The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill automatically derives review policy rules by analyzing repository documentation and source code, creating a surface where malicious instructions in these files could influence the generated artifacts.
Ingestion points: The skill reads the root AGENTS.md file, and several files within the documentation directory including ENGINEERING.md, AGENTS_CONTEXT.md, CONTEXT.md, CONTEXT-MAP.md, and Architecture Decision Records (ADRs) under adr/*.md. It also scans the entire repository's source code to mine inheritance conventions.
Boundary markers: The skill incorporates a structured interview phase where the user must explicitly confirm or reject any convention candidates mined from the code before they are promoted to rules. This creates a human-in-the-loop checkpoint.
Capability inventory: The skill possesses the capability to read any file within the repository, write new policy artifacts under <docs-dir>/review/policies/, and modify the root .gitignore file.
Sanitization: The skill implements a conflict resolution mechanism that flags contradictions between derived rules and established ADRs, marking them as needs decision for manual resolution by the user rather than applying them automatically.

review-policy-builder