The Agent Skills Directory

General Security (SAFE): The skill is primarily instructional and contains no malicious logic. The validation script uses yaml.safe_load() which prevents remote code execution via YAML deserialization.
Category 8: Indirect Prompt Injection (LOW): The skill possesses an attack surface as it is designed to process external untrusted data for safety filtering. * Ingestion points: Data enters via the task parameter and processed llm_response strings. * Boundary markers: None explicitly defined in the provided templates. * Capability inventory: Limited to internal monitoring and content filtering; no file-write or subprocess execution capabilities. * Sanitization: Employs the guardrails library for toxicity and PII detection.

agent-safety