Menu
Use before claiming work is complete — run verification commands and cite output before any success claim
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Appliquer les règles de développement et la mémoire technique durable de gba_translator/Pokémon Unbound. Utiliser pour toute tâche dans ce dépôt, notamment avant de planifier, modifier, tester, déboguer ou relire le pipeline ROM, les traductions FR/IT/DE, l'encodage CFRU, les patchs binaires, l'émulateur mGBA ou le workflow Git.
Use when fixing any bug or patch script in gba_translator — before closing the ticket, verify the fix does not break (or silently already affects) the other registered languages (FR/IT/DE/Indie). Covers deciding whether a change is shared or per-language, which test paths and rebuild targets to run, and how to check a base-ROM quirk against every language.
Use when building or rebuilding the French ROM of Pokémon Unbound end-to-end — from a translation_ready.json through make build-fr, the post-build patch chain, and validation.
Use before any new feature, pipeline change, or architectural decision — explores intent and design before implementation
Review a diff against project rules and acceptance criteria before merging
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes
Baseado na classificação ocupacional SOC
| name | verification-before-completion |
| description | Use before claiming work is complete — run verification commands and cite output before any success claim |
| when_to_use | Before any "done", commit, PR, or task completion claim |
Claiming work is complete without verification is dishonesty, not efficiency.
Core principle: Evidence before claims, always.
Project rule: 100% test pass rate is mandatory. "Looks good" is not evidence.
NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE
BEFORE claiming any status or expressing satisfaction:
1. IDENTIFY: What command proves this claim?
2. RUN: Execute the FULL command (fresh, complete)
3. READ: Full output, check exit code, count failures
4. VERIFY: Does output confirm the claim?
- If NO: State actual status with evidence
- If YES: State claim WITH evidence
5. ONLY THEN: Make the claim
Skip any step = lying, not verifying
# Full test suite — must show 0 failures, 0 errors
pytest tests/ -v
# Unit tests only
pytest tests/unit/ -v
# Integration tests (some require mGBA — skip if not available)
pytest tests/integration/ -v
# With coverage
pytest --cov=src tests/
# Check no stray files at repo root
ls *.py 2>/dev/null && echo "FAIL: py files at root" || echo "OK"
ls *.md 2>/dev/null | grep -v README.md && echo "FAIL: md files at root" || echo "OK"
| Claim | Requires | Not Sufficient |
|---|---|---|
| Tests pass | pytest tests/ → 0 failures | "Should pass", previous run |
| Feature works | Test exercising the feature passes | Code reviewed and looks correct |
| Bug fixed | Test reproducing the bug now passes | Code changed, assumed fixed |
| 100% pass rate | pytest tests/ → 100% of N tests | "All critical tests pass" |
| ROM output correct | Injection + decode round-trip passes | Manual visual inspection |
| No regression | Full pytest tests/ after change | Subset of tests passes |
pytest tests/tests/unit/ and claiming full passpytest tests/integration/ too..py or .md files landed at repo root.Run the command. Read the output. THEN claim the result.
✅ $ pytest tests/
534 passed, 0 failed, 0 errors in 12.3s
→ "All tests pass."
❌ "Tests should pass now."
❌ "I've verified the fix works."