fix(core): avoid persisting interrupted response placeholder

[Shivansh1980]

Summary

Fixes #28927.

After an interrupted tool-response turn, Gemini CLI persists this text as a synthetic model response:

[The previous response was interrupted before it completed.]

The model can then repeat that placeholder after later tool calls, leaving the session unable to continue normally. This change replaces the persistent model-authored turn with a transient user-side boundary used only for the next API request.

Details

Root cause

PR #28700 prevented the next user message from being fused into an unanswered tool response by inserting a synthetic model turn:

functionResponse -> model placeholder -> new user message

That preserves role alternation, but it also stores the placeholder as ordinary model output in agentHistory and repeatedly sends it back to the model.

Solution

When history ends with an unanswered functionResponse:

1. Do not append a synthetic model turn.
2. Preserve the tool result and real user message in durable history.
3. Clone the outbound request history.
4. Insert a user-authored interruption boundary between the tool result and all parts of the new user message.
5. Send that boundary only for the current logical request and its retries.

The outbound request becomes:

functionResponse
-> transient user-side interruption boundary
-> new user message

The boundary is applied to both standard curated history and context-managed apiHistoryOverride history. It is never stored in agentHistory, chat recordings, or resumed session data. Multipart user messages remain together and in order after the boundary.

Related Issues

• Fixes Model parotting "[The previous response was interrupted before it completed]" #28927.
• Regression introduced by fix(core): stop a new user message fusing into an unanswered tool response #28700.
• No other open PR currently targets this interruption-placeholder behavior.

How to Validate

npx vitest run packages/core/src/core/geminiChat.test.ts --silent

Test Files  1 passed (1)
Tests       111 passed (111)

Additional checks:

npx prettier --check \
  packages/core/src/core/geminiChat.ts \
  packages/core/src/core/geminiChat.test.ts

npx eslint \
  packages/core/src/core/geminiChat.ts \
  packages/core/src/core/geminiChat.test.ts \
  --max-warnings 0

npx tsc --noEmit -p packages/core/tsconfig.json

Coverage includes:

• Cancelled tool response followed by a multipart user message.
• Resumed session ending in an unanswered tool response.
• Context-managed API history override.
• Correct boundary ordering.
• Absence of the old model-authored placeholder from outbound requests.
• Confirmation that the transient boundary is not persisted.

Pre-Merge Checklist

• Updated relevant documentation and README (not needed; internal history-handling fix)
• Added/updated tests
• Noted breaking changes (none)
• Validated on required platforms/methods:
  • MacOS
    • npm run
    • npx
    • Docker
    • Podman
    • Seatbelt
  • Windows
    • npm run
    • npx
    • Docker
  • Linux
    • npm run
    • npx
    • Docker

Validated locally on Windows after merging current main. The focused suite, Prettier, file-scoped ESLint, core TypeScript check, and full build pass. Full npm run preflight reached lint setup and then stopped because Python/yamllint is not installed on this machine.

The full core suite completed with 7,857 passing and 73 skipped tests. Its 22 failures are unrelated Windows environment/integration failures: 15 symlink-permission EPERM failures, six shell/sandbox environment failures, and one background-log timing failure. No failures occurred in the changed files.

[github-actions]

📊 PR Size: size/M

• Lines changed: 174
• Additions: +125
• Deletions: -49
• Files changed: 2

[gemini-code-assist]

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 refines how the system handles interrupted model responses following tool results. By moving from a persistent synthetic model turn to a transient, context-injected boundary, the change ensures that tool results and role alternation are preserved without polluting the model's training context or durable chat history. This approach improves reliability during session resumption and stream failures.

Highlights

• Interruption Handling: Replaced the synthetic model-authored interruption turn with a transient user-side boundary to prevent the model from learning to repeat placeholder text.
• History Management: Implemented logic to inject the interruption boundary into both regular request history and context-managed API history overrides, while excluding it from durable agent history.
• Testing: Updated GeminiChat tests to verify that new user messages are correctly separated from cancelled tool responses without relying on persisted placeholders.

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.

Footnotes

1. 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 refactors how interrupted model responses (dangling tool responses) are handled in GeminiChat. Instead of appending a dummy model turn with a placeholder to close the unanswered tool response, it now inserts a transient interruption boundary text directly into the subsequent user message parts during history or API history override preparation. The corresponding unit tests in geminiChat.test.ts have been updated and expanded to verify this new boundary insertion behavior under different context management configurations. I have no additional feedback to provide.

[gemini-cli]

Hi there! Thank you for your interest in contributing to Gemini CLI.

To ensure we maintain high code quality and focus on our prioritized roadmap, we only guarantee review and consideration of pull requests for issues that are explicitly labeled as 'help wanted'.

This PR will be closed in 7 days if it remains without that designation. We encourage you to find and contribute to existing 'help wanted' issues in our backlog! Thank you for your understanding.

[Shivansh1980]

/gemini review

[gemini-code-assist]

Code Review

This pull request refactors the handling of interrupted or unanswered tool responses in GeminiChat. Instead of appending a dummy model turn placeholder to the history, it now inserts a transient interruption boundary message directly into the last user turn's parts, supporting both standard history and context-managed history overrides. The unit tests have been updated to reflect and validate this new behavior. There are no review comments to address, and the changes look solid.

[gemini-cli]

This pull request is being closed as it has been open for 14 days without a 'help wanted' designation. We encourage you to find and contribute to existing 'help wanted' issues in our backlog! Thank you for your understanding.