mirror of
https://github.com/l0ng-ai/openidea.git
synced 2026-11-08 08:52:12 +00:00
fix(agent): report Claude Code turns that fail instead of ending them as ready
Prompts are sent with the origin `human` and their results echo it, but `belongs` treated any result with an origin as another turn's. The result was ignored and the turn ended on `command_lifecycle`, which dropped the failure the turn had recorded. A signed-out Claude Code showed an empty reply and a Ready task, with no error. Only origins other than `human` now mark a result as foreign, so failures, usage limits and the `max_tokens` and `refusal` stop reasons reach the task. A turn ended by `command_lifecycle` also keeps its recorded failure.
This commit is contained in:
1 parent
67cad01ba7
commit
d6288ecb85
2 files changed
+52
-2
No files matched your search
@@ -915,7 +915,11 @@ impl Session {
|
||||
} else {
|
||||
"end_turn"
|
||||
};
|
||||
finished = st.turn.take().map(|t| (t, TurnEnd::Stop(stop)));
|
||||
finished = st.turn.take().map(|mut t| {
|
||||
// Claude Code reports a turn that failed as cancelled here.
|
||||
let failure = t.failure.take().filter(|_| !t.cancelled);
|
||||
(t, failure.unwrap_or(TurnEnd::Stop(stop)))
|
||||
});
|
||||
}
|
||||
}
|
||||
"assistant" if msg.get("parent_tool_use_id").is_none_or(Value::is_null) => {
|
||||
@@ -1139,7 +1143,8 @@ where
|
||||
/// Whether `result` ends `turn`: it answers the turn's last message, or (stopped) any of
|
||||
/// its messages, or names none (the CLI's own error before the turn began).
|
||||
fn belongs(turn: &Turn, result: &Value) -> bool {
|
||||
if result.get("origin").is_some_and(|o| !o.is_null()) {
|
||||
// Prompts are sent with the origin `human`, and their results echo it.
|
||||
if result.get("origin").is_some_and(|o| !o.is_null() && o.get("kind").and_then(Value::as_str) != Some("human")) {
|
||||
return false;
|
||||
}
|
||||
let mut uuids = strings(result.get("user_message_uuids"));
|
||||
|
||||
@@ -738,6 +738,50 @@ async fn a_stop_ahead_of_its_prompt_cancels_it() {
|
||||
assert!(cli.sent.lock()[0].iter().all(|m| m["type"] != "user"), "nothing was sent");
|
||||
}
|
||||
|
||||
#[tokio::test(flavor = "multi_thread")]
|
||||
async fn a_failed_turn_reports_its_failure() {
|
||||
let text = "Failed to authenticate: OAuth session expired and could not be refreshed";
|
||||
let lifecycle = |uuid: &str, state: &str| {
|
||||
Frame::Cli(json!({ "type": "command_lifecycle", "command_uuid": uuid, "state": state }))
|
||||
};
|
||||
let result = |uuid: &str| {
|
||||
Frame::Cli(json!({
|
||||
"type": "result", "subtype": "success", "is_error": true, "result": text, "stop_reason": "stop_sequence",
|
||||
"terminal_reason": "api_error", "origin": { "kind": "human" }, "user_message_uuids": [uuid],
|
||||
}))
|
||||
};
|
||||
// As Claude Code sends it, and with the turn's end ahead of its result.
|
||||
for lifecycle_first in [false, true] {
|
||||
let mut process = load_fixture("simple").remove(0);
|
||||
process.truncate(6);
|
||||
let Some(Frame::Shim(prompt)) = process.last() else { panic!("the fixture's prompt") };
|
||||
let uuid = prompt["uuid"].as_str().unwrap().to_owned();
|
||||
process.push(Frame::Cli(json!({
|
||||
"type": "assistant", "error": "authentication_failed", "parent_tool_use_id": null,
|
||||
"message": { "id": "m", "model": "<synthetic>", "content": [{ "type": "text", "text": text }] },
|
||||
})));
|
||||
if lifecycle_first {
|
||||
process.extend([lifecycle(&uuid, "cancelled"), result(&uuid)]);
|
||||
} else {
|
||||
process.extend([result(&uuid), lifecycle(&uuid, "cancelled")]);
|
||||
}
|
||||
let cli = Arc::new(Replay {
|
||||
processes: Mutex::new(VecDeque::from([process])),
|
||||
sent: Arc::default(),
|
||||
launched: Mutex::default(),
|
||||
errors: Arc::default(),
|
||||
});
|
||||
let (dir, cwd) = (tempfile::tempdir().unwrap(), tempfile::tempdir().unwrap());
|
||||
let h =
|
||||
Harness::connect(cli.clone(), dir.path().into(), dir.path().join("p"), cwd.path().into(), Vec::new()).await;
|
||||
let s = h.new_session().await;
|
||||
let Err(RpcError::Server { message, .. }) = h.prompt(&s, "hi").await else {
|
||||
panic!("the turn should fail (lifecycle first: {lifecycle_first})");
|
||||
};
|
||||
assert_eq!(message, format!("Claude Code authentication failed: {text}"));
|
||||
}
|
||||
}
|
||||
|
||||
/// Fork and undo against the real CLI, across connections, in a symlinked directory
|
||||
/// (macOS `/tmp`).
|
||||
#[tokio::test(flavor = "multi_thread")]
|
||||
@@ -829,6 +873,7 @@ fn which_result_ends_a_turn() {
|
||||
assert!(!belongs(&t, &json!({"user_message_uuids": ["a"]})), "a steer superseded it");
|
||||
assert!(belongs(&t, &json!({"subtype": "error_during_execution"})), "names no message");
|
||||
assert!(!belongs(&t, &json!({"origin": {"kind": "task-notification"}})), "Claude Code's own turn");
|
||||
assert!(belongs(&t, &json!({"origin": {"kind": "human"}, "user_message_uuids": ["b"]})), "the prompt's own result");
|
||||
assert!(belongs(&turn(&["a", "b"], true), &json!({"user_message_uuids": ["a"]})));
|
||||
}
|
||||
|
||||
|
||||
Reference in new issue
Block a user