fix: prevent HEARTBEAT_OK discard from swallowing steered user responses by Jo2234 · Pull Request #30449 · openclaw/openclaw

Jo2234 · 2026-03-01T08:19:10Z

Summary

Fix a bug where user messages sent during an active heartbeat run get silently swallowed when queue.mode is set to steer.

Problem

When steer mode injects a user message into a heartbeat run, the agent produces a combined response like:

HEARTBEAT_OK. Sure! Here is the answer to your question. Let me know if you need more details.

The current stripHeartbeatToken() logic strips HEARTBEAT_OK and then checks if the remaining text is ≤ maxAckChars (default: 300). Since many user responses are under 300 characters, the entire response is treated as a brief heartbeat ack and suppressed — the user never receives their answer.

Before (broken)

User sends message during heartbeat → Agent responds with HEARTBEAT_OK + answer
→ stripHeartbeatToken strips token, remaining text ≤ 300 chars
→ shouldSkip = true → delivery suppressed → user gets nothing

After (fixed)

User sends message during heartbeat → Agent responds with HEARTBEAT_OK + answer
→ stripHeartbeatToken strips token, remaining text ≤ 300 chars
→ BUT remaining text has multi-sentence structure → recognized as substantive response
→ shouldSkip = false → delivery proceeds → user gets their answer

Root Cause

In stripHeartbeatToken() (src/auto-reply/heartbeat.ts), the maxAckChars check does not distinguish between:

A brief heartbeat ack: "Nothing to report" (single fragment, no sentence structure)
A real user response: "Sure! Here is the answer. Let me know if you need more." (multi-sentence, addresses user)

Fix

Added a heuristic in the maxAckChars check: after stripping HEARTBEAT_OK, if the remaining text contains multi-sentence structure (sentence-ending punctuation [.!?] followed by more content), it is recognized as a substantive response and NOT suppressed.

Backward Compatibility

Input	Before	After
`HEARTBEAT_OK`	suppressed	suppressed
`HEARTBEAT_OK 🦞`	suppressed	suppressed
`HEARTBEAT_OK Nothing to report`	suppressed	suppressed
`HEARTBEAT_OK Sure! Here is the answer.`	suppressed	delivered
`HEARTBEAT_OK The server is fine. Deploy at 3:42.`	suppressed	delivered

Files Changed

File	Description
`src/auto-reply/heartbeat.ts`	Added multi-sentence heuristic in `stripHeartbeatToken()`
`src/auto-reply/heartbeat.test.ts`	3 new test cases for steered response scenarios

Test Cases Added

Multi-sentence steered response → preserved (not suppressed)
Single-fragment ack → still correctly suppressed
Question/answer pattern → preserved

Fixes #30197

When queue mode is set to 'steer', a user message that arrives during an active heartbeat run is injected into that run. The agent may respond with 'HEARTBEAT_OK' followed by a real answer to the user's question. Previously, stripHeartbeatToken() stripped the HEARTBEAT_OK token and then checked if the remaining text was <= maxAckChars (default 300). Since many user responses are under 300 characters, the entire response was treated as a brief heartbeat ack and suppressed — the user never received their answer. This commit adds a heuristic check: after stripping HEARTBEAT_OK, if the remaining text contains multi-sentence structure (sentence-ending punctuation followed by more content), it is recognized as a substantive response rather than a brief ack, and delivery proceeds normally. The heuristic preserves backward compatibility: - Pure HEARTBEAT_OK → still suppressed - HEARTBEAT_OK with emoji/short junk → still suppressed - HEARTBEAT_OK with single-fragment ack → still suppressed - HEARTBEAT_OK followed by multi-sentence response → NOW DELIVERED Added 3 test cases covering: - Multi-sentence steered response after HEARTBEAT_OK (preserved) - Single-fragment ack after HEARTBEAT_OK (still suppressed) - Question/answer pattern response (preserved) Fixes openclaw#30197

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9263b43c2f

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 · 2026-03-01T08:21:20Z

+      const hasMultipleSentences = /[.!?]\s+\S/.test(rest);
+      const hasSubstantiveContent = rest.length > 0 && hasMultipleSentences;


Exclude token-leading punctuation from substantive check

The new hasMultipleSentences heuristic treats punctuation left behind by start-token stripping as evidence of a multi-sentence answer, so heartbeat acks like HEARTBEAT_OK. Nothing to report are now delivered instead of suppressed. In stripTokenAtEdges(), stripping a leading token keeps the ./! that follows the token, and this line then matches /[.!?]\s+\S/, setting shouldSkip to false for short non-actionable heartbeat text. This will cause regular heartbeat chatter to leak to users whenever the model formats acknowledgements as HEARTBEAT_OK. plus a short sentence.

Useful? React with 👍 / 👎.

greptile-apps · 2026-03-01T08:25:49Z

Greptile Summary

Fixes bug where user messages sent during active heartbeat runs get silently swallowed in steer mode. Added multi-sentence detection heuristic to distinguish between brief heartbeat acknowledgments and substantive combined responses.

Changes:

Added regex-based check (/[.!?]\s+\S/) to detect multi-sentence structure in stripHeartbeatToken()
Responses with multi-sentence patterns are now preserved instead of being suppressed
Added 3 new test cases covering multi-sentence responses, single-fragment acks, and question/answer patterns

Behavior:

Brief acks like "Nothing to report" remain suppressed ✓
Multi-sentence responses like "Sure! Here is the answer. Let me know if you need more." are now delivered ✓
Maintains backward compatibility for existing heartbeat behavior

Confidence Score: 4/5

Safe to merge with minimal risk - addresses the reported issue effectively
Well-tested fix that solves the main problem (multi-sentence responses being swallowed). The heuristic could be more comprehensive to handle single-sentence responses, but the current implementation is solid for the described use case. Minor comment/code discrepancy noted but doesn't affect functionality.
No files require special attention

_{Last reviewed commit: 9263b43}

greptile-apps

_{2 files reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}

greptile-apps · 2026-03-01T08:25:53Z

+      // patterns, treat it as a real response and do NOT skip.
+      const hasMultipleSentences = /[.!?]\s+\S/.test(rest);
+      const hasSubstantiveContent = rest.length > 0 && hasMultipleSentences;


comment mentions "or starts with common response patterns" but only multi-sentence detection is implemented - single-sentence responses like "The answer is 42." would still be suppressed

Prompt To Fix With AI

This is a comment left during a code review. Path: src/auto-reply/heartbeat.ts Line: 174-176 Comment: comment mentions "or starts with common response patterns" but only multi-sentence detection is implemented - single-sentence responses like "The answer is 42." would still be suppressed How can I resolve this? If you propose a fix, please make it concise.

…user responses When 'steer' mode injects a user message into a heartbeat run, the agent produces a combined response like 'HEARTBEAT_OK Sure! Here is the answer.' The previous logic stripped HEARTBEAT_OK and then checked if the remaining text was ≤ maxAckChars (300). Since many user responses are under 300 chars, the entire response was treated as a brief heartbeat ack and suppressed. This fix adds a heuristic: after stripping HEARTBEAT_OK, if the remaining text contains multi-sentence structure (sentence-ending punctuation [.!?] followed by more content), it is recognized as a substantive response and NOT suppressed. Fixes openclaw#30449

openclaw-barnacle · 2026-04-05T04:27:50Z

This pull request has been automatically marked as stale due to inactivity.
Please add updates or it will be closed.

openclaw-barnacle · 2026-04-10T04:22:56Z

Closing due to inactivity.
If you believe this PR should be revived, post in #pr-thunderdome-dangerzone on Discord to talk to a maintainer.
That channel is the escape hatch for high-quality PRs that get auto-closed.

openclaw-barnacle Bot added the size: S label Mar 1, 2026

chatgpt-codex-connector Bot reviewed Mar 1, 2026

View reviewed changes

greptile-apps Bot reviewed Mar 1, 2026

View reviewed changes

openclaw-barnacle Bot added the stale Marked as stale due to inactivity label Apr 5, 2026

vincentkoc mentioned this pull request Apr 5, 2026

fix: deliver steered user replies swallowed by HEARTBEAT_OK discard #30304

Closed

15 tasks

openclaw-barnacle Bot closed this Apr 10, 2026

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix: prevent HEARTBEAT_OK discard from swallowing steered user responses#30449

fix: prevent HEARTBEAT_OK discard from swallowing steered user responses#30449
Jo2234 wants to merge 1 commit into
openclaw:mainfrom
Jo2234:fix/heartbeat-steer-response-swallowed

Jo2234 commented Mar 1, 2026

Uh oh!

chatgpt-codex-connector Bot left a comment

Uh oh!

chatgpt-codex-connector Bot Mar 1, 2026

Uh oh!

greptile-apps Bot commented Mar 1, 2026

Uh oh!

greptile-apps Bot left a comment

Uh oh!

greptile-apps Bot Mar 1, 2026

Uh oh!

openclaw-barnacle Bot commented Apr 5, 2026

Uh oh!

openclaw-barnacle Bot commented Apr 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

		const hasMultipleSentences = /[.!?]\s+\S/.test(rest);
		const hasSubstantiveContent = rest.length > 0 && hasMultipleSentences;

Uh oh!

Conversation

Jo2234 commented Mar 1, 2026

Summary

Problem

Before (broken)

After (fixed)

Root Cause

Fix

Backward Compatibility

Files Changed

Test Cases Added

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 1, 2026

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot commented Mar 1, 2026

Greptile Summary

Confidence Score: 4/5

Uh oh!

greptile-apps Bot left a comment

Choose a reason for hiding this comment

Uh oh!

greptile-apps Bot Mar 1, 2026

Choose a reason for hiding this comment

Uh oh!

openclaw-barnacle Bot commented Apr 5, 2026

Uh oh!

openclaw-barnacle Bot commented Apr 10, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant