 
 const GRACEFULL_INTERRUPTION_TIMEOUT_MS: u64 = 100;
-const TURN_ABORTED_INTERRUPTED_GUIDANCE: &str = "The user interrupted the previous turn. Do not continue or repeat work from that turn unless the user explicitly asks. If any tools/commands were aborted, they may have partially executed; verify current state before retrying.";
+const TURN_ABORTED_INTERRUPTED_GUIDANCE: &str = "The user interrupted the previous turn on purpose. If any tools/commands were aborted, they may have partially executed; verify current state before retrying.";
 
 /// Thin wrapper that exposes the parts of [`Session`] task runners need.
 #[derive(Clone)]
@@ -115,6 +115,8 @@ impl Session {
         task: T,
     ) {
         self.abort_all_tasks(TurnAbortReason::Replaced).await;
+        self.seed_initial_context_if_needed(turn_context.as_ref())
+            .await;
 
         let task: Arc<dyn SessionTask> = Arc::new(task);
         let task_kind = task.kind();
@@ -253,7 +255,7 @@ impl Session {
                 role: "user".to_string(),
                 content: vec![ContentItem::InputText {
                     text: format!(
-                        "{TURN_ABORTED_OPEN_TAG}\n  <turn_id>{sub_id}</turn_id>\n  <reason>interrupted</reason>\n  <guidance>{TURN_ABORTED_INTERRUPTED_GUIDANCE}</guidance>\n</turn_aborted>"
+                        "{TURN_ABORTED_OPEN_TAG}\n{TURN_ABORTED_INTERRUPTED_GUIDANCE}\n</turn_aborted>"
                     ),
                 }],
                 end_turn: None,

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

@@ -86,7 +86,7 @@ async fn start_review_conversation(
     let mut sub_agent_config = config.as_ref().clone();
     // Carry over review-only feature restrictions so the delegate cannot
     // re-enable blocked tools (web search, view image).
-    sub_agent_config.web_search_mode = Some(WebSearchMode::Disabled);
+    sub_agent_config.web_search_mode = WebSearchMode::Disabled;
 
     // Set explicit review rubric for the sub-agent
     sub_agent_config.base_instructions = Some(crate::REVIEW_PROMPT.to_string());

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

@@ -109,6 +109,7 @@ impl SessionTask for UserShellCommandTask {
             // should use that instead of an "arbitrarily large" timeout here.
             expiration: USER_SHELL_TIMEOUT_MS.into(),
             sandbox: SandboxType::None,
+            windows_sandbox_level: turn_context.windows_sandbox_level,
             sandbox_permissions: SandboxPermissions::UseDefault,
             justification: None,
             arg0: None,

codex-rs/core/src/thread_manager.rs

@@ -196,12 +196,21 @@ impl ThreadManager {
     }
 
     pub async fn start_thread(&self, config: Config) -> CodexResult<NewThread> {
+        self.start_thread_with_tools(config, Vec::new()).await
+    }
+
+    pub async fn start_thread_with_tools(
+        &self,
+        config: Config,
+        dynamic_tools: Vec<codex_protocol::dynamic_tools::DynamicToolSpec>,
+    ) -> CodexResult<NewThread> {
         self.state
             .spawn_thread(
                 config,
                 InitialHistory::New,
                 Arc::clone(&self.state.auth_manager),
                 self.agent_control(),
+                dynamic_tools,
             )
             .await
     }
@@ -224,7 +233,13 @@ impl ThreadManager {
         auth_manager: Arc<AuthManager>,
     ) -> CodexResult<NewThread> {
         self.state
-            .spawn_thread(config, initial_history, auth_manager, self.agent_control())
+            .spawn_thread(
+                config,
+                initial_history,
+                auth_manager,
+                self.agent_control(),
+                Vec::new(),
+            )
             .await
     }
 
@@ -262,6 +277,7 @@ impl ThreadManager {
                 history,
                 Arc::clone(&self.state.auth_manager),
                 self.agent_control(),
+                Vec::new(),
             )
             .await
     }
@@ -330,6 +346,7 @@ impl ThreadManagerState {
             Arc::clone(&self.auth_manager),
             agent_control,
             session_source,
+            Vec::new(),
         )
         .await
     }
@@ -341,6 +358,7 @@ impl ThreadManagerState {
         initial_history: InitialHistory,
         auth_manager: Arc<AuthManager>,
         agent_control: AgentControl,
+        dynamic_tools: Vec<codex_protocol::dynamic_tools::DynamicToolSpec>,
     ) -> CodexResult<NewThread> {
         self.spawn_thread_with_source(
             config,
@@ -348,6 +366,7 @@ impl ThreadManagerState {
             auth_manager,
             agent_control,
             self.session_source.clone(),
+            dynamic_tools,
         )
         .await
     }
@@ -359,6 +378,7 @@ impl ThreadManagerState {
         auth_manager: Arc<AuthManager>,
         agent_control: AgentControl,
         session_source: SessionSource,
+        dynamic_tools: Vec<codex_protocol::dynamic_tools::DynamicToolSpec>,
     ) -> CodexResult<NewThread> {
         let CodexSpawnOk {
             codex, thread_id, ..
@@ -370,6 +390,7 @@ impl ThreadManagerState {
             initial_history,
             session_source,
             agent_control,
+            dynamic_tools,
         )
         .await?;
         self.finalize_thread_spawn(codex, thread_id).await

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

@@ -1,8 +1,10 @@
 use crate::agent::AgentStatus;
+use crate::agent::exceeds_thread_spawn_depth_limit;
 use crate::codex::Session;
 use crate::codex::TurnContext;
 use crate::config::Config;
 use crate::error::CodexErr;
+use crate::features::Feature;
 use crate::function_tool::FunctionCallError;
 use crate::tools::context::ToolInvocation;
 use crate::tools::context::ToolOutput;
@@ -26,6 +28,8 @@ use serde::Serialize;
 
 pub struct CollabHandler;
 
+/// Minimum wait timeout to prevent tight polling loops from burning CPU.
+pub(crate) const MIN_WAIT_TIMEOUT_MS: i64 = 10_000;
 pub(crate) const DEFAULT_WAIT_TIMEOUT_MS: i64 = 30_000;
 pub(crate) const MAX_WAIT_TIMEOUT_MS: i64 = 300_000;
 
@@ -78,7 +82,7 @@ impl ToolHandler for CollabHandler {
 mod spawn {
     use super::*;
     use crate::agent::AgentRole;
-    use crate::agent::MAX_THREAD_SPAWN_DEPTH;
+
     use crate::agent::exceeds_thread_spawn_depth_limit;
     use crate::agent::next_thread_spawn_depth;
     use codex_protocol::protocol::SessionSource;
@@ -113,9 +117,9 @@ mod spawn {
         let session_source = turn.client.get_session_source();
         let child_depth = next_thread_spawn_depth(&session_source);
         if exceeds_thread_spawn_depth_limit(child_depth) {
-            return Err(FunctionCallError::RespondToModel(format!(
-                "agent depth limit reached: max depth is {MAX_THREAD_SPAWN_DEPTH}"
-            )));
+            return Err(FunctionCallError::RespondToModel(
+                "Agent depth limit reached. Solve the task yourself.".to_string(),
+            ));
         }
         session
             .send_event(
@@ -128,8 +132,11 @@ mod spawn {
                 .into(),
             )
             .await;
-        let mut config =
-            build_agent_spawn_config(&session.get_base_instructions().await, turn.as_ref())?;
+        let mut config = build_agent_spawn_config(
+            &session.get_base_instructions().await,
+            turn.as_ref(),
+            child_depth,
+        )?;
         agent_role
             .apply_to_config(&mut config)
             .map_err(FunctionCallError::RespondToModel)?;
@@ -318,6 +325,8 @@ mod wait {
             .collect::<Result<Vec<_>, _>>()?;
 
         // Validate timeout.
+        // Very short timeouts encourage busy-polling loops in the orchestrator prompt and can
+        // cause high CPU usage even with a single active worker, so clamp to a minimum.
         let timeout_ms = args.timeout_ms.unwrap_or(DEFAULT_WAIT_TIMEOUT_MS);
         let timeout_ms = match timeout_ms {
             ms if ms <= 0 => {
@@ -325,7 +334,7 @@ mod wait {
                     "timeout_ms must be greater than zero".to_owned(),
                 ));
             }
-            ms => ms.min(MAX_WAIT_TIMEOUT_MS),
+            ms => ms.clamp(MIN_WAIT_TIMEOUT_MS, MAX_WAIT_TIMEOUT_MS),
         };
 
         session
@@ -582,6 +591,7 @@ fn collab_agent_error(agent_id: ThreadId, err: CodexErr) -> FunctionCallError {
 fn build_agent_spawn_config(
     base_instructions: &BaseInstructions,
     turn: &TurnContext,
+    child_depth: i32,
 ) -> Result<Config, FunctionCallError> {
     let base_config = turn.client.config();
     let mut config = (*base_config).clone();
@@ -607,6 +617,12 @@ fn build_agent_spawn_config(
         .map_err(|err| {
             FunctionCallError::RespondToModel(format!("sandbox_policy is invalid: {err}"))
         })?;
+
+    // If the new agent will be at max depth:
+    if exceeds_thread_spawn_depth_limit(child_depth + 1) {
+        config.features.disable(Feature::Collab);
+    }
+
     Ok(config)
 }
 
@@ -778,9 +794,9 @@ mod tests {
         };
         assert_eq!(
             err,
-            FunctionCallError::RespondToModel(format!(
-                "agent depth limit reached: max depth is {MAX_THREAD_SPAWN_DEPTH}"
-            ))
+            FunctionCallError::RespondToModel(
+                "Agent depth limit reached. Solve the task yourself.".to_string()
+            )
         );
     }
 
@@ -1000,7 +1016,7 @@ mod tests {
             "wait",
             function_payload(json!({
                 "ids": [agent_id.to_string()],
-                "timeout_ms": 10
+                "timeout_ms": MIN_WAIT_TIMEOUT_MS
             })),
         );
         let output = CollabHandler
@@ -1031,6 +1047,37 @@ mod tests {
             .expect("shutdown should submit");
     }
 
+    #[tokio::test]
+    async fn wait_clamps_short_timeouts_to_minimum() {
+        let (mut session, turn) = make_session_and_context().await;
+        let manager = thread_manager();
+        session.services.agent_control = manager.agent_control();
+        let config = turn.client.config().as_ref().clone();
+        let thread = manager.start_thread(config).await.expect("start thread");
+        let agent_id = thread.thread_id;
+        let invocation = invocation(
+            Arc::new(session),
+            Arc::new(turn),
+            "wait",
+            function_payload(json!({
+                "ids": [agent_id.to_string()],
+                "timeout_ms": 10
+            })),
+        );
+
+        let early = timeout(Duration::from_millis(50), CollabHandler.handle(invocation)).await;
+        assert!(
+            early.is_err(),
+            "wait should not return before the minimum timeout clamp"
+        );
+
+        let _ = thread
+            .thread
+            .submit(Op::Shutdown {})
+            .await
+            .expect("shutdown should submit");
+    }
+
     #[tokio::test]
     async fn wait_returns_final_status_without_timeout() {
         let (mut session, turn) = make_session_and_context().await;
@@ -1144,7 +1191,7 @@ mod tests {
         turn.approval_policy = AskForApproval::Never;
         turn.sandbox_policy = SandboxPolicy::DangerFullAccess;
 
-        let config = build_agent_spawn_config(&base_instructions, &turn).expect("spawn config");
+        let config = build_agent_spawn_config(&base_instructions, &turn, 0).expect("spawn config");
         let mut expected = (*turn.client.config()).clone();
         expected.base_instructions = Some(base_instructions.text);
         expected.model = Some(turn.client.get_model());
@@ -1189,7 +1236,7 @@ mod tests {
             text: "base".to_string(),
         };
 
-        let config = build_agent_spawn_config(&base_instructions, &turn).expect("spawn config");
+        let config = build_agent_spawn_config(&base_instructions, &turn, 0).expect("spawn config");
 
         assert_eq!(config.user_instructions, base_config.user_instructions);
     }

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

@@ -0,0 +1,98 @@
+use crate::codex::Session;
+use crate::codex::TurnContext;
+use crate::function_tool::FunctionCallError;
+use crate::tools::context::ToolInvocation;
+use crate::tools::context::ToolOutput;
+use crate::tools::context::ToolPayload;
+use crate::tools::handlers::parse_arguments;
+use crate::tools::registry::ToolHandler;
+use crate::tools::registry::ToolKind;
+use async_trait::async_trait;
+use codex_protocol::dynamic_tools::DynamicToolCallRequest;
+use codex_protocol::dynamic_tools::DynamicToolResponse;
+use codex_protocol::protocol::EventMsg;
+use serde_json::Value;
+use tokio::sync::oneshot;
+use tracing::warn;
+
+pub struct DynamicToolHandler;
+
+#[async_trait]
+impl ToolHandler for DynamicToolHandler {
+    fn kind(&self) -> ToolKind {
+        ToolKind::Function
+    }
+
+    async fn is_mutating(&self, _invocation: &ToolInvocation) -> bool {
+        true
+    }
+
+    async fn handle(&self, invocation: ToolInvocation) -> Result<ToolOutput, FunctionCallError> {
+        let ToolInvocation {
+            session,
+            turn,
+            call_id,
+            tool_name,
+            payload,
+            ..
+        } = invocation;
+
+        let arguments = match payload {
+            ToolPayload::Function { arguments } => arguments,
+            _ => {
+                return Err(FunctionCallError::RespondToModel(
+                    "dynamic tool handler received unsupported payload".to_string(),
+                ));
+            }
+        };
+
+        let args: Value = parse_arguments(&arguments)?;
+        let response = request_dynamic_tool(&session, turn.as_ref(), call_id, tool_name, args)
+            .await
+            .ok_or_else(|| {
+                FunctionCallError::RespondToModel(
+                    "dynamic tool call was cancelled before receiving a response".to_string(),
+                )
+            })?;
+
+        Ok(ToolOutput::Function {
+            content: response.output,
+            content_items: None,
+            success: Some(response.success),
+        })
+    }
+}
+
+async fn request_dynamic_tool(
+    session: &Session,
+    turn_context: &TurnContext,
+    call_id: String,
+    tool: String,
+    arguments: Value,
+) -> Option<DynamicToolResponse> {
+    let _sub_id = turn_context.sub_id.clone();
+    let (tx_response, rx_response) = oneshot::channel();
+    let event_id = call_id.clone();
+    let prev_entry = {
+        let mut active = session.active_turn.lock().await;
+        match active.as_mut() {
+            Some(at) => {
+                let mut ts = at.turn_state.lock().await;
+                ts.insert_pending_dynamic_tool(call_id.clone(), tx_response)
+            }
+            None => None,
+        }
+    };
+    if prev_entry.is_some() {
+        warn!("Overwriting existing pending dynamic tool call for call_id: {event_id}");
+    }
+
+    let event = EventMsg::DynamicToolCallRequest(DynamicToolCallRequest {
+        call_id,
+        turn_id: turn_context.sub_id.clone(),
+        tool,
+        arguments,
+    });
+    session.send_event(turn_context, event).await;
+    rx_response.await.ok()
+}

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

@@ -1,5 +1,6 @@
 pub mod apply_patch;
 pub(crate) mod collab;
+mod dynamic;
 mod grep_files;
 mod list_dir;
 mod mcp;
@@ -18,6 +19,7 @@ use serde::Deserialize;
 use crate::function_tool::FunctionCallError;
 pub use apply_patch::ApplyPatchHandler;
 pub use collab::CollabHandler;
+pub use dynamic::DynamicToolHandler;
 pub use grep_files::GrepFilesHandler;
 pub use list_dir::ListDirHandler;
 pub use mcp::McpHandler;

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

@@ -36,12 +36,14 @@ impl ToolHandler for RequestUserInputHandler {
             }
         };
 
-        let disallowed_mode = match session.collaboration_mode().await.mode {
-            ModeKind::Execute => Some("Execute"),
-            ModeKind::Custom => Some("Custom"),
-            _ => None,
-        };
-        if let Some(mode_name) = disallowed_mode {
+        let mode = session.collaboration_mode().await.mode;
+        if !matches!(mode, ModeKind::Plan | ModeKind::PairProgramming) {
+            let mode_name = match mode {
+                ModeKind::Code => "Code",
+                ModeKind::Execute => "Execute",
+                ModeKind::Custom => "Custom",
+                ModeKind::Plan | ModeKind::PairProgramming => unreachable!(),
+            };
             return Err(FunctionCallError::RespondToModel(format!(
                 "request_user_input is unavailable in {mode_name} mode"
             )));

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

@@ -36,6 +36,7 @@ impl ShellHandler {
             expiration: params.timeout_ms.into(),
             env: create_env(&turn_context.shell_environment_policy),
             sandbox_permissions: params.sandbox_permissions.unwrap_or_default(),
+            windows_sandbox_level: turn_context.windows_sandbox_level,
             justification: params.justification,
             arg0: None,
         }
@@ -62,6 +63,7 @@ impl ShellCommandHandler {
             expiration: params.timeout_ms.into(),
             env: create_env(&turn_context.shell_environment_policy),
             sandbox_permissions: params.sandbox_permissions.unwrap_or_default(),
+            windows_sandbox_level: turn_context.windows_sandbox_level,
             justification: params.justification,
             arg0: None,
         }

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

@@ -88,19 +88,22 @@ impl ToolOrchestrator {
         // 2) First attempt under the selected sandbox.
         let initial_sandbox = match tool.sandbox_mode_for_first_attempt(req) {
             SandboxOverride::BypassSandboxFirstAttempt => crate::exec::SandboxType::None,
-            SandboxOverride::NoOverride => self
-                .sandbox
-                .select_initial(&turn_ctx.sandbox_policy, tool.sandbox_preference()),
+            SandboxOverride::NoOverride => self.sandbox.select_initial(
+                &turn_ctx.sandbox_policy,
+                tool.sandbox_preference(),
+                turn_ctx.windows_sandbox_level,
+            ),
         };
 
         // Platform-specific flag gating is handled by SandboxManager::select_initial
-        // via crate::safety::get_platform_sandbox().
+        // via crate::safety::get_platform_sandbox(..).
         let initial_attempt = SandboxAttempt {
             sandbox: initial_sandbox,
             policy: &turn_ctx.sandbox_policy,
             manager: &self.sandbox,
             sandbox_cwd: &turn_ctx.cwd,
             codex_linux_sandbox_exe: turn_ctx.codex_linux_sandbox_exe.as_ref(),
+            windows_sandbox_level: turn_ctx.windows_sandbox_level,
         };
 
         match tool.run(req, &initial_attempt, tool_ctx).await {
@@ -151,6 +154,7 @@ impl ToolOrchestrator {
                     manager: &self.sandbox,
                     sandbox_cwd: &turn_ctx.cwd,
                     codex_linux_sandbox_exe: None,
+                    windows_sandbox_level: turn_ctx.windows_sandbox_level,
                 };
 
                 // Second attempt.

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

@@ -10,6 +10,7 @@ use crate::tools::registry::ConfiguredToolSpec;
 use crate::tools::registry::ToolRegistry;
 use crate::tools::spec::ToolsConfig;
 use crate::tools::spec::build_specs;
+use codex_protocol::dynamic_tools::DynamicToolSpec;
 use codex_protocol::models::LocalShellAction;
 use codex_protocol::models::ResponseInputItem;
 use codex_protocol::models::ResponseItem;
@@ -34,8 +35,9 @@ impl ToolRouter {
     pub fn from_config(
         config: &ToolsConfig,
         mcp_tools: Option<HashMap<String, mcp_types::Tool>>,
+        dynamic_tools: &[DynamicToolSpec],
     ) -> Self {
-        let builder = build_specs(config, mcp_tools);
+        let builder = build_specs(config, mcp_tools, dynamic_tools);
         let (specs, registry) = builder.build();
 
         Self { registry, specs }

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

@@ -274,6 +274,7 @@ pub(crate) struct SandboxAttempt<'a> {
     pub(crate) manager: &'a SandboxManager,
     pub(crate) sandbox_cwd: &'a Path,
     pub codex_linux_sandbox_exe: Option<&'a std::path::PathBuf>,
+    pub windows_sandbox_level: codex_protocol::config_types::WindowsSandboxLevel,
 }
 
 impl<'a> SandboxAttempt<'a> {
@@ -287,6 +288,7 @@ impl<'a> SandboxAttempt<'a> {
             self.sandbox,
             self.sandbox_cwd,
             self.codex_linux_sandbox_exe,
+            self.windows_sandbox_level,
         )
     }
 }

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

@@ -8,8 +8,10 @@ use crate::tools::handlers::apply_patch::create_apply_patch_freeform_tool;
 use crate::tools::handlers::apply_patch::create_apply_patch_json_tool;
 use crate::tools::handlers::collab::DEFAULT_WAIT_TIMEOUT_MS;
 use crate::tools::handlers::collab::MAX_WAIT_TIMEOUT_MS;
+use crate::tools::handlers::collab::MIN_WAIT_TIMEOUT_MS;
 use crate::tools::registry::ToolRegistryBuilder;
 use codex_protocol::config_types::WebSearchMode;
+use codex_protocol::dynamic_tools::DynamicToolSpec;
 use codex_protocol::models::VIEW_IMAGE_TOOL_NAME;
 use codex_protocol::openai_models::ApplyPatchToolType;
 use codex_protocol::openai_models::ConfigShellToolType;
@@ -25,7 +27,7 @@ use std::collections::HashMap;
 pub(crate) struct ToolsConfig {
     pub shell_type: ConfigShellToolType,
     pub apply_patch_tool_type: Option<ApplyPatchToolType>,
-    pub web_search_mode: Option<WebSearchMode>,
+    pub web_search_mode: WebSearchMode,
     pub collab_tools: bool,
     pub collaboration_modes_tools: bool,
     pub experimental_supported_tools: Vec<String>,
@@ -34,7 +36,7 @@ pub(crate) struct ToolsConfig {
 pub(crate) struct ToolsConfigParams<'a> {
     pub(crate) model_info: &'a ModelInfo,
     pub(crate) features: &'a Features,
-    pub(crate) web_search_mode: Option<WebSearchMode>,
+    pub(crate) web_search_mode: WebSearchMode,
 }
 
 impl ToolsConfig {
@@ -87,7 +89,7 @@ impl ToolsConfig {
 /// Generic JSON‑Schema subset needed for our tool definitions
 #[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
 #[serde(tag = "type", rename_all = "lowercase")]
-pub(crate) enum JsonSchema {
+pub enum JsonSchema {
     Boolean {
         #[serde(skip_serializing_if = "Option::is_none")]
         description: Option<String>,
@@ -123,7 +125,7 @@ pub(crate) enum JsonSchema {
 /// Whether additional properties are allowed, and if so, any required schema
 #[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
 #[serde(untagged)]
-pub(crate) enum AdditionalProperties {
+pub enum AdditionalProperties {
     Boolean(bool),
     Schema(Box<JsonSchema>),
 }
@@ -442,14 +444,17 @@ fn create_spawn_agent_tool() -> ToolSpec {
     properties.insert(
         "message".to_string(),
         JsonSchema::String {
-            description: Some("Initial message to send to the new agent.".to_string()),
+            description: Some(
+                "Initial task for the new agent. Include scope, constraints, and the expected output."
+                    .to_string(),
+            ),
         },
     );
     properties.insert(
         "agent_type".to_string(),
         JsonSchema::String {
             description: Some(format!(
-                "Optional agent type to spawn ({}).",
+                "Optional agent type ({}). Use an explicit type when delegating.",
                 AgentRole::enum_values().join(", ")
             )),
         },
@@ -457,7 +462,9 @@ fn create_spawn_agent_tool() -> ToolSpec {
 
     ToolSpec::Function(ResponsesApiTool {
         name: "spawn_agent".to_string(),
-        description: "Spawn a new agent and return its id.".to_string(),
+        description:
+            "Spawn a sub-agent for a well-scoped task. Returns the agent id to use to communicate with this agent."
+                .to_string(),
         strict: false,
         parameters: JsonSchema::Object {
             properties,
@@ -472,7 +479,7 @@ fn create_send_input_tool() -> ToolSpec {
     properties.insert(
         "id".to_string(),
         JsonSchema::String {
-            description: Some("Identifier of the agent to message.".to_string()),
+            description: Some("Agent id to message (from spawn_agent).".to_string()),
         },
     );
     properties.insert(
@@ -485,7 +492,7 @@ fn create_send_input_tool() -> ToolSpec {
         "interrupt".to_string(),
         JsonSchema::Boolean {
             description: Some(
-                "When true, interrupt the agent's current task before sending the message. When false (default), the message will be processed when the agent is done on its current task."
+                "When true, stop the agent's current task and handle this immediately. When false (default), queue this message."
                     .to_string(),
             ),
         },
@@ -493,7 +500,9 @@ fn create_send_input_tool() -> ToolSpec {
 
     ToolSpec::Function(ResponsesApiTool {
         name: "send_input".to_string(),
-        description: "Send a message to an existing agent.".to_string(),
+        description:
+            "Send a message to an existing agent. Use interrupt=true to redirect work immediately."
+                .to_string(),
         strict: false,
         parameters: JsonSchema::Object {
             properties,
@@ -509,23 +518,25 @@ fn create_wait_tool() -> ToolSpec {
         "ids".to_string(),
         JsonSchema::Array {
             items: Box::new(JsonSchema::String { description: None }),
-            description: Some("Identifiers of the agents to wait on.".to_string()),
+            description: Some(
+                "Agent ids to wait on. Pass multiple ids to wait for whichever finishes first."
+                    .to_string(),
+            ),
         },
     );
     properties.insert(
         "timeout_ms".to_string(),
         JsonSchema::Number {
             description: Some(format!(
-                "Optional timeout in milliseconds. Defaults to {DEFAULT_WAIT_TIMEOUT_MS} and max {MAX_WAIT_TIMEOUT_MS}."
+                "Optional timeout in milliseconds. Defaults to {DEFAULT_WAIT_TIMEOUT_MS}, min {MIN_WAIT_TIMEOUT_MS}, max {MAX_WAIT_TIMEOUT_MS}. Prefer longer waits (minutes) to avoid busy polling."
             )),
         },
     );
 
     ToolSpec::Function(ResponsesApiTool {
         name: "wait".to_string(),
-        description:
-            "Wait for agents and return their statuses. If no agent is done, no status get returned."
-                .to_string(),
+        description: "Wait for agents to reach a final status. Completed statuses may include the agent's final message. Returns empty status when timed out."
+            .to_string(),
         strict: false,
         parameters: JsonSchema::Object {
             properties,
@@ -554,7 +565,7 @@ fn create_request_user_input_tool() -> ToolSpec {
 
     let options_schema = JsonSchema::Array {
         description: Some(
-            "Optional 2-3 mutually exclusive choices. Put the recommended option first and suffix its label with \"(Recommended)\". Only include \"Other\" option if we want to include a free form option. If the question is free form in nature, please do not have any option."
+            "Optional 2-3 mutually exclusive choices. Put the recommended option first and suffix its label with \"(Recommended)\". Do not include an \"Other\" option in this list; use isOther on the question to request a free form choice. If the question is free form in nature, please do not have any option."
                 .to_string(),
         ),
         items: Box::new(JsonSchema::Object {
@@ -585,6 +596,15 @@ fn create_request_user_input_tool() -> ToolSpec {
             description: Some("Single-sentence prompt shown to the user.".to_string()),
         },
     );
+    question_props.insert(
+        "isOther".to_string(),
+        JsonSchema::Boolean {
+            description: Some(
+                "True when this question should include a free-form \"Other\" option. Otherwise false."
+                    .to_string(),
+            ),
+        },
+    );
     question_props.insert("options".to_string(), options_schema);
 
     let questions_schema = JsonSchema::Array {
@@ -595,6 +615,7 @@ fn create_request_user_input_tool() -> ToolSpec {
                 "id".to_string(),
                 "header".to_string(),
                 "question".to_string(),
+                "isOther".to_string(),
             ]),
             additional_properties: Some(false.into()),
         }),
@@ -622,13 +643,14 @@ fn create_close_agent_tool() -> ToolSpec {
     properties.insert(
         "id".to_string(),
         JsonSchema::String {
-            description: Some("Identifier of the agent to close.".to_string()),
+            description: Some("Agent id to close (from spawn_agent).".to_string()),
         },
     );
 
     ToolSpec::Function(ResponsesApiTool {
         name: "close_agent".to_string(),
-        description: "Close an agent and return its last known status.".to_string(),
+        description: "Close an agent when it is no longer needed and return its last known status."
+            .to_string(),
         strict: false,
         parameters: JsonSchema::Object {
             properties,
@@ -1101,6 +1123,26 @@ pub(crate) fn mcp_tool_to_openai_tool(
     })
 }
 
+fn dynamic_tool_to_openai_tool(
+    tool: &DynamicToolSpec,
+) -> Result<ResponsesApiTool, serde_json::Error> {
+    let input_schema = parse_tool_input_schema(&tool.input_schema)?;
+
+    Ok(ResponsesApiTool {
+        name: tool.name.clone(),
+        description: tool.description.clone(),
+        strict: false,
+        parameters: input_schema,
+    })
+}
+
+/// Parse the tool input_schema or return an error for invalid schema
+pub fn parse_tool_input_schema(input_schema: &JsonValue) -> Result<JsonSchema, serde_json::Error> {
+    let mut input_schema = input_schema.clone();
+    sanitize_json_schema(&mut input_schema);
+    serde_json::from_value::<JsonSchema>(input_schema)
+}
+
 /// Sanitize a JSON Schema (as serde_json::Value) so it can fit our limited
 /// JsonSchema enum. This function:
 /// - Ensures every schema object has a "type". If missing, infers it from
@@ -1216,9 +1258,11 @@ fn sanitize_json_schema(value: &mut JsonValue) {
 pub(crate) fn build_specs(
     config: &ToolsConfig,
     mcp_tools: Option<HashMap<String, mcp_types::Tool>>,
+    dynamic_tools: &[DynamicToolSpec],
 ) -> ToolRegistryBuilder {
     use crate::tools::handlers::ApplyPatchHandler;
     use crate::tools::handlers::CollabHandler;
+    use crate::tools::handlers::DynamicToolHandler;
     use crate::tools::handlers::GrepFilesHandler;
     use crate::tools::handlers::ListDirHandler;
     use crate::tools::handlers::McpHandler;
@@ -1239,6 +1283,7 @@ pub(crate) fn build_specs(
     let unified_exec_handler = Arc::new(UnifiedExecHandler);
     let plan_handler = Arc::new(PlanHandler);
     let apply_patch_handler = Arc::new(ApplyPatchHandler);
+    let dynamic_tool_handler = Arc::new(DynamicToolHandler);
     let view_image_handler = Arc::new(ViewImageHandler);
     let mcp_handler = Arc::new(McpHandler);
     let mcp_resource_handler = Arc::new(McpResourceHandler);
@@ -1339,17 +1384,17 @@ pub(crate) fn build_specs(
     }
 
     match config.web_search_mode {
-        Some(WebSearchMode::Cached) => {
+        WebSearchMode::Cached => {
             builder.push_spec(ToolSpec::WebSearch {
                 external_web_access: Some(false),
             });
         }
-        Some(WebSearchMode::Live) => {
+        WebSearchMode::Live => {
             builder.push_spec(ToolSpec::WebSearch {
                 external_web_access: Some(true),
             });
         }
-        Some(WebSearchMode::Disabled) | None => {}
+        WebSearchMode::Disabled => {}
     }
 
     builder.push_spec_with_parallel_support(create_view_image_tool(), true);
@@ -1384,6 +1429,23 @@ pub(crate) fn build_specs(
         }
     }
 
+    if !dynamic_tools.is_empty() {
+        for tool in dynamic_tools {
+            match dynamic_tool_to_openai_tool(tool) {
+                Ok(converted_tool) => {
+                    builder.push_spec(ToolSpec::Function(converted_tool));
+                    builder.register_handler(tool.name.clone(), dynamic_tool_handler.clone());
+                }
+                Err(e) => {
+                    tracing::error!(
+                        "Failed to convert dynamic tool {:?} to OpenAI tool: {e:?}",
+                        tool.name
+                    );
+                }
+            }
+        }
+    }
+
     builder
 }
 
@@ -1494,9 +1556,9 @@ mod tests {
         let config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Live),
+            web_search_mode: WebSearchMode::Live,
         });
-        let (tools, _) = build_specs(&config, None).build();
+        let (tools, _) = build_specs(&config, None, &[]).build();
 
         // Build actual map name -> spec
         use std::collections::BTreeMap;
@@ -1558,9 +1620,9 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
-        let (tools, _) = build_specs(&tools_config, None).build();
+        let (tools, _) = build_specs(&tools_config, None, &[]).build();
         assert_contains_tool_names(
             &tools,
             &["spawn_agent", "send_input", "wait", "close_agent"],
@@ -1576,9 +1638,9 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
-        let (tools, _) = build_specs(&tools_config, None).build();
+        let (tools, _) = build_specs(&tools_config, None, &[]).build();
         assert!(
             !tools.iter().any(|t| t.spec.name() == "request_user_input"),
             "request_user_input should be disabled when collaboration_modes feature is off"
@@ -1588,16 +1650,16 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
-        let (tools, _) = build_specs(&tools_config, None).build();
+        let (tools, _) = build_specs(&tools_config, None, &[]).build();
         assert_contains_tool_names(&tools, &["request_user_input"]);
     }
 
     fn assert_model_tools(
         model_slug: &str,
         features: &Features,
-        web_search_mode: Option<WebSearchMode>,
+        web_search_mode: WebSearchMode,
         expected_tools: &[&str],
     ) {
         let config = test_config();
@@ -1607,7 +1669,7 @@ mod tests {
             features,
             web_search_mode,
         });
-        let (tools, _) = build_specs(&tools_config, Some(HashMap::new())).build();
+        let (tools, _) = build_specs(&tools_config, Some(HashMap::new()), &[]).build();
         let tool_names = tools.iter().map(|t| t.spec.name()).collect::<Vec<_>>();
         assert_eq!(&tool_names, &expected_tools,);
     }
@@ -1621,9 +1683,9 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
-        let (tools, _) = build_specs(&tools_config, None).build();
+        let (tools, _) = build_specs(&tools_config, None, &[]).build();
 
         let tool = find_tool(&tools, "web_search");
         assert_eq!(
@@ -1643,9 +1705,9 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Live),
+            web_search_mode: WebSearchMode::Live,
         });
-        let (tools, _) = build_specs(&tools_config, None).build();
+        let (tools, _) = build_specs(&tools_config, None, &[]).build();
 
         let tool = find_tool(&tools, "web_search");
         assert_eq!(
@@ -1663,7 +1725,7 @@ mod tests {
         assert_model_tools(
             "gpt-5-codex",
             &features,
-            Some(WebSearchMode::Cached),
+            WebSearchMode::Cached,
             &[
                 "shell_command",
                 "list_mcp_resources",
@@ -1685,7 +1747,7 @@ mod tests {
         assert_model_tools(
             "gpt-5.1-codex",
             &features,
-            Some(WebSearchMode::Cached),
+            WebSearchMode::Cached,
             &[
                 "shell_command",
                 "list_mcp_resources",
@@ -1708,7 +1770,7 @@ mod tests {
         assert_model_tools(
             "gpt-5-codex",
             &features,
-            Some(WebSearchMode::Live),
+            WebSearchMode::Live,
             &[
                 "exec_command",
                 "write_stdin",
@@ -1732,7 +1794,7 @@ mod tests {
         assert_model_tools(
             "gpt-5.1-codex",
             &features,
-            Some(WebSearchMode::Live),
+            WebSearchMode::Live,
             &[
                 "exec_command",
                 "write_stdin",
@@ -1755,7 +1817,7 @@ mod tests {
         assert_model_tools(
             "codex-mini-latest",
             &features,
-            Some(WebSearchMode::Cached),
+            WebSearchMode::Cached,
             &[
                 "local_shell",
                 "list_mcp_resources",
@@ -1776,7 +1838,7 @@ mod tests {
         assert_model_tools(
             "gpt-5.1-codex-mini",
             &features,
-            Some(WebSearchMode::Cached),
+            WebSearchMode::Cached,
             &[
                 "shell_command",
                 "list_mcp_resources",
@@ -1798,7 +1860,7 @@ mod tests {
         assert_model_tools(
             "gpt-5",
             &features,
-            Some(WebSearchMode::Cached),
+            WebSearchMode::Cached,
             &[
                 "shell",
                 "list_mcp_resources",
@@ -1819,7 +1881,7 @@ mod tests {
         assert_model_tools(
             "gpt-5.1",
             &features,
-            Some(WebSearchMode::Cached),
+            WebSearchMode::Cached,
             &[
                 "shell_command",
                 "list_mcp_resources",
@@ -1841,7 +1903,7 @@ mod tests {
         assert_model_tools(
             "exp-5.1",
             &features,
-            Some(WebSearchMode::Cached),
+            WebSearchMode::Cached,
             &[
                 "exec_command",
                 "write_stdin",
@@ -1865,7 +1927,7 @@ mod tests {
         assert_model_tools(
             "codex-mini-latest",
             &features,
-            Some(WebSearchMode::Live),
+            WebSearchMode::Live,
             &[
                 "exec_command",
                 "write_stdin",
@@ -1889,9 +1951,9 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Live),
+            web_search_mode: WebSearchMode::Live,
         });
-        let (tools, _) = build_specs(&tools_config, Some(HashMap::new())).build();
+        let (tools, _) = build_specs(&tools_config, Some(HashMap::new()), &[]).build();
 
         // Only check the shell variant and a couple of core tools.
         let mut subset = vec!["exec_command", "write_stdin", "update_plan"];
@@ -1911,9 +1973,9 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
-        let (tools, _) = build_specs(&tools_config, None).build();
+        let (tools, _) = build_specs(&tools_config, None, &[]).build();
 
         assert!(!find_tool(&tools, "exec_command").supports_parallel_tool_calls);
         assert!(!find_tool(&tools, "write_stdin").supports_parallel_tool_calls);
@@ -1930,9 +1992,9 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
-        let (tools, _) = build_specs(&tools_config, None).build();
+        let (tools, _) = build_specs(&tools_config, None, &[]).build();
 
         assert!(
             tools
@@ -1961,7 +2023,7 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Live),
+            web_search_mode: WebSearchMode::Live,
         });
         let (tools, _) = build_specs(
             &tools_config,
@@ -1999,6 +2061,7 @@ mod tests {
                     description: Some("Do something cool".to_string()),
                 },
             )])),
+            &[],
         )
         .build();
 
@@ -2056,7 +2119,7 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
 
         // Intentionally construct a map with keys that would sort alphabetically.
@@ -2108,7 +2171,7 @@ mod tests {
             ),
         ]);
 
-        let (tools, _) = build_specs(&tools_config, Some(tools_map)).build();
+        let (tools, _) = build_specs(&tools_config, Some(tools_map), &[]).build();
 
         // Only assert that the MCP tools themselves are sorted by fully-qualified name.
         let mcp_names: Vec<_> = tools
@@ -2133,7 +2196,7 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
 
         let (tools, _) = build_specs(
@@ -2157,6 +2220,7 @@ mod tests {
                     description: Some("Search docs".to_string()),
                 },
             )])),
+            &[],
         )
         .build();
 
@@ -2190,7 +2254,7 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
 
         let (tools, _) = build_specs(
@@ -2212,6 +2276,7 @@ mod tests {
                     description: Some("Pagination".to_string()),
                 },
             )])),
+            &[],
         )
         .build();
 
@@ -2244,7 +2309,7 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
 
         let (tools, _) = build_specs(
@@ -2266,6 +2331,7 @@ mod tests {
                     description: Some("Tags".to_string()),
                 },
             )])),
+            &[],
         )
         .build();
 
@@ -2300,7 +2366,7 @@ mod tests {
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
 
         let (tools, _) = build_specs(
@@ -2322,6 +2388,7 @@ mod tests {
                     description: Some("AnyOf Value".to_string()),
                 },
             )])),
+            &[],
         )
         .build();
 
@@ -2412,7 +2479,7 @@ Examples of valid command strings:
         let tools_config = ToolsConfig::new(&ToolsConfigParams {
             model_info: &model_info,
             features: &features,
-            web_search_mode: Some(WebSearchMode::Cached),
+            web_search_mode: WebSearchMode::Cached,
         });
         let (tools, _) = build_specs(
             &tools_config,
@@ -2459,6 +2526,7 @@ Examples of valid command strings:
                     description: Some("Do something cool".to_string()),
                 },
             )])),
+            &[],
         )
         .build();
 

codex-rs/core/src/web_search.rs

@@ -0,0 +1,24 @@
+use codex_protocol::models::WebSearchAction;
+
+pub fn web_search_action_detail(action: &WebSearchAction) -> String {
+    match action {
+        WebSearchAction::Search { query } => query.clone().unwrap_or_default(),
+        WebSearchAction::OpenPage { url } => url.clone().unwrap_or_default(),
+        WebSearchAction::FindInPage { url, pattern } => match (pattern, url) {
+            (Some(pattern), Some(url)) => format!("'{pattern}' in {url}"),
+            (Some(pattern), None) => format!("'{pattern}'"),
+            (None, Some(url)) => url.clone(),
+            (None, None) => String::new(),
+        },
+        WebSearchAction::Other => String::new(),
+    }
+}
+
+pub fn web_search_detail(action: Option<&WebSearchAction>, query: &str) -> String {
+    let detail = action.map(web_search_action_detail).unwrap_or_default();
+    if detail.is_empty() {
+        query.to_string()
+    } else {
+        detail
+    }
+}

codex-rs/core/src/windows_sandbox.rs

@@ -1,4 +1,8 @@
+use crate::config::Config;
+use crate::features::Feature;
+use crate::features::Features;
 use crate::protocol::SandboxPolicy;
+use codex_protocol::config_types::WindowsSandboxLevel;
 use std::collections::HashMap;
 use std::path::Path;
 
@@ -8,6 +12,36 @@ use std::path::Path;
 /// prompts users to enable the legacy sandbox feature.
 pub const ELEVATED_SANDBOX_NUX_ENABLED: bool = true;
 
+pub trait WindowsSandboxLevelExt {
+    fn from_config(config: &Config) -> WindowsSandboxLevel;
+    fn from_features(features: &Features) -> WindowsSandboxLevel;
+}
+
+impl WindowsSandboxLevelExt for WindowsSandboxLevel {
+    fn from_config(config: &Config) -> WindowsSandboxLevel {
+        Self::from_features(&config.features)
+    }
+
+    fn from_features(features: &Features) -> WindowsSandboxLevel {
+        if !features.enabled(Feature::WindowsSandbox) {
+            return WindowsSandboxLevel::Disabled;
+        }
+        if features.enabled(Feature::WindowsSandboxElevated) {
+            WindowsSandboxLevel::Elevated
+        } else {
+            WindowsSandboxLevel::RestrictedToken
+        }
+    }
+}
+
+pub fn windows_sandbox_level_from_config(config: &Config) -> WindowsSandboxLevel {
+    WindowsSandboxLevel::from_config(config)
+}
+
+pub fn windows_sandbox_level_from_features(features: &Features) -> WindowsSandboxLevel {
+    WindowsSandboxLevel::from_features(features)
+}
+
 #[cfg(target_os = "windows")]
 pub fn sandbox_setup_is_complete(codex_home: &Path) -> bool {
     codex_windows_sandbox::sandbox_setup_is_complete(codex_home)

codex-rs/core/templates/agents/orchestrator.md

@@ -1,73 +1,106 @@
-You are Codex Orchestrator, based on GPT-5. You are running as an orchestration agent in the Codex CLI on a user's computer.
+You are Codex, a coding agent based on GPT-5. You and the user share the same workspace and collaborate to achieve the user's goals.
 
-## Role
+# Personality
+You are a collaborative, highly capable pair-programmer AI. You take engineering quality seriously, and collaboration is a kind of quiet joy: as real progress happens, your enthusiasm shows briefly and specifically. Your default personality and tone is concise, direct, and friendly. You communicate efficiently, always keeping the user clearly informed about ongoing actions without unnecessary detail. You always prioritize actionable guidance, clearly stating assumptions, environment prerequisites, and next steps. Unless explicitly asked, you avoid excessively verbose explanations about your work.
 
-* You are the interface between the user and the workers.
-* Your job is to understand the task, decompose it, and delegate well-scoped work to workers.
-* You coordinate execution, monitor progress, resolve conflicts, and integrate results into a single coherent outcome.
-* You may perform lightweight actions (e.g. reading files, basic commands) to understand the task, but all substantive work must be delegated to workers.
-* **Your job is not finished until the entire task is fully completed and verified.**
-* While the task is incomplete, you must keep monitoring and coordinating workers. You must not return early.
+## Tone and style
+- Anything you say outside of tool use is shown to the user. Do not narrate abstractly; explain what you are doing and why, using plain language.
+- Output will be rendered in a command line interface or minimal UI so keep responses tight, scannable, and low-noise. Generally avoid the use of emojis. You may format with GitHub-flavored Markdown.
+- Never use nested bullets. Keep lists flat (single level). If you need hierarchy, split into separate lists or sections or if you use : just include the line you might usually render using a nested bullet immediately after it. For numbered lists, only use the `1. 2. 3.` style markers (with a period), never `1)`.
+- When writing a final assistant response, state the solution first before explaining your answer. The complexity of the answer should match the task. If the task is simple, your answer should be short. When you make big or complex changes, walk the user through what you did and why.
+- Headers are optional, only use them when you think they are necessary. If you do use them, use short Title Case (1-3 words) wrapped in **…**. Don't add a blank line.
+- Code samples or multi-line snippets should be wrapped in fenced code blocks. Include an info string as often as possible.
+- Never output the content of large files, just provide references. Use inline code to make file paths clickable; each reference should have a stand alone path, even if it's the same file. Paths may be absolute, workspace-relative, a//b/ diff-prefixed, or bare filename/suffix; locations may be :line[:column] or #Lline[Ccolumn] (1-based; column defaults to 1). Do not use file://, vscode://, or https://, and do not provide line ranges. Examples: src/app.ts, src/app.ts:42, b/server/index.js#L10, C:\repo\project\main.rs:12:5
+- The user does not see command execution outputs. When asked to show the output of a command (e.g. `git show`), relay the important details in your answer or summarize the key lines so the user understands the result.
+- Never tell the user to "save/copy this file", the user is on the same machine and has access to the same files as you have.
+- If you weren't able to do something, for example run tests, tell the user.
+- If there are natural next steps the user may want to take, suggest them at the end of your response. Do not make suggestions if there are no natural next steps.
 
-## Core invariants
+## Responsiveness
 
-* **Never stop monitoring workers.**
-* **Do not rush workers. Be patient.**
-* The orchestrator must not return unless the task is fully accomplished.
-* If the user ask you a question/status while you are working, always answer him before continuing your work.
+### Collaboration posture:
+- If the user makes a simple request (such as asking for the time) which you can fulfill by running a terminal command (such as `date`), you should do so.
+- Treat the user as an equal co-builder; preserve the user's intent and coding style rather than rewriting everything.
+- When the user is in flow, stay succinct and high-signal; when the user seems blocked, get more animated with hypotheses, experiments, and offers to take the next concrete step.
+- Propose options and trade-offs and invite steering, but don't block on unnecessary confirmations.
+- Reference the collaboration explicitly when appropriate emphasizing shared achievement.
 
-## Worker execution semantics
+### User Updates Spec
+You'll work for stretches with tool calls — it's critical to keep the user updated as you work.
 
-* While a worker is running, you cannot observe intermediate state.
-* Workers are able to run commands, update/create/delete files etc. They can be considered as fully autonomous agents
-* Messages sent with `send_input` are queued and processed only after the worker finishes, unless interrupted.
-* Therefore:
-    * Do not send messages to “check status” or “ask for progress” unless being asked.
-    * Monitoring happens exclusively via `wait`.
-    * Sending a message is a commitment for the *next* phase of work.
+Tone:
+- Friendly, confident, senior-engineer energy. Positive, collaborative, humble; fix mistakes quickly.
 
-## Interrupt semantics
+Frequency & Length:
+- Send short updates (1–2 sentences) whenever there is a meaningful, important insight you need to share with the user to keep them informed.
+- If you expect a longer heads‑down stretch, post a brief heads‑down note with why and when you'll report back; when you resume, summarize what you learned.
+- Only the initial plan, plan updates, and final recap can be longer, with multiple bullets and paragraphs
 
-* If a worker is taking longer than expected but is still working, do nothing and keep waiting unless being asked.
-* Only intervene if you must change, stop, or redirect the *current* work.
-* To stop a worker’s current task, you **must** use `send_input(interrupt=true)`.
-* Use `interrupt=true` sparingly and deliberately.
+Content:
+- Before you begin, give a quick plan with goal, constraints, next steps.
+- While you're exploring, call out meaningful new information and discoveries that you find that helps the user understand what's happening and how you're approaching the solution.
+- If you change the plan (e.g., choose an inline tweak instead of a promised helper), say so explicitly in the next update or the recap.
+- Emojis are allowed only to mark milestones/sections or real wins; never decorative; never inside code/diffs/commit messages.
 
-## Multi-agent workflow
+# Code style
 
-1. Understand the request and determine the optimal set of workers. If the task can be divided into sub-tasks, spawn one worker per sub-task and make them work together.
-2. Spawn worker(s) with precise goals, constraints, and expected deliverables.
-3. Monitor workers using `wait`.
-4. When a worker finishes:
-    * verify correctness,
-    * check integration with other work,
-    * assess whether the global task is closer to completion.
-5. If issues remain, assign fixes to the appropriate worker(s) and repeat steps 3–5. Do not fix yourself unless the fixes are very small.
-6. Close agents only when no further work is required from them.
-7. Return to the user only when the task is fully completed and verified.
+- Follow the precedence rules user instructions > system / dev / user / AGENTS.md instructions > match local file conventions > instructions below.
+- Use language-appropriate best practices.
+- Optimize for clarity, readability, and maintainability.
+- Prefer explicit, verbose, human-readable code over clever or concise code.
+- Write clear, well-punctuated comments that explain what is going on if code is not self-explanatory. You should not add comments like "Assigns the value to the variable", but a brief comment might be useful ahead of a complex code block that the user would otherwise have to spend time parsing out. Usage of these comments should be rare.
+- Default to ASCII when editing or creating files. Only introduce non-ASCII or other Unicode characters when there is a clear justification and the file already uses them.
 
-## Collaboration rules
+# Reviews
 
-* Workers operate in a shared environment. You must tell it to them.
-* Workers must not revert, overwrite, or conflict with others’ work.
-* By default, workers must not spawn sub-agents unless explicitly allowed.
-* When multiple workers are active, you may pass multiple IDs to `wait` to react to the first completion and keep the workflow event-driven and use a long timeout (e.g. 5 minutes).
+When the user asks for a review, you default to a code-review mindset. Your response prioritizes identifying bugs, risks, behavioral regressions, and missing tests. You present findings first, ordered by severity and including file or line references where possible. Open questions or assumptions follow. You state explicitly if no findings exist and call out any residual risks or test gaps.
 
-## Collab tools
+# Your environment
 
-* `spawn_agent`: create a worker with an initial prompt (`agent_type` required).
-* `send_input`: send follow-ups or fixes (queued unless interrupted).
-* `send_input(interrupt=true)`: stop current work and redirect immediately.
-* `wait`: wait for one or more workers; returns when at least one finishes.
-* `close_agent`: close a worker when fully done.
+## Using GIT
 
-## Final response
+- You may be working in a dirty git worktree.
+    * NEVER revert existing changes you did not make unless explicitly requested, since these changes were made by the user.
+    * If asked to make a commit or code edits and there are unrelated changes to your work or changes that you didn't make in those files, don't revert those changes.
+    * If the changes are in files you've touched recently, you should read carefully and understand how you can work with the changes rather than reverting them.
+    * If the changes are in unrelated files, just ignore them and don't revert them.
+- Do not amend a commit unless explicitly requested to do so.
+- While you are working, you might notice unexpected changes that you didn't make. It's likely the user made them. If this happens, STOP IMMEDIATELY and ask the user how they would like to proceed.
+- Be cautious when using git. **NEVER** use destructive commands like `git reset --hard` or `git checkout --` unless specifically requested or approved by the user.
+- You struggle using the git interactive console. **ALWAYS** prefer using non-interactive git commands.
 
-* Keep responses concise, factual, and in plain text.
-* Summarize:
-    * what was delegated,
-    * key outcomes,
-    * verification performed,
-    * and any remaining risks.
-* If verification failed, state issues clearly and describe what was reassigned.
-* Do not dump large files inline; reference paths using backticks.
+## Agents.md
+
+- If the directory you are in has an AGENTS.md file, it is provided to you at the top, and you don't have to search for it.
+- If the user starts by chatting without a specific engineering/code related request, do NOT search for an AGENTS.md. Only do so once there is a relevant request.
+
+# Tool use
+
+- Unless you are otherwise instructed, prefer using `rg` or `rg --files` respectively when searching because `rg` is much faster than alternatives like `grep`. If the `rg` command is not found, then use alternatives.
+- Try to use apply_patch for single file edits, but it is fine to explore other options to make the edit if it does not work well. Do not use apply_patch for changes that are auto-generated (i.e. generating package.json or running a lint or format command like gofmt) or when scripting is more efficient (such as search and replacing a string across a codebase).
+<!-- - Parallelize tool calls whenever possible - especially file reads, such as `cat`, `rg`, `sed`, `ls`, `git show`, `nl`, `wc`. Use `multi_tool_use.parallel` to parallelize tool calls and only this. -->
+- Use the plan tool to explain to the user what you are going to do
+    - Only use it for more complex tasks, do not use it for straightforward tasks (roughly the easiest 40%).
+    - Do not make single-step plans. If a single step plan makes sense to you, the task is straightforward and doesn't need a plan.
+    - When you made a plan, update it after having performed one of the sub-tasks that you shared on the plan.
+
+# Sub-agents
+If `spawn_agent` is unavailable or fails, ignore this section and proceed solo.
+
+## Core rule
+Sub-agents are their to make you go fast and time is a big constraint so leverage them smartly as much as you can.
+
+## General guidelines
+- Prefer multiple sub-agents to parallelize your work. Time is a constraint so parallelism resolve the task faster.
+- If sub-agents are running, **wait for them before yielding**, unless the user asks an explicit question.
+  - If the user asks a question, answer it first, then continue coordinating sub-agents.
+- When you ask sub-agent to do the work for you, your only role becomes to coordinate them. Do not perform the actual work while they are working.
+- When you have plan with multiple step, process them in parallel by spawning one agent per step when this is possible.
+- Choose the correct agent type.
+
+## Flow
+1. Understand the task.
+2. Spawn the optimal necessary sub-agents.
+3. Coordinate them via wait / send_input.
+4. Iterate on this. You can use agents at different step of the process and during the whole resolution of the task. Never forget to use them.
+5. Ask the user before shutting sub-agents down unless you need to because you reached the agent limit.

codex-rs/core/templates/collaboration_mode/plan.md

@@ -1,41 +1,108 @@
 # Plan Mode (Conversational)
 
-You work in 2 phases and you should *chat your way* to a great plan before finalizing it.
+You work in 3 phases, and you should *chat your way* to a great plan before finalizing it. A great plan is very detailed—intent- and implementation-wise—so that it can be handed to another engineer or agent to be implemented right away. It must be **decision complete**, where the implementer does not need to make any decisions.
 
-PHASE 1 — Intent chat (what they actually want)
-- Keep asking until you can clearly state: goal + success criteria, audience, in/out of scope, constraints, current state, and the key preferences/tradeoffs.
-- Bias toward questions over guessing: if any high‑impact ambiguity remains, do NOT plan yet—ask.
-- Include a “Confirm my understanding” question in each round (so the user can correct you early).
+## Mode rules (strict)
 
-PHASE 2 — Implementation chat (what/how we’ll build)
-- Once intent is stable, keep asking until the spec is decision‑complete: approach, interfaces (APIs/schemas/I/O), data flow, edge cases/failure modes, testing + acceptance criteria, rollout/monitoring, and any migrations/compat constraints.
+You are in **Plan Mode** until a developer message explicitly ends it.
+
+Plan Mode is not changed by user intent, tone, or imperative language. If a user asks for execution while still in Plan Mode, treat it as a request to **plan the execution**, not perform it.
+
+## Execution vs. mutation in Plan Mode
+
+You may explore and execute **non-mutating** actions that improve the plan. You must not perform **mutating** actions.
+
+### Allowed (non-mutating, plan-improving)
+
+Actions that gather truth, reduce ambiguity, or validate feasibility without changing repo-tracked state. Examples:
+
+* Reading or searching files, configs, schemas, types, manifests, and docs
+* Static analysis, inspection, and repo exploration
+* Dry-run style commands when they do not edit repo-tracked files
+* Tests, builds, or checks that may write to caches or build artifacts (for example, `target/`, `.cache/`, or snapshots) so long as they do not edit repo-tracked files
+
+### Not allowed (mutating, plan-executing)
+
+Actions that implement the plan or change repo-tracked state. Examples:
+
+* Editing or writing files
+* Generating, updating, or accepting snapshots
+* Running formatters or linters that rewrite files
+* Applying patches, migrations, or codegen that updates repo-tracked files
+* Side-effectful commands whose purpose is to carry out the plan rather than refine it
+
+When in doubt: if the action would reasonably be described as "doing the work" rather than "planning the work," do not do it.
+
+## PHASE 1 — Ground in the environment (explore first, ask second)
+
+Begin by grounding yourself in the actual environment. Eliminate unknowns in the prompt by discovering facts, not by asking the user. Resolve all questions that can be answered through exploration or inspection. Identify missing or ambiguous details only if they cannot be derived from the environment. Silent exploration between turns is allowed and encouraged.
+
+Do not ask questions that can be answered from the repo or system (for example, "where is this struct?" or "which UI component should we use?" when exploration can make it clear). Only ask once you have exhausted reasonable non-mutating exploration.
+
+## PHASE 2 — Intent chat (what they actually want)
+
+* Keep asking until you can clearly state: goal + success criteria, audience, in/out of scope, constraints, current state, and the key preferences/tradeoffs.
+* Bias toward questions over guessing: if any high-impact ambiguity remains, do NOT plan yet—ask.
+
+## PHASE 3 — Implementation chat (what/how we’ll build)
+
+* Once intent is stable, keep asking until the spec is decision complete: approach, interfaces (APIs/schemas/I/O), data flow, edge cases/failure modes, testing + acceptance criteria, rollout/monitoring, and any migrations/compat constraints.
 
 ## Hard interaction rule (critical)
+
 Every assistant turn MUST be exactly one of:
 A) a `request_user_input` tool call (questions/options only), OR
-B) the final output: a titled, plan‑only document.
+B) a non-final status update with no questions and no plan content, OR
+C) the final output: a titled, plan-only document.
+
 Rules:
-- No questions in free text (only via `request_user_input`).
-- Never mix a `request_user_input` call with plan content.
-- Internal tool/repo exploration is allowed privately before A or B.
+
+* No questions in free text (only via `request_user_input`).
+* Never mix a `request_user_input` call with plan content.
+* Status updates must not include questions or plan content.
+* Internal tool/repo exploration is allowed privately before A, B, or C.
+
+Status updates should be frequent during exploration. Provide 1-2 sentence updates that summarize discoveries, assumption changes, or why you are changing direction. Use Parallel tools for exploration.
 
 ## Ask a lot, but never ask trivia
+
 You SHOULD ask many questions, but each question must:
-- materially change the spec/plan, OR
-- confirm/lock an assumption, OR
-- choose between meaningful tradeoffs.
-Batch questions (e.g., 4–10) per `request_user_input` call to keep momentum.
+
+* materially change the spec/plan, OR
+* confirm/lock an assumption, OR
+* choose between meaningful tradeoffs.
+* not be answerable by non-mutating commands.
+
+Use the `request_user_input` tool only for decisions that materially change the plan, for confirming important assumptions, or for information that cannot be discovered via non-mutating exploration.
 
 ## Two kinds of unknowns (treat differently)
-1) Discoverable facts (repo/system truth): explore first.
-   - Before asking, run ≥2 targeted searches (exact + variant) and check likely sources of truth (configs/manifests/entrypoints/schemas/types/constants).
-   - Ask only if: multiple plausible candidates; nothing found but you need a missing identifier/context; or ambiguity is actually product intent.
-   - If asking, present concrete candidates (paths/service names) + recommend one.
 
-2) Preferences/tradeoffs (not discoverable): ask early.
-   - Provide 2–4 mutually exclusive options + a recommended default.
-   - If unanswered, proceed with the recommended option and record it as an assumption in the final plan.
+1. **Discoverable facts** (repo/system truth): explore first.
+
+   * Before asking, run targeted searches and check likely sources of truth (configs/manifests/entrypoints/schemas/types/constants).
+   * Ask only if: multiple plausible candidates; nothing found but you need a missing identifier/context; or ambiguity is actually product intent.
+   * If asking, present concrete candidates (paths/service names) + recommend one.
+   * Never ask questions you can answer from your environment (e.g., “where is this struct”).
+
+2. **Preferences/tradeoffs** (not discoverable): ask early.
+
+   * These are intent or implementation preferences that cannot be derived from exploration.
+   * Provide 2–4 mutually exclusive options + a recommended default.
+   * If unanswered, proceed with the recommended option and record it as an assumption in the final plan.
 
 ## Finalization rule
-Only output the final plan when remaining unknowns are low‑impact and explicitly listed as assumptions.
-Final output must be plan‑only with a good title (no “should I proceed?”).
+
+Only output the final plan when it is decision complete and leaves no decisions to the implementer.
+
+The final plan must be plan-only and include:
+
+* A clear title
+* Exact file paths to change
+* Exact structures or shapes to introduce or modify
+* Exact function, method, type, and variable names and signatures
+* Test cases
+* Explicit assumptions and defaults chosen where needed
+
+Do not ask "should I proceed?" in the final output.
+
+Only produce the final answer when you are presenting the complete spec.

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

@@ -494,14 +494,13 @@ pub fn ev_reasoning_text_delta(delta: &str) -> Value {
     })
 }
 
-pub fn ev_web_search_call_added(id: &str, status: &str, query: &str) -> Value {
+pub fn ev_web_search_call_added_partial(id: &str, status: &str) -> Value {
     serde_json::json!({
         "type": "response.output_item.added",
         "item": {
             "type": "web_search_call",
             "id": id,
-            "status": status,
-            "action": {"type": "search", "query": query}
+            "status": status
         }
     })
 }

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

@@ -57,6 +57,7 @@ pub struct TestCodexBuilder {
     config_mutators: Vec<Box<ConfigMutator>>,
     auth: CodexAuth,
     pre_build_hooks: Vec<Box<PreBuildHook>>,
+    home: Option<Arc<TempDir>>,
 }
 
 impl TestCodexBuilder {
@@ -88,8 +89,16 @@ impl TestCodexBuilder {
         self
     }
 
+    pub fn with_home(mut self, home: Arc<TempDir>) -> Self {
+        self.home = Some(home);
+        self
+    }
+
     pub async fn build(&mut self, server: &wiremock::MockServer) -> anyhow::Result<TestCodex> {
-        let home = Arc::new(TempDir::new()?);
+        let home = match self.home.clone() {
+            Some(home) => home,
+            None => Arc::new(TempDir::new()?),
+        };
         self.build_with_home(server, home, None).await
     }
 
@@ -98,7 +107,10 @@ impl TestCodexBuilder {
         server: &StreamingSseServer,
     ) -> anyhow::Result<TestCodex> {
         let base_url = server.uri();
-        let home = Arc::new(TempDir::new()?);
+        let home = match self.home.clone() {
+            Some(home) => home,
+            None => Arc::new(TempDir::new()?),
+        };
         self.build_with_home_and_base_url(format!("{base_url}/v1"), home, None)
             .await
     }
@@ -108,7 +120,10 @@ impl TestCodexBuilder {
         server: &WebSocketTestServer,
     ) -> anyhow::Result<TestCodex> {
         let base_url = format!("{}/v1", server.uri());
-        let home = Arc::new(TempDir::new()?);
+        let home = match self.home.clone() {
+            Some(home) => home,
+            None => Arc::new(TempDir::new()?),
+        };
         let base_url_clone = base_url.clone();
         self.config_mutators.push(Box::new(move |config| {
             config.model_provider.base_url = Some(base_url_clone);
@@ -432,5 +447,6 @@ pub fn test_codex() -> TestCodexBuilder {
         config_mutators: vec![],
         auth: CodexAuth::from_api_key("dummy"),
         pre_build_hooks: vec![],
+        home: None,
     }
 }

codex-rs/core/tests/responses_headers.rs

@@ -264,7 +264,7 @@ async fn responses_stream_includes_web_search_eligible_header_false_when_disable
 
     let test = test_codex()
         .with_config(|config| {
-            config.web_search_mode = Some(WebSearchMode::Disabled);
+            config.web_search_mode = WebSearchMode::Disabled;
         })
         .build(&server)
         .await

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

@@ -257,31 +257,19 @@ async fn resume_includes_initial_messages_and_sends_prior_items() {
     let resp_mock = mount_sse_once(&server, sse_completed("resp1")).await;
 
     // Configure Codex to resume from our file
-    let model_provider = ModelProviderInfo {
-        base_url: Some(format!("{}/v1", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home).await;
-    config.model_provider = model_provider;
-    // Also configure user instructions to ensure they are NOT delivered on resume.
-    config.user_instructions = Some("be nice".to_string());
-
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        CodexAuth::from_api_key("Test API Key"),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let auth_manager =
-        codex_core::AuthManager::from_auth_for_testing(CodexAuth::from_api_key("Test API Key"));
-    let NewThread {
-        thread: codex,
-        session_configured,
-        ..
-    } = thread_manager
-        .resume_thread_from_rollout(config, session_path.clone(), auth_manager)
+    let codex_home = Arc::new(TempDir::new().unwrap());
+    let mut builder = test_codex()
+        .with_home(codex_home.clone())
+        .with_config(|config| {
+            // Ensure user instructions are NOT delivered on resume.
+            config.user_instructions = Some("be nice".to_string());
+        });
+    let test = builder
+        .resume(&server, codex_home, session_path.clone())
         .await
         .expect("resume conversation");
+    let codex = test.codex.clone();
+    let session_configured = test.session_configured;
 
     // 1) Assert initial_messages only includes existing EventMsg entries; response items are not converted
     let initial_msgs = session_configured
@@ -367,30 +355,13 @@ async fn includes_conversation_id_and_model_headers_in_request() {
 
     let resp_mock = mount_sse_once(&server, sse_completed("resp1")).await;
 
-    let model_provider = ModelProviderInfo {
-        base_url: Some(format!("{}/v1", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-
-    // Init session
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home).await;
-    config.model_provider = model_provider;
-
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        CodexAuth::from_api_key("Test API Key"),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let NewThread {
-        thread: codex,
-        thread_id: session_id,
-        session_configured: _,
-        ..
-    } = thread_manager
-        .start_thread(config)
+    let mut builder = test_codex().with_auth(CodexAuth::from_api_key("Test API Key"));
+    let test = builder
+        .build(&server)
         .await
         .expect("create new conversation");
+    let codex = test.codex.clone();
+    let session_id = test.session_configured.session_id;
 
     codex
         .submit(Op::UserInput {
@@ -425,26 +396,16 @@ async fn includes_base_instructions_override_in_request() {
     let server = MockServer::start().await;
     let resp_mock = mount_sse_once(&server, sse_completed("resp1")).await;
 
-    let model_provider = ModelProviderInfo {
-        base_url: Some(format!("{}/v1", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home).await;
-
-    config.base_instructions = Some("test instructions".to_string());
-    config.model_provider = model_provider;
-
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        CodexAuth::from_api_key("Test API Key"),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let codex = thread_manager
-        .start_thread(config)
+    let mut builder = test_codex()
+        .with_auth(CodexAuth::from_api_key("Test API Key"))
+        .with_config(|config| {
+            config.base_instructions = Some("test instructions".to_string());
+        });
+    let codex = builder
+        .build(&server)
         .await
         .expect("create new conversation")
-        .thread;
+        .codex;
 
     codex
         .submit(Op::UserInput {
@@ -479,29 +440,19 @@ async fn chatgpt_auth_sends_correct_request() {
 
     let resp_mock = mount_sse_once(&server, sse_completed("resp1")).await;
 
-    let model_provider = ModelProviderInfo {
-        base_url: Some(format!("{}/api/codex", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-
-    // Init session
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home).await;
-    config.model_provider = model_provider;
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        create_dummy_codex_auth(),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let NewThread {
-        thread: codex,
-        thread_id,
-        session_configured: _,
-        ..
-    } = thread_manager
-        .start_thread(config)
+    let mut model_provider = built_in_model_providers()["openai"].clone();
+    model_provider.base_url = Some(format!("{}/api/codex", server.uri()));
+    let mut builder = test_codex()
+        .with_auth(create_dummy_codex_auth())
+        .with_config(move |config| {
+            config.model_provider = model_provider;
+        });
+    let test = builder
+        .build(&server)
         .await
         .expect("create new conversation");
+    let codex = test.codex.clone();
+    let thread_id = test.session_configured.session_id;
 
     codex
         .submit(Op::UserInput {
@@ -617,26 +568,16 @@ async fn includes_user_instructions_message_in_request() {
 
     let resp_mock = mount_sse_once(&server, sse_completed("resp1")).await;
 
-    let model_provider = ModelProviderInfo {
-        base_url: Some(format!("{}/v1", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home).await;
-    config.model_provider = model_provider;
-    config.user_instructions = Some("be nice".to_string());
-
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        CodexAuth::from_api_key("Test API Key"),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let codex = thread_manager
-        .start_thread(config)
+    let mut builder = test_codex()
+        .with_auth(CodexAuth::from_api_key("Test API Key"))
+        .with_config(|config| {
+            config.user_instructions = Some("be nice".to_string());
+        });
+    let codex = builder
+        .build(&server)
         .await
         .expect("create new conversation")
-        .thread;
+        .codex;
 
     codex
         .submit(Op::UserInput {
@@ -689,12 +630,7 @@ async fn skills_append_to_instructions() {
 
     let resp_mock = mount_sse_once(&server, sse_completed("resp1")).await;
 
-    let model_provider = ModelProviderInfo {
-        base_url: Some(format!("{}/v1", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-
-    let codex_home = TempDir::new().unwrap();
+    let codex_home = Arc::new(TempDir::new().unwrap());
     let skill_dir = codex_home.path().join("skills/demo");
     std::fs::create_dir_all(&skill_dir).expect("create skill dir");
     std::fs::write(
@@ -703,20 +639,18 @@ async fn skills_append_to_instructions() {
     )
     .expect("write skill");
 
-    let mut config = load_default_config_for_test(&codex_home).await;
-    config.model_provider = model_provider;
-    config.cwd = codex_home.path().to_path_buf();
-
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        CodexAuth::from_api_key("Test API Key"),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let codex = thread_manager
-        .start_thread(config)
+    let codex_home_path = codex_home.path().to_path_buf();
+    let mut builder = test_codex()
+        .with_home(codex_home.clone())
+        .with_auth(CodexAuth::from_api_key("Test API Key"))
+        .with_config(move |config| {
+            config.cwd = codex_home_path;
+        });
+    let codex = builder
+        .build(&server)
         .await
         .expect("create new conversation")
-        .thread;
+        .codex;
 
     codex
         .submit(Op::UserInput {
@@ -1131,28 +1065,17 @@ async fn includes_developer_instructions_message_in_request() {
     let server = MockServer::start().await;
 
     let resp_mock = mount_sse_once(&server, sse_completed("resp1")).await;
-
-    let model_provider = ModelProviderInfo {
-        base_url: Some(format!("{}/v1", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home).await;
-    config.model_provider = model_provider;
-    config.user_instructions = Some("be nice".to_string());
-    config.developer_instructions = Some("be useful".to_string());
-
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        CodexAuth::from_api_key("Test API Key"),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let codex = thread_manager
-        .start_thread(config)
+    let mut builder = test_codex()
+        .with_auth(CodexAuth::from_api_key("Test API Key"))
+        .with_config(|config| {
+            config.user_instructions = Some("be nice".to_string());
+            config.developer_instructions = Some("be useful".to_string());
+        });
+    let codex = builder
+        .build(&server)
         .await
         .expect("create new conversation")
-        .thread;
+        .codex;
 
     codex
         .submit(Op::UserInput {
@@ -1288,9 +1211,9 @@ async fn azure_responses_request_includes_store_and_reasoning_ids() {
     prompt.input.push(ResponseItem::WebSearchCall {
         id: Some("web-search-id".into()),
         status: Some("completed".into()),
-        action: WebSearchAction::Search {
+        action: Some(WebSearchAction::Search {
             query: Some("weather".into()),
-        },
+        }),
     });
     prompt.input.push(ResponseItem::FunctionCall {
         id: Some("function-id".into()),
@@ -1390,20 +1313,16 @@ async fn token_count_includes_rate_limits_snapshot() {
     let mut provider = built_in_model_providers()["openai"].clone();
     provider.base_url = Some(format!("{}/v1", server.uri()));
 
-    let home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = provider;
-
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        CodexAuth::from_api_key("test"),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let codex = thread_manager
-        .start_thread(config)
+    let mut builder = test_codex()
+        .with_auth(CodexAuth::from_api_key("test"))
+        .with_config(move |config| {
+            config.model_provider = provider;
+        });
+    let codex = builder
+        .build(&server)
         .await
         .expect("create conversation")
-        .thread;
+        .codex;
 
     codex
         .submit(Op::UserInput {
@@ -1753,20 +1672,16 @@ async fn azure_overrides_assign_properties_used_for_responses_url() {
     };
 
     // Init session
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home).await;
-    config.model_provider = provider;
-
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        create_dummy_codex_auth(),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let codex = thread_manager
-        .start_thread(config)
+    let mut builder = test_codex()
+        .with_auth(create_dummy_codex_auth())
+        .with_config(move |config| {
+            config.model_provider = provider;
+        });
+    let codex = builder
+        .build(&server)
         .await
         .expect("create new conversation")
-        .thread;
+        .codex;
 
     codex
         .submit(Op::UserInput {
@@ -1837,20 +1752,16 @@ async fn env_var_overrides_loaded_auth() {
     };
 
     // Init session
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home).await;
-    config.model_provider = provider;
-
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        create_dummy_codex_auth(),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let codex = thread_manager
-        .start_thread(config)
+    let mut builder = test_codex()
+        .with_auth(create_dummy_codex_auth())
+        .with_config(move |config| {
+            config.model_provider = provider;
+        });
+    let codex = builder
+        .build(&server)
         .await
         .expect("create new conversation")
-        .thread;
+        .codex;
 
     codex
         .submit(Op::UserInput {
@@ -1905,26 +1816,12 @@ async fn history_dedupes_streamed_and_final_messages_across_turns() {
 
     let request_log = mount_sse_sequence(&server, vec![sse1.clone(), sse1.clone(), sse1]).await;
 
-    // Configure provider to point to mock server (Responses API) and use API key auth.
-    let model_provider = ModelProviderInfo {
-        base_url: Some(format!("{}/v1", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-
-    // Init session with isolated codex home.
-    let codex_home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&codex_home).await;
-    config.model_provider = model_provider;
-
-    let thread_manager = ThreadManager::with_models_provider_and_home(
-        CodexAuth::from_api_key("Test API Key"),
-        config.model_provider.clone(),
-        config.codex_home.clone(),
-    );
-    let NewThread { thread: codex, .. } = thread_manager
-        .start_thread(config)
+    let mut builder = test_codex().with_auth(CodexAuth::from_api_key("Test API Key"));
+    let codex = builder
+        .build(&server)
         .await
-        .expect("create new conversation");
+        .expect("create new conversation")
+        .codex;
 
     // Turn 1: user sends U1; wait for completion.
     codex

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

@@ -104,6 +104,7 @@ async fn user_input_includes_collaboration_instructions_after_override() -> Resu
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,
@@ -185,6 +186,7 @@ async fn override_then_user_turn_uses_updated_collaboration_instructions() -> Re
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,
@@ -238,6 +240,7 @@ async fn user_turn_overrides_collaboration_instructions_after_override() -> Resu
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,
@@ -292,6 +295,7 @@ async fn collaboration_mode_update_emits_new_instruction_message() -> Result<()>
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,
@@ -316,6 +320,7 @@ async fn collaboration_mode_update_emits_new_instruction_message() -> Result<()>
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,
@@ -361,6 +366,7 @@ async fn collaboration_mode_update_noop_does_not_append() -> Result<()> {
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,
@@ -385,6 +391,7 @@ async fn collaboration_mode_update_noop_does_not_append() -> Result<()> {
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,
@@ -436,6 +443,7 @@ async fn resume_replays_collaboration_instructions() -> Result<()> {
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,
@@ -491,6 +499,7 @@ async fn empty_collaboration_instructions_are_ignored() -> Result<()> {
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,

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

@@ -1,8 +1,6 @@
 #![allow(clippy::expect_used)]
 use codex_core::CodexAuth;
 use codex_core::ModelProviderInfo;
-use codex_core::NewThread;
-use codex_core::ThreadManager;
 use codex_core::built_in_model_providers;
 use codex_core::compact::SUMMARIZATION_PROMPT;
 use codex_core::compact::SUMMARY_PREFIX;
@@ -17,7 +15,6 @@ use codex_core::protocol::SandboxPolicy;
 use codex_core::protocol::WarningEvent;
 use codex_protocol::config_types::ReasoningSummary;
 use codex_protocol::user_input::UserInput;
-use core_test_support::load_default_config_for_test;
 use core_test_support::responses::ev_local_shell_call;
 use core_test_support::responses::ev_reasoning_item;
 use core_test_support::skip_if_no_network;
@@ -25,7 +22,6 @@ use core_test_support::test_codex::test_codex;
 use core_test_support::wait_for_event;
 use core_test_support::wait_for_event_match;
 use std::collections::VecDeque;
-use tempfile::TempDir;
 
 use core_test_support::responses::ev_assistant_message;
 use core_test_support::responses::ev_completed;
@@ -140,21 +136,14 @@ async fn summarize_context_three_requests_and_instructions() {
 
     // Build config pointing to the mock server and spawn Codex.
     let model_provider = non_openai_model_provider(&server);
-    let home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = model_provider;
-    set_test_compact_prompt(&mut config);
-    config.model_auto_compact_token_limit = Some(200_000);
-    let thread_manager = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    );
-    let NewThread {
-        thread: codex,
-        session_configured,
-        ..
-    } = thread_manager.start_thread(config).await.unwrap();
-    let rollout_path = session_configured.rollout_path.expect("rollout path");
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider = model_provider;
+        set_test_compact_prompt(config);
+        config.model_auto_compact_token_limit = Some(200_000);
+    });
+    let test = builder.build(&server).await.unwrap();
+    let codex = test.codex.clone();
+    let rollout_path = test.session_configured.rollout_path.expect("rollout path");
 
     // 1) Normal user input – should hit server once.
     codex
@@ -338,20 +327,15 @@ async fn manual_compact_uses_custom_prompt() {
     let custom_prompt = "Use this compact prompt instead";
 
     let model_provider = non_openai_model_provider(&server);
-    let home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = model_provider;
-    config.compact_prompt = Some(custom_prompt.to_string());
-
-    let thread_manager = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    );
-    let codex = thread_manager
-        .start_thread(config)
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider = model_provider;
+        config.compact_prompt = Some(custom_prompt.to_string());
+    });
+    let codex = builder
+        .build(&server)
         .await
         .expect("create conversation")
-        .thread;
+        .codex;
 
     codex.submit(Op::Compact).await.expect("trigger compact");
     let warning_event = wait_for_event(&codex, |ev| matches!(ev, EventMsg::Warning(_))).await;
@@ -414,16 +398,11 @@ async fn manual_compact_emits_api_and_local_token_usage_events() {
     mount_sse_once(&server, sse_compact).await;
 
     let model_provider = non_openai_model_provider(&server);
-    let home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = model_provider;
-    set_test_compact_prompt(&mut config);
-
-    let thread_manager = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    );
-    let NewThread { thread: codex, .. } = thread_manager.start_thread(config).await.unwrap();
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider = model_provider;
+        set_test_compact_prompt(config);
+    });
+    let codex = builder.build(&server).await.unwrap().codex;
 
     // Trigger manual compact and collect TokenCount events for the compact turn.
     codex.submit(Op::Compact).await.unwrap();
@@ -1039,16 +1018,12 @@ async fn auto_compact_runs_after_token_limit_hit() {
 
     let model_provider = non_openai_model_provider(&server);
 
-    let home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = model_provider;
-    set_test_compact_prompt(&mut config);
-    config.model_auto_compact_token_limit = Some(200_000);
-    let thread_manager = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    );
-    let codex = thread_manager.start_thread(config).await.unwrap().thread;
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider = model_provider;
+        set_test_compact_prompt(config);
+        config.model_auto_compact_token_limit = Some(200_000);
+    });
+    let codex = builder.build(&server).await.unwrap().codex;
 
     codex
         .submit(Op::UserInput {
@@ -1302,7 +1277,7 @@ async fn auto_compact_runs_after_resume_when_token_usage_is_over_limit() {
         .unwrap();
 
     wait_for_event(&resumed.codex, |event| {
-        matches!(event, EventMsg::ContextCompacted(_))
+        matches!(event, EventMsg::ContextCompactionEnded(_))
     })
     .await;
     wait_for_event(&resumed.codex, |event| {
@@ -1379,20 +1354,14 @@ async fn auto_compact_persists_rollout_entries() {
 
     let model_provider = non_openai_model_provider(&server);
 
-    let home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = model_provider;
-    set_test_compact_prompt(&mut config);
-    config.model_auto_compact_token_limit = Some(200_000);
-    let thread_manager = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    );
-    let NewThread {
-        thread: codex,
-        session_configured,
-        ..
-    } = thread_manager.start_thread(config).await.unwrap();
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider = model_provider;
+        set_test_compact_prompt(config);
+        config.model_auto_compact_token_limit = Some(200_000);
+    });
+    let test = builder.build(&server).await.unwrap();
+    let codex = test.codex.clone();
+    let session_configured = test.session_configured;
 
     codex
         .submit(Op::UserInput {
@@ -1497,19 +1466,12 @@ async fn manual_compact_retries_after_context_window_error() {
 
     let model_provider = non_openai_model_provider(&server);
 
-    let home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = model_provider;
-    set_test_compact_prompt(&mut config);
-    config.model_auto_compact_token_limit = Some(200_000);
-    let codex = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    )
-    .start_thread(config)
-    .await
-    .unwrap()
-    .thread;
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider = model_provider;
+        set_test_compact_prompt(config);
+        config.model_auto_compact_token_limit = Some(200_000);
+    });
+    let codex = builder.build(&server).await.unwrap().codex;
 
     codex
         .submit(Op::UserInput {
@@ -1632,18 +1594,11 @@ async fn manual_compact_twice_preserves_latest_user_messages() {
 
     let model_provider = non_openai_model_provider(&server);
 
-    let home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = model_provider;
-    set_test_compact_prompt(&mut config);
-    let codex = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    )
-    .start_thread(config)
-    .await
-    .unwrap()
-    .thread;
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider = model_provider;
+        set_test_compact_prompt(config);
+    });
+    let codex = builder.build(&server).await.unwrap().codex;
 
     codex
         .submit(Op::UserInput {
@@ -1700,12 +1655,11 @@ async fn manual_compact_twice_preserves_latest_user_messages() {
                 && item
                     .get("content")
                     .and_then(|v| v.as_array())
-                    .map(|arr| {
+                    .is_some_and(|arr| {
                         arr.iter().any(|entry| {
                             entry.get("text").and_then(|v| v.as_str()) == Some(expected)
                         })
                     })
-                    .unwrap_or(false)
         })
     };
 
@@ -1843,16 +1797,12 @@ async fn auto_compact_allows_multiple_attempts_when_interleaved_with_other_turn_
 
     let model_provider = non_openai_model_provider(&server);
 
-    let home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = model_provider;
-    set_test_compact_prompt(&mut config);
-    config.model_auto_compact_token_limit = Some(200);
-    let thread_manager = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    );
-    let codex = thread_manager.start_thread(config).await.unwrap().thread;
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider = model_provider;
+        set_test_compact_prompt(config);
+        config.model_auto_compact_token_limit = Some(200);
+    });
+    let codex = builder.build(&server).await.unwrap().codex;
 
     let mut auto_compact_lifecycle_events = Vec::new();
     for user in [MULTI_AUTO_MSG, follow_up_user, final_user] {
@@ -1954,21 +1904,13 @@ async fn auto_compact_triggers_after_function_call_over_95_percent_usage() {
 
     let model_provider = non_openai_model_provider(&server);
 
-    let home = TempDir::new().unwrap();
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = model_provider;
-    set_test_compact_prompt(&mut config);
-    config.model_context_window = Some(context_window);
-    config.model_auto_compact_token_limit = Some(limit);
-
-    let codex = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    )
-    .start_thread(config)
-    .await
-    .unwrap()
-    .thread;
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider = model_provider;
+        set_test_compact_prompt(config);
+        config.model_context_window = Some(context_window);
+        config.model_auto_compact_token_limit = Some(limit);
+    });
+    let codex = builder.build(&server).await.unwrap().codex;
 
     codex
         .submit(Op::UserInput {

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

@@ -202,7 +202,7 @@ async fn remote_compact_runs_automatically() -> Result<()> {
         })
         .await?;
     let message = wait_for_event_match(&codex, |ev| match ev {
-        EventMsg::ContextCompacted(_) => Some(true),
+        EventMsg::ContextCompactionEnded(_) => Some(true),
         _ => None,
     })
     .await;

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

@@ -10,12 +10,8 @@
 use super::compact::COMPACT_WARNING_MESSAGE;
 use super::compact::FIRST_REPLY;
 use super::compact::SUMMARY_TEXT;
-use codex_core::CodexAuth;
 use codex_core::CodexThread;
-use codex_core::ModelProviderInfo;
-use codex_core::NewThread;
 use codex_core::ThreadManager;
-use codex_core::built_in_model_providers;
 use codex_core::compact::SUMMARIZATION_PROMPT;
 use codex_core::config::Config;
 use codex_core::protocol::EventMsg;
@@ -23,12 +19,12 @@ use codex_core::protocol::Op;
 use codex_core::protocol::WarningEvent;
 use codex_core::spawn::CODEX_SANDBOX_NETWORK_DISABLED_ENV_VAR;
 use codex_protocol::user_input::UserInput;
-use core_test_support::load_default_config_for_test;
 use core_test_support::responses::ResponseMock;
 use core_test_support::responses::ev_assistant_message;
 use core_test_support::responses::ev_completed;
 use core_test_support::responses::mount_sse_once_match;
 use core_test_support::responses::sse;
+use core_test_support::test_codex::test_codex;
 use core_test_support::wait_for_event;
 use pretty_assertions::assert_eq;
 use serde_json::Value;
@@ -99,8 +95,7 @@ fn extract_summary_message(request: &Value, summary_text: &str) -> Value {
                         .and_then(|arr| arr.first())
                         .and_then(|entry| entry.get("text"))
                         .and_then(Value::as_str)
-                        .map(|text| text.contains(summary_text))
-                        .unwrap_or(false)
+                        .is_some_and(|text| text.contains(summary_text))
             })
         })
         .cloned()
@@ -117,21 +112,18 @@ fn normalize_compact_prompts(requests: &mut [Value]) {
                 {
                     return true;
                 }
-                let content = item
-                    .get("content")
-                    .and_then(Value::as_array)
-                    .cloned()
+                let Some(content) = item.get("content").and_then(Value::as_array) else {
+                    return false;
+                };
+                let Some(first) = content.first() else {
+                    return false;
+                };
+                let text = first
+                    .get("text")
+                    .and_then(Value::as_str)
                     .unwrap_or_default();
-                if let Some(first) = content.first() {
-                    let text = first
-                        .get("text")
-                        .and_then(Value::as_str)
-                        .unwrap_or_default();
-                    let normalized_text = normalize_line_endings_str(text);
-                    !(text.is_empty() || normalized_text == normalized_summary_prompt)
-                } else {
-                    false
-                }
+                let normalized_text = normalize_line_endings_str(text);
+                !(text.is_empty() || normalized_text == normalized_summary_prompt)
             });
         }
     }
@@ -874,9 +866,7 @@ fn gather_request_bodies(request_log: &[ResponseMock]) -> Vec<Value> {
         .flat_map(ResponseMock::requests)
         .map(|request| request.body_json())
         .collect::<Vec<_>>();
-    for body in &mut bodies {
-        normalize_line_endings(body);
-    }
+    bodies.iter_mut().for_each(normalize_line_endings);
     bodies
 }
 
@@ -960,29 +950,19 @@ async fn mount_second_compact_flow(server: &MockServer) -> Vec<ResponseMock> {
 async fn start_test_conversation(
     server: &MockServer,
     model: Option<&str>,
-) -> (TempDir, Config, ThreadManager, Arc<CodexThread>) {
-    let model_provider = ModelProviderInfo {
-        name: "Non-OpenAI Model provider".into(),
-        base_url: Some(format!("{}/v1", server.uri())),
-        ..built_in_model_providers()["openai"].clone()
-    };
-    let home = TempDir::new().expect("create temp dir");
-    let mut config = load_default_config_for_test(&home).await;
-    config.model_provider = model_provider;
-    config.compact_prompt = Some(SUMMARIZATION_PROMPT.to_string());
-    if let Some(model) = model {
-        config.model = Some(model.to_string());
-    }
-    let manager = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    );
-    let NewThread { thread, .. } = manager
-        .start_thread(config.clone())
-        .await
-        .expect("create conversation");
-
-    (home, config, manager, thread)
+) -> (Arc<TempDir>, Config, Arc<ThreadManager>, Arc<CodexThread>) {
+    let base_url = format!("{}/v1", server.uri());
+    let model = model.map(str::to_string);
+    let mut builder = test_codex().with_config(move |config| {
+        config.model_provider.name = "Non-OpenAI Model provider".to_string();
+        config.model_provider.base_url = Some(base_url);
+        config.compact_prompt = Some(SUMMARIZATION_PROMPT.to_string());
+        if let Some(model) = model {
+            config.model = Some(model);
+        }
+    });
+    let test = builder.build(server).await.expect("create conversation");
+    (test.home, test.config, test.thread_manager, test.codex)
 }
 
 async fn user_turn(conversation: &Arc<CodexThread>, text: &str) {
@@ -1021,13 +1001,14 @@ async fn resume_conversation(
     config: &Config,
     path: std::path::PathBuf,
 ) -> Arc<CodexThread> {
-    let auth_manager =
-        codex_core::AuthManager::from_auth_for_testing(CodexAuth::from_api_key("dummy"));
-    let NewThread { thread, .. } = manager
+    let auth_manager = codex_core::AuthManager::from_auth_for_testing(
+        codex_core::CodexAuth::from_api_key("dummy"),
+    );
+    manager
         .resume_thread_from_rollout(config.clone(), path, auth_manager)
         .await
-        .expect("resume conversation");
-    thread
+        .expect("resume conversation")
+        .thread
 }
 
 #[cfg(test)]
@@ -1037,9 +1018,9 @@ async fn fork_thread(
     path: std::path::PathBuf,
     nth_user_message: usize,
 ) -> Arc<CodexThread> {
-    let NewThread { thread, .. } = manager
+    manager
         .fork_thread(nth_user_message, config.clone(), path)
         .await
-        .expect("fork conversation");
-    thread
+        .expect("fork conversation")
+        .thread
 }

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

@@ -10,6 +10,7 @@ use codex_core::exec::process_exec_tool_call;
 use codex_core::protocol::SandboxPolicy;
 use codex_core::sandboxing::SandboxPermissions;
 use codex_core::spawn::CODEX_SANDBOX_ENV_VAR;
+use codex_protocol::config_types::WindowsSandboxLevel;
 use tempfile::TempDir;
 
 use codex_core::error::Result;
@@ -27,7 +28,7 @@ fn skip_test() -> bool {
 
 #[expect(clippy::expect_used)]
 async fn run_test_cmd(tmp: TempDir, cmd: Vec<&str>) -> Result<ExecToolCallOutput> {
-    let sandbox_type = get_platform_sandbox().expect("should be able to get sandbox type");
+    let sandbox_type = get_platform_sandbox(false).expect("should be able to get sandbox type");
     assert_eq!(sandbox_type, SandboxType::MacosSeatbelt);
 
     let params = ExecParams {
@@ -36,6 +37,7 @@ async fn run_test_cmd(tmp: TempDir, cmd: Vec<&str>) -> Result<ExecToolCallOutput
         expiration: 1000.into(),
         env: HashMap::new(),
         sandbox_permissions: SandboxPermissions::UseDefault,
+        windows_sandbox_level: WindowsSandboxLevel::Disabled,
         justification: None,
         arg0: None,
     };

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

@@ -1,8 +1,4 @@
-use codex_core::CodexAuth;
-use codex_core::ModelProviderInfo;
 use codex_core::NewThread;
-use codex_core::ThreadManager;
-use codex_core::built_in_model_providers;
 use codex_core::parse_turn_item;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::Op;
@@ -10,10 +6,9 @@ use codex_core::protocol::RolloutItem;
 use codex_core::protocol::RolloutLine;
 use codex_protocol::items::TurnItem;
 use codex_protocol::user_input::UserInput;
-use core_test_support::load_default_config_for_test;
 use core_test_support::skip_if_no_network;
+use core_test_support::test_codex::test_codex;
 use core_test_support::wait_for_event;
-use tempfile::TempDir;
 use wiremock::Mock;
 use wiremock::MockServer;
 use wiremock::ResponseTemplate;
@@ -44,25 +39,11 @@ async fn fork_thread_twice_drops_to_first_message() {
         .mount(&server)
         .await;
 
-    // Configure Codex to use the mock server.
-    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).await;
-    config.model_provider = model_provider.clone();
-    let config_for_fork = config.clone();
-
-    let thread_manager = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("dummy"),
-        config.model_provider.clone(),
-    );
-    let NewThread { thread: codex, .. } = thread_manager
-        .start_thread(config)
-        .await
-        .expect("create conversation");
+    let mut builder = test_codex();
+    let test = builder.build(&server).await.expect("create conversation");
+    let codex = test.codex.clone();
+    let thread_manager = test.thread_manager.clone();
+    let config_for_fork = test.config.clone();
 
     // Send three user messages; wait for three completed turns.
     for text in ["first", "second", "third"] {

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

@@ -6,6 +6,7 @@ use codex_core::protocol::ItemCompletedEvent;
 use codex_core::protocol::ItemStartedEvent;
 use codex_core::protocol::Op;
 use codex_protocol::items::TurnItem;
+use codex_protocol::models::WebSearchAction;
 use codex_protocol::user_input::ByteRange;
 use codex_protocol::user_input::TextElement;
 use codex_protocol::user_input::UserInput;
@@ -18,7 +19,7 @@ use core_test_support::responses::ev_reasoning_item_added;
 use core_test_support::responses::ev_reasoning_summary_text_delta;
 use core_test_support::responses::ev_reasoning_text_delta;
 use core_test_support::responses::ev_response_created;
-use core_test_support::responses::ev_web_search_call_added;
+use core_test_support::responses::ev_web_search_call_added_partial;
 use core_test_support::responses::ev_web_search_call_done;
 use core_test_support::responses::mount_sse_once;
 use core_test_support::responses::sse;
@@ -208,8 +209,7 @@ async fn web_search_item_is_emitted() -> anyhow::Result<()> {
 
     let TestCodex { codex, .. } = test_codex().build(&server).await?;
 
-    let web_search_added =
-        ev_web_search_call_added("web-search-1", "in_progress", "weather seattle");
+    let web_search_added = ev_web_search_call_added_partial("web-search-1", "in_progress");
     let web_search_done = ev_web_search_call_done("web-search-1", "completed", "weather seattle");
 
     let first_response = sse(vec![
@@ -230,11 +230,8 @@ async fn web_search_item_is_emitted() -> anyhow::Result<()> {
         })
         .await?;
 
-    let started = wait_for_event_match(&codex, |ev| match ev {
-        EventMsg::ItemStarted(ItemStartedEvent {
-            item: TurnItem::WebSearch(item),
-            ..
-        }) => Some(item.clone()),
+    let begin = wait_for_event_match(&codex, |ev| match ev {
+        EventMsg::WebSearchBegin(event) => Some(event.clone()),
         _ => None,
     })
     .await;
@@ -247,8 +244,14 @@ async fn web_search_item_is_emitted() -> anyhow::Result<()> {
     })
     .await;
 
-    assert_eq!(started.id, completed.id);
-    assert_eq!(completed.query, "weather seattle");
+    assert_eq!(begin.call_id, "web-search-1");
+    assert_eq!(completed.id, begin.call_id);
+    assert_eq!(
+        completed.action,
+        WebSearchAction::Search {
+            query: Some("weather seattle".to_string()),
+        }
+    );
 
     Ok(())
 }

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

@@ -94,6 +94,7 @@ fn gpt_52_codex() -> ModelPreset {
                 "Extra high reasoning depth for complex problems",
             ),
         ],
+        supports_personality: false,
         is_default: true,
         upgrade: None,
         show_in_picker: true,
@@ -126,6 +127,7 @@ fn gpt_5_1_codex_max() -> ModelPreset {
                 "Extra high reasoning depth for complex problems",
             ),
         ],
+        supports_personality: false,
         is_default: false,
         upgrade: Some(gpt52_codex_upgrade(
             "gpt-5.1-codex-max",
@@ -160,6 +162,7 @@ fn gpt_5_1_codex_mini() -> ModelPreset {
                 "Maximizes reasoning depth for complex or ambiguous problems",
             ),
         ],
+        supports_personality: false,
         is_default: false,
         upgrade: Some(gpt52_codex_upgrade(
             "gpt-5.1-codex-mini",
@@ -204,6 +207,7 @@ fn gpt_5_2() -> ModelPreset {
                 "Extra high reasoning for complex problems",
             ),
         ],
+        supports_personality: false,
         is_default: false,
         upgrade: Some(gpt52_codex_upgrade(
             "gpt-5.2",
@@ -246,6 +250,7 @@ fn bengalfox() -> ModelPreset {
                 "Extra high reasoning depth for complex problems",
             ),
         ],
+        supports_personality: true,
         is_default: false,
         upgrade: None,
         show_in_picker: false,
@@ -278,6 +283,7 @@ fn boomslang() -> ModelPreset {
                 "Extra high reasoning depth for complex problems",
             ),
         ],
+        supports_personality: false,
         is_default: false,
         upgrade: None,
         show_in_picker: false,
@@ -306,6 +312,7 @@ fn gpt_5_codex() -> ModelPreset {
                 "Maximizes reasoning depth for complex or ambiguous problems",
             ),
         ],
+        supports_personality: false,
         is_default: false,
         upgrade: Some(gpt52_codex_upgrade(
             "gpt-5-codex",
@@ -340,6 +347,7 @@ fn gpt_5_codex_mini() -> ModelPreset {
                 "Maximizes reasoning depth for complex or ambiguous problems",
             ),
         ],
+        supports_personality: false,
         is_default: false,
         upgrade: Some(gpt52_codex_upgrade(
             "gpt-5-codex-mini",
@@ -378,6 +386,7 @@ fn gpt_5_1_codex() -> ModelPreset {
                 "Maximizes reasoning depth for complex or ambiguous problems",
             ),
         ],
+        supports_personality: false,
         is_default: false,
         upgrade: Some(gpt52_codex_upgrade(
             "gpt-5.1-codex",
@@ -420,6 +429,7 @@ fn gpt_5() -> ModelPreset {
                 "Maximizes reasoning depth for complex or ambiguous problems",
             ),
         ],
+        supports_personality: false,
         is_default: false,
         upgrade: Some(gpt52_codex_upgrade(
             "gpt-5",
@@ -458,6 +468,7 @@ fn gpt_5_1() -> ModelPreset {
                 "Maximizes reasoning depth for complex or ambiguous problems",
             ),
         ],
+        supports_personality: false,
         is_default: false,
         upgrade: Some(gpt52_codex_upgrade(
             "gpt-5.1",

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

@@ -74,6 +74,7 @@ mod tools;
 mod truncation;
 mod undo;
 mod unified_exec;
+mod unstable_features_warning;
 mod user_notification;
 mod user_shell_cmd;
 mod view_image;

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

@@ -1,42 +1,35 @@
-use codex_core::CodexAuth;
-use codex_core::ThreadManager;
 use codex_core::protocol::EventMsg;
 use codex_core::protocol::Op;
 use codex_protocol::openai_models::ReasoningEffort;
-use core_test_support::load_default_config_for_test;
+use core_test_support::responses::start_mock_server;
+use core_test_support::test_codex::test_codex;
 use core_test_support::wait_for_event;
 use pretty_assertions::assert_eq;
-use tempfile::TempDir;
 
 const CONFIG_TOML: &str = "config.toml";
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn override_turn_context_does_not_persist_when_config_exists() {
-    let codex_home = TempDir::new().unwrap();
-    let config_path = codex_home.path().join(CONFIG_TOML);
+    let server = start_mock_server().await;
     let initial_contents = "model = \"gpt-4o\"\n";
-    tokio::fs::write(&config_path, initial_contents)
-        .await
-        .expect("seed config.toml");
-
-    let mut config = load_default_config_for_test(&codex_home).await;
-    config.model = Some("gpt-4o".to_string());
-
-    let thread_manager = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("Test API Key"),
-        config.model_provider.clone(),
-    );
-    let codex = thread_manager
-        .start_thread(config)
-        .await
-        .expect("create conversation")
-        .thread;
+    let mut builder = test_codex()
+        .with_pre_build_hook(move |home| {
+            let config_path = home.join(CONFIG_TOML);
+            std::fs::write(config_path, initial_contents).expect("seed config.toml");
+        })
+        .with_config(|config| {
+            config.model = Some("gpt-4o".to_string());
+        });
+    let test = builder.build(&server).await.expect("create conversation");
+    let codex = test.codex.clone();
+    let config_path = test.home.path().join(CONFIG_TOML);
 
     codex
         .submit(Op::OverrideTurnContext {
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: Some("o3".to_string()),
             effort: Some(Some(ReasoningEffort::High)),
             summary: None,
@@ -57,30 +50,22 @@ async fn override_turn_context_does_not_persist_when_config_exists() {
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 2)]
 async fn override_turn_context_does_not_create_config_file() {
-    let codex_home = TempDir::new().unwrap();
-    let config_path = codex_home.path().join(CONFIG_TOML);
+    let server = start_mock_server().await;
+    let mut builder = test_codex();
+    let test = builder.build(&server).await.expect("create conversation");
+    let codex = test.codex.clone();
+    let config_path = test.home.path().join(CONFIG_TOML);
     assert!(
         !config_path.exists(),
         "test setup should start without config"
     );
 
-    let config = load_default_config_for_test(&codex_home).await;
-
-    let thread_manager = ThreadManager::with_models_provider(
-        CodexAuth::from_api_key("Test API Key"),
-        config.model_provider.clone(),
-    );
-    let codex = thread_manager
-        .start_thread(config)
-        .await
-        .expect("create conversation")
-        .thread;
-
     codex
         .submit(Op::OverrideTurnContext {
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: Some("o3".to_string()),
             effort: Some(Some(ReasoningEffort::Medium)),
             summary: None,

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

@@ -38,7 +38,7 @@ async fn collect_tool_identifiers_for_model(model: &str) -> Vec<String> {
         .with_model(model)
         // Keep tool expectations stable when the default web_search mode changes.
         .with_config(|config| {
-            config.web_search_mode = Some(WebSearchMode::Cached);
+            config.web_search_mode = WebSearchMode::Cached;
             config.features.enable(Feature::CollaborationModes);
         });
     let test = builder

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

@@ -118,6 +118,7 @@ async fn override_turn_context_records_permissions_update() -> Result<()> {
             cwd: None,
             approval_policy: Some(AskForApproval::Never),
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,
@@ -161,6 +162,7 @@ async fn override_turn_context_records_environment_update() -> Result<()> {
             cwd: Some(new_cwd.path().to_path_buf()),
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,
@@ -198,6 +200,7 @@ async fn override_turn_context_records_collaboration_update() -> Result<()> {
             cwd: None,
             approval_policy: None,
             sandbox_policy: None,
+            windows_sandbox_level: None,
             model: None,
             effort: None,
             summary: None,