Skip to content

fix(codex): distinguish command exits from wrapper completion #243

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
Abhijeet Prasad (AbhiPrasad) merged 3 commits into from
Oct 28, 2026
Merged

fix(codex): distinguish command exits from wrapper completion #243

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ef740c5
fix(codex): distinguish command exits from wrapper completion
AbhiPrasad Oct 28, 2026
0513faf
fix(codex): make command outcome checks portable in ci
AbhiPrasad Oct 28, 2026
9a542ce
fix(codex): retain rejected allsettled tool results
AbhiPrasad Oct 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
131 changes: 44 additions & 87 deletions bt-daemon/src/translate/codex.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,7 +20,7 @@
use super::git::GitMetadataCache;
use super::recent::{RecentMap, RecentSet};
use super::tool::{
error_text, explicit_skill_metadata, nonempty_error_text, tool_approval_metadata, ToolApproval,
add_tool_approval, explicit_skill_metadata, tool_approval_metadata, ToolApproval,
};
use super::{
local_username, AgentTranslator, SessionCtx, SpanOp, SpanRow, SpanType, TranslatorFactory,
Expand All @@ -35,6 +35,8 @@ use std::collections::HashMap;
use std::path::Path;
use std::sync::{Arc, OnceLock};

mod tool_output;

const SPAWN_AGENT_TOOL: &str = "spawn_agent";
const MISSING_TOOL_OUTPUT_ERROR: &str = "Tool output missing before turn ended";
const MAX_OPEN_TOOL_SPAN_MS: i64 = 24 * 60 * 60 * 1000;
Expand Down Expand Up @@ -169,6 +171,13 @@ struct OpenLlm {
output_preset: bool,
}

struct OpenTool {
span_id: String,
turn_id: String,
start_ms: i64,
name: String,
}

struct Scope {
path: String,
kind: ScopeKind,
Expand All @@ -182,7 +191,7 @@ struct Scope {
open_turns: Vec<OpenTurn>,
message_history: Vec<Value>,
open_llm: Option<OpenLlm>,
open_tools: HashMap<String, (String, String, i64)>, // call_id -> (tool span_id, turn_id, start_ms)
open_tools: HashMap<String, OpenTool>, // call_id -> native tool invocation
last_turn_end_ms: Option<i64>,
last_turn_output: Option<Value>,
turn_seq: u32,
Expand Down Expand Up @@ -1191,7 +1200,15 @@ impl CodexTranslator {
tags: if tags.is_empty() { None } else { Some(tags) },
..Default::default()
}));
scope.open_tools.insert(call_id, (span_id, turn_id, ts));
scope.open_tools.insert(
call_id,
OpenTool {
span_id,
turn_id,
start_ms: ts,
name: tool_name,
},
);
}

fn on_tool_output(
Expand All @@ -1206,29 +1223,33 @@ impl CodexTranslator {
return;
};
push_tool_result(scope, Some(&call_id), payload);
let Some((span_id, turn_id, _start_ms)) = scope.open_tools.remove(&call_id) else {
let Some(tool) = scope.open_tools.remove(&call_id) else {
return;
};
if let Some(turn) = scope
.open_turns
.iter_mut()
.find(|turn| turn.turn_id == turn_id)
.find(|turn| turn.turn_id == tool.turn_id)
{
turn.last_child_end_ms = Some(turn.last_child_end_ms.map_or(ts, |p| p.max(ts)));
}
let output = payload
.get("output")
.or_else(|| payload.get("result"))
.cloned();
let error = output.as_ref().and_then(classify_tool_output);
let mut outcome = output
.as_ref()
.map(|output| tool_output::classify(&tool.name, output))
.unwrap_or_default();
add_tool_approval(&mut outcome.metadata, Some(ToolApproval::Approved));
ops.push(SpanOp::Merge(SpanRow {
span_id,
span_id: tool.span_id,
root_span_id: self.effective_root_span_id.clone(),
parent_span_ids: self.turn_parent_span_ids(&turn_id),
parent_span_ids: self.turn_parent_span_ids(&tool.turn_id),
end_ms: Some(ts),
output,
metadata: Some(tool_approval_metadata(Some(ToolApproval::Approved))),
error,
metadata: Some(Value::Object(outcome.metadata)),
error: outcome.error,
..Default::default()
}));
}
Expand Down Expand Up @@ -1321,7 +1342,7 @@ impl CodexTranslator {
let has_open_tool = scope
.open_tools
.values()
.any(|(_, owner, _)| owner == &turn_id);
.any(|tool| tool.turn_id == turn_id);

if scope
.open_llm
Expand Down Expand Up @@ -1395,11 +1416,11 @@ impl CodexTranslator {
let call_ids: Vec<String> = scope
.open_tools
.iter()
.filter(|(_, (_, owner, _))| owner == turn_id)
.filter(|(_, tool)| tool.turn_id == turn_id)
.map(|(call_id, _)| call_id.clone())
.collect();
for call_id in call_ids {
if let Some((span_id, _, start_ms)) = scope.open_tools.remove(&call_id) {
if let Some(tool) = scope.open_tools.remove(&call_id) {
let mut metadata = tool_approval_metadata(Some(ToolApproval::Approved));
if let Some(reason) = interruption_reason {
if let Some(fields) = metadata.as_object_mut() {
Expand All @@ -1408,12 +1429,12 @@ impl CodexTranslator {
}
}
ops.push(SpanOp::Merge(SpanRow {
span_id,
span_id: tool.span_id,
root_span_id: self.effective_root_span_id.clone(),
parent_span_ids: parent_span_ids.clone(),
end_ms: Some(end_ms.map_or_else(
|| start_ms.saturating_add(MAX_OPEN_TOOL_SPAN_MS),
|end| end.min(start_ms.saturating_add(MAX_OPEN_TOOL_SPAN_MS)),
|| tool.start_ms.saturating_add(MAX_OPEN_TOOL_SPAN_MS),
|end| end.min(tool.start_ms.saturating_add(MAX_OPEN_TOOL_SPAN_MS)),
)),
metadata: Some(metadata),
error: interruption_reason
Expand Down Expand Up @@ -1586,29 +1607,18 @@ impl CodexTranslator {
..Default::default()
}));
}
let tools: Vec<(String, String, i64, String)> = scope
.open_tools
.drain()
.map(|(_, (span_id, turn_id, start_ms))| {
(
span_id,
turn_id,
start_ms,
MISSING_TOOL_OUTPUT_ERROR.to_string(),
)
})
.collect();
for (sid, turn_id, start_ms, error) in tools {
let tools: Vec<OpenTool> = scope.open_tools.drain().map(|(_, tool)| tool).collect();
for tool in tools {
ops.push(SpanOp::Merge(SpanRow {
span_id: sid,
span_id: tool.span_id,
root_span_id: self.effective_root_span_id.clone(),
parent_span_ids: self.turn_parent_span_ids(&turn_id),
parent_span_ids: self.turn_parent_span_ids(&tool.turn_id),
end_ms: Some(end_ms.map_or_else(
|| start_ms.saturating_add(MAX_OPEN_TOOL_SPAN_MS),
|end| end.min(start_ms.saturating_add(MAX_OPEN_TOOL_SPAN_MS)),
|| tool.start_ms.saturating_add(MAX_OPEN_TOOL_SPAN_MS),
|end| end.min(tool.start_ms.saturating_add(MAX_OPEN_TOOL_SPAN_MS)),
)),
metadata: Some(tool_approval_metadata(Some(ToolApproval::Approved))),
error: Some(error),
error: Some(MISSING_TOOL_OUTPUT_ERROR.to_string()),
..Default::default()
}));
}
Expand Down Expand Up @@ -1809,59 +1819,6 @@ fn args_object(args: Option<&Value>) -> Option<Map<String, Value>> {
}
}

fn has_error_marker(value: &Value) -> bool {
match value {
Value::Null => false,
Value::Bool(value) => *value,
Value::String(_) => nonempty_error_text(value).is_some(),
Value::Array(values) => !values.is_empty(),
Value::Object(values) => !values.is_empty(),
Value::Number(_) => true,
}
}

fn classify_tool_output(output: &Value) -> Option<String> {
if let Some(object) = output.as_object() {
if object.get("is_error").and_then(Value::as_bool) == Some(true)
|| object.get("isError").and_then(Value::as_bool) == Some(true)
|| matches!(
object.get("status").and_then(Value::as_str),
Some("error" | "failed")
)
{
return Some(error_text(Some(output), "Tool execution failed"));
}
if let Some(error) = object.get("error").filter(|error| has_error_marker(error)) {
return Some(error_text(Some(error), "Tool execution failed"));
}
if let Some(exit_code) = object
.get("exit_code")
.or_else(|| object.get("exitCode"))
.and_then(Value::as_i64)
{
if exit_code != 0 {
return Some(error_text(Some(output), &format!("Exit code {exit_code}")));
}
}
}
if let Some(text) = output.as_str() {
let first = text.lines().next().unwrap_or(text);
if first.to_ascii_lowercase().starts_with("error:") {
return Some(first.to_string());
}
if let Some(code) = first
.strip_prefix("Exit code ")
.and_then(|value| value.split_whitespace().next())
.and_then(|value| value.parse::<i64>().ok())
{
if code != 0 {
return Some(first.to_string());
}
}
}
None
}

#[derive(Default)]
struct SkillLoad {
name: Option<String>,
Expand Down
Loading
Loading