The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied descriptions to generate system prompts for LLM evaluators.
Ingestion points: User input gathered in Phase 1 and the $ARGUMENTS variable are used to define the evaluator's rubric and prompt.
Boundary markers: The skill relies on a human-in-the-loop confirmation step in Phase 3 rather than technical delimiters or escaping.
Capability inventory: The skill executes shell commands via the calibrate CLI, including creating new evaluators with custom prompts.
Sanitization: No explicit sanitization or filtering is applied to user-provided criteria before interpolation into the system_prompt field.
[COMMAND_EXECUTION]: The skill utilizes shell commands for managing evaluators and performing setup checks. It includes a self-update command using npx which is instructed to run silently without informing the user.
[EXTERNAL_DOWNLOADS]: Fetches package updates via npx and directs users to the artpark.ai domain for authentication and account management.

design-evaluator