AI Code Review

Use for

L

,

XL

, or explicit review requests. Default to one strong pass. Use a second pass only for high-risk changes.

---

Scope

• Only review files in git diff.
• Read the change, not just the final file state.
• Review by risk, not by style preference.

---

Focus Areas

Correctness

• Verify all callers still work correctly.
• Verify all inheritors / overrides are consistent.
• Verify abstract layer contracts are met.
• Check for missing null handling.
• Check for missing edge cases.
• Check for regressions in existing behavior.

Quality

• Code matches existing conventions in the repo.
• No unnecessary changes outside stated scope.
• No dead code introduced.
• Naming is clear and consistent.
• No performance regressions.
• Minimal code with no extra abstractions.

Risk

• Check whether validation matches the task size.
• Check whether tests are missing for risky behavior changes.
• Check whether user-visible flows or shared contracts changed silently.

---

Output Format

List findings first. Skip praise unless it adds real signal.

### [Title]

- **File**: path/to/file.dart:L42
- **Description**: one-line summary.
- **Impact**: which callers / modules affected.
- **Severity**: critical / warning / suggestion
- **Recommendation**: what to fix.

---

Summary

Files reviewed: N
Findings: X critical, Y warning, Z suggestion
Verdict: APPROVED / CHANGES_REQUESTED

---

Rules

• Prioritize real bugs, regressions, and risk gaps.
• Do not flag style preferences already handled by lint.
• Critical issues found -> return to IMPLEMENT.
• No critical issues -> APPROVED -> proceed to FINALIZE.