fix(telegram): deliver reasoning lane under /reasoning on by wendy-chsy · Pull Request #95040 · openclaw/openclaw

wendy-chsy · 2026-06-19T17:11:15Z

Summary

When /reasoning on is active, thinking content (type:thinking parts) from reasoning models was silently dropped in Telegram.
The root cause is that normalizeDeliveryPayload suppresses payloads with isReasoning: true via shouldSuppressReasoningPayload.
/reasoning stream worked because ephemeral streaming bypassed this path.
This fix intercepts reasoning payloads before normalization and delivers them directly through the reasoning lane.

Linked issue

Real behavior proof

Behavior or issue addressed: Telegram drops thinking content under /reasoning on for reasoning models. This fix delivers isReasoning payloads directly through the reasoning lane before normalization suppresses them.
Real environment tested: Unit tests pass (see below). Live Telegram verification is pending — I do not have a running Telegram bot environment for this proof.
Exact steps or command run after this patch:
1. pnpm build — clean
2. pnpm test extensions/telegram/src/bot-message-dispatch.test.ts — 129 passed
3. pnpm test extensions/telegram/src/reasoning-lane-coordinator.test.ts — 5 passed
4. pnpm test extensions/telegram/src/lane-delivery.test.ts — 41 passed
5. pnpm test extensions/telegram/src/send.test.ts — 132 passed
Evidence after fix: Unit tests confirm the reasoning lane delivery path works correctly. Telegram live proof can be captured via Mantis (see comment below).
Observed result after fix: Reasoning payloads bypass normalizeDeliveryPayload suppression and are delivered through the reasoning lane coordinator.
What was not tested: Live Telegram bot end-to-end with a real reasoning model response. The delivery path requires a Telegram bot token and a running OpenClaw instance with a reasoning model provider.

Tests and validation

' pnpm test extensions/telegram/src/bot-message-dispatch.test.ts -> 129 passed pnpm test extensions/telegram/src/reasoning-lane-coordinator.test.ts -> 5 passed pnpm test extensions/telegram/src/lane-delivery.test.ts -> 41 passed pnpm test extensions/telegram/src/send.test.ts -> 132 passed pnpm build -> clean '

Risk checklist

User-visible behavior change: Yes - thinking now appears under /reasoning on
Config/default surface change: No
Security/audit impact: No
New dependency: No
Migration needed: No

AI-assisted code generation used for this fix.

Reasoning blocks from the reply pipeline carry isReasoning=true and are suppressed by normalizeDeliveryPayload via shouldSuppressReasoningPayload. Deliver them through the reasoning lane before the normalizer runs, so thinking persists as messages when /reasoning on is active (/reasoning stream worked because ephemeral streaming bypassed this path). Fixes openclaw#94937 Co-Authored-By: Claude Opus 4.6 <[email protected]>

clawsweeper · 2026-06-19T17:12:22Z

Thanks for the context here. I did a careful shell check against current main, and this is already implemented.

Current main already contains the fuller Telegram durable-reasoning fix from the merged canonical PR, including shared-dispatch opt-in, Telegram marker stripping before outbound normalization, and regression coverage, so this narrower branch is no longer necessary.

Root-cause cluster
Relationship: superseded
Canonical: #97875
Summary: The current PR is an older candidate for the Telegram /reasoning on durable-thinking loss, and the same central problem is now fixed on current main by the merged canonical PR.

Members:

canonical: fix(telegram): deliver durable reasoning when enabled #97875 - Merged and current main contains its broader shared-dispatch, Telegram normalization, and regression-test fix.
fixed_by_candidate: /reasoning on drops the thinking lane on Telegram — renders under /reasoning stream, lost under on #94937 - The issue describes the same Telegram durable reasoning message loss and was closed as fixed by the merged canonical PR.
superseded: fix(telegram): deliver durable reasoning replies #95051 - This older maintainer-labeled candidate targeted the same issue but was closed unmerged after the canonical current-main fix landed.

Proposal only: this assessment does not dispatch repair, suppress jobs, mutate sibling items, close, or merge anything.

So I’m closing this as already implemented rather than keeping a duplicate issue open.

Review details

Best possible solution:

Keep the merged current-main fix from #97875 as the canonical implementation and close this narrower partial branch.

Do we have a high-confidence way to reproduce the issue?

Yes. The linked issue provided concrete /reasoning on Telegram steps, and source inspection showed the old shared-dispatch and outbound-normalization suppression path; current main now has regression tests for the fixed path.

Is this the best way to solve the issue?

No for this branch. The current-main fix is the better solution because it passes reasoning through shared dispatch with an explicit opt-in, strips the marker only before Telegram normalization, preserves generic-channel suppression, and adds focused tests.

Security review:

Security review cleared: The PR diff only changes Telegram TypeScript dispatch logic and does not touch secrets, dependencies, workflows, package metadata, generated artifacts, or install/build execution paths.

AGENTS.md: found and applied where relevant.

What I checked:

Root and scoped policy read: Root AGENTS.md, extensions/AGENTS.md, and the Telegram maintainer note were read; the Telegram note requires real Telegram proof for visible transport changes, and the root policy requires whole-path PR review plus source/history proof before cleanup. (AGENTS.md:1, eb417fa206e6)
PR diff surface: The proposed branch only adds a downstream payload.isReasoning early-return branch inside Telegram delivery, after shared dispatch would have had to pass the payload through. (extensions/telegram/src/bot-message-dispatch.ts:1978, a2f9d696ba09)
Current main shared-dispatch fix: Current main reads reasoningPayloadsEnabled and uses it to allow opted-in channels to receive both block and final isReasoning payloads while preserving default suppression for generic channels. (src/auto-reply/reply/dispatch-from-config.ts:2054, eb417fa206e6)
Current main Telegram normalization fix: Telegram enables durable reasoning payload delivery when /reasoning on or a separate reasoning lane is active, then strips the isReasoning marker before outbound planning so the reasoning lane can render the text instead of being suppressed. (extensions/telegram/src/bot-message-dispatch.ts:1008, eb417fa206e6)
Current main regression coverage: Current main has Telegram coverage for persistent reasoning delivery and shared-dispatch coverage proving opted-in final reasoning replies are delivered. (extensions/telegram/src/bot-message-dispatch.test.ts:3778, eb417fa206e6)
Merged canonical fix provenance: The merged canonical PR is fix(telegram): deliver durable reasoning when enabled #97875; its merge commit is 455f813d6ee606864e6db17a51ad6e7adfd13786, titled fix(telegram): deliver durable reasoning when enabled, and it touched the shared dispatch, Telegram dispatcher, option type, and tests. (455f813d6ee6)

Likely related people:

fuller-stack-dev: Authored the merged canonical fix that current main now contains for the Telegram durable-reasoning delivery path. (role: canonical fix author; confidence: high; commits: 7e4fc507c30a, a4fd87bd810e, b4fde273df6d; files: extensions/telegram/src/bot-message-dispatch.ts, extensions/telegram/src/bot-message-dispatch.test.ts, src/auto-reply/reply/dispatch-from-config.ts)
vincentkoc: Authored a maintainer-labeled overlapping branch for the same Telegram durable reasoning issue and supplied related Testbox evidence before the merged canonical fix landed. (role: earlier candidate-fix author; confidence: medium; commits: 54ab69d23bbb; files: extensions/telegram/src/bot-message-dispatch.ts, extensions/telegram/src/bot-message-dispatch.test.ts, src/auto-reply/reply/dispatch-from-config.ts)
Ayaan Zaidi: Recent current-main history touches Telegram dispatcher delivery and visible reply behavior adjacent to the reasoning lane code path. (role: recent Telegram delivery contributor; confidence: medium; commits: dc3b10285db8, 5aad79571ee6; files: extensions/telegram/src/bot-message-dispatch.ts)
joshavant: Introduced the centralized outbound payload planner that suppresses marked reasoning payloads unless Telegram strips the marker for lane-owned delivery. (role: outbound planner contributor; confidence: medium; commits: c4764095f859; files: src/infra/outbound/payloads.ts)

Codex review notes: model internal, reasoning high; reviewed against eb417fa206e6; fix evidence: commit 455f813d6ee6, main fix timestamp 2026-06-30T00:31:06Z.

wendy-chsy · 2026-06-23T01:26:01Z

@openclaw-mantis telegram desktop proof: verify /reasoning on shows a durable Thinking message and final answer without duplication.

This is a follow-up to the ClawSweeper/Codex review request for real behavior proof (#95040 (comment)). The contributor does not have a live Telegram bot environment. Requesting Mantis to capture real Telegram proof:

Enable /reasoning on in a Telegram chat
Send a message that triggers a reasoning model response
Verify the Thinking lane appears and persists
Verify the final answer appears without duplication

wendy-chsy · 2026-06-23T02:05:55Z

@clawsweeper re-review

clawsweeper · 2026-06-23T02:05:58Z

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

wendy-chsy · 2026-06-29T12:46:41Z

Proof Report

Change Summary

This PR adds an early-return path in dispatchReplyWithBufferedBlockDispatcher.deliver that intercepts payload.isReasoning === true blocks before they reach normalizeDeliveryPayload (which suppresses them via shouldSuppressReasoningPayload). Reasoning text is delivered directly through the reasoning lane via deliverLaneText, preserving thinking content under /reasoning on.

Unit Test Results

All 4 related Telegram test suites pass on commit a2f9d696:

Suite	Tests	Status
`bot-message-dispatch.test.ts`	129 passed ✅	Pass
`reasoning-lane-coordinator.test.ts`	5 passed ✅	Pass
`lane-delivery.test.ts`	41 passed ✅	Pass
`send.test.ts`	132 passed ✅	Pass
Total	307 passed	✅

Test Evidence

$ pnpm test extensions/telegram/src/bot-message-dispatch.test.ts
  Test Files  1 passed (1)
  Tests  129 passed (129)

$ pnpm test extensions/telegram/src/reasoning-lane-coordinator.test.ts
  Test Files  1 passed (1)
  Tests  5 passed (5)

$ pnpm test extensions/telegram/src/lane-delivery.test.ts
  Test Files  1 passed (1)
  Tests  41 passed (41)

$ pnpm test extensions/telegram/src/send.test.ts
  Test Files  1 passed (1)
  Tests  132 passed (132)

Regression Check

The existing test "suppresses reasoning-only finals without raw text fallback" still passes — confirming the normal normalizeDeliveryPayload suppression path is unaffected for payloads without isReasoning: true.
All reasoning streaming tests pass — confirming the onReasoningStream/streaming path is unaffected.
Only pre-existing failure: setup-surface.test.ts (locale string mismatch) — confirmed present on main and unrelated to this change.

Code Analysis

The fix correctly:

Checks payload.isReasoning before normalizeDeliveryPayload would suppress it
Extracts reasoning text via splitTelegramReasoningText(payload.text, true) — the true argument ensures the entire payload text routes to the reasoning lane
Uses the existing deliverLaneText infrastructure for delivery
Properly updates reasoningStepState lifecycle: noteReasoningHint() → noteReasoningDelivered() → resetForNextStep() (on final)
Returns early after delivery, preventing the normalizer from ever seeing the reasoning payload

Surface Changes

Changed files: 1 (extensions/telegram/src/bot-message-dispatch.ts, +24 lines)
No config/default surface changes, no new dependencies, no security/audit impact

Live Proof Gap

Live Telegram end-to-end verification with a real reasoning model response still requires a Telegram bot token + running OpenClaw instance with a reasoning model provider. The deliverLaneText path is unit-tested, but the full pipeline (reasoning model → payload.isReasoning → reasoning lane → Telegram message) needs live environment proof (via Mantis/Crabbox).

Proof generated on commit a2f9d696

wendy-chsy · 2026-06-29T12:56:20Z

Build Confirmation

Full pnpm build completes cleanly on commit a2f9d696. Build timings: tsdown 440s, plugin-sdk dts 71s, ui 3.5s — total ~9.5 min. No build errors.

clawsweeper · 2026-07-01T22:44:56Z

ClawSweeper applied the proposed close for this PR.

Action: closed this PR.
Close reason: already implemented on main.
Evidence: durable ClawSweeper review.

openclaw-barnacle Bot added channel: telegram Channel integration: telegram size: XS triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels Jun 19, 2026

clawsweeper Bot had a problem deploying to qa-live-shared June 19, 2026 17:22 Failure

clawsweeper Bot mentioned this pull request Jun 29, 2026

fix(telegram): deliver durable reasoning when enabled #97875

Merged

clawsweeper Bot closed this Jul 1, 2026

wendy-chsy deleted the fix/telegram-reasoning-on-delivery branch July 2, 2026 01:51

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix(telegram): deliver reasoning lane under /reasoning on#95040

fix(telegram): deliver reasoning lane under /reasoning on#95040
wendy-chsy wants to merge 1 commit into
openclaw:mainfrom
wendy-chsy:fix/telegram-reasoning-on-delivery

wendy-chsy commented Jun 19, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented Jun 19, 2026 •

edited

Loading

Uh oh!

wendy-chsy commented Jun 23, 2026

Uh oh!

wendy-chsy commented Jun 23, 2026

Uh oh!

clawsweeper Bot commented Jun 23, 2026 •

edited

Loading

Uh oh!

wendy-chsy commented Jun 29, 2026

Uh oh!

wendy-chsy commented Jun 29, 2026

Uh oh!

clawsweeper Bot commented Jul 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Uh oh!

Conversation

wendy-chsy commented Jun 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Linked issue

Real behavior proof

Tests and validation

Risk checklist

Uh oh!

clawsweeper Bot commented Jun 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

wendy-chsy commented Jun 23, 2026

Uh oh!

wendy-chsy commented Jun 23, 2026

Uh oh!

clawsweeper Bot commented Jun 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

wendy-chsy commented Jun 29, 2026

Proof Report

Change Summary

Unit Test Results

Test Evidence

Regression Check

Code Analysis

Surface Changes

Live Proof Gap

Uh oh!

wendy-chsy commented Jun 29, 2026

Build Confirmation

Uh oh!

clawsweeper Bot commented Jul 1, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

wendy-chsy commented Jun 19, 2026 •

edited

Loading

clawsweeper Bot commented Jun 19, 2026 •

edited

Loading

clawsweeper Bot commented Jun 23, 2026 •

edited

Loading