skill-eval-methodology
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides documentation and methodology for evaluating AI agent skills. It contains no executable code or malicious patterns. External links point to official documentation from trusted sources.
- [COMMAND_EXECUTION]: The documentation provides shell command examples for running evaluation scripts and processing data. These are intended for local execution by a developer or agent in a controlled environment to verify skill behavior.
- [INDIRECT_PROMPT_INJECTION]: The skill defines a methodology for processing outputs from other skills and user queries for testing purposes.
- Ingestion points: External query sets and traces of other skills as described in the trigger evaluation reference.
- Boundary markers: The methodology recommends using structured JSON and clear success criteria to isolate data.
- Capability inventory: Execution of shell commands and file system operations within the evaluation harness (e.g., run_eval.py).
- Sanitization: The methodology explicitly instructs starting trials from clean, isolated states and preventing contamination between concurrent tasks.
Audit Metadata