Summary

On Discord, a reasoning model's thinking is never presented to the user. With /reasoning on — which should surface the model's thinking as persistent messages — the thinking lane is dropped entirely: the thoughts are produced by the model and archived in the session, but no thinking content reaches the channel. Only the final answer is shown.

Environment

• OpenClaw 2026.6.8 (production)
• Provider/model: deepseek/deepseek-v4-flash (a reasoning model that reliably emits reasoning_content)
• Flags: /think high, /verbose off, /reasoning on

Steps to reproduce

1. /model deepseek/deepseek-v4-flash, then /think high, /verbose off, /reasoning on
2. Send a prompt that induces reasoning plus a tool call, e.g. "Reason it through first, then use a tool to list your working directory and tell me the single largest file."
3. Observe the Discord output.

Expected

With /reasoning on, the model's thinking should appear as persistent message(s), visually distinct from the final answer.

Actual

No thinking is shown at any point. During streaming the draft shows tool activity ephemerally, then collapses to the final answer only. The thinking is confirmed present in the session archive (multiple type:"thinking" parts per turn, ~100–195 characters each) but is never rendered.

Notes

• Discord has no per-message thinking-stream mode (unlike Telegram's /reasoning stream), so under /reasoning on there is currently no path that surfaces thinking on Discord at all.
• Reproduced with deepseek as a clean, reliable emitter; the gap affects reasoning models generally.
• Screen recordings of the before-state are available and can be attached.