feat: redact disabled tools from system prompt (#13597)#18613
feat: redact disabled tools from system prompt (#13597)#18613NTaylorMullen merged 1 commit intomainfrom
Conversation
Summary of ChangesHello @NTaylorMullen, 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 enhances the system prompt generation logic to be more adaptive to the available tools. By conditionally including or excluding references to Highlights
Changelog
Activity
Using Gemini Code AssistThe 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
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 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
|
There was a problem hiding this comment.
Code Review
This pull request effectively redacts references to disabled grep_search and glob tools from the system prompt, preventing the model from attempting to use unavailable tools. The changes are well-implemented with corresponding updates to prompt providers, snippets, and a new regression test to ensure correctness.
I've identified a couple of minor areas for improvement in the prompt construction logic to enhance clarity and correctness, which I've detailed in the review comments. Overall, this is a solid contribution.
|
Size Change: +681 B (0%) Total Size: 23.9 MB ℹ️ View Unchanged
|
2ea6b89 to
f7a6a6a
Compare
dbc81ef to
9ae951b
Compare
- Conditionally render 'grep_search' and 'glob' search tool mentions in system prompts - Only include tool mentions if the tools are enabled in the current registry - Update prompt provider to pass tool enablement status to snippets - Add tests to verify redaction when tools are disabled Fixes #13597
9ae951b to
23ced49
Compare
* 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>
Summary
Redact references to
grep_searchandglobfrom the system prompt when these tools are disabled.Details
Previously, the system prompt explicitly instructed the model to use
grep_searchandglobsearch tools extensively, regardless of whether they were available in the tool registry. This led to "unfortunate" situations where the model would attempt to use disabled tools.This PR:
enableGrepandenableGlobflags toPrimaryWorkflowsOptions.PromptProviderto detect if these tools are enabled.prompts.test.ts.Related Issues
Fixes #13597
How to Validate
Run the prompts tests:
npm test -w @google/gemini-cli-core -- src/core/prompts.test.tsThe test
should redact grep and glob from the system prompt when they are disabledspecifically verifies this behavior.Pre-Merge Checklist