Codex review: needs real behavior proof before merge.

Workflow note: Future ClawSweeper reviews update this same comment in place.

Summary
This PR suppresses late text-only tool/progress output after final reply delivery across shared dispatch, embedded runner, Codex projection, Telegram, and Discord while reserving raw tool-error detail for verbose full and updating docs/tests.

Reproducibility: yes. at source level: current main can still forward failed text-only tool output and terminal verbose error detail after final delivery, and the PR adds regression tests for those paths. I did not run a fresh live Telegram or Discord current-main repro in this read-only review.

PR rating
Overall: 🦐 gold shrimp
Proof: 🦐 gold shrimp
Patch quality: 🐚 platinum hermit
Summary: Patch quality looks solid from source review, but current-head real behavior proof is still incomplete for a visible Telegram/Discord delivery change.

Rank-up moves:

• Add a redacted current-head Telegram screenshot, recording, or live-output transcript showing the failed-tool repro no longer posts raw output after the final answer.
• Add Discord live proof if maintainers want parity beyond the focused Discord regression tests.

PR egg
🎁 Pass real behavior proof to wake the egg and unlock a hatchable treat.

Real behavior proof
Needs stronger real behavior proof before merge: The PR has logs, tests, deployment output, and an older Telegram screenshot, but the body says no fresh post-reload live Telegram or Discord repro was run on the final corrected head; add redacted current-head proof and update the PR body for re-review.

Mantis proof suggestion
A native Telegram recording would directly prove the current-head post-final transcript behavior that remains untested after the last correction. A maintainer can ask Mantis to capture proof by posting a new PR comment that starts with the OpenClaw Mantis account mention, followed by:

telegram desktop proof: verify on current PR head that a failed exec in regular verbose shows compact progress/final acknowledgement and no raw or duplicate failure warning appears after the final Telegram answer.

Risk before merge
Why this matters: - The latest head changed after the earlier screenshot proof, and the PR body explicitly says no fresh post-reload Telegram or Discord repro turn was run after the corrected reload.

• Regular /verbose on users lose raw failed-tool detail unless they switch to /verbose full, which is an intentional but compatibility-visible diagnostics change.
• The suppression is shared across dispatch plus Telegram and Discord timing, so maintainers should verify it does not hide legitimate late text-only progress in real transports.

Maintainer options:

1. Require current-head Telegram proof (recommended)
   Ask for a redacted live Telegram run on 42fccfb3eb550e80cc46951dfc41b36912163ce3 showing compact failed-tool progress, the final acknowledgement, and no post-final raw warning.
2. Accept the diagnostics compatibility change
   Maintainers can land with tests plus the documented /verbose full escape hatch if they intentionally accept that regular verbose no longer exposes raw failed-tool detail.
3. Pause for compatibility design
   If regular verbose must keep raw failed-tool diagnostics for existing operators, pause this PR and require a compatibility-preserving option or narrower suppression rule.

Next step before merge
A maintainer explicitly stopped automerge and left this for human review; the remaining blockers are proof and compatibility acceptance rather than a narrow repair.

Security
Cleared: The diff does not touch dependencies, CI, credentials, auth, permissions, release scripts, or other supply-chain surfaces.

Production/live proof update:

• Head SHA: 7b76ed87e0eb4d72a13b60921b2cda30e684dda7
• The PR is now ready for review.
• We are running the fix live in production, and the late raw failed-tool output has not reoccurred since.
• Telegram DM after-fix proof screenshot:

Focused validation rerun after the review fixes:

pnpm exec oxfmt --check --threads=1 <13 touched files>
All matched files use the correct format. Finished in 41ms on 13 files using 1 threads.

git diff --check HEAD~2..HEAD
(no output)

node scripts/run-vitest.mjs src/auto-reply/reply/dispatch-from-config.test.ts src/auto-reply/reply/agent-runner-execution.test.ts src/agents/pi-embedded-subscribe.subscribe-embedded-pi-session.suppresses-message-end-block-replies-message-tool.test.ts extensions/discord/src/monitor/message-handler.process.test.ts extensions/telegram/src/bot-message-dispatch.test.ts
Test Files 6 passed (6); Tests 400 passed (400); Duration 10.95s

Codex review was rerun with codex -s danger-full-access -a never --disable plugins review --commit HEAD; its three actionable P2 findings were accepted and fixed in the current head.

@clawsweeper automerge

🦞🔧
ClawSweeper automerge is enabled.

• Head: 0f97dc43635c
• Label: clawsweeper:automerge
• Action: repair worker queued. Run: https://github.com/openclaw/clawsweeper/actions/runs/26118611898
• Flow: review this head, repair/rebase only if needed, then re-review the exact repaired head before merge.

Draft PRs stay fix-only until GitHub marks them ready for review. Pause with /clawsweeper stop.

Automerge progress:

• 2026-05-19 18:55:24 UTC review queued 0f97dc43635c (queued)

• 2026-05-19 18:57:24 UTC repair queued 0f97dc43635c (autonomous) Run: https://github.com/openclaw/clawsweeper/actions/runs/26118611898

@clawsweeper stop

🦞✅
Got it. ClawSweeper will leave this item for human review.

I added clawsweeper:human-review, removed clawsweeper:automerge, and paused the automation trail until a maintainer asks again.

Follow-up proof for head dbe415b907821e97b280018b259068ec6919cd0e:

Root Cause

Before/RCA proof:

• The prior patch was loaded in the live gateway: the running gateway process was using built dist/index.js, and both source/dist contained the previous suppression markers for isError tool results and regular-vs-full verbose gating.
• The repro was therefore not stale code. The reported Telegram topic was on regular verbose (verboseDefault: "on", no per-session verboseLevel: "full").
• The latest live repro produced a failed bash/cat tool result, then the final assistant text acknowledged it as exit 1, but the session still emitted another post-final failed-tool warning.
• The missed path was src/agents/pi-embedded-runner/run/payloads.ts: final payload assembly still appends a synthetic mutating-tool warning unless it detects that the assistant already acknowledged the failed action. That detector handled phrases like “couldn't run the command,” but did not handle concise terminal-status acknowledgements like exit 1.

Real behavior proof

Behavior addressed: In regular verbose mode, when the final assistant answer already acknowledges an exec/bash failed exit status, OpenClaw should not append a second post-final failed-tool warning. Full verbose remains the raw/detail mode.

Real environment tested: Local tmp PR worktree on branch fix-channel-late-tool-output, head dbe415b907821e97b280018b259068ec6919cd0e, plus inspection of the live gateway/session that reproduced after the prior patch was loaded.

Exact steps or command run after this patch:

• pnpm exec oxfmt --check --threads=1 src/agents/pi-embedded-runner/run/payloads.ts src/agents/pi-embedded-runner/run/payloads.errors.test.ts
• git diff --check
• node scripts/run-vitest.mjs src/agents/pi-embedded-runner/run/payloads.errors.test.ts src/auto-reply/reply/dispatch-from-config.test.ts extensions/codex/src/app-server/event-projector.test.ts

Evidence after fix:

• oxfmt: All matched files use the correct format. Finished in 7ms on 2 files using 1 threads.
• git diff --check: no output.
• Vitest: Test Files 3 passed (3); Tests 235 passed (235); Duration 8.15s.
• New regression coverage proves the live-style final text Done — intentional missing-file check, exit \1`.suppresses the duplicatebash` warning in regular verbose mode.
• New negative coverage proves unrelated failure text such as The tests failed, so I stopped there. still surfaces write and bash warnings instead of hiding real tool failures.

Observed result after fix: The payload builder now returns only the final assistant payload for the live-style exit 1 acknowledgement, while preserving compact warnings for unrelated/vague failure text.

What was not tested: A fresh live Telegram topic send after dbe415b907821e97b280018b259068ec6919cd0e has not been run yet; I am running oc-update --skip-codex next so the live gateway can be tested against this exact head.

Deployment/update proof for head dbe415b907821e97b280018b259068ec6919cd0e:

• PR is open/ready for review, not draft.
• oc-update --skip-codex applied PR fix(channels): suppress late raw tool output #84178 on top of current origin/main, installed dependencies, rebuilt core/UI, reinstalled the daemon, and restarted the gateway.
• Live gateway is now active since 2026-05-19 15:38:27 EDT with /home/vac/openclaw/dist/index.js gateway --port 18789.
• Loaded-code proof: both source and built dist contain EXEC_FAILURE_TERMINAL_STATUS_PATTERN and the hasExplicitMutatingToolFailureAcknowledgement(cleanedText, params.lastToolError?.toolName) call.
• We are running the fix live in production. Since this restart/update, I have not observed the post-final failed-tool dump recur.
• Known unrelated update output: openclaw security audit still reports pre-existing local security/trust-model warnings, including reddit-modmail; those were not introduced by this PR.

Correction pushed and live.

What changed in this update:

• Removed dbe415b907 via acc513bdcb because the Pi-named payload regex was the wrong fix shape for this repro.
• Added 42fccfb3eb: dispatch-from-config now passes suppressToolErrorWarnings only when regular verbose compact progress is visible, while preserving verbose full.

RCA proof:

• The post-regex live repro was a Codex app-server run with a failed native bash/grep tool result, followed by a successful final answer, then an extra outbound message. That points to the shared terminal lastToolError warning fallback, not a Telegram adapter-only dump and not a missing exit 1 acknowledgement string.
• Source proof: Codex projector records lastToolError; agent-runner-execution passes suppressToolErrorWarnings; run.ts passes it to buildEmbeddedRunPayloads; payloads append the terminal warning after final if not suppressed.

Fix proof:

• pnpm exec oxfmt --check --threads=1 src/auto-reply/reply/dispatch-from-config.ts src/auto-reply/reply/dispatch-from-config.test.ts passed.
• git diff --check passed with no output.
• node scripts/run-vitest.mjs src/auto-reply/reply/dispatch-from-config.test.ts passed: 1 file, 132 tests.
• node scripts/run-vitest.mjs src/auto-reply/reply/dispatch-from-config.test.ts src/auto-reply/reply/agent-runner-execution.test.ts extensions/codex/src/app-server/event-projector.test.ts passed: 3 files, 295 tests.
• No codex-review rerun was performed per maintainer instruction.

Live production proof:

• oc-update --skip-codex applied PR fix(channels): suppress late raw tool output #84178, built core/UI, reinstalled/restarted the gateway, and health check completed.
• openclaw-gateway.service is active since Tue 2026-05-19 15:55:51 EDT, running /home/vac/openclaw/dist/index.js gateway --port 18789.
• Built artifact contains the corrected dispatch logic: dist/dispatch-8nUbnUg_.js has the regular-verbose suppressToolErrorWarnings handoff.
• openclaw health after reload: gateway event loop ok; Telegram and Discord configured.
• No post-reload recurrence observed at comment time; the latest reported-topic transcript timestamp I found was 2026-05-19 15:53:21 EDT, before this corrected reload.

Security audit warnings from oc-update are pre-existing local config/skill warnings, unrelated to this PR.

Closing this overloaded proof thread as superseded. I am opening a replacement PR from the same head/branch with the same code, full updated RCA, and the converted before/after PNG proof; I will link it back here once GitHub creates it.

Replacement PR: #84303

It uses the same branch/head with no code changes, and carries the cleaned full RCA plus the before/after PNG proof.

@clawsweeper hatch

🦞👀
ClawSweeper could not hatch this PR egg yet.

Reason: hatch requires an open pull request.