The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a detailed playbook of CLI commands for the agent to execute, ranging from environment inspection (comfy --json env) and job management (comfy jobs watch, comfy jobs cancel) to cloud configuration and authentication (comfy cloud login, comfy cloud set-key). These are standard administrative tasks for the managed tool.
[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to treat JSON 'envelopes' and their hint fields as authoritative sources of truth. This creates a surface for indirect injection if the tool output is influenced by untrusted data (e.g., custom node names or remote job error messages).
Ingestion points: JSON envelopes emitted by various comfy CLI commands, specifically the error, hint, and details fields.
Boundary markers: The skill emphasizes using the JSON format specifically for machine readability but does not include explicit delimiters or sanitization instructions for the content within the envelopes.
Capability inventory: The agent is granted the ability to install nodes (comfy node install), set API keys (comfy cloud set-key), and launch local servers (comfy launch).
Sanitization: No specific sanitization or validation of the JSON content is mentioned before the agent acts on the provided 'hints'.
[EXTERNAL_DOWNLOADS]: The skill mentions external resources consistent with its purpose, including a well-known service for token debugging (jwt.io) and a vendor-owned domain for test environments (testenvs.comfy.org). These references are informative and relevant to the developer workflow.

comfy-debug