docs: add OpenAI-compatible/local provider diagnostics playbook

[Yeachan-Heo]

Source

Discord #claw-code support threads around local Qwen, llama.cpp, Ollama, DeepSeek wrappers, and webchat-to-OpenAI-compatible adapters.

Problem

Many failures share the same shape:

• plain chat works
• slash/tool workflows fail or do nothing
• interactive mode shows only ✔ ✨ Done
• local wrappers strip tools, tool_choice, streaming tool-call deltas, or return text instead of message.tool_calls

Users need a focused diagnostic playbook before maintainers can tell whether the bug is Claw-side or wrapper-side.

Suggested scope

Add docs for checking raw /v1/chat/completions response shape:

• choices[0].message.content
• choices[0].message.tool_calls
• tool call IDs
• tools and tool_choice request preservation
• streaming tool-call deltas
• DeepSeek thinking fields such as reasoning_content or content[].thinking

Include copy-paste curl examples and the minimal report format:

• OS
• Claw version
• provider/model
• base URL shape, no secrets
• exact command
• raw response shape, redacted

Acceptance criteria

• Users can self-check whether a local/OpenAI-compatible provider supports tool calls.
• Bug reports include response-shape evidence instead of screenshots.
• Docs clearly distinguish wrapper bugs from Claw rendering bugs.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]