openclaw - 💡(How to fix) Fix Heartbeat / cron / exec wakeups submit empty user prompt to model providers in 4.25, causing 400 INVALID_ARGUMENT on Vertex/Gemini [1 comments, 2 participants]

islandpreneur007 · 2026-04-29T00:08:27Z

[openclaw] In 4.25, heartbeat/cron/exec wakeups are submitted as transient runtime context instead of visible user prompts per CHANGELOG 66496/ 66814 . However… In 4.25, heartbeat/cron/exec wakeups are submitted as transient runtime context instead of visible user prompts (per CHANGELOG #66496/#66814). However, the new code path in `selection-ABXC-aG3.js` `resolveRuntimeContextPromptParts()` returns `prompt: ""` when `transcriptPrompt` is empty and `effectivePrompt` is non-empty — submitting an empty user prompt to the model. Vertex/Gemini rejects this with `400 INVALID_ARGUMENT`. OpenRouter/xiaomi/mimo-v2-pro tolerates it (returns content normally). Reproduces consistently on every heartbeat tick in agents pinned to Vertex/Gemini. ## Fix / Workaround - Originally observed on OpenClaw `2026.4.25` (aa36ee6) - Status on 4.26 (be8c246): UNVERIFIED — source line refs below are from the 4.25 dist; 4.26 builds rename the `selection-*` file (different hash suffix), need to re-grep before filing - Vanilla runtime (no local patches) - Install: npm-global at `/home/openclawops/.npm-global/lib/node_modules/openclaw/openclaw.mjs` - Affected providers: Vertex/Gemini (rejects with 400 `Model input cannot be empty`) - Tolerated by: OpenRouter / xiaomi / mimo-v2-pro For agents pinned to Vertex/Gemini as their heartbeat model, every 5-min tick fails with the empty-input 400. We mitigated by swapping the heartbeat model away from Vertex/Gemini (currently OpenRouter/xiaomi/mimo-v2-pro, durable since 2026-04-27 22:11Z). Heartbeat empty-input storm stopped post-swap and has NOT reappeared on 4.26 — mitigation holds. ## Workaround (current) # Upstream bug report DRAFT — pending Trent blind-pass **Target:** github.com/openclaw/openclaw/issues **Status:** DRAFT v0 — not filed. Needs Trent review (RCA primarily lives in his workspace memory). --- ## Title Heartbeat / cron / exec wakeups submit empty user prompt to model providers in 4.25, causing 400 INVALID_ARGUMENT on Vertex/Gemini ## Environment - Originally observed on OpenClaw `2026.4.25` (aa36ee6) - Status on 4.26 (be8c246): UNVERIFIED — source line refs below are from the 4.25 dist; 4.26 builds rename the `selection-*` file (different hash suffix), need to re-grep before filing - Vanilla runtime (no local patches) - Install: npm-global at `/home/openclawops/.npm-global/lib/node_modules/openclaw/openclaw.mjs` - Affected providers: Vertex/Gemini (rejects with 400 `Model input cannot be empty`) - Tolerated by: OpenRouter / xiaomi / mimo-v2-pro ## Summary In 4.25, heartbeat/cron/exec wakeups are submitted as transient runtime context instead of visible user prompts (per CHANGELOG #66496/#66814). However, the new code path in `selection-ABXC-aG3.js` `resolveRuntimeContextPromptParts()` returns `prompt: ""` when `transcriptPrompt` is empty and `effectivePrompt` is non-empty — submitting an empty user prompt to the model. Vertex/Gemini rejects this with `400 INVALID_ARGUMENT`. OpenRouter/xiaomi/mimo-v2-pro tolerates it (returns content normally). Reproduces consistently on every heartbeat tick in agents pinned to Vertex/Gemini. ## Source chain ### 4.25 source (where bug was first traced) 1. `heartbeat-runner-CUi7jNCS.js` — builds non-empty `ctx.Body` 2. `get-reply-ltpHDntZ.js` line ~2261 — sets `transcriptBodyBase=""` for heartbeat 3. `runReplyAgent` — receives non-empty `commandBody` and empty `transcriptCommandBody` 4. Embedded path — passes `effectivePrompt=commandBody` and `transcriptPrompt=transcriptCommandBody` 5. `selection-ABXC-aG3.js` lines ~5590-5605 `resolveRuntimeContextPromptParts()` — converts that into `prompt:""` + `runtimeOnly:true` + populated `runtimeSystemContext` 6. `activeSession.prompt("")` — submits empty user input 7. Vertex/Gemini — rejects with `400 INVALID_ARGUMENT` / `Model input cannot be empty` ### 4.26 source (still present, file renamed) - `dist/runtime-context-prompt-BZWyPA2L.js:8-18` — `resolveRuntimeContextPromptParts`, `runtimeOnly: true` when transcript prompt is empty but runtime context exists - `dist/selection-D9uTvvsw.js:7426-7431` — runtime-context prompt parts resolved + runtime-only system context applied - `dist/selection-D9uTvvsw.js:7470-7477` — empty-prompt skip only applies when `!promptSubmission.runtimeOnly` - `dist/selection-D9uTvvsw.js:7600` — runtime-only path still submits `activeSession.prompt(promptSubmission.prompt)` even when prompt is `""` The 4.26 code path preserves the same empty-prompt submission behavior on the runtime-only path. Bug is unchanged across 4.25 → 4.26. ## Pre-condition for the bug `resolveRuntimeContextPromptParts` and `runtimeOnly` are NEW in 4.25. Verified zero hits in all 4.24 dist for both symbols. 4.24 has no `selection-*` file. Maps to CHANGELOG line 188 `#66496/#66814`: "submit heartbeat, cron, and exec wakeups as transient runtime context instead of visible user prompts." Other relevant CHANGELOG entries: - Line 512 `#69079/#69278` — stop injecting heartbeat system prompt into non-heartbeat runs

openclaw2026-04-29 00:08:27

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

openclaw/openclaw#73879•Fetched 2026-04-29 06:13:44

View on GitHub

Comments

Participants

Timeline

Reactions

Author

islandpreneur007

Participants

clawsweeper[bot]

islandpreneur007

Timeline (top)

closed ×1commented ×1

In 4.25, heartbeat/cron/exec wakeups are submitted as transient runtime context instead of visible user prompts (per CHANGELOG #66496/#66814). However, the new code path in selection-ABXC-aG3.js resolveRuntimeContextPromptParts() returns prompt: "" when transcriptPrompt is empty and effectivePrompt is non-empty — submitting an empty user prompt to the model. Vertex/Gemini rejects this with 400 INVALID_ARGUMENT. OpenRouter/xiaomi/mimo-v2-pro tolerates it (returns content normally).

Reproduces consistently on every heartbeat tick in agents pinned to Vertex/Gemini.

Root Cause

Reproduces consistently on every heartbeat tick in agents pinned to Vertex/Gemini.

Fix Action

Fix / Workaround

Originally observed on OpenClaw 2026.4.25 (aa36ee6)
Status on 4.26 (be8c246): UNVERIFIED — source line refs below are from the 4.25 dist; 4.26 builds rename the selection-* file (different hash suffix), need to re-grep before filing
Vanilla runtime (no local patches)
Install: npm-global at /home/openclawops/.npm-global/lib/node_modules/openclaw/openclaw.mjs
Affected providers: Vertex/Gemini (rejects with 400 Model input cannot be empty)
Tolerated by: OpenRouter / xiaomi / mimo-v2-pro

For agents pinned to Vertex/Gemini as their heartbeat model, every 5-min tick fails with the empty-input 400. We mitigated by swapping the heartbeat model away from Vertex/Gemini (currently OpenRouter/xiaomi/mimo-v2-pro, durable since 2026-04-27 22:11Z). Heartbeat empty-input storm stopped post-swap and has NOT reappeared on 4.26 — mitigation holds.

Workaround (current)

RAW_BUFFERClick to expand / collapse

Upstream bug report DRAFT — pending Trent blind-pass

Target: github.com/openclaw/openclaw/issues Status: DRAFT v0 — not filed. Needs Trent review (RCA primarily lives in his workspace memory).

Title

Heartbeat / cron / exec wakeups submit empty user prompt to model providers in 4.25, causing 400 INVALID_ARGUMENT on Vertex/Gemini

Environment

Originally observed on OpenClaw 2026.4.25 (aa36ee6)
Status on 4.26 (be8c246): UNVERIFIED — source line refs below are from the 4.25 dist; 4.26 builds rename the selection-* file (different hash suffix), need to re-grep before filing
Vanilla runtime (no local patches)
Install: npm-global at /home/openclawops/.npm-global/lib/node_modules/openclaw/openclaw.mjs
Affected providers: Vertex/Gemini (rejects with 400 Model input cannot be empty)
Tolerated by: OpenRouter / xiaomi / mimo-v2-pro

Summary

Reproduces consistently on every heartbeat tick in agents pinned to Vertex/Gemini.

Source chain

4.25 source (where bug was first traced)

heartbeat-runner-CUi7jNCS.js — builds non-empty ctx.Body
get-reply-ltpHDntZ.js line ~2261 — sets transcriptBodyBase="" for heartbeat
runReplyAgent — receives non-empty commandBody and empty transcriptCommandBody
Embedded path — passes effectivePrompt=commandBody and transcriptPrompt=transcriptCommandBody
selection-ABXC-aG3.js lines ~5590-5605 resolveRuntimeContextPromptParts() — converts that into prompt:"" + runtimeOnly:true + populated runtimeSystemContext
activeSession.prompt("") — submits empty user input
Vertex/Gemini — rejects with 400 INVALID_ARGUMENT / Model input cannot be empty

4.26 source (still present, file renamed)

dist/runtime-context-prompt-BZWyPA2L.js:8-18 — resolveRuntimeContextPromptParts, runtimeOnly: true when transcript prompt is empty but runtime context exists
dist/selection-D9uTvvsw.js:7426-7431 — runtime-context prompt parts resolved + runtime-only system context applied
dist/selection-D9uTvvsw.js:7470-7477 — empty-prompt skip only applies when !promptSubmission.runtimeOnly
dist/selection-D9uTvvsw.js:7600 — runtime-only path still submits activeSession.prompt(promptSubmission.prompt) even when prompt is ""

The 4.26 code path preserves the same empty-prompt submission behavior on the runtime-only path. Bug is unchanged across 4.25 → 4.26.

Pre-condition for the bug

resolveRuntimeContextPromptParts and runtimeOnly are NEW in 4.25. Verified zero hits in all 4.24 dist for both symbols. 4.24 has no selection-* file. Maps to CHANGELOG line 188 #66496/#66814: "submit heartbeat, cron, and exec wakeups as transient runtime context instead of visible user prompts."

Other relevant CHANGELOG entries:

Line 512 #69079/#69278 — stop injecting heartbeat system prompt into non-heartbeat runs
Line 1403 #66035
Line 1405 #66073/#63733/#35300
Line 1412 #66139

Suggested fix

When effectivePrompt is non-empty but transcriptPrompt is empty, the runtime should either:

Submit effectivePrompt as the user prompt (preserve historical behavior — visible prompt path), OR
Submit a documented sentinel (e.g. "<runtime-only>") that providers like Vertex/Gemini will accept, with the actual content carried via runtimeSystemContext

The current prompt: "" submission violates the implicit contract that user prompt is non-empty.

Production impact

Workaround (current)

Set heartbeat model to a provider that tolerates empty user prompts (e.g. OpenRouter/xiaomi/mimo-v2-pro). Vertex/Gemini cannot be used as heartbeat model on 4.25 or 4.26.

extent analysis

TL;DR

Modify the resolveRuntimeContextPromptParts function to submit a non-empty user prompt, such as the effectivePrompt or a documented sentinel, when transcriptPrompt is empty.

Guidance

Identify the resolveRuntimeContextPromptParts function in the selection-ABXC-aG3.js file (4.25) or dist/runtime-context-prompt-BZWyPA2L.js file (4.26) and modify it to handle the case where transcriptPrompt is empty.
Consider submitting the effectivePrompt as the user prompt to preserve historical behavior, or submit a documented sentinel (e.g. "<runtime-only>") that providers like Vertex/Gemini will accept.
Verify that the modified function correctly handles the case where transcriptPrompt is empty and effectivePrompt is non-empty.
Test the modified function with different input scenarios to ensure it behaves as expected.

Example

// Modified resolveRuntimeContextPromptParts function
function resolveRuntimeContextPromptParts(effectivePrompt, transcriptPrompt) {
  if (transcriptPrompt === "" && effectivePrompt!== "") {
    // Submit effectivePrompt as user prompt or a documented sentinel
    return { prompt: effectivePrompt, runtimeOnly: true };
  } else {
    // Existing logic
  }
}

Notes

The modification should be applied to the resolveRuntimeContextPromptParts function in the relevant file (4.25 or 4.26).
The exact implementation details may vary depending on the specific requirements and constraints of the system.

Recommendation

Apply the workaround by setting the heartbeat model to a provider that tolerates empty user prompts (e.g. OpenRouter/xiaomi/mimo-v2-pro) until the modified function is deployed and verified to work correctly.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#memory management #API rate limit #retriever error #indexing error #inference speed

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Data

Security

Network

Code

UI/UX

Text

System

Multimedia

Protocol

API

Engineering

openclaw - 💡(How to fix) Fix Heartbeat / cron / exec wakeups submit empty user prompt to model providers in 4.25, causing 400 INVALID_ARGUMENT on Vertex/Gemini [1 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fix / Workaround

Workaround (current)

Upstream bug report DRAFT — pending Trent blind-pass

Title

Environment

Summary

Source chain

4.25 source (where bug was first traced)

4.26 source (still present, file renamed)

Pre-condition for the bug

Suggested fix

Production impact

Workaround (current)

extent analysis

TL;DR

Guidance

Example

Notes

Recommendation

Still need to ship something?

TRENDING

openclaw - 💡(How to fix) Fix Heartbeat / cron / exec wakeups submit empty user prompt to model providers in 4.25, causing 400 INVALID_ARGUMENT on Vertex/Gemini [1 comments, 2 participants]

Recommended Tools

GitHub issue graph ai analysis

Root Cause

Fix Action

Fix / Workaround

Workaround (current)

Upstream bug report DRAFT — pending Trent blind-pass

Title

Environment

Summary

Source chain

4.25 source (where bug was first traced)

4.26 source (still present, file renamed)

Pre-condition for the bug

Suggested fix

Production impact

Workaround (current)

extent analysis

TL;DR

Guidance

Example

Notes

Recommendation

Still need to ship something?

RELATED_DISCOVERY

TRENDING