simmer-judge
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from 'Current candidate' (artifact text) and 'Evaluator output' (logs, metrics) to perform scoring and generate refinement instructions (ASI). It lacks explicit instructions to use boundary markers or to ignore instructions embedded within the data being analyzed. This creates a surface where a malicious candidate artifact or corrupted evaluator log could provide instructions that manipulate the agent's judgment or propagate to the next agent in the pipeline.
- Ingestion points: 'Current candidate' artifact files and 'Evaluator output' (stdout/stderr) referenced in SKILL.md.
- Boundary markers: No specific delimiters or 'ignore' instructions are defined for the untrusted content.
- Capability inventory: The agent produces scores and directional instructions (ASI) that guide the behavior of a downstream generator agent.
- Sanitization: No sanitization, validation, or filtering of the artifact or log content is described in the skill instructions.
Audit Metadata