Review a component change

• R-REVIEW-001: A review is read-only. Do not edit files, update snapshots, commit, or invoke an implementation workflow unless the user asks for fixes after receiving the review.
• R-REVIEW-002: Report only evidence-backed findings. Cite path:line, the violated rule ID or contract requirement ID, and whether the finding was verified (executed or confirmed in source) or inferred.

Workflow

1. Establish the scope: the named files, the working-tree diff, or the branch diff against its base.
2. Classify each changed surface with the classification guide: task intent, implementation model, public compatibility, and native obligations (R-CLASSIFY-001, R-CLASSIFY-002).

3. Load only the checklists for the surfaces present:

   Surface	Checklist
   v0	v0 review checklist
   v1	v1 review checklist
   Agentic higher-order	Package and higher-order component invariants; the component's SPEC.md and spec/; and the review checklists in types and slots, state and accessibility, styles and tokens, and rendering for the areas changed
   Agentic primitive	Primitive invariants, the primitive's CONTRACT.md, and primitive contracts
   Tests and stories	Tests and stories for agentic code; legacy test apps for v0 and v1
   Contract or provenance	Contract authoring workflow, the contracts reference, and only the source adapters matching the changed evidence
   Focus or keyboard	Required focus evidence and the affected platform file
   Native code or roots	Native overlay checklist; for iOS or macOS, also Apple and Office rules
   Framework-base pattern	Framework-base instructions

4. Check repository-wide rules: explicit exports (R-REPO-001), platform fork imports (R-REPO-002), changesets (R-REPO-003), no copied private source (R-REPO-004), and the code-scanning patterns: linear-time regular expressions and no values spliced into executed code (R-SCAN-001, R-SCAN-002).
5. Inspect validation claims. Run non-mutating checks only when they add evidence, preferring focused tests over the whole suite (R-VALID-004); never run snapshot-update or fix modes. Record each check as passed, failed, skipped, not run, or unsupported (R-VALID-002).

Output

1. Scope and classification: one line per changed surface.
2. Findings, highest severity first. For each: path:line, the problem, the rule or requirement ID, verified or inferred, and a suggested correction.
3. Validation: commands run with results, and checks not run with the reason.

When nothing qualifies, say so and list what was checked. Do not pad the review with style preferences.