Use captured step settings for tool planning and execution (#44242)

## Why

After a model change within a turn, tools could still use the turn's original model capabilities, output limits, and telemetry attribution.

## What changed

- Build each step's tool plan from its captured model settings.
- Use those settings for image support and detail, command and MCP output truncation, extension metadata, and tool telemetry.
- Preserve the issuing step's model for delayed command completion events while allowing later `write_stdin` calls to use their own step's output limits.

## Testing

Add regression coverage for model changes within a turn, including tool replanning, Code Mode execution, image handling, MCP tool and resource output limits, extension settings, and delayed command completion.

GitOrigin-RevId: ef991598e89a846889baf770a0c9e12e56886d5d
This commit is contained in:
Abhinav authored and copyberry committed 2026-09-09 17:45:37 +00:00
1 parent ed4ca07ba6
commit 205f3671e1
24 files changed
+1053 -241

No files matched your search

+9 -8
View File
@@ -239,7 +239,7 @@ pub(crate) async fn handle_mcp_tool_call(
let status = if result.is_ok() { "ok" } else { "error" };
let outcome = McpCallMetricOutcome::from_status(status);
emit_mcp_call_metrics(
turn_context.as_ref(),
&step_context.session_telemetry,
&outcome,
&server,
&tool_name,
@@ -348,7 +348,7 @@ pub(crate) async fn handle_mcp_tool_call(
let status = if result.is_ok() { "ok" } else { "error" };
let outcome = McpCallMetricOutcome::from_status(status);
emit_mcp_call_metrics(
turn_context.as_ref(),
&step_context.session_telemetry,
&outcome,
&server,
&tool_name,
@@ -489,7 +489,7 @@ async fn handle_approved_mcp_tool_call(
.map(|(connector_id, action_name)| HostedFileUploadContext {
connector_id: connector_id.clone(),
action_name: action_name.clone(),
model: turn_context.model_info().slug.clone(),
model: step_context.settings.model_info.slug.clone(),
});
let rewritten_arguments = rewrite_mcp_tool_arguments_for_openai_files(
sess,
@@ -545,7 +545,7 @@ async fn handle_approved_mcp_tool_call(
)
.await;
let result = sanitize_mcp_tool_result_for_model(
&turn_context.model_info().input_modalities,
&step_context.settings.model_info.input_modalities,
Ok(result),
)?;
Ok(maybe_request_codex_apps_auth_elicitation(
@@ -590,11 +590,11 @@ async fn handle_approved_mcp_tool_call(
truncate_mcp_tool_result_for_event(&result),
)
.await;
maybe_track_codex_app_used(sess, turn_context, &server, &metadata).await;
maybe_track_codex_app_used(sess, step_context, &server, &metadata).await;
let outcome = mcp_call_metric_outcome(&result);
emit_mcp_call_metrics(
turn_context,
&step_context.session_telemetry,
&outcome,
&server,
&tool_name,
@@ -1055,7 +1055,7 @@ async fn notify_mcp_tool_call_completed(
async fn maybe_track_codex_app_used(
sess: &Session,
turn_context: &TurnContext,
step_context: &StepContext,
server: &str,
metadata: &McpToolApprovalMetadata,
) {
@@ -1075,8 +1075,9 @@ async fn maybe_track_codex_app_used(
InvocationType::Implicit
};
let turn_context = &step_context.turn;
let tracking = build_track_events_context(
turn_context.model_info().slug.clone(),
step_context.settings.model_info.slug.clone(),
sess.thread_id.to_string(),
turn_context.sub_id.clone(),
turn_context.originator.clone(),