Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

Sustaining chain of thought after tool call hangs up randomly, due to missing thought_signature in the api call succeeding the tool output

Steps to reproduce

Not fully deterministic as this is not a hard enforcing from google point of view, docs: Though Signature

1. Open 2026.4.26 opneclaw
2. Setup Gemini API key (AI studio) and use thinking models (order of this commonly occuring 3.1 Pro > 2.5 Pro > Flash)
3. Run some multistep tool usage prompt (asking the main agents to spawn subagents for the task has >70% failure rate and the subagents simply not respond back due to this silent failure). My usage is mostly in physics research.

Expected behavior

Sustain chain of though agentic workflows smoothly.

Actual behavior

Random silent failure in chain of though agentic workflows.

OpenClaw version

2026.4.26

Operating system

Ubuntu 24.04.4 LTS

Install method

npm [email protected] (later update to 2026.4.26)

Model

Gemini

Provider / routing chain

openclaw -> Google API

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Google Generative AI API error (400): Function call is missing a thought_signature in functionCall parts. This is required for tools to work correctly, and missing thought_signature may lead to degraded model performance. Additional data, function call `def… [code=INVALID_ARGUMENT]

Although this sounds like its optional - its not, its a full fledge no response error.

Impact and severity

Affected: Gemini API Models
Severity: Blocks any agentic workflows
Frequency: intermittent
Consequence: No response / complete stall

Additional information

Fix suggestion:

Following the Vertex AI docs

• The fix should be as simple as storing the thought_signature of the output before the tool using and attaching it to the API call after the tool usage (that contain the tool outputs).

Thank you!