fix(evals): remove tool restrictions and add compile-time guards

[SandyTao520]

Summary

Remove tool restrictions (tools.core whitelists and excludeTools blacklists) from evals, and add compile-time type guards to prevent future evals from reintroducing them.

Details

Evals were artificially constraining available tools via settings.tools.core (whitelist) and configOverrides.excludeTools (blacklist). This makes evals less realistic — in production, the agent has access to all tools and must decide which to use. Removing these restrictions tests the agent's actual behavior.

What changed:

• save_memory.eval.ts: Removed tools: { core: ['save_memory'] } and tools: { core: ['save_memory', 'list_directory', 'read_file', 'run_shell_command'] } from all 11 test cases.
• model_steering.eval.ts: Removed excludeTools: ['run_shell_command', 'ls', 'google_web_search'] from both test cases.
• generalist_delegation.eval.ts: Removed excludeTools: ['run_shell_command'] from all 4 test cases.
• test-helper.ts: Added ForbiddenToolSettings type that marks tools.core as never in EvalCase.params.settings.
• app-test-helper.ts: Added EvalConfigOverrides type that marks excludeTools, coreTools, allowedTools, and mainAgentTools as never in AppEvalCase.configOverrides.

Tool behavioral settings (e.g., tools.shell.enableShellOutputEfficiency) remain allowed — only tool availability restrictions are blocked.

Related Issues

Closes #18007

How to Validate

Type guard verification

Any eval attempting to restrict tools will now fail at compile time:

// TS error: Type 'string[]' is not assignable to type 'undefined'
evalTest('ALWAYS_PASSES', {
  params: { settings: { tools: { core: ['save_memory'] } } },
  // ...
});

// TS error: Type 'string[]' is not assignable to type 'undefined'
appEvalTest('ALWAYS_PASSES', {
  configOverrides: { excludeTools: ['run_shell_command'] },
  // ...
});

Verify with npx tsc --noEmit --project tsconfig.json.

Eval pass rates

All 3 affected eval files were run on CI via evals-nightly.yml workflow on branch sandytao/eval-remove-tool-restrictions across 6 models × 3 attempts = 54 total jobs:

Eval file	Result
save_memory.eval.ts	18/18 ✅
model_steering.eval.ts	18/18 ✅
generalist_delegation.eval.ts	18/18 ✅

Additionally, save_memory.eval.ts was run 3× locally with all 11 tests passing each time (33/33).

Pre-Merge Checklist

• Updated relevant documentation and README (if needed)
• Noted breaking changes (if any)

[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 significantly improves the realism and consistency of evaluation tests by eliminating artificial constraints on tool availability. By removing explicit tool whitelists and blacklists from various evaluation files, agents are now tested in an environment that mirrors production, where all tools are accessible. Furthermore, compile-time type guards have been implemented to prevent these restrictions from being accidentally or intentionally reintroduced, ensuring that future evaluations maintain this realistic testing standard.

Highlights

• Removed Tool Restrictions: Artificial tool restrictions (whitelists and blacklists) were removed from save_memory.eval.ts, model_steering.eval.ts, and generalist_delegation.eval.ts to ensure evaluations test against the full, default tool set.
• Added Compile-Time Type Guards: New TypeScript type guards were introduced in evals/test-helper.ts and evals/app-test-helper.ts to prevent the future reintroduction of tool availability restrictions in evaluation configurations, enforcing realistic testing behavior.

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. ↩

[gundermanc]

If you haven't yet, can you try running it in GitHub actions to make sure the ones that previously passed still pass?

[SandyTao520]

Yes I ran:
https://github.com/google-gemini/gemini-cli/actions/runs/23364019511 and https://github.com/google-gemini/gemini-cli/actions/runs/23364019490

[gemini-code-assist]

Code Review

This pull request effectively removes tool restrictions from evaluation tests to make them more realistic, and cleverly introduces compile-time guards using TypeScript's never type to prevent these restrictions from being reintroduced. The changes are well-targeted and align with the goal of improving eval quality. I have one suggestion to make the type guard for settings-based tool restrictions even more robust, aligning it with our established policies on tool availability.

[github-actions]

Size Change: -4 B (0%)

Total Size: 26.1 MB

Filename	Size	Change
./bundle/chunk-QGRZDB3C.js	0 B	-3.64 MB (removed)	🏆
./bundle/chunk-Y4WT6QYH.js	0 B	-14.5 MB (removed)	🏆
./bundle/core-LCHJY526.js	0 B	-42.4 kB (removed)	🏆
./bundle/devtoolsService-OSKK3Q7G.js	0 B	-27.7 kB (removed)	🏆
./bundle/interactiveCli-DWPVPNMG.js	0 B	-1.61 MB (removed)	🏆
./bundle/oauth2-provider-QW3CAXQO.js	0 B	-9.16 kB (removed)	🏆
./bundle/chunk-CASRPLZB.js	3.64 MB	+3.64 MB (new file)	🆕
./bundle/chunk-LILCOQTC.js	14.5 MB	+14.5 MB (new file)	🆕
./bundle/core-24KY7H7Z.js	42.4 kB	+42.4 kB (new file)	🆕
./bundle/devtoolsService-KWINE2VM.js	27.7 kB	+27.7 kB (new file)	🆕
./bundle/interactiveCli-XJ7S7CQP.js	1.61 MB	+1.61 MB (new file)	🆕
./bundle/oauth2-provider-PDSAPT3X.js	9.16 kB	+9.16 kB (new file)	🆕

ℹ️ View Unchanged

Filename	Size
./bundle/chunk-34MYV7JD.js	2.45 kB
./bundle/chunk-5725SFQR.js	1.95 MB
./bundle/chunk-5AUYMPVF.js	858 B
./bundle/chunk-664ZODQF.js	124 kB
./bundle/chunk-DAHVX5MI.js	206 kB
./bundle/chunk-IUUIT4SU.js	56.5 kB
./bundle/chunk-RJTRUG2J.js	39.8 kB
./bundle/devtools-36NN55EP.js	696 kB
./bundle/dist-T73EYRDX.js	356 B
./bundle/gemini.js	519 kB
./bundle/getMachineId-bsd-TXG52NKR.js	1.55 kB
./bundle/getMachineId-darwin-7OE4DDZ6.js	1.55 kB
./bundle/getMachineId-linux-SHIFKOOX.js	1.34 kB
./bundle/getMachineId-unsupported-5U5DOEYY.js	1.06 kB
./bundle/getMachineId-win-6KLLGOI4.js	1.72 kB
./bundle/memoryDiscovery-OV4FUTHJ.js	922 B
./bundle/multipart-parser-KPBZEGQU.js	11.7 kB
./bundle/node_modules/@google/gemini-cli-devtools/dist/client/main.js	221 kB
./bundle/node_modules/@google/gemini-cli-devtools/dist/src/_client-assets.js	227 kB
./bundle/node_modules/@google/gemini-cli-devtools/dist/src/index.js	11.5 kB
./bundle/node_modules/@google/gemini-cli-devtools/dist/src/types.js	132 B
./bundle/sandbox-macos-permissive-open.sb	890 B
./bundle/sandbox-macos-permissive-proxied.sb	1.31 kB
./bundle/sandbox-macos-restrictive-open.sb	3.36 kB
./bundle/sandbox-macos-restrictive-proxied.sb	3.56 kB
./bundle/sandbox-macos-strict-open.sb	4.82 kB
./bundle/sandbox-macos-strict-proxied.sb	5.02 kB
./bundle/src-QVCVGIUX.js	47 kB
./bundle/tree-sitter-7U6MW5PS.js	274 kB
./bundle/tree-sitter-bash-34ZGLXVX.js	1.84 MB

_{compressed-size-action}