fix(eval): keep CLI-only commands out of the TUI menu

[jariy17]

Problem

The eval router and its batch-evaluation subgroup open an interactive TUI on a bare invocation (.default(renderTui(...))) but never called .supportedTuiCommands(...), so the menu (RouterScreen → isTuiCommandSupported) offered commands that have no interactive screen:

• eval menu listed ondemand (help-only default) and recommendation (no screen).
• eval batch-evaluation menu listed evaluate and simulate (imperative job-starting commands with required flags).

Selecting any of them navigated to a route with no screen. Every sibling with the same shape (batch-insights, ab-test, online-eval, dataset, config-bundle, …) already restricts the menu to ("get", "list"); these two were the outliers.

Fix

Add the missing allowlists, matching the existing batch-insights / gateway pattern:

• eval → the 8 groups that actually have an interactive screen (drops ondemand + recommendation).
• eval batch-evaluation → ("get", "list") (drops evaluate + simulate).

Behaviour of the commands themselves is unchanged — agentcore eval ondemand still prints help, batch-evaluation simulate still runs from the CLI. This only gates TUI-menu membership.

Tests

Mirror the gateway command-hierarchy suite (supportsTui helper + test.each for supported vs CLI-only):

• New src/handlers/eval/eval.test.tsx — the 8 screen-backed groups are offered; ondemand + recommendation are hidden.
• batch-evaluation.test.tsx — get/list (and bare group) offered; evaluate/simulate hidden.

bun test src/handlers/eval/ → 377 pass, 0 fail. oxlint + prettier clean on changed files.

Note: tsc --noEmit surfaces a pre-existing duplicate-identifier error in src/handlers/project/buildDeploy.screen.test.tsx (present on refactor before this branch; unrelated to these files).

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[agentcore-devx-automation]

AgentCore Harness Review

Verdict: Looks good

Small, focused fix that adds supportedTuiCommands to the eval and eval batch-evaluation routers so CLI-only commands (ondemand, recommendation, evaluate, simulate) stop showing up in the interactive menu. The categorization matches each subhandler's actual state (either no .default(renderTui...) or an explicit empty supportedTuiCommands()), and the pattern mirrors what's already in place for gateway, runtime, identity, etc.

The new tests validate menu membership through the real compiled router with TestCoreClient, consistent with gateway.test.tsx. No excessive mocking, no telemetry surface changed. LGTM.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.15%. Comparing base (a665955) to head (a3cc129).

Additional details and impacted files

@@            Coverage Diff            @@
##           refactor    #2191   +/-   ##
=========================================
  Coverage     97.15%   97.15%           
=========================================
  Files           542      542           
  Lines         37728    37740   +12     
=========================================
+ Hits          36653    36665   +12     
  Misses         1075     1075           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[nborges-aws]

LGTM