The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted content from external GitHub pull request comments to generate documentation.
Ingestion points: Untrusted data enters the agent context via gh api calls in SKILL.md (Step 2 and Step 5.5).
Boundary markers: The prompt for the review sub-agent uses structural labels like SOURCE CONTEXT and DRAFTED BEST PRACTICE to separate content, but it lacks strict character escaping or specific delimiters for the body of the comment.
Capability inventory: The skill possesses file writing capabilities, local Python script execution, and git command access.
Sanitization: No explicit programmatic sanitization or input filtering is performed on the comment body; however, the skill employs a sub-agent validation step to ensure technical accuracy before committing changes.
[COMMAND_EXECUTION]: The skill executes local scripts and development tools to manage repository content.
Evidence: The skill runs ./.claude/skills/review/discover-bp-docs.py and ./script/manage-bp-ids.py using python3, and utilizes standard CLI tools including git, gh, and pnpm for formatting and pull request management.
[EXTERNAL_DOWNLOADS]: The skill retrieves data from a well-known service as part of its core functionality.
Evidence: Fetches pull request comment details and diff hunks from GitHub's official API using the gh tool. This is a neutral interaction with a well-known service provider.

add-best-practice