eval
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow for evaluating outputs from other model runs ('runners'), which constitutes an ingestion point for untrusted data. A malicious response from a runner could potentially contain instructions designed to manipulate the 'judge' subagent or the supervisor agent during the evaluation phase.
- Ingestion points: Runner outputs and execution transcripts are ingested by the supervisor agent and the judge subagent as specified in SKILL.md.
- Boundary markers: The instructions do not define delimiters or specific 'ignore embedded instructions' warnings for the judge when processing runner-generated data.
- Capability inventory: The skill utilizes subagent spawning and a 'Skill tool' to access execution history and transcripts.
- Sanitization: The process described in SKILL.md does not include sanitization or filtering of the text content produced by runners before it is evaluated.
Audit Metadata