The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple external sources which are processed by the agent during its maintenance loop.
Ingestion points:
scripts/list-failing-checks.sh: Fetches and displays CI failure logs via gh run view --log-failed. These logs are generated by external build systems and can contain arbitrary text influenced by PR authors.
scripts/list-review-threads.sh: Fetches and prints the bodies of GitHub PR comments (specifically targeting CodeRabbit threads). Comment content is externally controlled and parsed by the agent.
scripts/sync-branch.sh: In the event of a merge conflict, the script dumps the full content of conflicted files (including conflict markers) into the agent's context using cat.
Boundary markers: The scripts do not implement explicit delimiters or instruction-bypass warnings when outputting this untrusted content to the agent.
Capability inventory: The skill is designed to coordinate the fix-all skill, which possesses capabilities to modify files, run build/test commands, and interact with GitHub APIs (as seen in scripts/gh-resolve-review-thread.sh).
Sanitization: No sanitization, filtering, or escaping is performed on the CI logs, comment bodies, or file contents before they are presented to the agent.

pr-maintenance-loop