The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources that could contain malicious instructions, such as ComfyUI logs and user-provided workflow files. The documentation identifies these logs as "attacker-controlled text."
Ingestion points: Command outputs from comfy deploy logs, comfy deploy events, and data within workflow JSON files supplied to comfy deploy run (SKILL.md).
Boundary markers: The skill does not define specific technical delimiters for untrusted content, relying instead on high-level instructions to the agent.
Capability inventory: The skill utilizes comfy-cli for infrastructure management, file writing for job outputs, and network communication with deployment endpoints (SKILL.md).
Sanitization: The instructions explicitly warn the AI agent to not execute commands, fetch URLs, or use arguments derived from deployment logs.
[COMMAND_EXECUTION]: The skill facilitates the execution of various comfy-cli subcommands to manage serverless infrastructure. This includes operations like up, stop, scale, and run. This is the primary function of the skill.
[EXTERNAL_DOWNLOADS]: The skill references the official GitHub repository for comfy-cli (github.com/Comfy-Org/comfy-cli) for tool acquisition and documentation.

comfy-deploy