fix(telegram): add reasoning filter to slash command delivery path by okuyam2y · Pull Request #38689 · openclaw/openclaw

okuyam2y · 2026-03-07T08:22:55Z

Summary

Problem: bot-native-commands.ts handles slash commands (/reset, /model, etc.) via dispatchReplyWithBufferedBlockDispatcher, but its deliver callback had no reasoning text filtering. Gemini Flash thinking content (delivered as <think> tags in text) leaks to Telegram users after slash commands, even when reasoning display is off.
Why it matters: Users see raw internal reasoning text (Reasoning:\n_thinking content_) in Telegram after /reset and other slash commands.
What changed: Added splitTelegramReasoningText filtering to the slash command deliver callback, matching the behavior already present in bot-message-dispatch.ts (regular messages). Reasoning-only payloads with media still deliver media with empty text.
What did NOT change (scope boundary): bot-message-dispatch.ts (regular message path), reasoning-lane-coordinator.ts (split logic), reasoning level configuration. Slash commands always suppress reasoning text (no /reasoning on support in this path).

Change Type (select all)

Bug fix

Scope (select all touched areas)

Integrations

Linked Issue/PR

User-visible / Behavior Changes

Slash command responses in Telegram no longer include leaked reasoning/thinking text from Gemini Flash models when reasoning display is off (the default).

Security Impact (required)

New permissions/capabilities? No
Secrets/tokens handling changed? No
New/changed network calls? No
Command/tool execution surface changed? No
Data access scope changed? No

Repro + Verification

Environment

OS: macOS
Runtime/container: Node 22, OpenClaw gateway via launchd
Model/provider: google/gemini-2.5-flash (Google AI native provider)
Integration/channel: Telegram DM
Relevant config: thinkingDefault: off, reasoningLevel: off (default)

Steps

Configure OpenClaw with Gemini 2.5 Flash and Telegram channel
Send /reset in Telegram DM
Observe the greeting response

Expected

Only the greeting text appears (e.g. "Hello! Is there anything I can help you with?")

Actual (before fix)

A separate message with Reasoning:\n_The user has started a new session..._ appears before the greeting

Evidence

Trace/log snippets

Raw stream log before fix shows <think> content in text_delta events, and the text delivered to Telegram includes reasoning prefix. After fix, only answer text is delivered.

The key diagnostic: appendRawStream debug logging in bot-message-dispatch.ts showed zero tg_ events for /reset responses, proving the delivery went through bot-native-commands.ts instead — which had no reasoning filter.

Human Verification (required)

Verified scenarios: /reset with Gemini 2.5 Flash on Telegram — reasoning text no longer appears
Edge cases checked: reasoning-only + media payloads (deliver media with empty text, matching bot-message-dispatch.ts behavior)
What I did not verify: Other slash commands with reasoning-capable models (expected same behavior since they share the same deliver callback)

Compatibility / Migration

Backward compatible? Yes
Config/env changes? No
Migration needed? No

Failure Recovery (if this breaks)

How to disable/revert this change quickly: Revert commit (single file, single function change)
Files/config to restore: src/telegram/bot-native-commands.ts
Known bad symptoms reviewers should watch for: Slash command responses being silently dropped (would indicate splitTelegramReasoningText false-positiving on non-reasoning text)

Risks and Mitigations

Risk: splitTelegramReasoningText could false-positive on legitimate text starting with "Reasoning:" prefix
- Mitigation: Same risk exists in bot-message-dispatch.ts which has been running in production. The split function uses specific pattern matching (Reasoning:\n_..._) not a broad substring match.
Risk: Slash commands always suppress reasoning regardless of /reasoning on setting
- Mitigation: Slash commands don't have reasoning level UI/state. If reasoning display support is added to slash commands in the future, this filter should be updated to check resolvedReasoningLevel.

bot-native-commands.ts handles slash commands (/reset, /model, etc.) via dispatchReplyWithBufferedBlockDispatcher, but its deliver callback had no reasoning text filtering. bot-message-dispatch.ts (regular messages) already filters via splitTelegramReasoningText. This causes Gemini Flash thinking content (delivered as <think> tags in text) to leak to Telegram users after slash commands like /reset, even when reasoning display is off. Add splitTelegramReasoningText filtering to the deliver callback: - Pure reasoning text → skip delivery (or deliver media with empty text) - Mixed reasoning + answer → deliver answer only Relates: openclaw#37890, openclaw#26466 Co-Authored-By: Claude Opus 4.6 <[email protected]>

…aw#38689) Match bot-message-dispatch.ts behavior: reasoning-only payloads with media deliver media with empty text instead of being silently dropped. Co-Authored-By: Claude Opus 4.6 <[email protected]>

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 8d7d6a73fa

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

chatgpt-codex-connector · 2026-03-07T08:26:54Z

src/telegram/bot-native-commands.ts

+                    } else {
+                      return;


Avoid dropping slash replies that start with "Reasoning:"

This new early return can silently suppress valid slash-command output because it relies on splitTelegramReasoningText, which currently classifies any text beginning with Reasoning:\n as reasoning-only (splitTelegramReasoningText in src/telegram/reasoning-lane-coordinator.ts, lines 71-75). In scenarios where a user prompt intentionally asks for a response section titled “Reasoning:” and no media is attached, this branch exits before deliverReplies is called, so Telegram receives no message at all.

Useful? React with 👍 / 👎.

greptile-apps · 2026-03-07T08:27:12Z

Greptile Summary

This PR fixes a reasoning text leak in Telegram's slash command delivery path. The fix adds splitTelegramReasoningText filtering to the deliver callback inside dispatchReplyWithBufferedBlockDispatcher, matching the filtering already present in bot-message-dispatch.ts for regular messages.

The implementation correctly handles the core cases:

No reasoning: payload passes through unchanged
Reasoning + answer: reasoning text stripped, answer delivered
Reasoning-only + media: media delivered with empty text
Reasoning-only + no media: early return, no delivery

The change is a single-function, targeted fix that mirrors an established production-validated pattern. Slash commands always suppress reasoning text (no /reasoning on support in this path), which is correctly documented in comments.

Confidence Score: 5/5

Safe to merge — targeted, single-function fix that mirrors an existing production-proven pattern from bot-message-dispatch.ts.
This is a well-scoped, low-risk change that adds reasoning text filtering to slash commands using the exact same mechanism already validated in production for regular messages. The filtering logic is correct for all practical use cases. The fix directly addresses the reported issue (reasoning text leaking to Telegram users after slash commands) without side effects or regressions.
No files require special attention.

_{Last reviewed commit: 8d7d6a7}

…aw#38689) Match bot-message-dispatch.ts behavior: reasoning-only payloads with media deliver media with empty text instead of being silently dropped. Co-Authored-By: Claude Opus 4.6 <[email protected]>

okuyam2y · 2026-03-08T15:07:00Z

Closing — this fix has been independently merged into upstream main (same code at bot-native-commands.ts L714-743, including the import at L74). Thanks!

…aw#38689) Match bot-message-dispatch.ts behavior: reasoning-only payloads with media deliver media with empty text instead of being silently dropped. Co-Authored-By: Claude Opus 4.6 <[email protected]>

openclaw-barnacle bot added channel: telegram Channel integration: telegram size: XS labels Mar 7, 2026

chatgpt-codex-connector bot reviewed Mar 7, 2026

View reviewed changes

who96 mentioned this pull request Mar 7, 2026

fix(auto-reply): add unified outbound text sanitizer for NO_REPLY and streaming leaks #38232

Open

4 tasks

okuyam2y closed this Mar 8, 2026

okuyam2y mentioned this pull request Mar 8, 2026

Gemini raw reasoning text leaks into answer during Telegram DM streaming #37890

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(telegram): add reasoning filter to slash command delivery path#38689

fix(telegram): add reasoning filter to slash command delivery path#38689
okuyam2y wants to merge 1 commit intoopenclaw:mainfrom
okuyam2y:fix/telegram-slash-reasoning-filter

okuyam2y commented Mar 7, 2026 •

edited

Loading

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

chatgpt-codex-connector bot Mar 7, 2026

Uh oh!

greptile-apps bot commented Mar 7, 2026

Uh oh!

okuyam2y commented Mar 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Conversation

okuyam2y commented Mar 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Change Type (select all)

Scope (select all touched areas)

Linked Issue/PR

User-visible / Behavior Changes

Security Impact (required)

Repro + Verification

Environment

Steps

Expected

Actual (before fix)

Evidence

Human Verification (required)

Compatibility / Migration

Failure Recovery (if this breaks)

Risks and Mitigations

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector bot Mar 7, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot commented Mar 7, 2026

Greptile Summary

Confidence Score: 5/5

Uh oh!

okuyam2y commented Mar 8, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

okuyam2y commented Mar 7, 2026 •

edited

Loading