Skip to content

Comments

fix: allow ask_user tool in interactive yolo mode#18541

Merged
jackwotherspoon merged 10 commits intomainfrom
ask-user/yolo
Feb 10, 2026
Merged

fix: allow ask_user tool in interactive yolo mode#18541
jackwotherspoon merged 10 commits intomainfrom
ask-user/yolo

Conversation

@jackwotherspoon
Copy link
Contributor

Summary

Allow ask_user in interactive yolo mode, previously tool was being auto-accepted with empty answers.

Moved yolo to priority 998 and made ask_user priority 999.

Details

  • YOLO Mode Logic: Updated PromptProvider to inject a new yolo section into the system prompt when
    ApprovalMode.YOLO is active and the session is interactive.
  • Prompt Snippets: Added renderYoloMode to both snippets.ts and snippets.legacy.ts containing the specific
    instructions for autonomous operation.
  • Priority Management: Ensured that while YOLO mode auto-approves most tools, the ask_user tool remains at priority = 999 with a decision = "ask_user". This guarantees that the model still requires explicit user confirmation when it chooses use ask_user, preventing it from "auto-answering" its own questions.
  • Validation: Added tests in prompts.test.ts to ensure the instructions are only included when
    appropriate (YOLO mode active + interactive).

Related Issues

Fixes #18540

How to Validate

  1. Run unit tests for the prompt provider:

    npm test -w @google/gemini-cli-core -- src/core/prompts.test.ts
  2. Build the project:

    npm run build
  3. (Optional) Toggle yolo mode on and ask agent to ask you questions it will. But ask it to implement a feature and it won't unless necessary

Pre-Merge Checklist

  • Updated relevant documentation and README (if needed)
  • Added/updated tests (if needed)
  • Noted breaking changes (if any)
  • Validated on required platforms/methods:
    • MacOS
      • npm run
      • npx
      • Docker
      • Podman
      • Seatbelt
    • Windows
      • npm run
      • npx
      • Docker
    • Linux
      • npm run
      • npx
      • Docker

@jackwotherspoon jackwotherspoon requested a review from a team as a code owner February 7, 2026 21:29
@gemini-code-assist
Copy link
Contributor

Summary of Changes

Hello @jackwotherspoon, 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 the behavior of the ask_user tool within the system's 'YOLO' (You Only Live Once) autonomous mode. The core objective is to ensure that even when operating with minimal interruption, the system still correctly seeks user input when the ask_user tool is invoked, rather than proceeding with an empty or default response. This enhancement improves the reliability and user-friendliness of autonomous operations by guaranteeing critical user confirmations are not bypassed.

Highlights

  • YOLO Mode ask_user Tool Fix: The ask_user tool now correctly prompts the user for input in interactive YOLO mode, preventing it from being auto-accepted with empty answers, which was previously an issue.
  • Priority Adjustment for Tools: The general allow rule for YOLO mode has been assigned a priority of 998, while the ask_user tool has been given a higher priority of 999 with an explicit ask_user decision, ensuring it takes precedence.
  • System Prompt Enhancement: A new 'Autonomous Mode (YOLO)' section has been added to the system prompt. This section provides specific instructions to the model on when it should and should not use the ask_user tool in YOLO mode.
  • Comprehensive Testing: New unit tests have been introduced to prompts.test.ts and policy-engine.test.ts to validate that YOLO mode instructions are included in the prompt only when appropriate and that the ask_user tool behaves as expected in YOLO mode.
Changelog
  • packages/core/src/core/prompts.test.ts
    • Added tests to verify that YOLO mode instructions are included in the system prompt only when ApprovalMode.YOLO is active and the session is interactive.
    • Added tests to confirm YOLO mode instructions are NOT included in non-interactive mode or for ApprovalMode.DEFAULT.
  • packages/core/src/policy/config.test.ts
    • Updated the expected priority for the general ALLOW rule in YOLO mode from 1.999 to 1.998 in test assertions, reflecting the new priority scheme.
  • packages/core/src/policy/policies/yolo.toml
    • Introduced a specific rule for toolName = "ask_user" with decision = "ask_user" and priority = 999 for YOLO mode.
    • Modified the general allow rule for YOLO mode to have priority = 998.
  • packages/core/src/policy/policy-engine.test.ts
    • Added a new test suite 'YOLO mode with ask_user tool' to confirm that ask_user tool results in PolicyDecision.ASK_USER even in YOLO mode.
    • Added a test to ensure other tools correctly result in PolicyDecision.ALLOW in YOLO mode.
  • packages/core/src/prompts/promptProvider.ts
    • Introduced isYoloMode variable to detect if ApprovalMode.YOLO is active.
    • Conditionally added a yoloMode section to the prompt context, dependent on isYoloMode and interactiveMode.
  • packages/core/src/prompts/snippets.legacy.ts
    • Added yoloMode?: boolean; to the SystemPromptOptions interface.
    • Integrated renderYoloMode(options.yoloMode) into the main prompt template.
    • Implemented the renderYoloMode function to generate the 'Autonomous Mode (YOLO)' instructions for the system prompt.
  • packages/core/src/prompts/snippets.ts
    • Added yoloMode?: boolean; to the SystemPromptOptions interface.
    • Integrated renderYoloMode(options.yoloMode) into the main prompt template.
    • Implemented the renderYoloMode function to generate the 'Autonomous Mode (YOLO)' instructions for the system prompt.
Activity
  • The pull request was created by jackwotherspoon to address an issue with the ask_user tool in YOLO mode.
  • The author provided a detailed summary and specific steps for validation, including running unit tests and an optional manual validation flow.
  • Tests have been added and updated, and the changes have been validated on MacOS.
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.

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-cli gemini-cli bot added the area/non-interactive Issues related to GitHub Actions, SDK, 3P Integrations, Shell Scripting, Command line automation label Feb 7, 2026
Copy link
Contributor

@gemini-code-assist gemini-code-assist bot left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request effectively addresses the issue of the ask_user tool being auto-accepted in YOLO mode. The solution is well-implemented through a two-pronged approach: adjusting policy priorities in yolo.toml to create a specific exception for ask_user, and adding clear instructions to the system prompt for the language model on how to handle this tool in autonomous mode. The changes are supported by comprehensive tests that validate both the policy engine behavior and the conditional inclusion of the new prompt section. The code is clean, correct, and follows existing patterns. I have no major concerns with this implementation.

@github-actions
Copy link

github-actions bot commented Feb 7, 2026

Size Change: +1.79 kB (+0.01%)

Total Size: 23.9 MB

Filename Size Change
./bundle/gemini.js 23.9 MB +1.79 kB (+0.01%)
ℹ️ View Unchanged
Filename Size
./bundle/sandbox-macos-permissive-closed.sb 1.03 kB
./bundle/sandbox-macos-permissive-open.sb 890 B
./bundle/sandbox-macos-permissive-proxied.sb 1.31 kB
./bundle/sandbox-macos-restrictive-closed.sb 3.29 kB
./bundle/sandbox-macos-restrictive-open.sb 3.36 kB
./bundle/sandbox-macos-restrictive-proxied.sb 3.56 kB

compressed-size-action

@gemini-cli gemini-cli bot added the 🔒 maintainer only ⛔ Do not contribute. Internal roadmap item. label Feb 7, 2026
Copy link
Contributor

@jerop jerop left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

left some suggestions to make it clear that this is for interactive yolo mode

return `
# Autonomous Mode (YOLO)
You are operating in **autonomous mode**. The user has requested minimal interruption.
Copy link
Member

@gundermanc gundermanc Feb 10, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It'd be great to have 1-3 evals to cover the major scenarios.

  • "Significant rework"
  • "Ambiguous request"
  • "Explicitly asked"

These can be a bit of a pain to implement so I think it's also fine if you just wanted to add one.

You are operating in **autonomous mode**. The user has requested minimal interruption.
**Only use the \`${ASK_USER_TOOL_NAME}\` tool if:**
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We may find we get better behavior with at least one example.

@jackwotherspoon jackwotherspoon added this pull request to the merge queue Feb 10, 2026
@github-merge-queue github-merge-queue bot removed this pull request from the merge queue due to failed status checks Feb 10, 2026
@jackwotherspoon jackwotherspoon added this pull request to the merge queue Feb 10, 2026
@github-merge-queue github-merge-queue bot removed this pull request from the merge queue due to failed status checks Feb 10, 2026
@jackwotherspoon jackwotherspoon added this pull request to the merge queue Feb 10, 2026
Merged via the queue into main with commit 740f0e4 Feb 10, 2026
26 checks passed
@jackwotherspoon jackwotherspoon deleted the ask-user/yolo branch February 10, 2026 19:06
krsjenmt added a commit to krsjenmt/gemini-cli that referenced this pull request Feb 11, 2026
* Fix newline insertion bug in replace tool (google-gemini#18595)

* fix(evals): update save_memory evals and simplify tool description (google-gemini#18610)

* chore(evals): update validation_fidelity_pre_existing_errors to USUALLY_PASSES (google-gemini#18617)

* fix: shorten tool call IDs and fix duplicate tool name in truncated output filenames (google-gemini#18600)

* feat(cli): implement atomic writes and safety checks for trusted folders (google-gemini#18406)

* Remove relative docs links (google-gemini#18650)

* docs: add legacy snippets convention to GEMINI.md (google-gemini#18597)

* fix(chore): Support linting for cjs (google-gemini#18639)

Co-authored-by: Gal Zahavi <38544478+galz10@users.noreply.github.com>

* feat: move shell efficiency guidelines to tool description (google-gemini#18614)

* Added "" as default value, since getText() used to expect a string only and thus crashed when undefined...  Fixes google-gemini#18076   (google-gemini#18099)

* Allow @-includes outside of workspaces (with permission) (google-gemini#18470)

* chore: make `ask_user` header description more clear (google-gemini#18657)

* bug(core): Fix minor bug in migration logic. (google-gemini#18661)

* Harded code assist converter. (google-gemini#18656)

* refactor(core): model-dependent tool definitions (google-gemini#18563)

* feat: enable plan mode experiment in settings (google-gemini#18636)

* refactor: push isValidPath() into parsePastedPaths() (google-gemini#18664)

* fix(cli): correct 'esc to cancel' position and restore duration display (google-gemini#18534)

* feat(cli): add DevTools integration with gemini-cli-devtools (google-gemini#18648)

* chore: remove unused exports and redundant hook files (google-gemini#18681)

* Fix number of lines being reported in rewind confirmation dialog (google-gemini#18675)

* feat(cli): disable folder trust in headless mode (google-gemini#18407)

* Disallow unsafe type assertions (google-gemini#18688)

* Change event type for release (google-gemini#18693)

* feat: handle multiple dynamic context filenames in system prompt (google-gemini#18598)

* Properly parse at-commands with narrow non-breaking spaces (google-gemini#18677)

* refactor(core): centralize core tool definitions and support model-specific schemas (google-gemini#18662)

* feat(core): Render memory hierarchically in context. (google-gemini#18350)

* feat: Ctrl+O to expand paste placeholder (google-gemini#18103)

* fix(cli): Improve header spacing (google-gemini#18531)

* Feature/quota visibility 16795 (google-gemini#18203)

* docs: remove TOC marker from Plan Mode header (google-gemini#18678)

* Inline thinking bubbles with summary/full modes (google-gemini#18033)

Co-authored-by: Jacob Richman <jacob314@gmail.com>

* fix(ui): remove redundant newlines in Gemini messages (google-gemini#18538)

* test(cli): fix AppContainer act() warnings and improve waitFor resilience (google-gemini#18676)

* refactor(core): refine Security & System Integrity section in system prompt (google-gemini#18601)

* Fix layout rounding. (google-gemini#18667)

* docs(skills): enhance pr-creator safety and interactivity (google-gemini#18616)

* test(core): remove hardcoded model from TestRig (google-gemini#18710)

* feat(core): optimize sub-agents system prompt intro (google-gemini#18608)

* feat(cli): update approval mode labels and shortcuts per latest UX spec (google-gemini#18698)

* fix(plan): update persistent approval mode setting (google-gemini#18638)

Co-authored-by: Sandy Tao <sandytao520@icloud.com>

* fix: move toasts location to left side (google-gemini#18705)

* feat(routing): restrict numerical routing to Gemini 3 family (google-gemini#18478)

* fix(ide): fix ide nudge setting (google-gemini#18733)

* fix(core): standardize tool formatting in system prompts (google-gemini#18615)

* chore: consolidate to green in ask user dialog (google-gemini#18734)

* feat: add `extensionsExplore` setting to enable extensions explore UI. (google-gemini#18686)

* feat(cli): defer devtools startup and integrate with F12 (google-gemini#18695)

* ui: update & subdue footer colors and animate progress indicator (google-gemini#18570)

* test: add model-specific snapshots for coreTools (google-gemini#18707)

Co-authored-by: matt korwel <matt.korwel@gmail.com>

* ci: shard windows tests and fix event listener leaks (google-gemini#18670)

* fix: allow `ask_user` tool in yolo mode (google-gemini#18541)

* feat: redact disabled tools from system prompt (google-gemini#13597) (google-gemini#18613)

* Update Gemini.md to use the curent year on creating new files (google-gemini#18460)

* Code review cleanup for thinking display (google-gemini#18720)

* fix(cli): hide scrollbars when in alternate buffer copy mode (google-gemini#18354)

Co-authored-by: Jacob Richman <jacob314@gmail.com>

* Fix issues with rip grep (google-gemini#18756)

* fix(cli): fix history navigation regression after prompt autocomplete (google-gemini#18752)

* chore: cleanup unused and add unlisted dependencies in packages/cli (google-gemini#18749)

* Fix issue where Gemini CLI creates tests in a new file (google-gemini#18409)

* feat(telemetry): Ensure experiment IDs are included in OpenTelemetry logs (google-gemini#18747)

* feat(ux): added text wrapping capabilities to markdown tables (google-gemini#18240)

Co-authored-by: jacob314 <jacob314@gmail.com>

* Revert "fix(mcp): ensure MCP transport is closed to prevent memory leaks" (google-gemini#18771)

* chore(release): bump version to 0.30.0-nightly.20260210.a2174751d (google-gemini#18772)

* chore: cleanup unused and add unlisted dependencies in packages/core (google-gemini#18762)

* chore(core): update activate_skill prompt verbiage to be more direct (google-gemini#18605)

* Add autoconfigure memory usage setting to the dialog (google-gemini#18510)

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

* fix(core): prevent race condition in policy persistence (google-gemini#18506)

Co-authored-by: Allen Hutchison <adh@google.com>

* fix(evals): prevent false positive in hierarchical memory test (google-gemini#18777)

* test(evals): mark all `save_memory` evals as `USUALLY_PASSES` due to unreliability (google-gemini#18786)

* feat(cli): add setting to hide shortcuts hint UI (google-gemini#18562)

* feat(core): formalize 5-phase sequential planning workflow (google-gemini#18759)

* Introduce limits for search results. (google-gemini#18767)

---------

Co-authored-by: Andrew Garrett <andrewgarrett@google.com>
Co-authored-by: N. Taylor Mullen <ntaylormullen@google.com>
Co-authored-by: Sandy Tao <sandytao520@icloud.com>
Co-authored-by: Gal Zahavi <38544478+galz10@users.noreply.github.com>
Co-authored-by: christine betts <chrstn@uw.edu>
Co-authored-by: Aswin Ashok <aswwwin@google.com>
Co-authored-by: Abhijith V Ashok <abhi2349jith@gmail.com>
Co-authored-by: Tommaso Sciortino <sciortino@gmail.com>
Co-authored-by: Jack Wotherspoon <jackwoth@google.com>
Co-authored-by: joshualitt <joshualitt@google.com>
Co-authored-by: Jacob Richman <jacob314@gmail.com>
Co-authored-by: Aishanee Shah <aishaneeshah@gmail.com>
Co-authored-by: Jerop Kipruto <jerop@google.com>
Co-authored-by: Adib234 <30782825+Adib234@users.noreply.github.com>
Co-authored-by: Christian Gunderman <gundermanc@gmail.com>
Co-authored-by: g-samroberts <158088236+g-samroberts@users.noreply.github.com>
Co-authored-by: Spencer <spencertang@google.com>
Co-authored-by: Dmitry Lyalin <dmitry.lyalin@lyalin.com>
Co-authored-by: matt korwel <matt.korwel@gmail.com>
Co-authored-by: Shreya Keshive <shreyakeshive@google.com>
Co-authored-by: Sri Pasumarthi <111310667+sripasg@users.noreply.github.com>
Co-authored-by: Keith Guerin <keithguerin@gmail.com>
Co-authored-by: Sehoon Shon <sshon@google.com>
Co-authored-by: Adam Weidman <65992621+adamfweidman@users.noreply.github.com>
Co-authored-by: Kevin Ramdass <ramdass.kevin@gmail.com>
Co-authored-by: Dev Randalpura <devrandalpura@google.com>
Co-authored-by: gemini-cli-robot <gemini-cli-robot@google.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Brad Dux <959674+braddux@users.noreply.github.com>
Co-authored-by: Allen Hutchison <adh@google.com>
Co-authored-by: Abhijit Balaji <abhijitbalaji@google.com>
@jerop jerop changed the title fix: allow ask_user tool in yolo mode fix: allow ask_user tool in interactive yolo mode Feb 19, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

area/non-interactive Issues related to GitHub Actions, SDK, 3P Integrations, Shell Scripting, Command line automation 🔒 maintainer only ⛔ Do not contribute. Internal roadmap item.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Ask User tool being auto-approved with empty answers in YOLO mode

3 participants