skill-eval
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from evaluation outputs which is then used to build prompts for grading subagents. Ingestion points: Evaluation output files in the workspace. Boundary markers: Prompt templates use simple variable substitution without specialized delimiters. Capability inventory: File system access (read/write) and subagent management. Sanitization: No specific sanitization of evaluation outputs was observed before prompt interpolation.
- [DYNAMIC_EXECUTION]: The skill dynamically generates dispatch instructions for subagents using templates located in the references directory. This mechanism is used to automate the multi-step evaluation process and requires the user to manually dispatch the agents.
Audit Metadata