The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Pull Request titles, bodies, and comments, creating a potential surface for prompt injection attacks. 1. Ingestion points: PR title, body, and review threads are read in Step 1. 2. Boundary markers: No explicit delimiters or safety instructions are used to isolate the ingested PR data. 3. Capability inventory: The skill can draft reviews via gh api and invoke multiple sub-agents for code analysis. 4. Sanitization: PR content is not sanitized before being passed to sub-agents or drafted into review comments. 5. Mitigation: The workflow requires a manual 'Triage' step where a user must approve every finding before it is drafted.
[COMMAND_EXECUTION]: The skill executes GitHub CLI commands to draft and submit reviews. Evidence: Step 7 describes the use of gh api repos/<owner>/<repo>/pulls/<number>/reviews --input <file>.
[EXTERNAL_DOWNLOADS]: The skill references documentation and code from a public GitHub repository. Evidence: CREDITS.md provides a link to the show-me skill in the humanlayer/skills GitHub repository.

review-pr