BigPlayer Prompt Debug

BigPlayer Prompt Debug is the transparency node in an otherwise opaque-seeming pack. It takes the shared session from BigPlayer Natural Language Root and outputs two strings: request_text and response_text. They're the raw provider-owned debug records captured at the provider boundary - exactly what your graph sent to Grok and exactly what came back, before any local validation touched it.

The request text is worth actually reading once. It shows the model, the system prompt (the "you convert user natural language prose into structured, production-ready ComfyUI workflow data" boilerplate), the user prompt with all your prose and preset context inline, and the full JSON schema the provider is forced to return - a strict object with additionalProperties: false and every capability field enumerated. That's the mechanism behind the pack's "structured output first" claim: the LLM isn't asked to produce freeform JSON and hope; it's locked to a schema, and this node lets you see the lock.

The two gotchas that matter

• It only captures if it's attached before the run. The README is explicit: "If this node is not attached to the root session, no provider debug text is captured or stored." Debug capture happens at the provider call, so you can't attach it after the fact and expect history. Attach it, run, and the strings are there.
• It never echoes your API key. Redaction happens upstream of the debug record, so the request_text and response_text outputs are safe to paste into a bug report or a Discord thread without leaking a secret. That's a real courtesy in a pack where the key is a literal widget.

When you'd reach for it

When the root fails with "provider response failed schema validation," the error names the problem but not the shape. This node shows you the actual response so you can see whether the model drifted off-schema, whether it invented a field, or whether your ComfyUI list changed under a cached session. It's also just a great way to understand what this pack actually does - five minutes with request_text will teach you more about the architecture than the README will. Wire the outputs into a Show Text / text-viewer node and read them.

Install and troubleshooting

Pack install: ComfyUI Manager, search "BigPlayer Prompting", or git clone https://github.com/josh-ent/comfyui-bigplayer-llm-nodes into ComfyUI/custom_nodes, then restart. Deps are httpx and pydantic; no model downloads. If you get "Prompt Debug was not attached to the Natural Language Root for this session," it means the node wasn't directly connected to the root when the graph ran - or it's reading a cached session from before you attached it.