fix(translator): stabilize tool_call finish_reason #865

wxyzh · 2026-01-05T07:53:35Z

This PR stabilizes finish_reason handling in streaming responses, particularly for tool call scenarios.

Issues Fixed:

Tool call finish_reason being overridden by subsequent STOP chunks
Missing finish_reason in streaming responses causing client confusion
Unfiltered upstream finish_reason values passed through to clients
Changes
Commit 1: Fix stop_reason mapping in streaming responses
Author: @HsnSaboor
Fix Codex → Claude to prioritize upstream stop_reason
Fix Gemini/Gemini CLI → Claude to handle MAX_TOKENS correctly
Commit 2: Stabilize streaming finish_reason mapping
Delay finish_reason setting to prevent STOP chunk override in tool call flows
Simplify switch statements in Antigravity → Claude translator
Add FunctionIndex == 0 condition to control finish_reason emission
Commit 3: Add finish_reason whitelist filter
Whitelist filter for OpenAI format: stop/STOP/max_tokens/tool_calls
Whitelist filter for Claude format: end_turn/max_tokens/stop/tool_use
Prevent invalid upstream values from breaking client behavior
Affected Translators
OpenAI Format:

Gemini → OpenAI
Gemini CLI → OpenAI
Antigravity → OpenAI
Claude Format:

Codex → Claude
Gemini → Claude
Gemini CLI → Claude
Antigravity → Claude

gemini-code-assist · 2026-01-05T07:53:56Z

Summary of Changes

Hello @wxyzh, 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 focuses on enhancing the robustness and accuracy of finish_reason and stop_reason handling across various API translators. The changes ensure that streaming responses consistently provide correct termination reasons, especially in tool call contexts, by preventing overrides, filling in missing values, and filtering out invalid ones. This leads to more predictable and reliable client interactions with the translated API responses.

Highlights

Stabilized Finish Reason Handling: Addressed issues where finish_reason was being incorrectly overridden by subsequent STOP chunks in streaming responses, particularly for tool call scenarios.
Resolved Missing Finish Reasons: Fixed instances where finish_reason was missing in streaming responses, which previously caused confusion for client applications.
Implemented Finish Reason Whitelisting: Introduced whitelist filters for finish_reason values to prevent invalid upstream values from being passed through to clients, ensuring consistent and expected behavior across different API formats (OpenAI and Claude).
Improved Stop Reason Mapping: Enhanced the mapping logic for stop_reason in streaming responses, prioritizing upstream values and correctly handling MAX_TOKENS for various translators (Codex, Gemini, Gemini CLI, Antigravity).
Simplified Translator Logic: Refactored and simplified switch statements in several translators (e.g., Antigravity to Claude) and delayed finish_reason setting to prevent premature overrides.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

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 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 counter productive. 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.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

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. ↩

gemini-code-assist

Code Review

This pull request introduces several improvements to stabilize finish_reason handling in streaming responses, particularly for tool call scenarios across various translators. The changes correctly prioritize response.stop_reason, delay setting the finish_reason to prevent it from being overridden by STOP chunks in tool call flows, and introduce whitelisting to filter finish_reason values for both OpenAI and Claude formats. Additionally, some logic has been simplified by replacing switch statements with if conditions, which enhances code readability.

The implementation is solid, but I've identified a recurring logical error in the OpenAI translators where a check for an uppercase "STOP" is performed on a variable that has already been converted to lowercase. I've left comments with suggestions to fix this. Overall, these changes are a good step towards making the translators more robust.

gemini-code-assist · 2026-01-05T07:55:34Z

internal/translator/antigravity/openai/chat-completions/antigravity_openai_response.go

 		template, _ = sjson.Set(template, "choices.0.native_finish_reason", "tool_calls")
+	} else if finishReason != "" && (*param).(*convertCliResponseToOpenAIChatParams).FunctionIndex == 0 {
+		// Only pass through specific finish reasons
+		if finishReason == "max_tokens" || finishReason == "STOP" || finishReason == "stop" {


The finishReason variable is converted to lowercase on line 92. Therefore, the condition finishReason == "STOP" will always evaluate to false and is redundant.

Suggested change

if finishReason == "max_tokens" || finishReason == "STOP" || finishReason == "stop" {

if finishReason == "max_tokens" || finishReason == "stop" {

gemini-code-assist · 2026-01-05T07:55:34Z

internal/translator/gemini-cli/openai/chat-completions/gemini-cli_openai_response.go

 		template, _ = sjson.Set(template, "choices.0.native_finish_reason", "tool_calls")
+	} else if finishReason != "" && (*param).(*convertCliResponseToOpenAIChatParams).FunctionIndex == 0 {
+		// Only pass through specific finish reasons
+		if finishReason == "max_tokens" || finishReason == "STOP" || finishReason == "stop" {


The finishReason variable is converted to lowercase on line 90. Therefore, the condition finishReason == "STOP" will always evaluate to false and is redundant.

Suggested change

if finishReason == "max_tokens" || finishReason == "STOP" || finishReason == "stop" {

if finishReason == "max_tokens" || finishReason == "stop" {

gemini-code-assist · 2026-01-05T07:55:34Z

internal/translator/gemini/openai/chat-completions/gemini_openai_response.go

 		template, _ = sjson.Set(template, "choices.0.native_finish_reason", "tool_calls")
+	} else if finishReason != "" && (*param).(*convertGeminiResponseToOpenAIChatParams).FunctionIndex == 0 {
+		// Only pass through specific finish reasons
+		if finishReason == "max_tokens" || finishReason == "STOP" || finishReason == "stop" {


The finishReason variable is converted to lowercase on line 94. Therefore, the condition finishReason == "STOP" will always evaluate to false and is redundant.

Suggested change

if finishReason == "max_tokens" || finishReason == "STOP" || finishReason == "stop" {

if finishReason == "max_tokens" || finishReason == "stop" {

Co-authored-by: Amp <[email protected]> Amp-Thread-ID: https://ampcode.com/threads/T-019b8d07-bb78-775f-acea-61caa9f26dc7

HsnSaboor and others added 2 commits January 5, 2026 15:40

fix(translator): correctly map stop_reason in streaming responses

4b838b0

fix(translator): stabilize streaming finish_reason mapping

9fd678a

gemini-code-assist bot reviewed Jan 5, 2026

View reviewed changes

fix(translator): add finish_reason whitelist to prevent request loops

7847813

Co-authored-by: Amp <[email protected]> Amp-Thread-ID: https://ampcode.com/threads/T-019b8d07-bb78-775f-acea-61caa9f26dc7

wxyzh force-pushed the fix-stop-reason-mapping branch from a51ffab to 7847813 Compare January 5, 2026 08:33

This was referenced Jan 5, 2026

gemini 模型 tool_calls 问题 #866

Open

fix(antigravity): Streaming finish_reason 'tool_calls' overwritten by 'stop' - breaks Claude Code tool detection #876

Open

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

fix(translator): stabilize tool_call finish_reason #865

fix(translator): stabilize tool_call finish_reason #865

Uh oh!

wxyzh commented Jan 5, 2026

Uh oh!

gemini-code-assist bot commented Jan 5, 2026

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Jan 5, 2026

Uh oh!

gemini-code-assist bot Jan 5, 2026

Uh oh!

gemini-code-assist bot Jan 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	if finishReason == "max_tokens" \|\| finishReason == "STOP" \|\| finishReason == "stop" {
	if finishReason == "max_tokens" \|\| finishReason == "stop" {

Uh oh!

fix(translator): stabilize tool_call finish_reason #865

Are you sure you want to change the base?

fix(translator): stabilize tool_call finish_reason #865

Uh oh!

Conversation

wxyzh commented Jan 5, 2026

Uh oh!

gemini-code-assist bot commented Jan 5, 2026

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Jan 5, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 5, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 5, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants