The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes GitHub CLI (gh) commands to retrieve metadata and logs from GitHub Actions.
Evidence: Shell commands like gh pr checks and gh api are used to resolve failed jobs and fetch their associated logs from the GitHub API.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of CI job logs which could be used to manipulate the agent's output.
Ingestion points: The skill fetches logs from GitHub repositories via the gh api command in Step 2 of the execution flow.
Boundary markers: There are no explicit markers used to separate the external log data from the agent's instructions, though it uses grep to filter for errors.
Capability inventory: The skill has the capability to propose shell commands (e.g., git push, pnpm install, uv lock) to the user.
Sanitization: No sanitization is performed on the log content, but the risk is mitigated by a human-in-the-loop requirement.
[SAFE]: The skill implements a robust security guardrail by design.
Evidence: The skill explicitly states in its description and execution section that it will 'NEVER apply without explicit user approval' and requires the user to reply 'go' to ship any proposed fixes.

ci-debug