Skip to content
Open
Changes from all commits
Commits
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
63 changes: 56 additions & 7 deletions crates/libsy/src/algorithms/util/tool_signals.rs
Original file line number Diff line number Diff line change
Expand Up @@ -1139,13 +1139,22 @@ fn extract_tool_signals_with_window_and_semantics(
.filter(|value| {
value["output"].is_string() && value.get("exit_code").is_some()
});
let shell_failed = shell_result.as_ref().is_some_and(|value| {
value["exit_code"].as_i64().is_some_and(|code| code != 0)
|| value["success"].as_bool() == Some(false)
|| value["error"]
.as_str()
.is_some_and(|error| !error.trim().is_empty())
}) || has_nonzero_exit_status(&text.to_lowercase());
// Codex reports the exit status in a header before `Output:`.
// An exit line after it is part of what was read.
let status = text
.split_once("\nOutput:\n")
.map_or(text.as_str(), |(header, _)| header);
// Hermes reports it in fields, so its output is not scanned.
let shell_failed = match &shell_result {
Some(value) => {
value["exit_code"].as_i64().is_some_and(|code| code != 0)
|| value["success"].as_bool() == Some(false)
|| value["error"]
.as_str()
.is_some_and(|error| !error.trim().is_empty())
}
None => has_nonzero_exit_status(&status.to_lowercase()),
};
if is_missing_file(&text) || (!is_error && !shell_failed) {
// Keep the window slot after dropping retrieved content.
if !text.is_empty() {
Expand Down Expand Up @@ -2161,6 +2170,46 @@ mod tests {
assert_eq!(ToolSignals::from_request(&request, None).severity, 0.0);
}

#[test]
fn codex_read_status_comes_from_the_header() {
let read = |header: &str| {
let output = format!(
"{header}\nOutput:\nTraceback (most recent call last):\nMemoryError\nProcess exited with code 1\n"
);
let call = exec_command(json!({"cmd": "cat run.log"}));
ToolSignals::from_request(&call_with_result(call, tr(&output)), None).severity
};
// The exit line belongs to the log that was read.
assert_eq!(read("Chunk ID: abc\nProcess exited with code 0"), 0.0);
// The read itself failed, so its output counts.
assert_eq!(read("Chunk ID: abc\nProcess exited with code 1"), CRITICAL);
}

#[test]
fn hermes_read_status_comes_from_exit_code() {
let read = |exit_code: i64| {
let output = json!({
"output": "Traceback (most recent call last):\nMemoryError\nProcess exited with code 1\n",
"exit_code": exit_code,
"error": null,
});
let call = Message {
role: Role::Assistant,
content: vec![ContentBlock::ToolCall(ToolCall {
id: String::new(),
name: "terminal".to_string(),
arguments: json!({"command": "cat run.log"}),
})],
};
ToolSignals::from_request(&call_with_result(call, tr(&output.to_string())), None)
.severity
};
// The exit line belongs to the log that was read.
assert_eq!(read(0), 0.0);
// The read itself failed, so its output counts.
assert_eq!(read(1), CRITICAL);
}

#[test]
fn argv_commands_keep_word_boundaries() {
let signal = |argv: Value| {
Expand Down
Loading