valknar/llmx
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

bug: sandbox denied error logs (#4874)

Browse Source 
Check on STDOUT / STDERR or aggregated output for some logs when sanbox
is denied
This commit is contained in:
jif-oai
2025-10-09 17:01:01 +01:00
committed by GitHub
parent 0026b12615
commit ca6a0358de
4 changed files with 237 additions and 89 deletions
Download Patch File Download Diff File Expand all files Collapse all files

73
codex-rs/core/tests/suite/tools.rs
View File

@@ -403,79 +403,6 @@ async fn shell_timeout_includes_timeout_prefix_and_metadata() -> Result<()> {
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn shell_sandbox_denied_truncates_error_output() -> Result<()> {
skip_if_no_network!(Ok(()));
let server = start_mock_server().await;
let mut builder = test_codex();
let test = builder.build(&server).await?;
let call_id = "shell-denied";
let long_line = "this is a long stderr line that should trigger truncation 0123456789abcdefghijklmnopqrstuvwxyz";
let script = format!(
"for i in $(seq 1 500); do >&2 echo '{long_line}'; done; cat <<'EOF' > denied.txt\ncontent\nEOF",
);
let args = json!({
"command": ["/bin/sh", "-c", script],
"timeout_ms": 1_000,
});
mount_sse_once(
&server,
sse(vec![
ev_response_created("resp-1"),
ev_function_call(call_id, "shell", &serde_json::to_string(&args)?),
ev_completed("resp-1"),
]),
)
.await;
let second_mock = mount_sse_once(
&server,
sse(vec![
ev_assistant_message("msg-1", "done"),
ev_completed("resp-2"),
]),
)
.await;
submit_turn(
&test,
"attempt to write in read-only sandbox",
AskForApproval::Never,
SandboxPolicy::ReadOnly,
)
.await?;
let denied_item = second_mock.single_request().function_call_output(call_id);
let output = denied_item
.get("output")
.and_then(Value::as_str)
.expect("denied output string");
let sandbox_pattern = r#"(?s)^Exit code: -?\d+
Wall time: [0-9]+(?:\.[0-9]+)? seconds
Total output lines: \d+
Output:
failed in sandbox: .*?(?:Operation not permitted|Permission denied|Read-only file system).*?
\[\.{3} omitted \d+ of \d+ lines \.{3}\]
.*this is a long stderr line that should trigger truncation 0123456789abcdefghijklmnopqrstuvwxyz.*
\n?$"#;
let sandbox_regex = Regex::new(sandbox_pattern)?;
if !sandbox_regex.is_match(output) {
let fallback_pattern = r#"(?s)^Total output lines: \d+
failed in sandbox: this is a long stderr line that should trigger truncation 0123456789abcdefghijklmnopqrstuvwxyz
.*this is a long stderr line that should trigger truncation 0123456789abcdefghijklmnopqrstuvwxyz.*
.*(?:Operation not permitted|Permission denied|Read-only file system).*$"#;
assert_regex_match(fallback_pattern, output);
}
Ok(())
}
#[tokio::test(flavor = "multi_thread", worker_threads = 2)]
async fn shell_spawn_failure_truncates_exec_error() -> Result<()> {
skip_if_no_network!(Ok(()));