openai/codex
1
0
Fork 0
mirror of https://github.com/openai/codex.git synced 2026-11-07 20:30:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

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

Browse Source 
## 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
2026-09-09 17:45:37 +00:00
committed by copyberry
parent ed4ca07ba6
commit 205f3671e1
24 changed files with 1053 additions and 241 deletions
Download Patch File Download Diff File Expand all files Collapse all files
codex-rs/core/src/unified_exec/async_watcher.rs
+12 -3
View File
@@ -24,6 +24,7 @@ use crate::unified_exec::head_tail_buffer::HeadTailBuffer;
use codex_core_plugins::PluginCommandAttribution;
use codex_protocol::exec_output::ExecToolCallOutput;
use codex_protocol::exec_output::StreamOutput;
use codex_protocol::openai_models::ModelInfo;
use codex_protocol::protocol::EventMsg;
use codex_protocol::protocol::ExecCommandOutputDeltaEvent;
use codex_protocol::protocol::ExecCommandSource;
@@ -164,9 +165,7 @@ pub(crate) fn start_streaming_output(
#[allow(clippy::too_many_arguments)]
pub(crate) fn spawn_exit_watcher(
process: Arc<UnifiedExecProcess>,
session_ref: Arc<Session>,
turn_ref: Arc<TurnContext>,
call_id: String,
context: &UnifiedExecContext,
command: Vec<String>,
cwd: PathUri,
process_id: i32,
@@ -176,6 +175,10 @@ pub(crate) fn spawn_exit_watcher(
network_denial_monitor: Option<tokio::task::JoinHandle<()>>,
plugin_metrics_sidecar: Option<SharedPluginMetricsSidecar>,
) {
let session_ref = Arc::clone(&context.session);
let turn_ref = Arc::clone(&context.step_context.turn);
let model_info = Arc::clone(&context.step_context.settings.model_info);
let call_id = context.call_id.clone();
let exit_token = process.cancellation_token();
let output_drained = process.output_drained_notify();
let interaction_lock = process.interaction_lock();
@@ -200,6 +203,7 @@ pub(crate) fn spawn_exit_watcher(
emit_failed_exec_end_for_unified_exec(
session_ref,
turn_ref,
model_info,
call_id,
command,
cwd,
@@ -225,6 +229,7 @@ pub(crate) fn spawn_exit_watcher(
emit_exec_end_for_unified_exec(
session_ref,
turn_ref,
model_info,
call_id,
command,
cwd,
@@ -337,6 +342,7 @@ impl Emitter {
pub(crate) async fn emit_exec_end_for_unified_exec(
session_ref: Arc<Session>,
turn_ref: Arc<TurnContext>,
model_info: Arc<ModelInfo>,
call_id: String,
command: Vec<String>,
cwd: PathUri,
@@ -360,6 +366,7 @@ pub(crate) async fn emit_exec_end_for_unified_exec(
let event_ctx = ToolEventCtx::new(
session_ref.as_ref(),
turn_ref.as_ref(),
&model_info,
&call_id,
/*turn_diff_tracker*/ None,
);
@@ -385,6 +392,7 @@ pub(crate) async fn emit_exec_end_for_unified_exec(
pub(crate) async fn emit_failed_exec_end_for_unified_exec(
session_ref: Arc<Session>,
turn_ref: Arc<TurnContext>,
model_info: Arc<ModelInfo>,
call_id: String,
command: Vec<String>,
cwd: PathUri,
@@ -416,6 +424,7 @@ pub(crate) async fn emit_failed_exec_end_for_unified_exec(
let event_ctx = ToolEventCtx::new(
session_ref.as_ref(),
turn_ref.as_ref(),
&model_info,
&call_id,
/*turn_diff_tracker*/ None,
);
codex-rs/core/src/unified_exec/async_watcher_tests.rs
+15 -4
View File
@@ -208,7 +208,7 @@ async fn exit_watcher_waits_for_late_network_denial_before_classifying_end() ->
stdout_tx,
exit_tx,
transcript,
context,
mut context,
rx_event,
} = streaming_output_harness().await?;
@@ -226,11 +226,15 @@ async fn exit_watcher_waits_for_late_network_denial_before_classifying_end() ->
#[allow(deprecated)]
let cwd = context.step_context.turn.cwd.clone().into();
let step = Arc::get_mut(&mut context.step_context).expect("unshared test step");
let model_info = Arc::make_mut(&mut Arc::make_mut(&mut step.settings).model_info);
model_info.truncation_policy = codex_protocol::openai_models::TruncationPolicyConfig {
mode: codex_protocol::openai_models::TruncationMode::Bytes,
limit: 4,
};
spawn_exit_watcher(
Arc::clone(&process),
Arc::clone(&context.session),
Arc::clone(&context.step_context.turn),
context.call_id,
&context,
vec!["proof".to_string()],
cwd,
/*process_id*/ 123,
@@ -266,6 +270,13 @@ async fn exit_watcher_waits_for_late_network_denial_before_classifying_end() ->
Some("LATE_DENIAL")
)
);
assert_eq!(
item.formatted_output,
Some(codex_utils_output_truncation::formatted_truncate_text(
"LATE_DENIAL",
codex_utils_output_truncation::TruncationPolicy::Bytes(4),
))
);
assert!(
elapsed >= Duration::from_millis(10) && elapsed < TRAILING_OUTPUT_GRACE,
"completion should wait for denial without falling back to the output grace: {elapsed:?}"
codex-rs/core/src/unified_exec/process_manager.rs
+6 -5
View File
@@ -406,6 +406,7 @@ async fn emit_failed_initial_exec_end_if_unstored(
emit_failed_exec_end_for_unified_exec(
Arc::clone(&context.session),
Arc::clone(&context.step_context.turn),
Arc::clone(&context.step_context.settings.model_info),
context.call_id.clone(),
request.command.clone(),
cwd,
@@ -532,6 +533,7 @@ impl UnifiedExecProcessManager {
let event_ctx = ToolEventCtx::new(
context.session.as_ref(),
context.step_context.turn.as_ref(),
&context.step_context.settings.model_info,
&context.call_id,
/*turn_diff_tracker*/ None,
);
@@ -766,6 +768,7 @@ impl UnifiedExecProcessManager {
emit_exec_end_for_unified_exec(
Arc::clone(&context.session),
Arc::clone(&context.step_context.turn),
Arc::clone(&context.step_context.settings.model_info),
context.call_id.clone(),
request.command.clone(),
cwd.clone(),
@@ -796,8 +799,8 @@ impl UnifiedExecProcessManager {
raw_output: collected,
truncation_policy: context
.step_context
.turn
.model_info()
.settings
.model_info
.truncation_policy
.into(),
max_output_tokens: request.max_output_tokens,
@@ -1175,9 +1178,7 @@ impl UnifiedExecProcessManager {
spawn_exit_watcher(
Arc::clone(&process),
Arc::clone(&context.session),
Arc::clone(&context.step_context.turn),
context.call_id.clone(),
context,
command.to_vec(),
cwd,
process_id,