The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill inherently processes untrusted data as part of its primary function, including implementation reports and code changes. This creates a surface where embedded instructions in those files could attempt to influence the agent's auditing logic.
Ingestion points: Reads implementations/<folder>/report.md, spec.md, seed.md, and code files via git show.
Boundary markers: The instructions do not explicitly require the agent to ignore or delimit embedded instructions within the files it reviews.
Capability inventory: The skill uses git show, git grep, and has the ability to write findings bundles to the .pending-reviews/ directory.
Sanitization: No sanitization or escaping of the ingested file content is performed before interpretation.

review-implementation