The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the comfy CLI to manage cloud infrastructure, involving commands like comfy deploy up, status, stop, and delete. The instructions include safeguards requiring the agent to obtain user confirmation before executing actions that spend money or delete resources.
[INDIRECT_PROMPT_INJECTION]: The skill identifies and mitigates risks associated with processing external data.
Ingestion points: The agent reads deployment logs via comfy deploy logs and processes user-provided workflow JSON files.
Capability inventory: The agent can trigger cloud deployments, upload local assets, and delete records.
Boundary markers: The instructions explicitly require the agent to disclose cost models and technical parameters (GPU, region, worker bounds) to the user before deployment.
Sanitization: The skill warns that deployment logs are attacker-controlled and strictly forbids using log content as instructions, command arguments, or URLs.
[EXTERNAL_DOWNLOADS]: The comfy deploy run command triggers the downloading of workflow outputs from a remote serverless endpoint to the local ./outputs/ directory.

comfy-deploy