The Agent Skills Directory

Indirect Prompt Injection (SAFE): The skill is intended to process untrusted codebase data through static analysis tools, which is its primary purpose. This inherent vulnerability surface is managed by the use of established, read-only tools.
Ingestion points: Files in the target codebase analyzed via ripgrep and Semgrep.
Boundary markers: The methodology directs the agent to start with exact matches and iterate incrementally.
Capability inventory: Read-only search (rg) and pattern matching (semgrep).
Sanitization: Not explicitly defined, but the workflow encourages manual verification of all findings.
Dynamic Execution (SAFE): Documentation and Semgrep rules (e.g., resources/semgrep/javascript.yaml) reference dangerous functions like eval(), exec(), and system(), but these are used as identification targets in external code and are not executed by the skill itself.

variant-analysis