The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, or dangerous command executions were detected in the skill instructions.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to audit external datasets, which represents an indirect prompt injection surface. However, the severity is safe as the skill ships no code and relies on the agent's existing sandbox. 1. Ingestion points: Dataset grain, source, and owner identification in SKILL.md. 2. Boundary markers: None present. 3. Capability inventory: The skill specifies a workflow but includes no scripts or subprocess calls. 4. Sanitization: None present.

data-qa