fix: prevent compression endless loop in agentic sessions

[Rahu378]

Summary

Prevent endless auto-compression loop in long agentic sessions ("Ralph loops").
When tool-call chains trigger multiple processTurn calls within a single user
turn, tryCompressChat was firing on every call — even after already compressing
once — causing an infinite compression loop. Priority fix before the 20%
compression flag rolls out to users.

Details

Two changes:

1. GeminiClient - added hasCompressedThisTurn boolean flag that is set to
   true after the first successful compression and reset to false at the
   start of each new user turn (in startChat). Auto-compression is skipped
   entirely if this flag is already set.

2. ChatCompressionService - added stillAboveThreshold check after
   compression completes. If the new token count is still above the compression
   threshold (even though it didn't inflate), the service now returns
   COMPRESSION_FAILED_INFLATED_TOKEN_COUNT instead of COMPRESSED. This sets
   hasFailedCompressionAttempt = true in the client, stopping further retries.

Related Issues

Fixes #23907
Parent: #23905

How to Validate

1. Run a "Ralph loop" with compression threshold set to 20%:

   • In settings, set compressionThreshold: 0.2
   • Run a long agentic script that fills context (e.g. repeatedly read/summarise
     files in the repo)
   • Without fix: compression triggers on every tool-call follow-up turn
   • With fix: compression fires at most once per user turn, loop stops

2. Check logs for:
   [compression] Skipping auto-compression: already compressed this turn.

3. Manual compress command (/compress) should still work — it passes
   force: true which bypasses the guard entirely.

Pre-Merge Checklist

• Updated relevant documentation and README (if needed)
• [ x ] Added/updated tests (if needed) : existing compression tests pass;
  unit tests for the guard can be added in follow-up
• Noted breaking changes (if any): no breaking changes
• Validated on required platforms/methods:

  • [ x] MacOS

    • [ x] npm run

[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 addresses an issue where long-running agentic sessions could enter an infinite auto-compression loop. By implementing a per-turn guard and stricter threshold validation, the system now correctly identifies when further compression is ineffective or unnecessary, ensuring stability during complex tool-call chains.

Highlights

• Compression Guard: Introduced a hasCompressedThisTurn flag in GeminiClient to ensure auto-compression triggers at most once per user turn, preventing infinite loops.
• Threshold Validation: Updated ChatCompressionService to treat compression attempts that remain above the token threshold as failures, effectively stopping further retry attempts.

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.

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 introduces measures to prevent endless compression loops and adds safety evaluation tests for shell commands. In GeminiClient, a new flag hasCompressedThisTurn ensures compression only happens once per user turn. Additionally, ChatCompressionService now bails out if the token count remains above the threshold after compression. A new evaluation file shell_command_safety.eval.ts was added to test shell tool usage and safety constraints. I have no feedback to provide.

_{Note: Security Review is unavailable for this PR.}