agent-eval-loop
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a process for the agent to ingest and evaluate untrusted data in the form of test queries.
- Ingestion points: Queries are loaded from
evals/<agent-name>/golden-set.jsonl. - Boundary markers: The instructions lack guidance on using delimiters to separate query data from instructions.
- Capability inventory: The workflow utilizes the agent's native file access and external monitoring tools (Langfuse/RAGAS).
- Sanitization: The skill does not prescribe any validation or filtering for the query content.
Audit Metadata