The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes git and gh (GitHub CLI) commands to fetch repository data, analyze pull request diffs, and manage a local cache. These operations are performed using Node.js spawnSync and execFileSync without shell interpolation, mitigating common command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes Pull Request descriptions, titles, and commit messages which originate from external contributors. These sources could potentially contain malicious instructions intended to manipulate the LLM's triage judgment.
Ingestion points: The scripts/build.mjs script ingests the PR body, title, and commits into the agent's context.
Boundary markers: Instructions in SKILL.md explicitly direct the agent to "Judge the change, not its size, title, or author" and provide a specific judge brief.
Capability inventory: The skill allows the agent to read file contents, search codebases via grep, and generate reports, but restricts actions on GitHub to read-only.
Sanitization: The scripts/annotate.mjs script acts as a deterministic validator that checks every claim in the model-written notes against the actual diff and refuses any content containing merge approval keywords (e.g., "LGTM", "safe to merge").
[EXTERNAL_DOWNLOADS]: The skill references and downloads resources from well-known and trusted services.
Fonts: The scripts/render.mjs script links to https://fonts.googleapis.com to style the generated HTML reports.
Tooling: The scripts/repos.mjs script provides links to the official https://git-scm.com and https://cli.github.com websites for dependency installation.

pr-triage