Add individual evaluator selection with multi-select dropdown (Vibe Kanban)

[cteyton]

Summary

• Replace the 3-option category filter (All / Errors Only / Suggestions Only) with a multi-select dropdown listing all 17 evaluators individually, grouped by issue type
• Users can now pick any combination of evaluators, with at least 1 required
• By default all evaluators are selected, preserving existing behavior
• Cloud mode is unchanged (UI hidden, always runs all evaluators)

Why

Previously, the frontend only allowed filtering evaluators by broad category ("errors" or "suggestions"). Users had no way to run a specific subset of evaluators — for example, only "Security" and "Outdated Documentation" — without running the entire category. This change gives fine-grained control over which evaluators to run when not in cloud mode.

Implementation details

Backend (additive, backward-compatible)

• New selectedEvaluators?: string[] option on IEvaluationOptions — when set and non-empty, overrides evaluatorFilter. The existing evaluatorFilter type ("all" | "errors" | "suggestions") and CLI --evaluator-filter flag are unchanged.
• New filterEvaluatorsByIds() function in evaluator-types.ts — filters the evaluator list by specific IDs (filenames without .md), preserving original order.
• Runner integration — both runAllEvaluators() and runUnifiedEvaluation() accept and use selectedEvaluators, falling back to category-based filtering when not set.
• Engine pass-through — selectedEvaluators is explicitly forwarded at both runner call sites in engine.ts (per the documented Configuration Options Flow pattern in CLAUDE.md).
• API enrichment — GET /api/evaluators now returns issueType ("error" | "suggestion") per evaluator, enabling the frontend to group them.

Frontend

• Multi-select dropdown in RepositoryUrlInput.tsx replaces the old <select>:
  • Button shows "{N}/{total} evaluators" with chevron
  • Dropdown panel with "Select All" / "Deselect All" controls
  • "Errors (13)" and "Suggestions (4)" group headers with group-level checkboxes (supports indeterminate state)
  • Individual checkboxes per evaluator with minimum-1 constraint
  • Click-outside-to-close via useRef + mousedown listener
• useEvaluationApi — submitJob() accepts and sends selectedEvaluators in the POST body
• Progress tracking — getFilteredEvaluatorCount() accepts optional selectedCount for accurate progress bar initialization
• App.tsx — forwards selectedEvaluators through the submission flow

Tests

• 7 new tests for filterEvaluatorsByIds covering: specific selection, order preservation, empty input, unknown IDs, full selection, and mixed error/suggestion selection

Files changed (11)

Layer	File	Change
Types	src/shared/types/evaluation.ts	Add selectedEvaluators field
Backend	src/shared/evaluation/evaluator-types.ts	Add filterEvaluatorsByIds()
Backend	src/shared/evaluation/runner.ts	Wire selectedEvaluators in both runners
Backend	src/shared/evaluation/engine.ts	Pass-through at 2 call sites
API	src/api/utils/evaluator-utils.ts	Add issueType to evaluator response
Frontend	frontend/src/types/evaluator.ts	Add issueType field
Frontend	frontend/src/hooks/useEvaluationApi.ts	Add selectedEvaluators param
Frontend	frontend/src/lib/formatters.ts	Accept selected count
Frontend	frontend/src/components/RepositoryUrlInput.tsx	Multi-select dropdown UI
Frontend	frontend/src/App.tsx	Forward selectedEvaluators
Tests	src/shared/evaluation/evaluator-types.test.ts	7 new tests

---

This PR was written using Vibe Kanban