The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted source code and configuration data. It mitigates indirect prompt injection by instructing the agent to treat all audited content as non-authoritative data and to disregard any instructions found within the code being reviewed.
[DYNAMIC_EXECUTION]: The assets/examples/ and evals/ directories contain intentionally vulnerable code samples demonstrating unsafe deserialization (via Pickle and YAML) and shell injection. These are clearly documented as educational benchmarks and are not executed by the skill's functional components.
[CREDENTIALS_UNSAFE]: The evaluation suite includes hardcoded example AWS credentials (e.g., AKIAIOSFODNN7EXAMPLE) used solely for testing the scan_secrets.py tool. These are well-known documentation placeholders and do not represent a compromise of production systems.

find-vulnerabilities