/edge-case-hunter — Edge Case Discovery

Protocol

Given code or a diff, enumerate ALL execution paths systematically:

1. Happy path — nominal flow, expected inputs
2. Error paths — what can throw, what can fail
3. Boundary conditions — off-by-one, min/max values, empty vs single vs many
4. Concurrent access — race conditions, shared state, parallel mutations
5. Null/undefined — missing fields, optional chaining gaps, uninitialized state
6. Type coercion — implicit JS coercion, NaN propagation, string/number confusion
7. RTL edge cases — bidirectional text, mixed Hebrew/English, number direction
8. Numeric overflow — integer limits, float precision, currency arithmetic
9. Empty collections — zero-length arrays, empty maps, null vs empty distinction
10. Async/await — unhandled rejections, race conditions, stale closures, cancellation

Must find at least 10 edge cases or explicitly explain why fewer exist.

Format

Output one JSON object per edge case:

{
  "location": "file.ts:42 — functionName()",
  "trigger_condition": "When user.profile is null and role check is called",
  "guard_snippet": "if (!user?.profile?.role) return null;",
  "potential_consequence": "TypeError: Cannot read property 'role' of null — crashes auth middleware",
  "severity": "BLOCKER"
}

Severity levels: BLOCKER | CRITICAL | MAJOR | MINOR

After the JSON list, provide a summary table:

Rules

• Minimum 10 edge cases per review or justify why fewer
• Check every function parameter for null/undefined/wrong-type
• Trace async flows end-to-end — find every await that can reject silently
• For RTL: check every string that might contain Hebrew, every directional icon, every left/right CSS value
• For React: check every useEffect dependency array, every conditional hook call
• For Supabase: check every RLS policy gap, every missing .error check
• Never say "this looks safe" — prove it or flag it