credo-evaluation
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, including user-provided questions, target populations, system inputs/outputs, and existing evaluator definitions, which presents an attack surface for indirect prompt injection.\n
- Ingestion points: Data is ingested through the 'Define the measurement' section in SKILL.md, where the agent recovers user questions and system outputs.\n
- Boundary markers: There are no explicit delimiters (e.g., XML tags or unique markers) or instructions to ignore embedded commands within the processed data.\n
- Capability inventory: The skill interacts with other tools like credo-dataset and evaluation-operations and can implement code for scorers and harnesses, creating a pathway for malicious data to influence system-level actions.\n
- Sanitization: The instructions do not define any validation, filtering, or sanitization steps for the data being evaluated.\n- [EXTERNAL_DOWNLOADS]: The skill references academic research papers from reputable external repositories.\n
- Evidence: Links to the ACL Anthology (aclanthology.org) and arXiv (arxiv.org) are provided in references/sources.md for methodology context.
Audit Metadata