fix(matrix): recognize MiniMax mm: namespaced reasoning tags in monitor replies by zhangguiping-xydt · Pull Request #94038 · openclaw/openclaw

zhangguiping-xydt · 2026-06-17T08:46:08Z

Related: #89473

What Problem This Solves

Fixes an issue where Matrix users could see hidden MiniMax reasoning text in monitor replies when a model reply used mm: thinking tags such as <mm:think>, <mm:thinking>, or <mm:thought>.

Without this patch, Matrix could suppress replies that were entirely reasoning, but mixed reasoning-plus-visible replies could still deliver the hidden reasoning block into the room along with the visible answer.

Why This Change Was Made

Matrix now strips recognized thinking blocks before the Matrix send boundary, instead of only using them to decide whether a whole reply is reasoning-only.
The stripping uses the public shared openclaw/plugin-sdk/text-chunking reasoning-tag sanitizer so literal <mm:think> or <antml:think> examples inside inline/fenced code are preserved without importing core private files from the Matrix package.
The follow-up regression covers the waiting-on-author review request for mixed mm:/antml: reasoning blocks in text replies and locks the code-example preservation boundary.

User Impact

Matrix users no longer receive hidden reasoning text when a monitor reply mixes visible answer text with MiniMax or Anthropic-style thinking blocks. Visible reply text still sends normally, and answers that discuss reasoning tags as literal code examples are not corrupted.

Evidence

Real Matrix room proof from the earlier synced head remains relevant for the original reasoning-only path: a disposable unencrypted Matrix room on matrix.org received exactly one visible Matrix event while three mm: reasoning-only replies were suppressed at the Matrix monitor delivery boundary.
Additional mixed-path delivery-boundary proof on this head used the real deliverMatrixReplies -> sendMessageMatrix -> client.sendMessage Matrix send path with a fake Matrix client capturing final event content. It verified hidden mm: and antml: reasoning blocks were removed before outbound Matrix payloads, while a literal code example was preserved:
```
Matrix mixed reasoning delivery proof passed
{
  "bodies": [
    "Visible answer",
    "Second answer",
    "Literal example: `<mm:think>draft</mm:think>`"
  ],
  "eventCount": 3
}
```

Focused Matrix regression test with the public SDK sanitizer import:

$ node scripts/test-projects.mjs extensions/matrix/src/matrix/monitor/replies.test.ts
[test] starting test/vitest/vitest.extension-matrix.config.ts
The `envFile` option is deprecated, please use `envDir: false` instead.

 RUN  v4.1.9 [local path redacted]


 Test Files  1 passed (1)
      Tests  9 passed (9)
   Start at  22:00:17
   Duration  2.74s (transform 2.57s, setup 1.18s, import 817ms, tests 378ms, environment 0ms)

[test] passed 1 Vitest shard in 15.60s

Whitespace/conflict-marker check:

git diff --check origin/main...HEAD
Validation result: pass, exit_code=0, 143ms

Changed-file checks passed locally after refreshing dependencies with pnpm install --frozen-lockfile --ignore-scripts to avoid this host's node-llama-cpp postinstall/CMake 3.19 issue:

$ node scripts/check-changed.mjs -- extensions/matrix/src/matrix/monitor/replies.ts extensions/matrix/src/matrix/monitor/replies.test.ts
[check:changed] lanes=extensions, extensionTests
[check:changed] extensions/matrix/src/matrix/monitor/replies.test.ts: extension test
[check:changed] extensions/matrix/src/matrix/monitor/replies.ts: extension production
[check:changed] typecheck extensions
[check:changed] typecheck extension tests
[check:changed] runtime import cycles
Import cycle check: 0 runtime value cycle(s).

What was not rerun: an external Matrix homeserver room proof for the new mixed-block scenario. The previous disposable Matrix credentials and proof harness are not present in this workspace, so the new evidence is a local delivery-boundary run through the real Matrix send code with a fake client capturing the exact outbound event bodies.

clawsweeper · 2026-06-18T19:35:06Z

Codex review: needs changes before merge. Reviewed July 8, 2026, 4:09 AM ET / 08:09 UTC.

Summary
The PR expands Matrix monitor reply reasoning-tag regexes to recognize mm: and antml: thinking tags and adds a regression test for dropping mm: reasoning-only replies while sending a visible reply.

PR surface: Source +1, Tests +25. Total +26 across 2 files.

Reproducibility: yes. source inspection gives a high-confidence reproduction: a deliverMatrixReplies payload like <mm:think>secret</mm:think>Visible answer falls through the reasoning-only check and sends raw text. I did not run a local test because this review is read-only.

Review metrics: none identified.

Merge readiness
Overall: 🧂 unranked krab
Proof: 🦞 diamond lobster
Patch quality: 🧂 unranked krab
Result: blocked by patch quality or review findings.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

[P2] Strip recognized reasoning blocks before Matrix send and add a mixed mm: regression test.
Keep or refresh the live Matrix proof after the send-boundary change.

Risk before merge

[P1] Merging as-is could mark the Matrix mm:/antml: tag leak fixed while mixed reasoning-plus-visible replies still deliver hidden reasoning blocks into Matrix rooms.
[P1] The live proof demonstrates reasoning-only suppression, but it does not exercise the mixed block path that source inspection shows still reaches sendMessageMatrix.

Maintainer options:

Strip Mixed Reasoning Before Merge (recommended)
Update Matrix delivery to remove recognized reasoning blocks from mixed visible replies and add regression coverage for that exact path.
Merge Only The Narrow Drop Fix
Maintainers could intentionally land only reasoning-only suppression, but the mixed leak should remain tracked as unresolved on the canonical issue.
Pause For A Canonical Replacement
If maintainers prefer a different Matrix PR to carry the broader sanitizer, pause this PR and route the duplicate PRs to that canonical path.

Copy recommended automerge instruction

@clawsweeper automerge

Special instructions:
Strip recognized reasoning tags from Matrix reply text before chunking or media caption send, preserve reasoning-only suppression, and add a regression test for `<mm:think>secret</mm:think>Visible answer` delivering only visible text.

Next step before merge

[P2] A narrow automated repair can strip recognized reasoning blocks from Matrix outbound text and add mixed-content regression coverage.

Security
Needs attention: No supply-chain changes were found, but the patch still leaves a concrete hidden-reasoning leak path in Matrix delivery.

Review findings

[P1] Strip mixed reasoning before Matrix send — extensions/matrix/src/matrix/monitor/replies.ts:8-11

Review details

Best possible solution:

Strip recognized reasoning tags from Matrix outbound text before chunking or media caption delivery, while preserving the existing drop behavior for replies that become empty after stripping.

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

Yes, source inspection gives a high-confidence reproduction: a deliverMatrixReplies payload like <mm:think>secret</mm:think>Visible answer falls through the reasoning-only check and sends raw text. I did not run a local test because this review is read-only.

Is this the best way to solve the issue?

No. Expanding the regex is a necessary narrow step, but the best Matrix fix also strips recognized reasoning blocks from mixed visible replies using the existing SDK helper or equivalent send-boundary sanitizer.

Full review comments:

[P1] Strip mixed reasoning before Matrix send — extensions/matrix/src/matrix/monitor/replies.ts:8-11
This still leaves the previous blocker open. With the expanded regex, a reply like <mm:think>secret</mm:think>Visible answer is recognized as containing reasoning, but shouldSuppressReasoningReplyText only drops the reply when the stripped remainder is empty; otherwise rawText is sent unchanged. Please strip recognized reasoning tags before chunking/sending and add a mixed reasoning+visible regression test.
Confidence: 0.92

Overall correctness: patch is incorrect
Overall confidence: 0.9

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 4ae8d735bf54.

Label changes

Label changes:

add merge-risk: 🚨 message-delivery: The patch still allows mixed recognized reasoning blocks to be delivered as Matrix message text.
add merge-risk: 🚨 security-boundary: Hidden model reasoning can contain internal context, and this patch does not close the mixed reasoning leak path before merge.
add proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes live Matrix external-room output and the Real behavior proof check succeeded on the current head; it proves the reasoning-only case, not the mixed leak finding.
add rating: 🧂 unranked krab: Overall readiness is 🧂 unranked krab; proof is 🦞 diamond lobster and patch quality is 🧂 unranked krab.
add status: ⏳ waiting on author: ClawSweeper has contributor-facing work open and is waiting for author action. Sufficient (live_output): The PR body includes live Matrix external-room output and the Real behavior proof check succeeded on the current head; it proves the reasoning-only case, not the mixed leak finding.

Label justifications:

P1: The PR targets a real user-facing Matrix reasoning leak, and the remaining blocker is still a high-priority channel privacy/message-delivery bug.
merge-risk: 🚨 message-delivery: The patch still allows mixed recognized reasoning blocks to be delivered as Matrix message text.
merge-risk: 🚨 security-boundary: Hidden model reasoning can contain internal context, and this patch does not close the mixed reasoning leak path before merge.
rating: 🧂 unranked krab: Overall readiness is 🧂 unranked krab; proof is 🦞 diamond lobster and patch quality is 🧂 unranked krab.
status: ⏳ waiting on author: ClawSweeper has contributor-facing work open and is waiting for author action. Sufficient (live_output): The PR body includes live Matrix external-room output and the Real behavior proof check succeeded on the current head; it proves the reasoning-only case, not the mixed leak finding.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes live Matrix external-room output and the Real behavior proof check succeeded on the current head; it proves the reasoning-only case, not the mixed leak finding.

Evidence reviewed

PR surface:

Source +1, Tests +25. Total +26 across 2 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	1	3	2	+1
Tests	1	25	0	+25
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	2	28	2	+26

Security concerns:

[medium] Mixed reasoning can still leak to Matrix — extensions/matrix/src/matrix/monitor/replies.ts:8
The new tag recognition does not strip mixed reasoning blocks before send, so internal reasoning can still be delivered alongside visible text.
Confidence: 0.9

Acceptance criteria:

[P1] pnpm test extensions/matrix/src/matrix/monitor/replies.test.ts.
[P1] pnpm check:changed -- extensions/matrix/src/matrix/monitor/replies.ts extensions/matrix/src/matrix/monitor/replies.test.ts.
[P1] git diff --check origin/main...HEAD.

What I checked:

Repository policy: Read the full root AGENTS.md and scoped extensions/AGENTS.md; their review-depth and plugin-boundary guidance applied to this Matrix plugin review. (AGENTS.md:1, 4ae8d735bf54)
Current shipped Matrix behavior: Latest release v2026.6.11 still has Matrix-local reasoning-only regexes that match bare tags and antthinking, but not mm: or antml: namespaced tags. (extensions/matrix/src/matrix/monitor/replies.ts:8, e085fa1a3ffd)
PR head behavior: At PR head, the regexes recognize mm: and antml:, but shouldSuppressReasoningReplyText only returns true when removing thinking blocks leaves no visible text. (extensions/matrix/src/matrix/monitor/replies.ts:8, 62e38b010fda)
Send path: When suppression returns false, Matrix assigns rawText = reply.text ?? "" and sends the raw chunks to sendMessageMatrix, so mixed <mm:think>secret</mm:think>Visible answer still leaks the reasoning block. (extensions/matrix/src/matrix/monitor/replies.ts:83, 4ae8d735bf54)
Existing strip helper: The plugin SDK exports stripReasoningTagsFromText, and the shared helper already strips mixed mm: reasoning blocks while preserving visible text. (src/plugin-sdk/text-chunking.ts:62, 4ae8d735bf54)
Regression coverage gap: The added test covers three reasoning-only mm: replies plus a visible reply, but it does not cover a single mixed reasoning-plus-visible reply. (extensions/matrix/src/matrix/monitor/replies.test.ts:198, 62e38b010fda)

Likely related people:

justinhuangcode: Introduced Matrix's reasoning-only suppression helper and the original bare thinking-tag guard. (role: introduced behavior; confidence: high; commits: 1298bd4e1bdb; files: extensions/matrix/src/matrix/monitor/replies.ts)
vincentkoc: Recently touched the current Matrix replies/test files during the current tree import/release-history path. (role: recent area contributor; confidence: medium; commits: 3dbe98a1f737, e085fa1a3ffd; files: extensions/matrix/src/matrix/monitor/replies.ts, extensions/matrix/src/matrix/monitor/replies.test.ts)
steipete: Recent history shows Matrix test-boundary and reply-payload refactor work on this path. (role: adjacent owner; confidence: medium; commits: 605cb60586eb, 62edfdffbdd0; files: extensions/matrix/src/matrix/monitor/replies.ts)
gumadeiras: Recent Matrix monitor/runtime seam commits touched the same monitor module family. (role: adjacent owner; confidence: medium; commits: 3ceec929df0f, d077faab1a1f; files: extensions/matrix/src/matrix/monitor/replies.ts, extensions/matrix/src/matrix/monitor/handler.ts)

What the crustacean ranks mean

🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

How this review workflow works

ClawSweeper keeps one durable marker-backed review comment per issue or PR.
Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
Maintainers can also comment @clawsweeper review to request a fresh review only.
Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

Review history (3 earlier review cycles)

reviewed 2026-06-26T18:33:35.987Z sha 0317d3a :: needs maintainer review before merge. :: none
reviewed 2026-07-07T00:56:49.297Z sha 605178e :: needs maintainer review before merge. :: none
reviewed 2026-07-08T07:29:24.974Z sha 0d77028 :: needs changes before merge. :: [P1] Strip mixed reasoning before Matrix send

…or replies Mirrors openclaw#93874 (slack monitor) for the matrix extension. The matrix reply delivery path checks shouldSuppressReasoningReplyText to drop reasoning-only payloads before they reach the Matrix room; the THINKING_TAG_RE / THINKING_BLOCK_RE patterns only matched bare, antml:, or antthinking variants, so MiniMax <mm:think>…</mm:think> payloads were treated as ordinary replies and the reasoning text was sent into the Matrix room (a reasoning leak, not just a missed preview). Align the regexes with src/shared/text/reasoning-tags.ts and openclaw#93767's precedent by adding an optional (?:antml:|mm:)? prefix on think/thinking/thought; antthinking is left unprefixed. Adds a regression test that streams three mm: namespaced reasoning-only replies and asserts only the visible answer reaches sendMessageMatrix.

zhangguiping-xydt · 2026-07-08T07:49:22Z

@clawsweeper re-review

Updated this PR after syncing with current main:

Head is now 0d77028d38, merged from origin/main at 7e0324263b.
The PR body was refreshed with live Matrix room proof plus current synced-head validation.
Real behavior proof passed on the latest head.
Relevant Matrix and quality checks passed: QA Smoke CI (matrix), check-test-types, check-lint, build-artifacts, and Critical Quality (channel-runtime-boundary).
The only failing CI job is checks-node-compact-small-whole-2, where src/tui/tui-pty-local.e2e.test.ts > creates and adopts a fresh session through the real Gateway backend timed out after 150s. This PR only changes extensions/matrix/src/matrix/monitor/replies.ts and extensions/matrix/src/matrix/monitor/replies.test.ts, so that failure is outside this Matrix diff.

Could you please refresh the proof/status labels and rerun the unrelated failed shard if needed?

clawsweeper · 2026-07-08T07:49:25Z

🦞🧹
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.

openclaw-barnacle Bot added channel: matrix Channel integration: matrix size: XS proof: supplied External PR includes structured after-fix real behavior proof. labels Jun 17, 2026

openclaw-barnacle Bot added triage: mock-only-proof Candidate: PR proof only shows tests, mocks, snapshots, lint, typecheck, or CI. and removed proof: supplied External PR includes structured after-fix real behavior proof. labels Jun 19, 2026

This was referenced Jun 19, 2026

[Bug]: Reasoning tokens leak to chat channels when models stream interleaved text/thinking blocks #89473

Open

fix(matrix): recognize MiniMax mm: namespaced reasoning tags in monitor suppression #93926

Closed

clawsweeper Bot mentioned this pull request Jun 20, 2026

fix(imessage): recognize MiniMax mm: reasoning tags in reflection guard (completes #93767) #93820

Merged

zhangguiping-xydt force-pushed the followup/pr-93874 branch from 339e6fc to 0317d3a Compare June 20, 2026 01:56

clawsweeper Bot added rating: 🐚 platinum hermit Good normal PR readiness with ordinary maintainer review expected. and removed rating: 🦞 diamond lobster Very strong PR readiness with only minor maintainer review expected. labels Jul 7, 2026

Merge branch 'main' into followup/pr-93874

5556237

clawsweeper Bot mentioned this pull request Jul 7, 2026

fix(matrix): recognize mm: and antml: reasoning tags in reply suppression #101794

Closed

3 tasks

Merge remote-tracking branch 'origin/main' into pr-94038-sync

0d77028

zhangguiping-xydt added 2 commits July 8, 2026 15:50

Merge remote-tracking branch 'origin/main' into pr-94038-sync

c3434cb

Merge remote-tracking branch 'origin/main' into pr-94038-sync

62e38b0

fix(matrix): use public reasoning sanitizer import

47ec23f

openclaw-barnacle Bot added size: S and removed size: XS labels Jul 9, 2026

zhangguiping-xydt closed this Jul 9, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix(matrix): recognize MiniMax mm: namespaced reasoning tags in monitor replies#94038

fix(matrix): recognize MiniMax mm: namespaced reasoning tags in monitor replies#94038
zhangguiping-xydt wants to merge 7 commits into
openclaw:mainfrom
zhangguiping-xydt:followup/pr-93874

zhangguiping-xydt commented Jun 17, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented Jun 18, 2026 •

edited

Loading

Uh oh!

zhangguiping-xydt commented Jul 8, 2026

Uh oh!

clawsweeper Bot commented Jul 8, 2026 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Uh oh!

Uh oh!

Conversation

zhangguiping-xydt commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What Problem This Solves

Why This Change Was Made

User Impact

Evidence

Uh oh!

clawsweeper Bot commented Jun 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

zhangguiping-xydt commented Jul 8, 2026

Uh oh!

clawsweeper Bot commented Jul 8, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

zhangguiping-xydt commented Jun 17, 2026 •

edited

Loading

clawsweeper Bot commented Jun 18, 2026 •

edited

Loading

clawsweeper Bot commented Jul 8, 2026 •

edited

Loading