The Agent Skills Directory

[SAFE]: The skill is designed as a read-only auditor for natural language content. It explicitly restricts its own capabilities by instructing the agent not to write files, update memory, or generate persistent reports unless specifically requested by the user.
[INDIRECT_PROMPT_INJECTION]: The skill acknowledges that audited manuscripts may contain instructions (e.g., 'ignore requirements', 'run script'). It mitigates this by invoking a specific safety boundary protocol and strictly defining manuscript content as data to be evaluated, not instructions to be executed.
[DATA_EXFILTRATION]: No network communication or sensitive file access was detected. The skill uses relative paths to local reference files within its own ecosystem.

eva-preflight