model-gap
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill describes a standard developer workflow for AI model evaluation and optimization. No malicious behavior or suspicious activity was detected.
- [COMMAND_EXECUTION]: The skill instructs the agent to run a local script using
node ../skill-diet/scripts/eval.mjs. This script is part of the skill's relative project environment and is used to perform the core evaluation tasks described in the instructions. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process other skill files and task definitions to measure model performance.
- Ingestion points: Accesses and evaluates skill files (
SKILL.md) and task data as defined in the referencedeval.mddocumentation. - Boundary markers: The skill references an external
adapter.mdfor defining the model execution environment, though it does not specify explicit prompt delimiters within the instruction text. - Capability inventory: Uses local shell commands, including
nodeand ascoreutility, to execute tests and aggregate results. - Sanitization: The instructions focus on performance measurement and do not explicitly mention sanitization of the skill content being evaluated.
Audit Metadata