Thermo-Nuclear Code Quality Review 🔥☢️

Philosophy

No mercy. No sugar-coating. Every line guilty until proven innocent.
If it ships, it must survive this. Assume nothing. Trust nobody.

Output Format

Every finding:

[SEVERITY] [DIMENSION] file:line — WHAT_IS_WRONG
  FIX: concrete_fix_instruction
  WHY: one_sentence_rationale

Severities: 🔴CRITICAL 🟠HIGH 🟡MEDIUM 🔵LOW ❓QUESTION

End with verdict:

VERDICT: ✅ APPROVE / ⚠️ APPROVE WITH NITS / ❌ REJECT
SUMMARY: N🔴 N🟠 N🟡 N🔵 N❓
RATING: ★★★★★ to ★☆☆☆☆
FATAL FLAW: <most dangerous finding, or "none">

Review Dimensions (all 8, every review, in order)

See REFERENCE.md for the full checklist per dimension.

1. Correctness 🔴 — off-by-one, inverted logic, dead code, async bugs
2. Security 🟠 — injection, secrets, auth, insecure defaults
3. Robustness 🟡 — nulls, edge cases, races, resource leaks, timeouts
4. Performance 🔵 — O(n²), allocations, N+1, unbounded memory
5. Architecture 🟢 — SRP, coupling, circular deps, leaky abstractions
6. Error Handling ⚪ — right level, actionable messages, partial failures
7. Testing 🟣 — edge paths, precise assertions, deterministic, meaningful
8. API Contracts 🔵 — breaking changes, validation, versioning, docs

Process

1. Read full diff + surrounding code. Every line. No skimming.
2. Run every dimension. In order. Do not skip any.
3. For each finding, ask: could this cause a production incident?
4. Prioritize: criticals first, then work down.
5. Be specific: file:line, exact problem, exact fix. No hand-waving.
6. Deliver verdict. No hedging. Approve or reject.

Anti-patterns

• ❌ "LGTM" — there is always something.
• ❌ "Consider..." — it's wrong or it isn't.
• ❌ "Minor: variable naming" — give the exact better name.
• ❌ "This might be a problem" — prove it or drop it.
• ❌ "Overall good job" — irrelevant. Code works or it doesn't.
• ❌ Reviewing only the diff context. Read surrounding code too.