Use captured step settings for request metadata and tool hooks (#45248)

## Why

Model and reasoning effort updates during a turn can leave request metadata and tool hooks reporting the turn's initial settings. Metadata should describe the step that issued the request or tool call.

## What changed

- Share captured execution metadata across Responses, MCP, and extension tool calls, including model, reasoning effort, and automatic review and Node REPL flags.
- Build Responses tool inventory metadata from the issuing step's finalized tool router. Attach the finalized inventory separately for remote compaction.
- Use captured step settings for pre- and post-tool hooks, and captured review settings for permission-request hooks.

## Testing

Add and extend regression tests for model and effort changes during a turn, MCP metadata, pre-tool hook model attribution, captured review flags, and tool inventory matching the issuing request.

GitOrigin-RevId: 9dee46a8b4839de8b434cdad4ed441cc0bb6ff9b
This commit is contained in:
Abhinav
2026-09-13 15:47:59 +00:00
committed by copyberry
parent 36f0dbe796
commit 16537b20a5
20 changed files with 545 additions and 255 deletions
+1 -5
View File
@@ -13,7 +13,6 @@ use crate::hook_runtime::PreCompactHookOutcome;
use crate::hook_runtime::run_post_compact_hooks;
use crate::hook_runtime::run_pre_compact_hooks;
use crate::responses_metadata::CodexResponsesMetadata;
use crate::responses_metadata::CodexResponsesRequestKind;
use crate::responses_metadata::CompactionTurnMetadata;
use crate::session::RequestEffortUsage;
use crate::session::session::Session;
@@ -258,10 +257,7 @@ async fn run_compact_task_inner_impl(
// request tracking)
// survives retries within this compact turn.
let responses_metadata = sess
.responses_metadata(
turn_context.as_ref(),
CodexResponsesRequestKind::Compaction(compaction_metadata),
)
.compaction_responses_metadata(turn_context.as_ref(), compaction_metadata)
.await;
let compaction_response_id = loop {
@@ -6,7 +6,6 @@ use crate::Prompt;
use crate::client::ModelClientSession;
use crate::compact::CompactionAnalyticsDetails;
use crate::compact_remote_history::trim_function_call_history_to_fit_context_window;
use crate::responses_metadata::CodexResponsesRequestKind;
use crate::responses_metadata::CompactionTurnMetadata;
use crate::session::session::Session;
use crate::session::step_context::StepContext;
@@ -87,12 +86,10 @@ pub(super) async fn run_remote_compact_v2_attempt(
cyber_access_program: turn_context.cyber_access_program,
};
let responses_metadata = sess
.responses_metadata(
turn_context.as_ref(),
CodexResponsesRequestKind::Compaction(compaction_metadata),
)
let mut responses_metadata = sess
.compaction_responses_metadata(turn_context.as_ref(), compaction_metadata)
.await;
responses_metadata.tool_namespaces_info = tool_router.tool_namespaces_info().cloned();
let trace_attempt = compaction_trace.start_attempt(&serde_json::json!({
"model": turn_context.model_info().slug.as_str(),
"instructions": prompt.base_instructions.text.as_str(),
+1 -1
View File
@@ -98,7 +98,7 @@ pub(crate) async fn finalize(
ReasoningSummary::None,
/*service_tier*/ None,
&session
.responses_metadata(&step.turn, CodexResponsesRequestKind::Turn)
.responses_metadata(step, CodexResponsesRequestKind::Turn)
.await,
)?;
let mut existing = super::request_budget::estimate_request_tokens(&request)
+21 -21
View File
@@ -61,6 +61,7 @@ use tracing::instrument;
use crate::context::ContextualUserFragment;
use crate::context::HookAdditionalContext;
use crate::event_mapping::parse_turn_item;
use crate::guardian::GuardianReviewContext;
use crate::session::TurnInput;
use crate::session::session::Session;
use crate::session::step_context::StepContext;
@@ -68,7 +69,7 @@ use crate::session::turn_context::TurnContext;
use crate::state::TurnState;
use crate::tools::hook_names::HookToolName;
use crate::tools::sandboxing::PermissionRequestPayload;
use crate::turn_metadata::McpTurnMetadataContext;
use crate::turn_metadata::ExecutionMetadata;
pub(crate) struct HookRuntimeOutcome {
pub should_stop: bool,
@@ -155,7 +156,7 @@ pub(crate) async fn run_pending_session_start_hooks(
cwd: turn_context.cwd.clone(),
transcript_path: sess.hook_transcript_path().await,
model: turn_context.model_info().slug.clone(),
permission_mode: hook_permission_mode(turn_context),
permission_mode: hook_permission_mode(turn_context.approval_policy()),
target,
};
let hooks = sess.hooks();
@@ -184,11 +185,12 @@ pub(crate) async fn run_pending_session_start_hooks(
/// handlers.
pub(crate) async fn run_pre_tool_use_hooks(
sess: &Arc<Session>,
turn_context: &Arc<TurnContext>,
step_context: &StepContext,
tool_use_id: String,
tool_name: &HookToolName,
tool_input: &Value,
) -> PreToolUseHookResult {
let turn_context = &step_context.turn;
let request = PreToolUseRequest {
session_id: sess.session_id().into(),
turn_id: turn_context.sub_id.clone(),
@@ -196,8 +198,8 @@ pub(crate) async fn run_pre_tool_use_hooks(
#[allow(deprecated)]
cwd: turn_context.cwd.clone(),
transcript_path: sess.hook_transcript_path().await,
model: turn_context.model_info().slug.clone(),
permission_mode: hook_permission_mode(turn_context),
model: step_context.settings.model_info.slug.clone(),
permission_mode: hook_permission_mode(step_context.settings.approval_policy()),
tool_name: tool_name.name().to_string(),
matcher_aliases: tool_name.matcher_aliases().to_vec(),
tool_use_id,
@@ -246,10 +248,11 @@ pub(crate) async fn run_pre_tool_use_hooks(
// tool input or post-run state.
pub(crate) async fn run_permission_request_hooks(
sess: &Arc<Session>,
turn_context: &Arc<TurnContext>,
review_context: &GuardianReviewContext,
run_id_suffix: &str,
payload: PermissionRequestPayload,
) -> Option<PermissionRequestDecision> {
let turn_context = review_context.turn();
let request = PermissionRequestRequest {
session_id: sess.session_id().into(),
turn_id: turn_context.sub_id.clone(),
@@ -257,8 +260,8 @@ pub(crate) async fn run_permission_request_hooks(
#[allow(deprecated)]
cwd: turn_context.cwd.to_path_buf(),
transcript_path: sess.hook_transcript_path().await,
model: turn_context.model_info().slug.clone(),
permission_mode: hook_permission_mode(turn_context),
model: review_context.model_info.slug.clone(),
permission_mode: hook_permission_mode(review_context.approval_policy),
tool_name: payload.tool_name.name().to_string(),
matcher_aliases: payload.tool_name.matcher_aliases().to_vec(),
run_id_suffix: run_id_suffix.to_string(),
@@ -285,13 +288,14 @@ pub(crate) async fn run_permission_request_hooks(
/// matchers and hook logs.
pub(crate) async fn run_post_tool_use_hooks(
sess: &Arc<Session>,
turn_context: &Arc<TurnContext>,
step_context: &StepContext,
tool_use_id: String,
tool_name: String,
matcher_aliases: Vec<String>,
tool_input: Value,
tool_response: Value,
) -> PostToolUseOutcome {
let turn_context = &step_context.turn;
let request = PostToolUseRequest {
session_id: sess.session_id().into(),
turn_id: turn_context.sub_id.clone(),
@@ -299,8 +303,8 @@ pub(crate) async fn run_post_tool_use_hooks(
#[allow(deprecated)]
cwd: turn_context.cwd.clone(),
transcript_path: sess.hook_transcript_path().await,
model: turn_context.model_info().slug.clone(),
permission_mode: hook_permission_mode(turn_context),
model: step_context.settings.model_info.slug.clone(),
permission_mode: hook_permission_mode(step_context.settings.approval_policy()),
tool_name,
matcher_aliases,
tool_use_id,
@@ -367,11 +371,7 @@ fn build_request_metadata(
.unwrap_or(turn_context.initial_settings.as_ref());
turn_context
.turn_metadata_state
.current_meta_value_for_mcp_request(McpTurnMetadataContext {
model: settings.model_info.slug.as_str(),
reasoning_effort: settings.effective_reasoning_effort(),
node_repl_disabled: settings.model_info.node_repl_disabled,
})
.current_meta_value_for_mcp_request(ExecutionMetadata::from_settings(settings))
.map(|turn_metadata| {
Map::from_iter([(
crate::X_CODEX_TURN_METADATA_HEADER.to_string(),
@@ -445,7 +445,7 @@ pub(crate) async fn run_turn_stop_hooks(
cwd: turn_context.cwd.clone(),
transcript_path,
model: turn_context.model_info().slug.clone(),
permission_mode: hook_permission_mode(turn_context),
permission_mode: hook_permission_mode(turn_context.approval_policy()),
request_metadata: (!request_metadata.is_empty()).then_some(request_metadata),
stop_hook_active,
last_assistant_message,
@@ -522,7 +522,7 @@ pub(crate) async fn run_turn_interrupt_hooks(
cwd: turn_context.cwd.clone(),
transcript_path: sess.hook_transcript_path().await,
model: turn_context.model_info().slug.clone(),
permission_mode: hook_permission_mode(turn_context),
permission_mode: hook_permission_mode(turn_context.approval_policy()),
request_metadata: (!request_metadata.is_empty()).then_some(request_metadata),
};
if let Err(err) = sess.flush_rollout().await {
@@ -682,7 +682,7 @@ pub(crate) async fn inspect_pending_input(
cwd: turn_context.cwd.clone(),
transcript_path: sess.hook_transcript_path().await,
model: turn_context.model_info().slug.clone(),
permission_mode: hook_permission_mode(turn_context),
permission_mode: hook_permission_mode(turn_context.approval_policy()),
prompt: UserMessageItem::new(content).message(),
};
let hooks = sess.hooks();
@@ -1020,8 +1020,8 @@ fn hook_run_metric_tags(run: &HookRunSummary) -> [(&'static str, &'static str);
]
}
fn hook_permission_mode(turn_context: &TurnContext) -> String {
match turn_context.approval_policy() {
fn hook_permission_mode(approval_policy: AskForApproval) -> String {
match approval_policy {
AskForApproval::Never => "bypassPermissions",
AskForApproval::UnlessTrusted | AskForApproval::OnRequest | AskForApproval::Granular(_) => {
"default"
+4 -6
View File
@@ -20,7 +20,7 @@ use crate::tools::hook_names::HookToolName;
use crate::tools::lifecycle::process_mcp_tool_result;
use crate::tools::sandboxing::ApprovalAction;
use crate::tools::sandboxing::ToolError;
use crate::turn_metadata::McpTurnMetadataContext;
use crate::turn_metadata::ExecutionMetadata;
use codex_analytics::AppInvocation;
use codex_analytics::InvocationType;
use codex_analytics::build_track_events_context;
@@ -1252,11 +1252,9 @@ fn build_mcp_tool_call_request_meta(
if let Some(turn_metadata) = step_context
.turn
.turn_metadata_state
.current_meta_value_for_mcp_request(McpTurnMetadataContext {
model: step_context.settings.model_info.slug.as_str(),
reasoning_effort: step_context.settings.effective_reasoning_effort(),
node_repl_disabled: step_context.settings.model_info.node_repl_disabled,
})
.current_meta_value_for_mcp_request(ExecutionMetadata::from_settings(
&step_context.settings,
))
{
request_meta.insert(
crate::X_CODEX_TURN_METADATA_HEADER.to_string(),
+21 -20
View File
@@ -14,7 +14,7 @@ use crate::session::turn_context::TurnEnvironment;
use crate::state::ActiveTurn;
use crate::test_support::models_manager_with_provider;
use crate::tools::hook_names::HookToolName;
use crate::turn_metadata::McpTurnMetadataContext;
use crate::turn_metadata::ExecutionMetadata;
use codex_config::CONFIG_TOML_FILE;
use codex_config::config_toml::ConfigToml;
use codex_config::types::AppConfig;
@@ -189,11 +189,16 @@ fn non_apps_tool_does_not_require_account_metadata(meta: JsonValue) {
);
}
fn mcp_turn_metadata_context(turn_context: &TurnContext) -> McpTurnMetadataContext<'_> {
McpTurnMetadataContext {
fn mcp_turn_metadata_context(turn_context: &TurnContext) -> ExecutionMetadata<'_> {
ExecutionMetadata {
model: turn_context.model_info().slug.as_str(),
reasoning_effort: turn_context.effective_reasoning_effort(),
node_repl_disabled: turn_context.model_info().node_repl_disabled,
auto_review_enabled: crate::guardian::routes_approval_policy_to_guardian(
turn_context.approval_policy(),
turn_context.config.approvals_reviewer,
),
node_repl_auto_review_required: turn_context.model_info().node_repl_auto_review_required,
}
}
@@ -1160,21 +1165,27 @@ async fn mcp_tool_call_request_meta_includes_turn_metadata_for_custom_server() {
turn_metadata
.get("model")
.and_then(serde_json::Value::as_str),
Some(turn_context.model_info().slug.as_str())
Some(step_context.settings.model_info.slug.as_str())
);
assert_eq!(
turn_metadata["node_repl_auto_review_required"],
serde_json::Value::Bool(turn_context.model_info().node_repl_auto_review_required),
serde_json::Value::Bool(
step_context
.settings
.model_info
.node_repl_auto_review_required
),
);
assert_eq!(
turn_metadata["node_repl_disabled"],
serde_json::Value::Bool(turn_context.model_info().node_repl_disabled),
serde_json::Value::Bool(step_context.settings.model_info.node_repl_disabled),
);
assert_eq!(
turn_metadata
.get("reasoning_effort")
.and_then(serde_json::Value::as_str),
turn_context
step_context
.settings
.effective_reasoning_effort()
.map(|effort| effort.to_string())
.as_deref()
@@ -1214,7 +1225,7 @@ async fn mcp_tool_call_request_meta_uses_the_issuing_step(
model.slug = "model-b".to_string();
model.default_reasoning_level = Some(ReasoningEffortConfig::High);
model.node_repl_disabled = true;
// This separate policy remains turn-owned in this migration.
// Node review requirements are captured with the issuing step.
model.node_repl_auto_review_required =
!turn_context.model_info().node_repl_auto_review_required;
@@ -1222,6 +1233,8 @@ async fn mcp_tool_call_request_meta_uses_the_issuing_step(
expected["model"] = serde_json::json!("model-b");
expected["reasoning_effort"] = serde_json::json!(expected_effort);
expected["node_repl_disabled"] = serde_json::json!(true);
expected["node_repl_auto_review_required"] =
serde_json::json!(step_b.settings.model_info.node_repl_auto_review_required);
assert_eq!(
build_mcp_tool_call_request_meta(&step_b, "node_repl", "call-b", /*metadata*/ None),
Some(serde_json::json!({
@@ -1234,18 +1247,6 @@ async fn mcp_tool_call_request_meta_uses_the_issuing_step(
build_mcp_tool_call_request_meta(&step_a, "node_repl", "call-a", /*metadata*/ None),
original_meta,
);
assert_eq!(
turn_context
.turn_metadata_state
.to_responses_metadata(
"installation".to_string(),
"window".to_string(),
crate::responses_metadata::CodexResponsesRequestKind::Turn,
)
.turn_metadata_value()
.expect("Responses turn metadata")["node_repl_disabled"],
serde_json::json!(false),
);
}
#[tokio::test]
-14
View File
@@ -3812,20 +3812,6 @@ impl Session {
)
.or_cancel(cancellation_token)
.await??;
// Publish inventory only after the step's tool plan has been finalized.
if turn_context
.config
.tool_registry
.turn_metadata_includes_tool_info
&& turn_context.model_info().use_responses_lite
{
turn_context.turn_metadata_state.set_tool_namespaces_info(
tool_router
.tool_namespaces_info()
.cloned()
.unwrap_or_default(),
);
}
Ok(Arc::new(StepContext {
settings,
token_budget,
+54 -2
View File
@@ -1,5 +1,6 @@
use super::input_queue::InputQueue;
use super::mcp_refresh::McpRefresh;
use super::step_context::StepContext;
use super::step_settings::ModelInfoOverrides;
use super::step_settings::StepSettings;
use super::step_settings::StepSettingsConstraints;
@@ -15,9 +16,11 @@ use crate::hook_mcp_executor::CoreHookMcpExecutor;
use crate::mcp_tool_call::McpToolApprovalMetadata;
use crate::responses_metadata::CodexResponsesMetadata;
use crate::responses_metadata::CodexResponsesRequestKind;
use crate::responses_metadata::CompactionTurnMetadata;
use crate::shell_snapshot::ShellSnapshot;
use crate::shell_snapshot::SnapshotCredentialBrokerState;
use crate::state::ActiveTurn;
use crate::turn_metadata::ExecutionMetadata;
use codex_extension_api::ExtensionDataInit;
use codex_http_client::ClientRouteClass;
use codex_http_client::RouteAwareClientPool;
@@ -644,15 +647,64 @@ impl Session {
pub(crate) async fn responses_metadata(
&self,
turn_context: &TurnContext,
step_context: &StepContext,
request_kind: CodexResponsesRequestKind,
) -> CodexResponsesMetadata {
let (window_id, window_number, context_window_id) = self.current_window().await;
let mut responses_metadata = step_context.turn.turn_metadata_state.to_responses_metadata(
self.installation_id.clone(),
window_id,
request_kind,
);
ExecutionMetadata::from_settings(&step_context.settings).apply_to(&mut responses_metadata);
responses_metadata.tool_namespaces_info = if step_context
.turn
.config
.tool_registry
.turn_metadata_includes_tool_info
&& step_context.settings.model_info.use_responses_lite
{
step_context.tool_router.tool_namespaces_info().cloned()
} else {
None
};
self.with_window_and_fork_metadata(
&step_context.turn,
responses_metadata,
window_number,
context_window_id,
)
}
// TODO(CDXENT-454): Build the compaction request and metadata from the captured execution.
// Remote compaction currently attaches only finalized tool inventory because the rest of the
// request remains turn-backed; local compaction does not have a finalized request inventory.
pub(crate) async fn compaction_responses_metadata(
&self,
turn_context: &TurnContext,
compaction_metadata: CompactionTurnMetadata,
) -> CodexResponsesMetadata {
let (window_id, window_number, context_window_id) = self.current_window().await;
let responses_metadata = turn_context.turn_metadata_state.to_responses_metadata(
self.installation_id.clone(),
window_id,
request_kind,
CodexResponsesRequestKind::Compaction(compaction_metadata),
);
self.with_window_and_fork_metadata(
turn_context,
responses_metadata,
window_number,
context_window_id,
)
}
fn with_window_and_fork_metadata(
&self,
turn_context: &TurnContext,
responses_metadata: CodexResponsesMetadata,
window_number: u64,
context_window_id: uuid::Uuid,
) -> CodexResponsesMetadata {
CodexResponsesMetadata {
window_number: Some(window_number),
context_window_id: Some(context_window_id),
+3 -2
View File
@@ -5862,16 +5862,17 @@ pub(crate) async fn build_world_state_from_turn_context(
#[tokio::test]
async fn responses_metadata_uses_selected_harness_analytics_client() {
let (mut session, mut turn_context) = make_session_and_context().await;
for enabled in [true, false] {
let (mut session, mut turn_context) = make_session_and_context().await;
session.services.analytics_events_client = AnalyticsEventsClient::new(
Arc::clone(&session.services.auth_manager),
turn_context.config.chatgpt_base_url.clone(),
Some(enabled),
);
Arc::make_mut(&mut turn_context.config).analytics_enabled = Some(!enabled);
let step_context = StepContext::for_test(Arc::new(turn_context));
let metadata = session
.responses_metadata(&turn_context, CodexResponsesRequestKind::Turn)
.responses_metadata(&step_context, CodexResponsesRequestKind::Turn)
.await;
assert_eq!(metadata.analytics_enabled, Some(enabled));
}
+1 -1
View File
@@ -517,7 +517,7 @@ pub(crate) async fn run_turn(
.await;
let responses_metadata = sess
.responses_metadata(turn_context.as_ref(), CodexResponsesRequestKind::Turn)
.responses_metadata(step_context.as_ref(), CodexResponsesRequestKind::Turn)
.await;
run_sampling_request(
Arc::clone(&sess),
+1 -1
View File
@@ -316,7 +316,7 @@ async fn schedule_startup_prewarm_inner(
/*status*/ None,
);
let responses_metadata = session
.responses_metadata(&startup_turn_context, CodexResponsesRequestKind::Prewarm)
.responses_metadata(step_context.as_ref(), CodexResponsesRequestKind::Prewarm)
.await;
let mut client_session = session.services.model_client.new_session();
let websocket_warmup_started_at = Instant::now();
+1 -1
View File
@@ -495,7 +495,7 @@ impl Session {
// 2. If StrictAutoReview || Guardian enabled, then Guardian. Else, user.
let resolution = match run_permission_request_hooks(
self,
ctx.review_context.turn(),
&ctx.review_context,
&permission_request_run_id,
action.permission_request_payload(),
)
@@ -26,7 +26,7 @@ use crate::tools::handlers::apply_granted_turn_permissions;
use crate::tools::lifecycle::extension_tool_call_source;
use crate::tools::registry::CoreToolRuntime;
use crate::tools::registry::ToolExecutor;
use crate::turn_metadata::McpTurnMetadataContext;
use crate::turn_metadata::ExecutionMetadata;
pub(crate) struct ExtensionToolAdapter(
Arc<dyn for<'call> codex_tools::ToolExecutor<ExtensionToolCall<'call>>>,
@@ -170,11 +170,9 @@ async fn to_extension_call(invocation: &ToolInvocation) -> ExtensionToolCall<'_>
let codex_turn_metadata = invocation
.turn
.turn_metadata_state
.current_meta_value_for_mcp_request(McpTurnMetadataContext {
model: settings.model_info.slug.as_str(),
reasoning_effort: settings.effective_reasoning_effort(),
node_repl_disabled: settings.model_info.node_repl_disabled,
})
.current_meta_value_for_mcp_request(ExecutionMetadata::from_settings(
&invocation.step_context.settings,
))
.and_then(|metadata| to_ascii_json_string(&metadata).ok());
let mut environments = Vec::new();
for environment in invocation.step_context.environments.turn_environments() {
+2 -2
View File
@@ -567,7 +567,7 @@ impl ToolRegistry {
if let Some(pre_tool_use_payload) = tool.pre_tool_use_payload(&invocation) {
match run_pre_tool_use_hooks(
&invocation.session,
&invocation.turn,
invocation.step_context.as_ref(),
invocation.call_id.clone(),
&pre_tool_use_payload.tool_name,
&pre_tool_use_payload.tool_input,
@@ -683,7 +683,7 @@ impl ToolRegistry {
Some(
run_post_tool_use_hooks(
&invocation.session,
&invocation.turn,
invocation.step_context.as_ref(),
post_tool_use_payload.tool_use_id,
post_tool_use_payload.tool_name.name().to_string(),
post_tool_use_payload.tool_name.matcher_aliases().to_vec(),
@@ -45,7 +45,6 @@ use serde_json::json;
use crate::WaitForEnvironmentToolConfig;
use crate::config::CurrentTimeReminderConfig;
use crate::environment_selection::TurnEnvironmentState;
use crate::responses_metadata::CodexResponsesRequestKind;
use crate::responses_metadata::TurnToolFunctionInfo;
use crate::responses_metadata::TurnToolNamespacesInfo;
use crate::responses_metadata::TurnToolSource;
@@ -1596,13 +1595,6 @@ async fn candidate_model_plan_leaves_selected_model_and_inventory_unchanged() {
turn.model_info(),
ToolPlanInputs::default(),
));
let selected_inventory = selected
.tool_namespaces_info
.clone()
.expect("selected plan inventory");
turn.turn_metadata_state
.set_tool_namespaces_info(selected_inventory.clone());
let mut candidate_model = selected_model.as_ref().clone();
candidate_model.tool_mode = Some(ToolMode::CodeModeOnly);
candidate_model.shell_type = ConfigShellToolType::UnifiedExec;
@@ -1633,12 +1625,6 @@ async fn candidate_model_plan_leaves_selected_model_and_inventory_unchanged() {
source: TurnToolSource::Harness,
}
);
let metadata = turn.turn_metadata_state.to_responses_metadata(
"installation".to_string(),
"window".to_string(),
CodexResponsesRequestKind::Turn,
);
assert_eq!(metadata.tool_namespaces_info, Some(selected_inventory));
assert_eq!(turn.model_info(), &selected_model);
assert_eq!(
ToolPlanProbe::from_router(plan_with_model(
+43 -35
View File
@@ -20,12 +20,12 @@ use crate::responses_metadata::MAX_EXTRA_METADATA_VALUE_BYTES;
use crate::responses_metadata::PARENT_TURN_ID_KEY;
use crate::responses_metadata::ROOT_TURN_ID_KEY;
use crate::responses_metadata::TurnMetadataWorkspace;
use crate::responses_metadata::TurnToolNamespacesInfo;
use crate::responses_metadata::filter_extra_metadata;
use crate::responses_metadata::subagent_header_value;
use crate::responses_metadata::subagent_metadata_kind;
use crate::sandbox_tags::SandboxTags;
use crate::sandbox_tags::record_policy_metadata;
use crate::session::step_settings::ResolvedStepSettings;
use codex_git_utils::SanitizedGitUrl;
use codex_git_utils::get_git_remote_urls_assume_git_repo;
use codex_git_utils::get_has_changes_in_repo;
@@ -50,10 +50,48 @@ const REASONING_EFFORT_KEY: &str = "reasoning_effort";
const USER_INPUT_REQUESTED_DURING_TURN_KEY: &str = "user_input_requested_during_turn";
const WORKSPACE_KIND_KEY: &str = "workspace_kind";
pub(crate) struct McpTurnMetadataContext<'a> {
/// Captured execution settings shared by Responses and MCP metadata.
pub(crate) struct ExecutionMetadata<'a> {
pub(crate) model: &'a str,
pub(crate) reasoning_effort: Option<ReasoningEffortConfig>,
pub(crate) node_repl_disabled: bool,
pub(crate) auto_review_enabled: bool,
pub(crate) node_repl_auto_review_required: bool,
}
impl<'a> ExecutionMetadata<'a> {
pub(crate) fn from_settings(settings: &'a ResolvedStepSettings) -> Self {
Self {
model: settings.model_info.slug.as_str(),
reasoning_effort: settings.effective_reasoning_effort(),
node_repl_disabled: settings.model_info.node_repl_disabled,
auto_review_enabled: crate::guardian::routes_approval_policy_to_guardian(
settings.approval_policy(),
settings.approvals_reviewer(),
),
node_repl_auto_review_required: settings.model_info.computer_use_review_required(),
}
}
pub(crate) fn apply_to(&self, metadata: &mut CodexResponsesMetadata) {
metadata.auto_review_enabled = Some(self.auto_review_enabled);
metadata.node_repl_auto_review_required = Some(self.node_repl_auto_review_required);
metadata.node_repl_disabled = Some(self.node_repl_disabled);
metadata
.extra
.insert(MODEL_KEY.to_string(), self.model.to_string());
match &self.reasoning_effort {
Some(reasoning_effort) => {
metadata.extra.insert(
REASONING_EFFORT_KEY.to_string(),
reasoning_effort.to_string(),
);
}
None => {
metadata.extra.remove(REASONING_EFFORT_KEY);
}
}
}
}
#[derive(Clone, Debug, Default)]
@@ -136,7 +174,6 @@ pub(crate) struct TurnMetadataState {
node_repl_auto_review_required: bool,
node_repl_disabled: bool,
enriched_workspaces: RwLock<Option<BTreeMap<String, TurnMetadataWorkspace>>>,
tool_namespaces_info: RwLock<Option<TurnToolNamespacesInfo>>,
turn_started_at_unix_ms: RwLock<Option<i64>>,
responses_api_metadata: RwLock<BTreeMap<String, String>>,
responsesapi_client_metadata: RwLock<BTreeMap<String, String>>,
@@ -222,7 +259,6 @@ impl TurnMetadataState {
node_repl_auto_review_required: model_info.computer_use_review_required(),
node_repl_disabled: model_info.node_repl_disabled,
enriched_workspaces: RwLock::new(None),
tool_namespaces_info: RwLock::new(None),
turn_started_at_unix_ms: RwLock::new(None),
responses_api_metadata: RwLock::new(BTreeMap::new()),
responsesapi_client_metadata: RwLock::new(BTreeMap::new()),
@@ -234,11 +270,10 @@ impl TurnMetadataState {
pub(crate) fn current_meta_value_for_mcp_request(
&self,
context: McpTurnMetadataContext<'_>,
execution: ExecutionMetadata<'_>,
) -> Option<serde_json::Value> {
let mut responses_metadata = self.mcp_metadata_template();
// Use the issuing step's Node REPL restriction.
responses_metadata.node_repl_disabled = Some(context.node_repl_disabled);
execution.apply_to(&mut responses_metadata);
// Never serialize harness-owned tool inventory for external MCP servers.
responses_metadata.tool_namespaces_info = None;
let Value::Object(mut metadata) = responses_metadata.turn_metadata_value()? else {
@@ -247,25 +282,10 @@ impl TurnMetadataState {
metadata.remove(AGENT_NAME_KEY);
metadata.remove(PARENT_TURN_ID_KEY);
metadata.remove(ROOT_TURN_ID_KEY);
metadata.insert(
MODEL_KEY.to_string(),
Value::String(context.model.to_string()),
);
metadata.insert(
CODEX_VERSION_KEY.to_string(),
Value::String(env!("CARGO_PKG_VERSION").to_string()),
);
match context.reasoning_effort {
Some(reasoning_effort) => {
metadata.insert(
REASONING_EFFORT_KEY.to_string(),
Value::String(reasoning_effort.to_string()),
);
}
None => {
metadata.remove(REASONING_EFFORT_KEY);
}
}
if self
.user_input_requested_during_turn
.load(Ordering::Relaxed)
@@ -299,14 +319,6 @@ impl TurnMetadataState {
.store(true, Ordering::Relaxed);
}
pub(crate) fn set_tool_namespaces_info(&self, tool_namespaces_info: TurnToolNamespacesInfo) {
*self
.tool_namespaces_info
.write()
.unwrap_or_else(std::sync::PoisonError::into_inner) =
(!tool_namespaces_info.is_empty()).then_some(tool_namespaces_info);
}
pub(crate) fn set_parent_turn_id(&self, parent_turn_id: String) {
if parent_turn_id.trim().is_empty() {
return;
@@ -430,11 +442,7 @@ impl TurnMetadataState {
node_repl_auto_review_required: Some(self.node_repl_auto_review_required),
node_repl_disabled: Some(self.node_repl_disabled),
workspaces: self.current_workspaces(),
tool_namespaces_info: self
.tool_namespaces_info
.read()
.unwrap_or_else(std::sync::PoisonError::into_inner)
.clone(),
tool_namespaces_info: None,
turn_started_at_unix_ms: self.current_turn_started_at_unix_ms(),
extra,
..CodexResponsesMetadata::new(
+66 -69
View File
@@ -3,6 +3,7 @@ use super::*;
use crate::responses_metadata::ANALYTICS_ENABLED_KEY;
use crate::responses_metadata::AUTO_REVIEW_ENABLED_KEY;
use crate::responses_metadata::CONTEXT_WINDOW_ID_KEY;
use crate::responses_metadata::CodexResponsesMetadata;
use crate::responses_metadata::CodexResponsesRequestKind;
use crate::responses_metadata::CompactionTurnMetadata;
use crate::responses_metadata::FORKED_FROM_ORDINAL_EXCLUSIVE_KEY;
@@ -15,12 +16,12 @@ use crate::responses_metadata::ROOT_TURN_ID_KEY;
use crate::responses_metadata::SANDBOX_MODE_KEY;
use crate::responses_metadata::TOOL_NAMESPACES_INFO_KEY;
use crate::responses_metadata::TURN_TRIGGER_KEY;
use crate::responses_metadata::TurnToolFunctionInfo;
use crate::responses_metadata::TurnToolNamespaceInfo;
use crate::responses_metadata::TurnToolSource;
use crate::responses_metadata::WINDOW_ID_KEY;
use crate::responses_metadata::WINDOW_NUMBER_KEY;
use crate::responses_metadata::validate_extra_metadata;
use crate::session::step_context::StepContext;
use crate::session::tests::make_session_and_context;
use crate::session::tests::update_turn_settings_for_test;
use codex_analytics::CompactionImplementation;
use codex_analytics::CompactionPhase;
use codex_analytics::CompactionReason;
@@ -46,31 +47,31 @@ use std::time::Duration;
use tempfile::TempDir;
use tokio::process::Command;
fn test_mcp_turn_metadata_context() -> McpTurnMetadataContext<'static> {
McpTurnMetadataContext {
fn test_mcp_turn_metadata_context() -> ExecutionMetadata<'static> {
ExecutionMetadata {
model: "gpt-5.4",
reasoning_effort: Some(ReasoningEffortConfig::High),
node_repl_disabled: false,
auto_review_enabled: true,
node_repl_auto_review_required: true,
}
}
fn test_responses_metadata_json(
fn test_turn_responses_metadata(
state: &TurnMetadataState,
window_id: &str,
request_kind: CodexResponsesRequestKind,
) -> String {
state
.to_responses_metadata(
"installation-a".to_string(),
window_id.to_string(),
request_kind,
)
.turn_metadata_json()
.expect("turn metadata json")
) -> CodexResponsesMetadata {
state.to_responses_metadata(
"installation-a".to_string(),
window_id.to_string(),
CodexResponsesRequestKind::Turn,
)
}
fn test_turn_responses_metadata_json(state: &TurnMetadataState, window_id: &str) -> String {
test_responses_metadata_json(state, window_id, CodexResponsesRequestKind::Turn)
test_turn_responses_metadata(state, window_id)
.turn_metadata_json()
.expect("turn metadata json")
}
fn test_compaction_responses_metadata_json(
@@ -78,11 +79,14 @@ fn test_compaction_responses_metadata_json(
window_id: &str,
compaction: CompactionTurnMetadata,
) -> String {
test_responses_metadata_json(
state,
window_id,
CodexResponsesRequestKind::Compaction(compaction),
)
state
.to_responses_metadata(
"installation-a".to_string(),
window_id.to_string(),
CodexResponsesRequestKind::Compaction(compaction),
)
.turn_metadata_json()
.expect("turn metadata json")
}
fn test_turn_metadata_header(state: &TurnMetadataState) -> String {
@@ -547,12 +551,19 @@ fn turn_metadata_state_includes_model_and_reasoning_effort_only_in_request_meta(
assert!(meta.get("request_kind").is_none());
assert_eq!(meta["model"].as_str(), Some("gpt-5.4"));
assert_eq!(meta["reasoning_effort"].as_str(), Some("high"));
assert_eq!(meta[AUTO_REVIEW_ENABLED_KEY].as_bool(), Some(true));
assert_eq!(
meta[NODE_REPL_AUTO_REVIEW_REQUIRED_KEY].as_bool(),
Some(true)
);
let meta_without_reasoning_effort = state
.current_meta_value_for_mcp_request(McpTurnMetadataContext {
.current_meta_value_for_mcp_request(ExecutionMetadata {
model: "gpt-5.4",
reasoning_effort: None,
node_repl_disabled: false,
auto_review_enabled: false,
node_repl_auto_review_required: false,
})
.expect("turn metadata should be present");
assert_eq!(
@@ -566,6 +577,35 @@ fn turn_metadata_state_includes_model_and_reasoning_effort_only_in_request_meta(
);
}
#[tokio::test]
async fn execution_metadata_applies_captured_settings_to_responses_metadata() {
let (_session, mut turn) = make_session_and_context().await;
let config = Arc::make_mut(&mut turn.config);
config.approvals_reviewer = codex_protocol::config_types::ApprovalsReviewer::AutoReview;
config.permissions.approval_policy =
crate::config::Constrained::allow_any(codex_protocol::protocol::AskForApproval::OnRequest);
update_turn_settings_for_test(&mut turn, |settings| {
let model_info = Arc::make_mut(&mut settings.model_info);
model_info.slug = "captured-step-model".to_string();
model_info.node_repl_auto_review_required = true;
model_info.node_repl_disabled = true;
});
let step_context = StepContext::for_test(Arc::new(turn));
let mut metadata = step_context.turn.turn_metadata_state.to_responses_metadata(
"installation-a".to_string(),
"window-a".to_string(),
CodexResponsesRequestKind::Turn,
);
ExecutionMetadata::from_settings(&step_context.settings).apply_to(&mut metadata);
let metadata = metadata.turn_metadata_value().expect("turn metadata");
assert_eq!(metadata["model"], "captured-step-model");
assert_eq!(metadata[AUTO_REVIEW_ENABLED_KEY], true);
assert_eq!(metadata[NODE_REPL_AUTO_REVIEW_REQUIRED_KEY], true);
assert_eq!(metadata[NODE_REPL_DISABLED_KEY], true);
}
#[test]
fn turn_metadata_state_marks_user_input_requested_during_turn_only_for_mcp_request_meta() {
let temp_dir = TempDir::new().expect("temp dir");
@@ -816,25 +856,6 @@ fn turn_metadata_state_merges_client_metadata_without_replacing_reserved_fields(
),
]));
state.set_turn_started_at_unix_ms(/*turn_started_at_unix_ms*/ 1_700_000_000_123);
state.set_tool_namespaces_info(BTreeMap::from([(
"mcp__calendar".to_string(),
TurnToolNamespaceInfo {
name: "mcp__calendar".to_string(),
functions: BTreeMap::from([(
"lookup".to_string(),
TurnToolFunctionInfo {
name: "lookup".to_string(),
direct: true,
code_mode_name: Some("mcp__calendar__lookup".to_string()),
deferred: false,
source: TurnToolSource::Mcp {
server_name: "calendar".to_string(),
},
},
)]),
},
)]));
let header = test_turn_metadata_header(&state);
assert!(header.is_ascii());
assert!(!header.contains("東京"));
@@ -851,26 +872,7 @@ fn turn_metadata_state_merges_client_metadata_without_replacing_reserved_fields(
assert_eq!(json["thread_id"].as_str(), Some("thread-a"));
assert!(json.get(LEGACY_CODE_MODE_TOOL_NAMES_KEY).is_none());
assert_eq!(json["agent_name"].as_str(), Some("/root"));
assert_eq!(
json[TOOL_NAMESPACES_INFO_KEY],
serde_json::json!({
"mcp__calendar": {
"name": "mcp__calendar",
"functions": {
"lookup": {
"name": "lookup",
"direct": true,
"code_mode_name": "mcp__calendar__lookup",
"deferred": false,
"source": {
"kind": "mcp",
"server_name": "calendar",
},
},
},
},
})
);
assert!(json.get(TOOL_NAMESPACES_INFO_KEY).is_none());
assert!(json.get(INSTALLATION_ID_KEY).is_none());
assert!(json.get("x-codex-installation-id").is_none());
assert!(json.get("x-codex-parent-thread-id").is_none());
@@ -921,13 +923,8 @@ fn turn_metadata_state_merges_client_metadata_without_replacing_reserved_fields(
Some("thread-a:1")
);
let compatibility_headers = state
.to_responses_metadata(
"installation-a".to_string(),
"thread-a:1".to_string(),
CodexResponsesRequestKind::Turn,
)
.compatibility_headers();
let compatibility_headers =
test_turn_responses_metadata(&state, "thread-a:1").compatibility_headers();
let compatibility_metadata: Value = serde_json::from_str(
compatibility_headers
.get("x-codex-turn-metadata")
+146 -1
View File
@@ -1,4 +1,3 @@
use std::collections::HashMap;
use std::fs;
use std::path::Path;
@@ -20,6 +19,7 @@ use codex_history::InitialHistory;
use codex_history::RolloutItem;
use codex_model_provider_info::ModelProviderInfo;
use codex_model_provider_info::built_in_model_providers;
use codex_models_manager::bundled_models_response;
use codex_plugin::PluginHookSource;
use codex_plugin::PluginId;
use codex_protocol::items::parse_hook_prompt_fragment;
@@ -28,6 +28,8 @@ use codex_protocol::models::ContentItem;
use codex_protocol::models::NetworkPermissions;
use codex_protocol::models::PermissionProfile;
use codex_protocol::models::ResponseItem;
use codex_protocol::openai_models::ModelsResponse;
use codex_protocol::openai_models::ReasoningEffort;
use codex_protocol::permissions::NetworkSandboxPolicy;
use codex_protocol::protocol::AskForApproval;
use codex_protocol::protocol::EventMsg;
@@ -36,9 +38,13 @@ use codex_protocol::protocol::Op;
use codex_protocol::protocol::SessionSource;
use codex_protocol::protocol::SubAgentSource;
use codex_protocol::protocol::ThreadSettingsOverrides;
use codex_protocol::protocol::TurnSettingsUpdate;
use codex_protocol::protocol::TurnSettingsUpdateOutcome;
use codex_protocol::request_permissions::PermissionGrantScope;
use codex_protocol::request_permissions::RequestPermissionProfile;
use codex_protocol::request_permissions::RequestPermissionsResponse;
use codex_protocol::request_user_input::RequestUserInputAnswer;
use codex_protocol::request_user_input::RequestUserInputResponse;
use codex_protocol::user_input::UserInput;
use codex_thread_store::InMemoryThreadStore;
use codex_utils_absolute_path::AbsolutePathBuf;
@@ -70,10 +76,12 @@ use core_test_support::test_codex::test_codex;
use core_test_support::test_codex::turn_permission_fields;
use core_test_support::test_target_os;
use core_test_support::wait_for_event;
use core_test_support::wait_for_event_match;
use core_test_support::wait_for_mcp_server;
use pretty_assertions::assert_eq;
use serde_json::Value;
use serde_json::json;
use std::collections::HashMap;
use std::sync::Arc;
use std::time::Duration;
use tempfile::TempDir;
@@ -3605,6 +3613,143 @@ async fn pre_tool_use_json_deny_blocks_exec_command_before_execution() -> Result
Ok(())
}
#[tokio::test]
async fn pre_tool_use_hook_model_tracks_step_after_a_turn_update() -> Result<()> {
skip_if_no_network!(Ok(()));
let server = start_mock_server().await;
let request_user_input_call_id = "pause-before-pretooluse-exec-command";
let call_id = "pretooluse-exec-command";
let model_a = "pretooluse-attribution-a";
let model_b = "pretooluse-attribution-b";
let command = "echo attribution";
let args = serde_json::json!({ "cmd": command });
let responses = mount_sse_sequence(
&server,
vec![
sse(vec![
ev_response_created("resp-1"),
ev_function_call(
request_user_input_call_id,
"request_user_input",
r#"{"questions":[{"id":"continue","header":"Continue","question":"Continue?","options":[{"label":"Yes (Recommended)","description":"Run it."},{"label":"No","description":"Stop."}]}]}"#,
),
ev_completed("resp-1"),
]),
sse(vec![
ev_response_created("resp-2"),
core_test_support::responses::ev_function_call(
call_id,
"exec_command",
&serde_json::to_string(&args)?,
),
ev_completed("resp-2"),
]),
sse(vec![
ev_response_created("resp-3"),
ev_assistant_message("msg-1", "hook blocked it"),
ev_completed("resp-3"),
]),
],
)
.await;
let base_model = bundled_models_response()
.expect("bundled models should parse")
.models
.into_iter()
.find(|model| model.slug == "gpt-5.4")
.expect("bundled gpt-5.4 model");
let models = [model_a, model_b]
.into_iter()
.map(|slug| {
let mut model = base_model.clone();
model.slug = slug.to_string();
model
})
.collect();
let mut builder = test_codex()
.with_model(model_a)
.with_pre_build_hook(|home| {
write_pre_tool_use_hook(home, Some("^Bash$"), "json_deny", "blocked by pre hook")
.expect("failed to write pre tool use hook test fixture");
})
.with_config(move |config| {
trust_discovered_hooks(config);
for feature in [
Feature::StepModelSwitching,
Feature::DefaultModeRequestUserInput,
] {
config
.features
.enable(feature)
.expect("test config should allow feature update");
}
config.model_catalog = Some(ModelsResponse { models });
});
let test = builder.build(&server).await?;
let (sandbox_policy, permission_profile) =
turn_permission_fields(PermissionProfile::Disabled, test.config.cwd.as_path());
test.codex
.start_or_steer_turn(
TurnInputRequest::user_input(vec![UserInput::Text {
text: "run the blocked shell command".to_string(),
text_elements: Vec::new(),
}])
.with_thread_settings(ThreadSettingsOverrides {
approval_policy: Some(AskForApproval::Never),
sandbox_policy: Some(sandbox_policy),
permission_profile,
..Default::default()
}),
)
.await?;
let request = wait_for_event_match(&test.codex, |event| match event {
EventMsg::RequestUserInput(request) => Some(request.clone()),
_ => None,
})
.await;
assert_eq!(request.call_id, request_user_input_call_id);
let (reply, outcome) = oneshot::channel();
test.codex
.submit(Op::TurnSettings {
turn_id: request.turn_id.clone(),
update: TurnSettingsUpdate {
model: Some(model_b.to_string()),
effort: Some(Some(ReasoningEffort::High)),
..Default::default()
},
reply,
})
.await?;
assert_eq!(outcome.await?, TurnSettingsUpdateOutcome::Applied);
test.codex
.submit(Op::UserInputAnswer {
id: request.turn_id,
response: RequestUserInputResponse {
answers: HashMap::from([(
"continue".to_string(),
RequestUserInputAnswer {
answers: vec!["Yes (Recommended)".to_string()],
},
)]),
},
})
.await?;
wait_for_event(&test.codex, |event| {
matches!(event, EventMsg::TurnComplete(_))
})
.await;
assert_eq!(responses.requests().len(), 3);
let hook_inputs = read_pre_tool_use_hook_inputs(test.codex_home_path())?;
assert_eq!(hook_inputs.len(), 1);
assert_eq!(hook_inputs[0]["model"], model_b);
Ok(())
}
#[tokio::test]
async fn pre_tool_use_records_additional_context_for_exec_command() -> Result<()> {
skip_if_no_network!(Ok(()));
+86 -23
View File
@@ -8,6 +8,7 @@ use codex_core::TurnInputRequest;
use codex_core::config::Config;
use codex_features::Feature;
use codex_history::RolloutItem;
use codex_models_manager::bundled_models_response;
use codex_protocol::approvals::ElicitationRequest;
use codex_protocol::config_types::ApprovalsReviewer;
use codex_protocol::config_types::CollaborationMode;
@@ -17,12 +18,16 @@ use codex_protocol::items::McpToolCallStatus;
use codex_protocol::items::TurnItem;
use codex_protocol::models::NetworkPermissions;
use codex_protocol::models::PermissionProfile;
use codex_protocol::openai_models::ModelsResponse;
use codex_protocol::openai_models::ReasoningEffort;
use codex_protocol::protocol::AskForApproval;
use codex_protocol::protocol::ElicitationAction;
use codex_protocol::protocol::EventMsg;
use codex_protocol::protocol::Op;
use codex_protocol::protocol::ThreadSettingsOverrides;
use codex_protocol::protocol::TurnEnvironmentSelections;
use codex_protocol::protocol::TurnSettingsUpdate;
use codex_protocol::protocol::TurnSettingsUpdateOutcome;
use codex_protocol::request_permissions::PermissionGrantScope;
use codex_protocol::request_permissions::RequestPermissionProfile;
use codex_protocol::request_permissions::RequestPermissionsResponse;
@@ -325,6 +330,65 @@ async fn submit_user_turn(
Ok(())
}
fn attribution_models(model_slugs: [&str; 2]) -> Vec<codex_protocol::openai_models::ModelInfo> {
let base_model = bundled_models_response()
.expect("bundled models should parse")
.models
.into_iter()
.find(|model| model.slug == "gpt-5.4")
.expect("bundled gpt-5.4 model");
model_slugs
.into_iter()
.map(|slug| {
let mut model = base_model.clone();
model.slug = slug.to_string();
model
})
.collect()
}
async fn apply_turn_attribution_update(
test: &TestCodex,
request_user_input_call_id: &str,
model: &str,
) -> Result<()> {
let request = wait_for_event_match(&test.codex, |event| match event {
EventMsg::RequestUserInput(request) => Some(request.clone()),
_ => None,
})
.await;
assert_eq!(request.call_id, request_user_input_call_id);
let (reply, outcome) = tokio::sync::oneshot::channel();
test.codex
.submit(Op::TurnSettings {
turn_id: request.turn_id.clone(),
update: TurnSettingsUpdate {
model: Some(model.to_string()),
effort: Some(Some(ReasoningEffort::High)),
..Default::default()
},
reply,
})
.await?;
assert_eq!(outcome.await?, TurnSettingsUpdateOutcome::Applied);
test.codex
.submit(Op::UserInputAnswer {
id: request.turn_id,
response: RequestUserInputResponse {
answers: HashMap::from([(
"confirm_path".to_string(),
RequestUserInputAnswer {
answers: vec!["Yes (Recommended)".to_string()],
},
)]),
},
})
.await?;
Ok(())
}
async fn wait_for_mcp_tool_call_item(
test: &TestCodex,
call_id: &str,
@@ -1311,7 +1375,7 @@ async fn apps_default_writes_prompts_for_writes_but_not_reads() -> Result<()> {
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn mcp_tool_call_metadata_records_prior_request_user_input_tool() -> Result<()> {
async fn mcp_tool_call_metadata_uses_captured_step_after_request_user_input() -> Result<()> {
skip_if_no_network!(Ok(()));
let server = start_mock_server().await;
@@ -1368,16 +1432,26 @@ async fn mcp_tool_call_metadata_records_prior_request_user_input_tool() -> Resul
)
.await;
let model_a = "mcp-metadata-a";
let model_b = "mcp-metadata-b";
let models = attribution_models([model_a, model_b]);
let mut builder = search_capable_apps_builder(apps_server.chatgpt_base_url.clone())
.with_config(|config| {
.with_model(model_a)
.with_config(move |config| {
set_calendar_approval_mode(config, AppToolApproval::Approve);
config
.features
.enable(Feature::StepModelSwitching)
.expect("test config should allow feature update");
config.model_catalog = Some(ModelsResponse { models });
config.model_reasoning_effort = Some(ReasoningEffort::Low);
});
let test = builder.build(&server).await?;
submit_user_turn(
&test,
"Ask for confirmation, then create a calendar event.",
AskForApproval::Never,
AskForApproval::OnRequest,
PermissionProfile::Disabled,
Some(CollaborationMode {
mode: ModeKind::Plan,
@@ -1390,26 +1464,7 @@ async fn mcp_tool_call_metadata_records_prior_request_user_input_tool() -> Resul
)
.await?;
let request = wait_for_event_match(&test.codex, |event| match event {
EventMsg::RequestUserInput(request) => Some(request.clone()),
_ => None,
})
.await;
assert_eq!(request.call_id, request_user_input_call_id);
test.codex
.submit(Op::UserInputAnswer {
id: request.turn_id,
response: RequestUserInputResponse {
answers: HashMap::from([(
"confirm_path".to_string(),
RequestUserInputAnswer {
answers: vec!["Yes (Recommended)".to_string()],
},
)]),
},
})
.await?;
apply_turn_attribution_update(&test, request_user_input_call_id, model_b).await?;
let EventMsg::McpToolCallBegin(begin) = wait_for_event(&test.codex, |event| {
matches!(event, EventMsg::McpToolCallBegin(_))
@@ -1433,6 +1488,14 @@ async fn mcp_tool_call_metadata_records_prior_request_user_input_tool() -> Resul
.pointer("/params/_meta/x-codex-turn-metadata/user_input_requested_during_turn"),
Some(&json!(true))
);
assert_eq!(
apps_tool_call.pointer("/params/_meta/x-codex-turn-metadata/model"),
Some(&json!(model_b))
);
assert_eq!(
apps_tool_call.pointer("/params/_meta/x-codex-turn-metadata/reasoning_effort"),
Some(&json!("high"))
);
Ok(())
}
+87 -25
View File
@@ -169,14 +169,6 @@ fn direct_tool_settings_test() -> TestCodexBuilder {
})
}
fn advertises_apply_patch(request: &Value) -> bool {
request["tools"]
.as_array()
.expect("provider tools")
.iter()
.any(|tool| tool["type"] == "custom" && tool["name"] == "apply_patch")
}
fn paused_response(response_id: &str, call_id: &str) -> String {
sse(vec![
ev_response_created(response_id),
@@ -288,6 +280,15 @@ fn request_turn_id(request: &ResponsesRequest) -> String {
.to_string()
}
fn request_turn_metadata(request: &ResponsesRequest) -> Value {
serde_json::from_str(
request.body_json()["client_metadata"]["x-codex-turn-metadata"]
.as_str()
.expect("request should include turn metadata"),
)
.expect("valid turn metadata")
}
// Dynamic tools return the original payload, so handler truncation cannot hide a recorder bug.
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn tool_result_history_keeps_originating_model_across_switch_and_replay() -> Result<()> {
@@ -1569,7 +1570,9 @@ async fn captured_model_replans_tools_and_retains_the_issuing_request_router() -
for feature in [Feature::ShellTool, Feature::UnifiedExec] {
config.features.enable(feature).expect("enable shell tools");
}
config.tool_registry.turn_metadata_includes_tool_info = true;
for model in &mut config.model_catalog.as_mut().expect("models").models {
model.use_responses_lite = true;
model.shell_type = if model.slug == MODEL_B {
ConfigShellToolType::Disabled
} else {
@@ -1635,7 +1638,35 @@ async fn captured_model_replans_tools_and_retains_the_issuing_request_router() -
assert_eq!(
requests
.iter()
.map(advertises_apply_patch)
.map(|request| {
core_test_support::responses::namespace_child_tool(
&request["input"][0],
"functions",
"apply_patch",
)
.is_some()
})
.collect::<Vec<_>>(),
vec![false, true, false],
);
let metadata = requests
.iter()
.map(|request| {
serde_json::from_str::<Value>(
request["client_metadata"]["x-codex-turn-metadata"]
.as_str()
.expect("request metadata"),
)
})
.collect::<std::result::Result<Vec<_>, _>>()?;
assert_eq!(
metadata
.iter()
.map(|metadata| {
metadata["tool_namespaces_info"]["functions"]["functions"]
.get("apply_patch")
.is_some()
})
.collect::<Vec<_>>(),
vec![false, true, false],
);
@@ -1643,12 +1674,16 @@ async fn captured_model_replans_tools_and_retains_the_issuing_request_router() -
requests
.iter()
.map(|request| {
request["tools"]
.as_array()
.expect("provider tools")
.iter()
.filter_map(|tool| tool["name"].as_str())
.filter(|name| matches!(*name, "exec_command" | "write_stdin"))
["exec_command", "write_stdin"]
.into_iter()
.filter(|name| {
core_test_support::responses::namespace_child_tool(
&request["input"][0],
"functions",
name,
)
.is_some()
})
.collect::<Vec<_>>()
})
.collect::<Vec<_>>(),
@@ -1779,7 +1814,7 @@ async fn captured_model_enables_and_executes_code_mode() -> Result<()> {
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn response_metadata_uses_the_captured_model_after_a_turn_update() -> Result<()> {
async fn response_metadata_uses_the_captured_step_after_a_turn_update() -> Result<()> {
skip_if_no_network!(Ok(()));
let server = start_mock_server().await;
@@ -1800,15 +1835,19 @@ async fn response_metadata_uses_the_captured_model_after_a_turn_update() -> Resu
.await;
let test = step_settings_test().build_with_auto_env(&server).await?;
let request = start_paused_turn(&test.codex).await?;
apply_turn_settings(
&test.codex,
&request.turn_id,
TurnSettingsUpdate {
model: Some(MODEL_B.to_string()),
..Default::default()
},
)
.await?;
assert_eq!(
submit_turn_settings(
&test.codex,
&request.turn_id,
TurnSettingsUpdate {
model: Some(MODEL_B.to_string()),
effort: Some(Some(ReasoningEffort::High)),
..Default::default()
},
)
.await?,
TurnSettingsUpdateOutcome::Applied
);
answer_paused_turn(&test.codex, &request.turn_id).await?;
let mut reroutes = Vec::new();
@@ -1836,6 +1875,29 @@ async fn response_metadata_uses_the_captured_model_after_a_turn_update() -> Resu
.collect::<Vec<_>>(),
vec![json!(MODEL_A), json!(MODEL_B)]
);
assert_eq!(
response_mock
.requests()
.iter()
.map(request_turn_metadata)
.map(|metadata| {
json!({
"model": metadata["model"],
"reasoning_effort": metadata["reasoning_effort"],
})
})
.collect::<Vec<_>>(),
vec![
json!({
"model": MODEL_A,
"reasoning_effort": "low",
}),
json!({
"model": MODEL_B,
"reasoning_effort": "high",
}),
]
);
// B's matching response header is not a reroute from the turn's initial A.
// Buffering metadata likewise belongs to the captured B step.
assert_eq!(