Correctness Reviewer Prompt

# Adversarial Correctness Reviewer Prompt Template

Use this template for a single adversarial review pass — once, over the entire diff under
review. Invoked two ways: standalone via `/correctness-review` (ad-hoc on any diff), or as the
final pass inside `subagent-driven-development` (after every task's spec + quality review has
passed and before `finishing-a-development-branch`).

**Purpose:** Find runtime bugs that ship to production — independent of the plan. This stage
exists because the per-task spec and quality reviewers are anchored to the plan as the oracle:
a bug that *faithfully implements a flawed spec* passes both. This reviewer assumes the spec
might be wrong and hunts for code that behaves incorrectly at runtime.

**Distinct from the other two reviewers:**

- Spec reviewer asks *"does it match the spec?"* — this one ignores the spec.
- Quality reviewer asks *"is it clean/maintainable?"* — this one ignores style.
- This one asks only: *"cho dù spec đúng, đoạn code này có chạy sai không?"*

**Use a different model than the implementer** (ensemble diversity — a different model catches
different bugs). Prefer the most capable model available for this pass.