The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external GitHub contributors which could contain malicious instructions or deceptive content designed to influence the agent.
Ingestion points: SKILL.md uses gh pr view, gh pr diff, and gh api to fetch pull request metadata, commit messages, and code diffs.
Boundary markers: No explicit delimiters or instructions to ignore embedded content are provided when the agent is asked to read the diffs or PR metadata.
Capability inventory: The skill utilizes shell commands (gh, git), including the ability to fetch and push commits to the repository.
Sanitization: While jq is used for structured data extraction, the raw pull request diffs and commit messages are processed by text-processing tools (grep, awk, wc), which does not sanitize against prompt-based attacks targeting the LLM.
[COMMAND_EXECUTION]: The skill relies on local command execution to perform repository audits and pull request management.
Evidence: The instructions guide the agent to execute gh (GitHub CLI), git, grep, awk, and wc commands to analyze PRs, verify sign-offs, and carry commits between branches.

review-a-contributor-pr