The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub pull requests, including code diffs and review comments, which may contain adversarial instructions.
Ingestion points: Fetches pull request metadata, file content, and review threads (SKILL.md, Workflow step 1 and 2).
Boundary markers: Directs the agent to distrust cached state and review the complete diff (SKILL.md, Invariants).
Capability inventory: Authorized to merge pull requests, push to branches, and execute local build/test scripts (SKILL.md, Workflow step 3 and 4).
Sanitization: Relies on the agent's internal review step; no explicit content filtering.
[DYNAMIC_EXECUTION]: The skill executes repository-native commands such as build scripts and tests found within the pull request code.
Evidence: The 'Verify or repair' section of SKILL.md directs the agent to 'Run repository-native formatting, lint, typecheck, build, and tests'.
[COMMAND_EXECUTION]: The skill heavily utilizes CLI tools to interact with GitHub and the local file system.
Evidence: Uses the gh CLI for GitHub API interactions and git for branch management and pushing updates.

pr-drain