pick-model
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a vendor-specific script
pick-model-state(located in the vendor's script directory) to retrieve session metadata such as token usage and cache hit rates. - [COMMAND_EXECUTION]: Uses the
awkcommand-line utility to calculate a 'switch penalty' based on session tokens and model pricing constants. - [INDIRECT_PROMPT_INJECTION]: The skill ingests user prompts to classify them, which presents an indirect injection surface.
- Ingestion points: The user's intended prompt is provided as input to the classification workflow in
SKILL.md. - Boundary markers: The skill contains an explicit instruction to "Do NOT execute" the input prompt, which serves as a protective boundary to prevent the agent from following instructions embedded in the user data.
- Capability inventory: The skill has the ability to run the
pick-model-statescript andawkcommand. - Sanitization: No programmatic sanitization or escaping is applied to the user input; the skill relies entirely on negative constraints in the system prompt.
Audit Metadata