The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands using the git and gh (GitHub CLI) tools to manage the development workflow. This includes querying CI status, downloading logs, and committing/pushing code changes. These actions are within the intended scope of a development automation tool.
[EXTERNAL_DOWNLOADS]: Communicates with GitHub's official APIs via the gh command-line tool to retrieve logs and workflow run metadata. These downloads are from a well-known service and are necessary for the skill's functionality.
[PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it processes output from GitHub Actions logs, which could potentially contain malicious instructions embedded by an attacker (e.g., in a failing test case).
Ingestion points: gh run view --log in SKILL.md.
Boundary markers: The skill uses specific failure markers (FAIL, error:, etc.) to scope its analysis rather than treating the entire log as instruction.
Capability inventory: Includes the ability to write to the local file system and push commits to remote repositories.
Sanitization: The risk is mitigated by a mandatory user confirmation step ([Y/n]) before any git push operation is performed, ensuring a human-in-the-loop safety check.

github-fix-action-error