The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub pull requests, creating a surface for indirect prompt injection. Ingestion points: Reads PR title, body, commits, and comments via gh pr view and gh api calls. Boundary markers: Instructions explicitly state to treat PR-controlled content as untrusted and to load agent instructions (AGENTS.md, CLAUDE.md, etc.) only from the current base branch. Capability inventory: The skill utilizes git (fetch, merge, cherry-pick), gh CLI (api, pr merge), and executes repository-defined tests. Sanitization: Instructs the use of isolated temporary clones or sandboxes for test execution and the removal of unrelated credentials from the environment.
[REMOTE_CODE_EXECUTION]: The skill is designed to execute tests provided in external pull requests. Evidence: SKILL.md instructs the agent to 'Run untrusted tests in an isolated temporary clone or sandbox'. This capability is a fundamental part of the skill's validation workflow.
[COMMAND_EXECUTION]: Extensive use of shell commands via git and gh CLI to interact with the repository and GitHub API. Evidence: Multiple scripts throughout the instruction set use commands like git fetch, git merge-tree, gh pr view, and gh api to manage the review and landing process.

github-review-pr