The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on various shell commands (git, gh) and local Python scripts (discover-bp-docs.py, chunk-best-practices.py) to automate the code review workflow. It uses these tools to fetch repository status, diffs, and interact with the GitHub API for issue tracking and review posting.
[INDIRECT_PROMPT_INJECTION]: The skill has a defined attack surface for indirect prompt injection as it ingests and processes content from external sources.
Ingestion points: External data enters the agent context via gh pr view (PR title, body, files), gh issue view (Issue title, body, comments), and the code diffs retrieved via gh pr diff or git diff.
Boundary markers: Diffs and rules are wrapped in markdown code blocks (triple backticks) when passed to subagents to provide structural separation.
Capability inventory: The skill possesses capabilities including reading all files in the repository, executing shell commands, running Python scripts, and performing POST requests to GitHub APIs to create comments and reviews.
Sanitization: The skill does not perform specific sanitization or filtering of the text content fetched from PRs or issues before it is included in the model's prompt context.
[DATA_EXFILTRATION]: The skill includes functionality to post review comments and overall verdicts back to GitHub PRs. This involves transmitting analysis results from the local environment to an external service. This risk is mitigated by explicit instructions requiring user approval before posting and the use of visible AI watermarks for transparency.

review