feat: add multi-persona review panel for document critique

[rjwalters]

Closes #4

Note: Builder completed changes but exited before creating a PR. PR created via direct completion.

Changes

packages/review-panel/package.json                |  27 +++
 packages/review-panel/src/aggregator.ts           | 199 +++++++++++++++++++
 packages/review-panel/src/index.ts                |  56 ++++++
 packages/review-panel/src/personas.ts             | 135 +++++++++++++
 packages/review-panel/src/runner.ts               | 117 +++++++++++
 packages/review-panel/src/test/aggregator.test.ts | 227 ++++++++++++++++++++++
 packages/review-panel/src/test/runner.test.ts     | 163 ++++++++++++++++
 packages/review-panel/src/types.ts                | 121 ++++++++++++
 packages/review-panel/tsconfig.json               |  15 ++
 9 files changed, 1060 insertions(+)

Commits

• 12c2df9 feat: add multi-persona review panel package

Test plan

• Verify changes match issue requirements
• Confirm tests pass

[rjwalters]

✅ Approved!

Thorough evaluation of the multi-persona review panel implementation. Excellent work.

Evaluation

Acceptance Criteria (Issue #4):

• ✅ Persona format defined (Persona interface with id, name, perspective, dimensions, promptTemplate)
• ✅ Four default personas: Critical Editor, Domain Expert, General Reader, Style Reviewer
• ✅ Independent execution via Promise.all in runAllPersonas()
• ✅ Aggregation with consensus clustering and ranking by strength (4/4 > 3/4 > 2/4 > 1/4)
• ✅ Disagreements explicitly separated (consensusItems vs disagreements)
• ✅ User-configurable personas via ReviewPanelOptions.personas

Code Quality:

• Clean TypeScript with strict: true
• Well-structured types with thorough JSDoc documentation
• Robust JSON parsing with markdown fence stripping and graceful fallback
• Input validation for categories and severities
• Clean public API via reviewDocument() entry point

Testing:

• 20 tests across 9 suites — all pass
• Good coverage of edge cases (invalid JSON, missing descriptions, same-persona clustering prevention)
• Tests for custom/user-defined personas

Design Notes:

• Text similarity using Jaccard index is a pragmatic choice with appropriate comment noting embeddings would improve accuracy
• Single-linkage greedy clustering is reasonable for the expected scale
• callModel callback pattern keeps the package backend-agnostic — good design

Test Execution

Test plan from PR description:

1. Verify changes match issue requirements — ✅ All 6 acceptance criteria met
2. Confirm tests pass — ✅ 20/20 tests pass, type checking clean

Test Scoping

Strategy: Full build + test (new package, no pre-existing tests)
Result: tsc --noEmit clean, 20 tests pass in 322ms