The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data (code diffs and pull request metadata) which presents a standard attack surface for indirect prompt injection.
Ingestion points: The skill identifies changed files, commits, and PR titles/bodies as primary inputs in SKILL.md.
Boundary markers: The instructions recommend using an "isolated reviewer" for material changes, which serves as a context boundary to prevent implementation conversations from affecting the review.
Capability inventory: The skill outlines an analytical review and reporting process; it does not define or request autonomous tool execution or file-write capabilities.
Sanitization: No explicit input sanitization or escaping of the diff content is prescribed within the instructions.
[SAFE]: The skill instructions explicitly direct the agent to evaluate security, privacy, and integrity boundaries as part of its review passes. It also prioritizes safety-critical findings and encourages independent review for high-risk changes.

code-review