skill-model-gap
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to run local evaluation and scoring tasks using a Node.js script located at
../skill-diet/scripts/eval.mjs. These commands are used to measure performance gaps and verify improvements during the optimization process. - [INDIRECT_PROMPT_INJECTION]: The skill analyzes and modifies external skill files and task definitions, creating a potential surface for indirect prompt injection if those processed files contain adversarial content.
- Ingestion points: Reads content from
SKILL.mdand evaluation tasks defined ineval.mdwithin the../skill-diet/directory. - Boundary markers: The instructions do not specify the use of delimiters or specific safety markers when the agent processes the content of the target skills.
- Capability inventory: The skill facilitates shell command execution via Node.js for evaluation purposes and performs file editing as part of the tuning process.
- Sanitization: The skill processes existing skill text and task logs directly to identify and fix behavioral differences without explicit sanitization steps.
Audit Metadata