work-implement-queue
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill includes a dedicated 'Author authority' section that instructs the agent to treat third-party text (issues, comments, metadata) strictly as data and never as instructions, explicitly addressing and mitigating indirect prompt injection.
- [SAFE]: Uses a local single-flight lock mechanism via 'mkdir' in the git directory to ensure mutually exclusive execution within a checkout.
- [SAFE]: The provided 'resolve-config.sh' script safely manages configuration merging using 'jq', with no external network dependencies or unsafe command interpolation.
- [SAFE]: Implementation tasks are delegated to fresh workers using the platform's 'Agent' or 'task' tools, which provides isolation between the main loop and individual issue processing.
- [SAFE]: Mandatory 'cap' and 'concurrency' settings are enforced to prevent resource exhaustion and ensure the agent operates within defined bounds.
Audit Metadata