The Agent Skills Directory

[SAFE]: No malicious patterns detected. The skill is entirely composed of Markdown and YAML files providing instructions and reference material for RAG evaluation. It does not include scripts, tools, or external dependencies.
[PROMPT_INJECTION]: Indirect prompt injection surface identified. Ingestion points: 'references/generation-metrics.md' contains templates that ingest untrusted RAG chunks and model responses. Boundary markers: The prompt template uses explicit labels like 'Context:' and 'Response:' to delimit external data. Capability inventory: No dangerous capabilities such as subprocess calls, file-writes, or network operations are present in the skill's code or instructions. Sanitization: No explicit sanitization or escaping of external content is defined. The risk is considered minimal and inherent to the skill's purpose.

rag-auditor