The Agent Skills Directory

Indirect Prompt Injection: The skill processes untrusted external data from issue trackers, which could contain adversarial instructions. Ingestion occurs through issue bodies and comments (Steps 1 and 2). The skill mitigates this via Golden Rule 6 (security screening for sensitive content), instructions to treat data as non-executable (Category 8b), and mandatory human confirmation for all state-changing actions.
Command Execution: The skill invokes the gh CLI and a local Python module (setup_preflight) for environment verification and issue management. These operations are restricted to triage tasks and utilize the user's existing authenticated session.
Human-in-the-Loop Controls: All write operations, such as posting comments or closing issues, require explicit review and confirmation by a human maintainer, ensuring a significant layer of oversight for all actions.
Shell Injection Mitigation: The skill uses a file-based pattern for comment bodies to prevent potential shell injection when handling external strings.

stale-sweep