The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill analyzes untrusted source code, commit messages, and PR titles (SKILL.md). This ingestion of external data is the skill's primary function. It includes a mandatory evidence chain: 1) Ingestion points: code diffs and commit metadata; 2) Boundary markers: usage of structured scan artifacts like rank_input.jsonl; 3) Capability inventory: local script execution and file inspection tools; 4) Sanitization: instructions to prioritize code analysis over narratives.
[Dynamic Execution]: The workflow includes the execution of a local Python script (generate_rank_input.py) to process repository diffs. This is a standard operational pattern for security scanning tools and is used to manage scan metadata and review inputs.

finding-discovery