iframe-proxy

amelidev · 2026-10-19T15:15:12Z

Summary

Fixes an issue where multi-turn agent conversations fail with HTTP 400 Bad Request (Function call is missing a thought_signature in functionCall parts) when historical model tool calls precede subsequent user prompts.

Details

Root Cause

In GeminiChat.ensureActiveLoopHasThoughtSignatures, thought signature enforcement was only applied to turns beginning at activeLoopStartIndex (the most recent user text prompt). Any historical model turn containing a function call prior to activeLoopStartIndex (such as a tool invoked earlier in a coding session, e.g. at position 58) was forwarded to the Gemini API without a thoughtSignature.

The Gemini API requires that every model turn containing function calls in the contents history payload must have a thought_signature on its first functionCall part. Consequently, once a session progressed past the initial tool turn, the Gemini API rejected the subsequent request with HTTP 400 INVALID_ARGUMENT.

Fix

Updated ensureActiveLoopHasThoughtSignatures in packages/core/src/core/geminiChat.ts to iterate across all model turns in requestContents rather than restricting to activeLoopStartIndex.
Guaranteed that the first functionCall part of every model turn in history has a thoughtSignature (SYNTHETIC_THOUGHT_SIGNATURE if not already present).
Preserved existing valid signatures and left parallel function calls (after the first in a turn) unsigned per Gemini API specifications.
Updated and added regression test coverage in packages/core/src/core/geminiChat.test.ts.

How to Validate

Run the targeted unit and regression tests in @google/gemini-cli-core:
```
npm test -w @google/gemini-cli-core -- src/core/geminiChat.test.ts
```
Expected result: All 134 tests pass.
Run a2a-server test suite:
```
npm test -w @google/gemini-cli-a2a-server
```
Expected result: All 16 test files pass (177 tests).
Run workspace typecheck and linter:
```
npm run typecheck
npm run lint
```
Expected result: 0 errors, 0 warnings.
Manual verification across 8 cases:
- Case 1: Standard Active Loop Tool Call — Verified first function call is signed.
- Case 2: Multi-Turn Historical Tool Call — Verified historical Turn 1 and active Turn 5 both signed.
- Case 3: Parallel Tool Calls — Verified first function call signed, subsequent calls unsigned.
- Case 4: Existing Real Signature Preservation — Verified real encrypted signatures are preserved.
- Case 5: History Without User Text — Verified model function call is signed.
- Case 6: Deep Turn 58 Reproduction — Verified position 58 function call is signed.
- Case 7: Text-Only and Empty Turns — Verified no spurious fields added.
- Case 8: User Turns — Verified user parts remain unaffected.

Pre-Merge Checklist

… signatures

gemini-code-assist · 2026-10-19T15:15:25Z

Summary of Changes

Hello, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request addresses an issue where multi-turn conversations with the Gemini API would fail with a 400 Bad Request error. The root cause was that historical model turns containing function calls were being sent without the required thought_signature. The changes ensure that all model turns across the entire conversation history are properly signed, maintaining API compliance throughout the session.

Highlights

Global Thought Signature Enforcement: Updated ensureActiveLoopHasThoughtSignatures to apply thought signatures to all historical model turns, rather than limiting enforcement to the current active loop.
API Compatibility: Ensures that every model turn containing function calls includes a thought_signature to prevent HTTP 400 Bad Request errors from the Gemini API.
Regression Testing: Added comprehensive unit and integration tests to verify that historical tool calls are correctly signed and that signatures are preserved.

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize the Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counterproductive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for GitHub and other Google products, sign up here.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

github-actions · 2026-10-19T15:15:29Z

📊 PR Size: size/M

Lines changed: 189
Additions: +147
Deletions: -42
Files changed: 2

gemini-code-assist

Code Review

This pull request updates ensureActiveLoopHasThoughtSignatures in GeminiChat to apply thoughtSignature (using SYNTHETIC_THOUGHT_SIGNATURE) to the first function call of every model turn across the entire conversation history, rather than limiting it to the active loop. This prevents 400 Bad Request errors from the Gemini API, which validates all historical function calls. Corresponding unit tests and reproduction test cases have been updated and added to verify this behavior. I have no further feedback to provide as the changes are clean and well-tested.

amelidev · 2026-10-19T15:27:45Z

/gemini review

gemini-code-assist

Code Review

This pull request updates ensureActiveLoopHasThoughtSignatures in geminiChat.ts to ensure that the first function call in every model turn across the entire conversation history has a thoughtSignature property, resolving potential 400 Bad Request errors from the Gemini API. It also adds comprehensive unit and reproduction tests to verify this behavior. The review feedback recommends defining a local interface for the undocumented thoughtSignature property and adding a detailed comment explaining the rationale for using this internal SDK property to maintain type safety and document the workaround.

…ure property

amelidev · 2026-10-19T18:48:46Z

/gemini review

gemini-code-assist

Code Review

This pull request updates the ensureActiveLoopHasThoughtSignatures method in GeminiChat to apply a synthetic thoughtSignature to the first function call of all model turns across the entire conversation history, rather than restricting it to the active loop, preventing 400 Bad Request errors from the Gemini API. Corresponding unit and reproduction tests have been updated and added. The review feedback correctly identifies a performance optimization opportunity to avoid redundant array and object allocations by lazily copying the history and parts arrays only when modifications are actually required.

…tures

amelidev · 2026-10-19T20:26:05Z

/gemini review

gemini-code-assist

Code Review

This pull request updates ensureActiveLoopHasThoughtSignatures in geminiChat.ts to ensure that the first function call in every model turn across the entire conversation history has a thoughtSignature property, preventing 400 Bad Request errors from the Gemini API. It also optimizes the method to lazily clone arrays only when modifications are needed, and adds comprehensive unit tests to verify this behavior. The feedback suggests defensively checking if part is defined before accessing its properties to prevent potential runtime TypeError crashes.

…reActiveLoopHasThoughtSignatures

amelidev · 2026-10-19T20:47:48Z

gemini-code-assist

Code Review

This pull request updates the ensureActiveLoopHasThoughtSignatures method in GeminiChat to apply synthetic thought signatures to the first function call of every model turn across the entire conversation history, rather than limiting it to the active loop. This change prevents HTTP 400 Bad Request errors from the Gemini API, which validates all historical function calls. Additionally, the implementation has been optimized to lazily clone the history array only when modifications are necessary, returning the original reference otherwise. Comprehensive unit tests and reproduction cases have been added to verify this behavior. There are no review comments to address, and the changes look solid.

fix(core): ensure all model turns with function calls receive thought…

2015a0a

… signatures

amelidev requested a review from a team as a code owner October 19, 2026 15:15

github-actions Bot added the size/m A medium sized PR label Oct 19, 2026