The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub Pull Request reviews and comments to identify bug candidates, which triggers actions like issue creation and thread resolution. This creates a surface for indirect prompt injection where malicious instructions embedded in a PR comment could influence the agent's behavior.
Ingestion points: The collect command in scripts/collect-review.mjs fetches reviews, inline comments, and conversations using GitHub's REST and GraphQL APIs.
Boundary markers: The prompt in SKILL.md instructs the agent to treat external content as evidence rather than instructions, providing a defensive boundary.
Capability inventory: The skill performs GitHub thread resolutions (resolve command) and posts summary comments (summary command) using the GitHub CLI, and delegates to a sibling add-issue skill for bug reporting.
Sanitization: The raw body of reviews and comments is passed to the handoff payload without explicit filtering or escaping in the scripts.

collect-review