What problem does it solve?
When a comfy deploy command is refused, a deployment never reaches ready, or a submitted job fails, the raw error code alone does not tell you what to do next. This Skill maps every deploy_* error code to its meaning and the exact command or action that clears it, so you stop guessing and stop paying for duplicate billed jobs.
Core Features & Use Cases
- Error code reference: Covers all deploy lifecycle codes (build, quota, billing, conflicts) and workflow submission codes (format validation, asset uploads, idempotency, job failures) with the specific move that resolves each one.
- Failure triage order: Prescribes reading
comfy deploy status, then events, then logs to distinguish billing problems from transient states and stuck deployments like stop_failed or unhealthy.
- Cost-safety guardrails: Explains why
deploy_job_submit_unknown must never trigger an automatic resubmit, why endpoint trust refusals must not be bypassed, and how to probe class registration with a minimal billed test job.
- Use Case: Your deployment is stuck in
provisioning and a job submission timed out. Use this Skill to read the deployment events, identify deploy_compute_unavailable, pick another GPU/region pair, and avoid double-billing on the ambiguous submission.
Quick Start
Ask the assistant to read the comfy-deploy-failures skill and explain why my deployment is stuck and what command fixes the deploy error code I received.
