The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local commands using the GitHub CLI (gh) and Git (git) to fetch pull request metadata and manage local commits. It also invokes project-standard linting and testing tools (e.g., npm test, cargo test) to verify fixes as part of its approved workflow.
[DATA_EXFILTRATION]: The agent fetches pull request descriptions, review summaries, and comment threads from GitHub's servers via the gh CLI. These operations target a well-known service and are limited to the data necessary for the skill's primary function of PR triage.
[PROMPT_INJECTION]: The skill processes untrusted external data in the form of PR review comments from both humans and third-party AI bots. While this represents an indirect prompt injection surface, the risk is mitigated by a human-in-the-loop workflow that requires the user to manually review and approve every proposal before any action is taken.
Ingestion points: PR comments and descriptions fetched via GraphQL in SKILL.md (Step 2).
Boundary markers: None specified in the instructions for separating comment text from instructions.
Capability inventory: File system read/write, local command execution (git, gh, test runners) in SKILL.md (Step 1 and Step 5).
Sanitization: No explicit sanitization or filtering of comment content is performed before processing.
[SAFE]: The skill explicitly forbids automatic code pushing or thread resolution on GitHub, and avoids adding AI attribution to commits, ensuring that the user maintains full control over the final state of the repository and the public PR discussion.

pr-triage