The Agent Skills Directory

[SAFE]: The skill implements a security-focused process for auditing financial scripts. It utilizes a two-stage approach: a deterministic static analysis (Stage 1) and a qualitative manual review (Stage 2) guided by a comprehensive safety checklist.
[COMMAND_EXECUTION]: The skill invokes a local utility via node to perform static audits. This tool is used to parse and validate the file structure according to strict safety rules (e.g., forbidding network access, wallet access, and dynamic code) without executing the untrusted script content.
[INDIRECT_PROMPT_INJECTION]: The skill processes external script files provided by users. It mitigates potential injection risks by instructing the agent to evaluate the code against a specific checklist that includes checks for obfuscation, misdirection, and unauthorized actions. The Stage 1 static analysis rules also act as a primary sanitization layer by preventing the ingestion of files containing known-dangerous patterns.

feast-audit