The Agent Skills Directory

[COMMAND_EXECUTION]: The skill extensively uses the comet CLI tool for project lifecycle management. Commands such as comet creator guide, comet creator generate, and comet publish distribute are used to initialize projects, record authoring evidence, and manage skill distribution.
[DYNAMIC_EXECUTION]: The skill generates and executes local JavaScript files (.mjs) to manage workflow states and guardrails. It specifically instructs the agent to run scripts using variables, such as node "$WORKFLOW_STATE" status, which represents a dynamic code execution pattern.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a factory for new instructions, processing user goals and existing skill content to generate new logic.
Ingestion points: User-provided workflow goals and content from candidate skills read via comet skill show (documented in SKILL.md).
Boundary markers: The authoring process is structured into sequential waves (wave1, wave2) defined in reference/authoring-protocol.json to separate content generation from script creation.
Capability inventory: Includes file system writes through comet creator authoring-record, CLI command execution, and JavaScript execution via node (documented in reference/subagents/script-author.md).
Sanitization: Employs a mandatory skill-review barrier as described in reference/subagents/skill-reviewer.md, which utilizes a multi-vote LLM process and automated evaluation (comet eval) to validate the generated skill bundle before it is marked as ready.

comet-any