The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from GitHub PR comments and review threads which could potentially contain malicious instructions intended to influence the agent's behavior during the feedback loop.\n
Ingestion points: External review data is ingested in SKILL.md (Step 2) via gh api graphql and gh pr view calls.\n
Boundary markers: The skill enforces a strong security boundary through a mandatory human approval gate; Rule 1 and Step 4 require the user to approve all planned fixes and replies before any remote operations are executed.\n
Capability inventory: The skill has capabilities to modify the local filesystem (Step 5), execute shell commands via git and gh (Steps 1, 6, 7), and update remote repositories by invoking the /commit-push-pr skill.\n
Sanitization: No automated sanitization of fetched comment bodies is performed, relying instead on the user's manual review of the disposition list.\n- [COMMAND_EXECUTION]: The skill uses standard command-line tools to interact with the repository and the GitHub platform.\n
Evidence: SKILL.md documents the use of git for branch management and gh (GitHub CLI) for querying and interacting with the GitHub API.

pr-feedback