The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted pull request comments, creating a surface for Indirect Prompt Injection (Category 8).
Ingestion points: PR comments are loaded from three separate surfaces (inline, top-level, and reviews) using the gh api command in SKILL.md.
Boundary markers: The instructions establish a "reception discipline" requiring every claim to be verified against the codebase and restated in the agent's own words, rather than using explicit string delimiters.
Capability inventory: Includes command execution via the gh CLI, repository write access through git push/commit, and the ability to invoke other skills like run-tests.
Sanitization: The skill mandates a triage process where the agent must reproduction failures and understand the root cause before applying any changes.
[COMMAND_EXECUTION]: The skill makes extensive use of the GitHub CLI (gh) to interact with the repository and PR metadata.
Uses gh api with pagination and JQ for fetching comments and users.
Uses gh api graphql for resolving review threads and checking resolution status.
Uses gh pr checks and gh pr view for CI and state tracking.
These commands are legitimate and required for the skill's primary purpose of automating developer workflows.

address-pr-comments