openclaw - ✅(Solved) Fix [Bug] Reasoning mode 'On' sends only thinking content, no final response in Telegram [2 pull requests, 1 participants]

huhtreebot · 2026-03-12T07:32:54Z

[openclaw] When using /reasoning on in Telegram, only the thinking content is displayed/sent, and the final response message is never delivered to the user. When using `/reasoning on` in Telegram, only the thinking content is displayed/sent, and the final response message is never delivered to the user. # PR #43885: fix: restore final response delivery when Telegram reasoning mode is "on" - Repository: openclaw/openclaw - Author: jepson-liu - State: closed | merged: False - Link: https://github.com/openclaw/openclaw/pull/43885 ## Description (problem / solution / changelog) ## Summary Fixes #43807 When using `/reasoning on` in Telegram, users only received the thinking content — the final response was silently dropped and never delivered. `/reasoning off` and `/reasoning stream` both worked correctly. Root cause: two separate bugs in the reply pipeline: 1. **Reasoning payloads suppressed on the wrong path.** The generic `onBlockReply` handler in `dispatch-from-config.ts` unconditionally dropped all payloads where `isReasoning: true`. A comment in that file claimed Telegram had its own dispatch path, but in `reasoning="on"` mode Telegram was actually going through this generic path. The reasoning payload was swallowed before it could reach the Telegram lane coordinator, which meant the pipeline registered a "did stream" state but never forwarded the final answer, leaving the user with no response at all. 2. **Answer fragmented across multiple messages.** `blockReplyBreak` defaulted to `"text_end"`, so in `reasoning="on"` mode the answer was split into many small Telegram messages rather than one coherent reply. ## Behavior Changes - **`/reasoning on`**: now delivers both the reasoning block (as a separate Telegram message) **and** the final answer (as a single consolidated message). Previously only the reasoning block appeared. - **`/reasoning off` / `/reasoning stream`**: behavior unchanged. - **Other channels** (WhatsApp, web, etc.): unaffected — reasoning payloads are still suppressed on those paths. ## Existing Functionality Check - [x] Searched for all call sites of `shouldSuppressReasoningPayload` and `onBlockReply` to confirm no other channel is affected. - [x] Verified `forceBlockStreamingForReasoning` is only `true` when `reasoningLevel === "on"`, so `allowReasoningPayloads` is never set for other reasoning levels or other channels. - [x] Confirmed `blockReplyBreak = "message_end"` override is scoped to `resolvedReasoningLevel === "on"` only. ## What Changed - **`src/auto-reply/types.ts`** - Added `allowReasoningPayloads?: boolean` to `GetReplyOptions`. Channels with a dedicated reasoning lane (Telegram `reasoning="on"`) set this to `true` to bypass generic reasoning suppression. - **`src/auto-reply/reply/dispatch-from-config.ts`** - The `shouldSuppressReasoningPayload` guard is now conditional: suppression is skipped when `params.replyOptions?.allowReasoningPayloads` is `true`. - **`src/auto-reply/reply/get-reply-directives.ts`** - `resolvedBlockStreamingBreak` is forced to `"message_end"` when `resolvedReasoningLevel === "on"`, ensuring the answer is delivered as one complete message instead of many fragments. - **`src/telegram/bot-message-dispatch.ts`** - Passes `allowReasoningPayloads: forceBlockStreamingForReasoning` in `replyOptions` so the new guard in `dispatch-from-config.ts` is activated for Telegram `reasoning="on"` sessions. - **`src/telegram/bot-message-dispatch.test.ts`** - Extended the `"keeps block streaming enabled when session reasoning level is on"` test to assert: - `replyOptions` includes `allowReasoningPayloads: true` - `deliverReplies` is called with the reasoning block (`"Reasoning:\n_step_"`) - `deliverReplies` is also called with the final answer (`"Hello"`) ## Tests - `src/telegram/bot-message-dispatch.test.ts` — **66 tests passed** - `src/auto-reply/reply/dispatch-from-config.test.ts` — **44 tests passed** ## Files Changed - `src/auto-reply/types.ts` (+4 lines) - `src/auto-reply/reply/dispatch-from-config.ts` (+5 lines, -4 lines) - `src/auto-reply/reply/get-reply-directives.ts` (+3 lines, -1 line) - `src/telegram/bot-message-dispatch.ts` (+1 line) - `src/telegram/bot-message-dispatch.test.ts` (+6 lines) ## Checklist - [x] Test locally with your OpenClaw instance - [x] Run tests: `node_modules/.bin/vitest.CMD run src/telegram/bot-message-dispatch.test.ts` ✅ - [x] Run tests: `node_modules/.bin/vitest.CMD run src/auto-reply/reply/dispatch-from-config.test.ts` ✅ - [x] Keep PRs focused (one thing per PR) ✅ - [x] Describe what & why ✅ **Sign-Off** - Models used: Claude - Submitter effort: AI-assisted with manual code review - Agent notes: Fix is minimal and surgical — 19 lines changed across 5 files. No new abstractions introduced; the `allowReasoningPayloads` flag follows the same opt-in pattern already used by `disableBlockStreaming` and `onPartialReply` in `GetReplyOptions`. ## Changed files - `src/auto-reply/reply/dispatch-from-config.ts` (modified, +5/-4) - `src/auto-reply

openclaw2026-03-12 07:32:54

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

openclaw/openclaw#43807•Fetched 2026-04-08 00:17:52

View on GitHub

Comments

Participants

Timeline

Reactions

Author

huhtreebot

Participants

huhtreebot

Timeline (top)

cross-referenced ×2referenced ×1subscribed ×1

When using /reasoning on in Telegram, only the thinking content is displayed/sent, and the final response message is never delivered to the user.

Root Cause

When using /reasoning on in Telegram, only the thinking content is displayed/sent, and the final response message is never delivered to the user.

RAW_BUFFERClick to expand / collapse

Description

When using /reasoning on in Telegram, only the thinking content is displayed/sent, and the final response message is never delivered to the user.

Environment

OpenClaw version: 2026.3.8
Channel: Telegram
Model: MiniMax-M2.5 (via minimax-portal)
Reasoning mode: On (via /reasoning command)

Steps to Reproduce

In a Telegram DM with the OpenClaw bot, send /reasoning on
Send any message that requires a response
Observe: only thinking/reasoning content appears
Expected: both thinking AND final response should be sent

Current Behavior

/reasoning off: Works correctly, sends only the final response
/reasoning on: Only shows thinking content, final response is missing
/reasoning stream: Works correctly - shows thinking in real-time preview AND sends final response

Additional Notes

This issue makes the "On" mode unusable as users cannot receive the actual response, only the internal thinking process.

The "Stream" mode appears to work correctly and is currently the recommended workaround.

extent analysis

Fix Summary

Bug: In the Telegram channel the “reasoning‑on” mode only sends the intermediate “thinking” message and never dispatches the final answer.
Fix: Adjust the Telegram adapter so that after the reasoning text is posted it also sends the final response (or edits the thinking message). The logic used by the “stream” mode can be reused – the only difference is that we don’t stream token‑by‑token, we just send the whole reasoning block first and then the final answer.

Step‑by‑Step Fix Plan

1. Locate the Telegram channel implementation

File (example): src/channels/telegram.ts (or telegram_adapter.py if Python).

You’ll find a switch on reasoningMode that builds a Message object:

switch (reasoningMode) {
  case 'off':
    // send final only
    break;
  case 'on':
    // currently: sendThinkingOnly()
    break;
  case 'stream':
    // sendThinkingLive() + sendFinal()
    break;
}

2. Add a “send final” step for `reasoningMode === 'on'`

TypeScript / Node example

// Existing helper – sends the reasoning block as a separate message
async function sendThinking(chatId: string, thinking: string) {
  await bot.sendMessage(chatId, `🧠 *Thinking...*\n${thinking}`, {
    parse_mode: 'Markdown',
    disable_web_page_preview: true,
  });
}

// New helper – sends the final answer
async function sendFinal(chatId: string, answer: string) {
  await bot.sendMessage(chatId, answer, {
    parse_mode: 'Markdown',
    disable_web_page_preview: true,
  });
}

// Updated case
case 'on': {
  // 1️⃣ Send the reasoning block
  await sendThinking(chatId, reasoningText);

  // 2️⃣ After the model finishes, send the final answer
  await sendFinal(chatId, finalAnswer);
  break;
}

Python example (if the bot is written in Python)

async def _send_thinking(chat

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#api #ssr #installation #tensor shape #autograd error #agent execution #callback error #memory management #API rate limit #retriever error

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

openclaw - ✅(Solved) Fix [Bug] Reasoning mode 'On' sends only thinking content, no final response in Telegram [2 pull requests, 1 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fix / Workaround

PR fix notes

PR #43885: fix: restore final response delivery when Telegram reasoning mode is "on"

Description (problem / solution / changelog)

Summary

Behavior Changes

Existing Functionality Check

What Changed

Tests

Files Changed

Checklist

Changed files

PR #43902: fix: restore final response delivery when Telegram reasoning mode is "on"

Description (problem / solution / changelog)

Summary

Behavior Changes

Existing Functionality Check

What Changed

Tests

Files Changed

Checklist

Changed files

Description

Environment

Steps to Reproduce

Current Behavior

Additional Notes

extent analysis

Fix Summary

Step‑by‑Step Fix Plan

1. Locate the Telegram channel implementation

2. Add a “send final” step for reasoningMode === 'on'

TypeScript / Node example

Python example (if the bot is written in Python)

Still need to ship something?

RELATED_DISCOVERY

TRENDING

2. Add a “send final” step for `reasoningMode === 'on'`