openclaw - 💡(How to fix) Fix [Bug]: Feishu sends duplicate replies when DeepSeek V4 reasoning_content is present

Summary

When using DeepSeek V4 Flash/Pro with thinking=high on the Feishu channel, the model"s reasoning_content leaks into delivery as a separate message, causing visible duplicate replies in the chat.

Environment

• OpenClaw: v2026.5.2
• Channel: Feishu (websocket)
• Model: deepseek/deepseek-v4-flash with thinkingDefault: "high"
• Config: blockStreamingDefault: "off"

Symptoms

1. Log shows dispatch complete (queuedFinal=true, replies=2) or replies=3 where normal is replies=1
2. User sees the model"s chain-of-thought reasoning text as a separate message followed by the actual answer
3. Makes it look like the bot sent duplicate replies

Root Cause

The problem is in the payload assembly pipeline:

• DeepSeek returns reasoning_content as a separate field alongside content in the API response (per their docs)
• OpenClaw"s openai-completions provider correctly captures reasoning_content into type: "thinking" content blocks
• However, in buildEmbeddedRunPayloads() (src/agents/pi-embedded-runner/run/payloads.ts), the .map() return only propagates isError, not isReasoning
• Since isReasoning never reaches downstream filters, Feishu dispatcher treats type: "thinking" blocks as normal text payloads and sends them as separate messages

Related Issues

• #73186 (open): Thinking/reasoning content leaks into cron announce delivery for Matrix/Feishu
• #67791 (closed): Feishu blockStreamingCoalesce regression (different root cause)
• #80033 (merged): add supportedReasoningEfforts to DeepSeek V4 model entries

Suggested Fix

The isReasoning flag needs to be propagated through the payload pipe so downstream channel dispatchers can filter or merge thinking blocks instead of sending them as separate messages.