ara-rigor-reviewer
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from research artifacts to generate a review report. * Ingestion points: The skill reads multiple markdown and YAML files (PAPER.md, logic/claims.md, trace/exploration_tree.yaml, etc.) from an artifact directory to perform its evaluation. * Boundary markers: The instructions do not specify any delimiters or explicit warnings to prevent the agent from following instructions potentially embedded within the research data. * Capability inventory: The agent is limited to native file tools (Read, Write, Glob, Grep) and is explicitly prohibited from network access or code execution. * Sanitization: No sanitization is performed on ingested data before it is evaluated by the model.
Audit Metadata