iblai-api-agent-eval
Pass
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data such as JSON items, CSV uploads, and chat traces to build evaluation datasets that are later processed by an LLM-as-Judge. This presents a potential surface for indirect prompt injection where instructions embedded in the data could attempt to influence the grading logic.\n
- Ingestion points: Dataset items are added via JSON arrays, CSV file uploads, or existing chat traces as defined in SKILL.md.\n
- Boundary markers: The instructions do not specify any delimiters or safety markers for the content being evaluated.\n
- Capability inventory: The skill triggers background API tasks for running experiments and performing LLM-based evaluations.\n
- Sanitization: No explicit sanitization or filtering of the dataset items is mentioned in the documentation.\n- [SAFE]: The skill communicates exclusively with the vendor's API endpoint (api.iblai.app) and uses platform-provided environment variables for authentication, which is standard and safe behavior.\n- [SAFE]: The skill explicitly instructs the agent to confirm with the user before performing destructive delete operations, maintaining proper user oversight.
Audit Metadata