fix(imessage): recognize MiniMax mm: reasoning tags in reflection guard (completes #93767) by Alix-007 · Pull Request #93820 · openclaw/openclaw

Alix-007 · 2026-06-17T00:16:27Z

What Problem This Solves

The iMessage reflection guard (extensions/imessage/src/monitor/reflection-guard.ts) failed to recognize MiniMax mm:-namespaced reasoning tags (<mm:think>, <mm:thinking>, <mm:thought>). Its THINKING_TAG_RE only matched bare/antthinking reasoning tags, so a MiniMax reply carrying <mm:think>…</mm:think> was not treated as reflected assistant reasoning. Because detectReflectedContent feeds the inbound drop decision (inbound-processing.ts:718), any user on a MiniMax-backed agent would have the model's internal reasoning leak through to the iMessage thread (and risk echo-amplification) instead of being suppressed. This continues the cleanup flagged by #93767, which added mm: support to the central reasoning-tag handling and noted the same pattern was duplicated in other monitor paths — this guard was one of them (it was missing even the antml: prefix).

Evidence

Driven against the real exported detectReflectedContent (no mocks — it is a deterministic function and importing it exercises the exact code path inbound-processing.ts:718 uses), local, Node v22.22.0, node --import tsx:

After fix: "<mm:think>secret reasoning</mm:think>visible" → isReflection=true, labels=["thinking-tag"], content suppressed before reaching the downstream inbound pipeline. <mm:thinking>/<mm:thought> variants all flagged too.
No regression: bare <think>/<thinking> tags still isReflection=true; mm:think inside inline code still isReflection=false (code-fence exemption preserved); the phrase "thinking about your minimax question" still isReflection=false (no misclassification).
Negative control: reverting only the source regex to its pre-fix form and re-running the same proof flips the MiniMax case to isReflection=false — the reasoning is delivered downstream (leaks past the guard); restoring the fix returns it to suppressed. This pins the behavior change to exactly this patch.
Tests: reflection-guard.test.ts → 20 pass, including the 2 new namespaced cases (Anthropic <think> and MiniMax <mm:think>). oxlint + oxfmt --check clean.

Summary

#93767 added MiniMax mm: namespace support to reasoning-tag handling, and its own follow-up note flagged that the same tag pattern is duplicated in several other monitor paths. The iMessage reflection guard is one of them: THINKING_TAG_RE in extensions/imessage/src/monitor/reflection-guard.ts:7 only matched bare/antthinking reasoning tags, so a MiniMax <mm:think>…</mm:think> reply was not recognized as reflected reasoning. detectReflectedContent feeds the inbound drop decision (inbound-processing.ts:718), so unrecognized MiniMax reasoning slips through reflection detection and can be echo-amplified.

Changes

extensions/imessage/src/monitor/reflection-guard.ts: (?:think(?:ing)?|thought|antthinking) → (?:(?:antml:|mm:)?(?:think(?:ing)?|thought)|antthinking), matching fix(reasoning-tags): strip MiniMax mm: namespaced reasoning tags #93767's exact pattern (this guard had been missing even antml: — aligned to the shared reasoning-tag contract fix(reasoning-tags): strip MiniMax mm: namespaced reasoning tags #93767 established). +8/-1 + 2 regression tests.

Real behavior proof

Behavior addressed: the iMessage reflection guard recognizes MiniMax mm:-namespaced reasoning tags (<mm:think>, <mm:thinking>, <mm:thought>) the same as bare ones, so MiniMax reasoning is suppressed by the inbound reflection guard instead of being forwarded downstream.

Real environment tested: local, Node v22.22.0, node --import tsx, driving the real exported detectReflectedContent from the changed file (which uses the real openclaw/plugin-sdk/text-chunking code-region logic). No mocks — this is a deterministic reflection-detection function, so importing it exercises the exact production code path inbound-processing.ts:718 relies on. A downstream-delivery wrapper models the readback boundary ("what actually reaches the downstream inbound pipeline": a flagged reflection is suppressed; an unflagged one is delivered/leaks).

Exact steps or command run after this patch:

git worktree add -f /tmp/pwt fork/fix/minimax-mm-channel-monitor-guards
cd /tmp/pwt && node --import tsx proof.mts            # drives real detectReflectedContent
node scripts/run-vitest.mjs extensions/imessage/src/monitor/reflection-guard.test.ts --run

Evidence after fix (real function output):

[PASS] MiniMax mm: namespaced reasoning tag
       input="<mm:think>secret reasoning</mm:think>visible"
       isReflection=true  labels=["thinking-tag"]
       downstream delivered=<suppressed by guard>
[PASS] bare <think>/<thinking> tags         -> isReflection=true  (no regression)
[PASS] mm:think inside inline code           -> isReflection=false (code-fence exemption preserved)
[PASS] "thinking about your minimax question"-> isReflection=false (no misclassification)
<mm:thinking>/<mm:thought> variants          -> isReflection=true  (all mm: forms covered)

Observed result after fix: MiniMax <mm:think>…</mm:think> content is now flagged as a reflection (isReflection=true, label thinking-tag) and suppressed before reaching the downstream inbound pipeline; bare-tag and code-fence-exemption behavior is unchanged. reflection-guard.test.ts: 20 pass (incl. 2 new namespaced cases). oxlint + oxfmt --check clean.

Negative control: reverting only the source regex to its pre-fix form (git show <parent>:reflection-guard.ts) and re-running the same proof flips the MiniMax case to isReflection=false — "<mm:think>secret reasoning</mm:think>visible" is delivered downstream (leaks past the guard) instead of being suppressed; restoring the fix returns it to suppressed. This pins the behavior change to exactly this patch.

What was not tested: no live iMessage round-trip; the change is in the deterministic reflection-detection regex, driven via the real exported function.

Real behavior proof

Behavior addressed: The iMessage reflection guard did not recognize MiniMax mm: namespaced reasoning tags such as mm:think, mm:thinking, and mm:thought. This patch makes the guard flag those tags as reflected reasoning so they are suppressed instead of being forwarded downstream.

Real environment tested: Local Linux runtime with Node v22.22.0 and node --import tsx, driving the real exported detectReflectedContent from extensions/imessage/src/monitor/reflection-guard.ts. The function imports and uses the real openclaw/plugin-sdk text-chunking code-region logic. A small downstream-delivery wrapper read back whether flagged content would be suppressed or delivered.

Exact steps or command run after this patch: In PR head fork/fix/minimax-mm-channel-monitor-guards, ran node --import tsx proof.mts to drive real detectReflectedContent, then ran node scripts/run-vitest.mjs extensions/imessage/src/monitor/reflection-guard.test.ts --run.

Evidence after fix: Live terminal output from the proof script:

[PASS] MiniMax mm: namespaced reasoning tag
       input="<mm:think>secret reasoning</mm:think>visible"
       isReflection=true  labels=["thinking-tag"]
       downstream delivered=<suppressed by guard>
[PASS] bare <think>/<thinking> tags         -> isReflection=true  (no regression)
[PASS] mm:think inside inline code           -> isReflection=false (code-fence exemption preserved)
[PASS] "thinking about your minimax question"-> isReflection=false (no misclassification)
<mm:thinking>/<mm:thought> variants          -> isReflection=true  (all mm: forms covered)

Negative control reverting only the source regex to the pre-fix form flipped the MiniMax case to isReflection=false and delivered the mm:think content downstream; restoring the fix suppressed it again.

Observed result after fix: MiniMax mm:think...</mm:think> content is now flagged as a reflection with label thinking-tag and suppressed before reaching the downstream inbound pipeline. Bare-tag behavior and code-fence exemption behavior remain unchanged. reflection-guard.test.ts passed 20 tests, with oxlint and oxfmt --check clean.

What was not tested: No live iMessage round trip was run. The changed behavior is a deterministic reflection-detection regex and was driven through the real exported function rather than a phone/device integration.

clawsweeper · 2026-06-17T00:32:06Z

Codex review: needs maintainer review before merge. Reviewed June 30, 2026, 3:28 PM ET / 19:28 UTC.

Summary
The PR broadens the iMessage reflection guard's thinking-tag regex to recognize antml: and mm: namespaced reasoning tags and adds table-driven regression tests.

PR surface: Source +3, Tests +9. Total +12 across 2 files.

Reproducibility: yes. Current main and v2026.6.11 do not match mm: namespaced tags in the iMessage reflection guard, and the inbound caller drops only when that guard returns a reflection result; I did not run tests in this read-only review.

Review metrics: 1 noteworthy metric.

Drop classifier change: 1 regex broadened. The changed regex controls which inbound iMessage texts are suppressed before reaching the agent pipeline.

Root-cause cluster
Relationship: canonical
Canonical: #93820
Summary: This PR is the canonical open iMessage branch for the MiniMax namespaced reasoning-tag reflection-guard gap; related items cover shared, source, Slack, Matrix, and broader reasoning-leak slices.

Members:

partial_overlap: fix(reasoning-tags): strip MiniMax mm: namespaced reasoning tags #93767 - Merged shared and Telegram MiniMax mm: reasoning-tag support established the contract but did not update the iMessage reflection guard.
same_root_cause: fix(reasoning-tags): strip MiniMax mm: tags on silent-reply and streaming paths missed by #93767 #93806 - Merged follow-up fixed omitted MiniMax namespace recognition on silent-reply and streaming paths outside iMessage.
same_root_cause: fix(slack): recognize MiniMax mm: namespaced reasoning tags in monitor preview #93874 - Merged Slack monitor follow-up fixed the same omitted namespace pattern in a sibling channel-local reasoning matcher.
duplicate: fix(imessage): detect namespaced thinking reflections #95690 - Closed unmerged duplicate targeted the same iMessage reflection-guard namespaced-tag behavior and the same two files.
adjacent_distinct: fix(matrix): recognize MiniMax mm: namespaced reasoning tags in monitor replies #94038 - Open Matrix PR targets the same MiniMax namespace family in Matrix reply suppression, not the iMessage inbound reflection guard.
1 more in the report.

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

Merge readiness
Overall: 🦞 diamond lobster
Proof: 🦞 diamond lobster
Patch quality: 🦞 diamond lobster
Result: ready for maintainer review.

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

Risk before merge

[P1] The broadened iMessage classifier will drop non-code inbound messages containing known antml: or mm: reasoning markup before agent processing, so maintainers should explicitly accept that message-delivery tradeoff.
[P1] The proof is strong parser/drop-boundary output with a negative control, but it is not a live iMessage device round trip.

Maintainer options:

Accept Known-Tag Suppression (recommended)
Land the PR after explicitly accepting that iMessage will now suppress non-code inbound messages containing the known assistant reasoning tags as reflected content.
Ask For Live iMessage Proof
Request a short iMessage monitor or device-boundary proof if deterministic parser and caller evidence is not enough for this classifier change.
Pause For Shared Matcher Consolidation
Hold or close this branch only if maintainers want a shared plugin-SDK reasoning-tag matcher before accepting another channel-local regex alignment.

Next step before merge

No automated repair remains; maintainers should decide whether to accept the iMessage drop-classifier expansion and whether deterministic proof is enough before merge.

Security
Cleared: The diff changes only a local iMessage regex and focused tests, with no dependency, workflow, secret, permission, or supply-chain changes.

Review details

Best possible solution:

Land the narrow iMessage guard alignment after maintainer acceptance of the known-tag suppression boundary, while leaving broader cross-channel reasoning taxonomy cleanup with #89473.

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

Yes. Current main and v2026.6.11 do not match mm: namespaced tags in the iMessage reflection guard, and the inbound caller drops only when that guard returns a reflection result; I did not run tests in this read-only review.

Is this the best way to solve the issue?

Yes. The iMessage plugin owns this inbound reflection guard, and aligning its local matcher with the existing shared reasoning-tag contract is the narrow fix; a shared matcher is a broader follow-up, not a blocker.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 5a89484eb31d.

Label changes

Label justifications:

P2: This is a normal-priority iMessage-specific bug fix with limited blast radius but real message-delivery impact.
merge-risk: 🚨 message-delivery: The diff changes which inbound iMessage texts are classified as reflected assistant content and dropped before the agent pipeline.
rating: 🦞 diamond lobster: Overall readiness is 🦞 diamond lobster; proof is 🦞 diamond lobster and patch quality is 🦞 diamond lobster.
status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Sufficient (live_output): The PR body includes live output from the real exported reflection parser plus a negative control; no live iMessage round trip was run, but the changed behavior is deterministic and at the production drop boundary.
proof: sufficient: Contributor real behavior proof is sufficient. The PR body includes live output from the real exported reflection parser plus a negative control; no live iMessage round trip was run, but the changed behavior is deterministic and at the production drop boundary.

Evidence reviewed

PR surface:

Source +3, Tests +9. Total +12 across 2 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	1	4	1	+3
Tests	1	9	0	+9
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	2	13	1	+12

What I checked:

PR head broadens iMessage reflection tag detection: At PR head, THINKING_TAG_RE accepts antml: and mm: prefixes for think/thinking/thought tags while keeping the existing antthinking alternative. (extensions/imessage/src/monitor/reflection-guard.ts:9, 53eb0ed7c871)
PR head adds namespaced regression coverage: The test table covers both antml:think and mm:think and asserts the reflection label remains thinking-tag. (extensions/imessage/src/monitor/reflection-guard.test.ts:40, 53eb0ed7c871)
Current main still lacks namespaced matching: Current main's iMessage reflection guard only matches bare think/thinking/thought and antthinking, so this PR is not obsolete on main. (extensions/imessage/src/monitor/reflection-guard.ts:7, 5a89484eb31d)
Latest release also lacks namespaced matching: v2026.6.11 has the same non-namespaced iMessage regex, so the fix is not already shipped in the latest release. (extensions/imessage/src/monitor/reflection-guard.ts:7, e085fa1a3ffd)
Inbound caller drops on reflection result: The iMessage inbound path calls detectReflectedContent(messageText) and returns a drop decision when the guard reports reflected assistant content. (extensions/imessage/src/monitor/inbound-processing.ts:718, 5a89484eb31d)
Shared reasoning-tag contract already includes MiniMax: The shared reasoning-tag sanitizer accepts antml: and mm: prefixes, supporting the PR's alignment approach for the iMessage-local guard. (src/shared/text/reasoning-tags.ts:8, 5a89484eb31d)

Likely related people:

joelnishanth: Authored the merged iMessage echo-loop hardening PR that introduced the reflection guard and inbound reflected-content drop behavior now being adjusted. (role: introduced behavior; confidence: high; commits: adb9234d0375; files: src/imessage/monitor/reflection-guard.ts, src/imessage/monitor/inbound-processing.ts)
scoootscooob: Moved the iMessage implementation into extensions/imessage, establishing the current plugin-owned paths touched by this PR. (role: owner-boundary refactor contributor; confidence: high; commits: 0ce23dc62d37; files: extensions/imessage/src/monitor/reflection-guard.ts, extensions/imessage/src/monitor/inbound-processing.ts)
DrHack1: Authored the merged shared MiniMax mm: reasoning-tag support that this PR aligns the iMessage guard with. (role: recent related contract contributor; confidence: medium; commits: 4f860bfab09c; files: src/shared/text/reasoning-tags.ts, extensions/telegram/src/reasoning-lane-coordinator.ts)
Alix-007: Beyond this PR, authored merged follow-ups for MiniMax mm: recognition in sibling silent-reply, streaming, and Slack monitor paths. (role: recent related follow-up contributor; confidence: high; commits: f8e7d66ae6b0, d2bf67f4b709; files: src/auto-reply/tokens.ts, src/agents/embedded-agent-subscribe.handlers.messages.ts, extensions/slack/src/monitor/message-handler/dispatch.ts)
steipete: Is assigned to this PR and authored the latest branch refactor commit tightening the iMessage coverage shape. (role: current reviewer and branch refactor author; confidence: medium; commits: 53eb0ed7c871; files: extensions/imessage/src/monitor/reflection-guard.ts, extensions/imessage/src/monitor/reflection-guard.test.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.

Alix-007 · 2026-06-20T00:43:32Z

@clawsweeper re-review — Added a Real behavior proof section. It drives the real detectReflectedContent reflection guard; post-patch a <mm:think> namespaced reasoning tag is detected and suppressed (not delivered downstream), while a negative control on the pre-patch regex lets the mm: reasoning leak past the guard. Bare tags and code-fence exemptions unchanged. Vitest (20 tests) passes.

clawsweeper · 2026-06-20T00:43:35Z

🦞🧹
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.

Alix-007 · 2026-06-21T06:27:48Z

@clawsweeper re-review — Added the What Problem This Solves and Evidence headings the proof workflow expects (proof content unchanged, just restructured under the required sections). This should clear needs-pr-context.

Alix-007 · 2026-06-23T13:33:02Z

@clawsweeper re-review — Updated the PR body with a machine-readable ## Real behavior proof section using the exact field names required by the current proof policy. I verified the updated body locally against evaluateRealBehaviorProof and the latest GitHub Real behavior proof check is now passing.

…tion guard The iMessage reflection guard's THINKING_TAG_RE only matched bare `<think>/<thought>` tags, so MiniMax's `<mm:think>` namespaced reasoning tags (and Anthropic's `antml:` prefix) were not recognized as reflected assistant content. When such reasoning leaked back inbound, the guard treated it as a normal user message instead of dropping it, allowing recursive echo amplification. Accept the known reasoning-tag namespace prefixes `(?:antml:|mm:)?` on the think/thought alternatives, mirroring the shared reasoning-tag contract in src/shared/text/reasoning-tags.ts (PR openclaw#93767). This is the remaining channel-monitor complement of openclaw#93767 for the iMessage path; the optional prefix is a pure recognition enhancement with no change to the bare-tag, code-region exemption, or other reflection patterns. Verified by driving the real detectReflectedContent export: <mm:think>secret</mm:think>visible -> isReflection=true (was false) <think>...</think> -> isReflection=true <think>secret</think> -> isReflection=true (no regression) "thinking about minimax algorithms" -> isReflection=false (no false-positive) `<mm:think>x</mm:think>` in code -> isReflection=false (code exemption kept)

steipete · 2026-06-30T19:41:27Z

Land-ready maintainer pass on exact head 53eb0ed7c8710b18494fad63c9c3378ce8b9610a:

Tightened the new invariant comment and table-drove the duplicate namespace cases; final diff is +13/-1 across the guard and its tests.
Focused proof: node scripts/run-vitest.mjs extensions/imessage/src/monitor/reflection-guard.test.ts — 1 file, 20 tests passed.
Direct runtime proof: antml:think, mm:think, mm:thinking, and mm:thought classify as thinking-tag; inline-code and normal-prose controls remain unclassified.
Fresh autoreview: clean, no accepted/actionable findings (0.99 confidence).
Exact-head hosted CI: run 28470084930, all relevant checks green. Its unrelated otel-trace-smoke QA failure passed on a single-job rerun.
Repository landing verifier: exact-head hosted CI/Testbox gates passed.

No live iMessage device round trip was run; the changed boundary is a deterministic parser used directly by the inbound decision path.

steipete · 2026-06-30T19:41:49Z

Merged via squash.

Prepared head SHA: 53eb0ed7c8710b18494fad63c9c3378ce8b9610a
Landed commit: 076da567f434388e03153692fdce099f1e6c5ca5

…rd (completes openclaw#93767) (openclaw#93820) * fix(imessage): detect MiniMax mm: namespaced reasoning tags in reflection guard The iMessage reflection guard's THINKING_TAG_RE only matched bare `<think>/<thought>` tags, so MiniMax's `<mm:think>` namespaced reasoning tags (and Anthropic's `antml:` prefix) were not recognized as reflected assistant content. When such reasoning leaked back inbound, the guard treated it as a normal user message instead of dropping it, allowing recursive echo amplification. Accept the known reasoning-tag namespace prefixes `(?:antml:|mm:)?` on the think/thought alternatives, mirroring the shared reasoning-tag contract in src/shared/text/reasoning-tags.ts (PR openclaw#93767). This is the remaining channel-monitor complement of openclaw#93767 for the iMessage path; the optional prefix is a pure recognition enhancement with no change to the bare-tag, code-region exemption, or other reflection patterns. Verified by driving the real detectReflectedContent export: <mm:think>secret</mm:think>visible -> isReflection=true (was false) <think>...</think> -> isReflection=true <think>secret</think> -> isReflection=true (no regression) "thinking about minimax algorithms" -> isReflection=false (no false-positive) `<mm:think>x</mm:think>` in code -> isReflection=false (code exemption kept) * chore: retrigger CI for real behavior proof check * refactor(imessage): tighten reflection guard coverage --------- Co-authored-by: Peter Steinberger <[email protected]>

…rd (completes openclaw#93767) (openclaw#93820) * fix(imessage): detect MiniMax mm: namespaced reasoning tags in reflection guard The iMessage reflection guard's THINKING_TAG_RE only matched bare `<think>/<thought>` tags, so MiniMax's `<mm:think>` namespaced reasoning tags (and Anthropic's `antml:` prefix) were not recognized as reflected assistant content. When such reasoning leaked back inbound, the guard treated it as a normal user message instead of dropping it, allowing recursive echo amplification. Accept the known reasoning-tag namespace prefixes `(?:antml:|mm:)?` on the think/thought alternatives, mirroring the shared reasoning-tag contract in src/shared/text/reasoning-tags.ts (PR openclaw#93767). This is the remaining channel-monitor complement of openclaw#93767 for the iMessage path; the optional prefix is a pure recognition enhancement with no change to the bare-tag, code-region exemption, or other reflection patterns. Verified by driving the real detectReflectedContent export: <mm:think>secret</mm:think>visible -> isReflection=true (was false) <think>...</think> -> isReflection=true <think>secret</think> -> isReflection=true (no regression) "thinking about minimax algorithms" -> isReflection=false (no false-positive) `<mm:think>x</mm:think>` in code -> isReflection=false (code exemption kept) * chore: retrigger CI for real behavior proof check * refactor(imessage): tighten reflection guard coverage --------- Co-authored-by: Peter Steinberger <[email protected]> (cherry picked from commit 076da56)

…eplies The THINKING_TAG_RE and THINKING_BLOCK_RE regex patterns in the Matrix monitor replies module were missing the mm: namespace prefix for detecting MiniMax reasoning tags. This caused mm:think/mm:thought reasoning content to leak into Matrix room messages. Add (?:antml:|mm:)? prefix to think(?:ing)?|thought alternatives in both regexes, consistent with the pattern already applied in reasoning-tags.ts and other channel modules (see openclaw#93767, openclaw#93806, openclaw#93820). Fixes openclaw#99412

…rd (completes openclaw#93767) (openclaw#93820) * fix(imessage): detect MiniMax mm: namespaced reasoning tags in reflection guard The iMessage reflection guard's THINKING_TAG_RE only matched bare `<think>/<thought>` tags, so MiniMax's `<mm:think>` namespaced reasoning tags (and Anthropic's `antml:` prefix) were not recognized as reflected assistant content. When such reasoning leaked back inbound, the guard treated it as a normal user message instead of dropping it, allowing recursive echo amplification. Accept the known reasoning-tag namespace prefixes `(?:antml:|mm:)?` on the think/thought alternatives, mirroring the shared reasoning-tag contract in src/shared/text/reasoning-tags.ts (PR openclaw#93767). This is the remaining channel-monitor complement of openclaw#93767 for the iMessage path; the optional prefix is a pure recognition enhancement with no change to the bare-tag, code-region exemption, or other reflection patterns. Verified by driving the real detectReflectedContent export: <mm:think>secret</mm:think>visible -> isReflection=true (was false) <think>...</think> -> isReflection=true <think>secret</think> -> isReflection=true (no regression) "thinking about minimax algorithms" -> isReflection=false (no false-positive) `<mm:think>x</mm:think>` in code -> isReflection=false (code exemption kept) * chore: retrigger CI for real behavior proof check * refactor(imessage): tighten reflection guard coverage --------- Co-authored-by: Peter Steinberger <[email protected]> (cherry picked from commit 076da56)

openclaw-barnacle Bot added channel: imessage Channel integration: imessage size: XS triage: needs-real-behavior-proof Candidate: external PR needs after-fix proof from a real setup. labels Jun 17, 2026

This was referenced Jun 17, 2026

fix(slack): recognize MiniMax mm: namespaced reasoning tags in monitor preview #93874

Merged

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

Closed

zhangguiping-xydt mentioned this pull request Jun 17, 2026

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

Closed

Alix-007 mentioned this pull request Jun 17, 2026

fix(control-ui): recognize namespaced reasoning tags in thinking extraction #94165

Closed

Alix-007 force-pushed the fix/minimax-mm-channel-monitor-guards branch from 2b7eba4 to a57ec19 Compare June 18, 2026 02:14

openclaw-barnacle Bot added the triage: needs-pr-context Candidate: external PR body lacks required problem context or evidence. label Jun 20, 2026

clawsweeper Bot mentioned this pull request Jun 20, 2026

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

Open

openclaw-barnacle Bot removed the triage: needs-pr-context Candidate: external PR body lacks required problem context or evidence. label Jun 21, 2026

clawsweeper Bot added status: 🔁 re-review loop A fresh ClawSweeper review was explicitly requested after the latest review. and removed status: 👀 ready for maintainer look ClawSweeper has no concrete contributor-facing blocker left for this PR. labels Jun 21, 2026

Pick-cat mentioned this pull request Jun 22, 2026

fix(imessage): detect namespaced thinking reflections #95690

Closed

clawsweeper Bot added status: 👀 ready for maintainer look ClawSweeper has no concrete contributor-facing blocker left for this PR. and removed status: 🔁 re-review loop A fresh ClawSweeper review was explicitly requested after the latest review. labels Jun 23, 2026

steipete self-assigned this Jun 30, 2026

steipete force-pushed the fix/minimax-mm-channel-monitor-guards branch 2 times, most recently from bde8976 to 08df499 Compare June 30, 2026 19:13

Alix-007 and others added 3 commits June 30, 2026 12:22

chore: retrigger CI for real behavior proof check

6bad92b

refactor(imessage): tighten reflection guard coverage

53eb0ed

steipete force-pushed the fix/minimax-mm-channel-monitor-guards branch from 08df499 to 53eb0ed Compare June 30, 2026 19:22

steipete merged commit 076da56 into openclaw:main Jun 30, 2026
140 of 141 checks passed

github-actions Bot mentioned this pull request Jun 30, 2026

📡 Upstream Digest — 2026-06-30 19:44 UTC curtismercier/openclaw-mods#1178

Open

sheyanmin mentioned this pull request Jul 3, 2026

fix(matrix): THINKING_TAG_RE missing mm: namespace prefix in monitor replies #99412

Open

krissding mentioned this pull request Jul 3, 2026

fix(matrix): add mm:/antml: namespace prefix to THINKING regex in monitor replies #99415

Closed

sheyanmin mentioned this pull request Jul 3, 2026

fix(matrix): add mm: namespace prefix to THINKING_TAG_RE in monitor replies #99417

Closed

9 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

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

fix(imessage): recognize MiniMax mm: reasoning tags in reflection guard (completes #93767)#93820
steipete merged 3 commits into
openclaw:mainfrom
Alix-007:fix/minimax-mm-channel-monitor-guards

Alix-007 commented Jun 17, 2026 •

edited

Loading

Uh oh!

clawsweeper Bot commented Jun 17, 2026 •

edited

Loading

Uh oh!

Alix-007 commented Jun 20, 2026

Uh oh!

clawsweeper Bot commented Jun 20, 2026 •

edited

Loading

Uh oh!

Alix-007 commented Jun 21, 2026

Uh oh!

Alix-007 commented Jun 23, 2026

Uh oh!

steipete commented Jun 30, 2026

Uh oh!

Uh oh!

steipete commented Jun 30, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Uh oh!

Conversation

Alix-007 commented Jun 17, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What Problem This Solves

Evidence

Summary

Changes

Real behavior proof

Real behavior proof

Uh oh!

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

Uh oh!

Uh oh!

Alix-007 commented Jun 20, 2026

Uh oh!

clawsweeper Bot commented Jun 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Alix-007 commented Jun 21, 2026

Uh oh!

Alix-007 commented Jun 23, 2026

Uh oh!

steipete commented Jun 30, 2026

Uh oh!

Uh oh!

steipete commented Jun 30, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Alix-007 commented Jun 17, 2026 •

edited

Loading

clawsweeper Bot commented Jun 17, 2026 •

edited

Loading

clawsweeper Bot commented Jun 20, 2026 •

edited

Loading