The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the comfy CLI to perform system-level operations such as environment installation, launching background processes, and stopping instances.
[REMOTE_CODE_EXECUTION]: The tool facilitates the installation of 'custom nodes' from a registry. These nodes are third-party Python scripts that execute within the ComfyUI server context to extend its functionality.
[EXTERNAL_DOWNLOADS]: The skill downloads software dependencies, ComfyUI core, and AI models from external sources including GitHub, Hugging Face, CivitAI, and user-specified URLs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data files that could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Processes workflow.json and snapshot.json files in SKILL.md.
Boundary markers: No specific boundary markers or 'ignore' instructions are used for the processed JSON payloads.
Capability inventory: Capabilities include network requests (comfy model download), file system writes, and subprocess execution (comfy launch, comfy node install).
Sanitization: The skill relies on the structure of the input JSON; no explicit sanitization logic for embedded strings is described.

comfy-cli