Finding category taxonomy on AI review findings

[JSONbored]

Phase 1 — Review output, noise & trust

Context

AI findings (blockers/nits) carry a severity but no category, which limits inline labeling and blocks category-level analytics.

Goal

Tag every finding with a category (e.g. security, correctness, performance, maintainability, tests, style).

Requirements

• Add a category enum to the AI finding schema (src/services/ai-review.ts parseModelReview) with a safe default; surface it in inline-comment labels and the unified comment; feed the analytics dashboard (Phase 3).
• Deterministic fallback categorization when the model omits it (keyword/path heuristics).

Deliverables

• Schema + parser + rendering + tests.

Expected outcome

• Cleaner, categorized findings; unlocks category-level acceptance analytics.

Effort

S.

[linear-code]

JSO-2038