The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's implementation playbook contains Python code (PRAnalyzer class) that uses subprocess.run to execute local git commands such as git rev-parse and git diff. These are used to identify changed files and calculate line statistics.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data in the form of git diffs to generate summaries and review checklists. A malicious actor could include specific strings in code comments or diffs designed to manipulate the generated PR description or bypass the review checklist logic.
Ingestion points: Git diff output is ingested via the analyze_changes method in resources/implementation-playbook.md and the manual workflow in SKILL.md.
Boundary markers: The PR Description Template in SKILL.md uses markdown headers but lacks explicit 'ignore' instructions for the content of the diffs themselves.
Capability inventory: The skill uses subprocess.run to interact with the local filesystem's git repository and generates structured text based on file contents.
Sanitization: No explicit sanitization or filtering of diff content is implemented before it is processed by the AI for summary generation.
[METADATA_POISONING]: The YAML frontmatter in SKILL.md includes a field risk: critical. While this might be a platform-specific classification or a self-reported tag, it is a significant metadata claim that does not appear to reflect the actual functional risk of the skill's logic.

comprehensive-review-pr-enhance