wisdom-accumulation

Wisdom Accumulation — Cross-Wave Learning

Use this skill to extract learnings after each implementation wave and pass them to the next wave. Prevents repeating mistakes and ensures consistent patterns across backend, frontend, and database layers.

---

The Core Principle

Learn once, apply everywhere.

When Hermes discovers a pattern, Aphrodite should know about it. When Demeter finds a gotcha, Hermes should avoid it. Wisdom Accumulation makes this happen automatically.

---

What Gets Extracted

After each wave completes, extract learnings into 5 categories:

1. Conventions

Patterns and standards discovered during implementation:

- Use async/await em todos os I/O
- Validação com Pydantic v2
- Repository pattern com SQLAlchemy 2.0
- API retorna snake_case, frontend converte para camelCase

2. Successes

What worked well and should be repeated:

- Factory pattern para testes de usuário funcionou bem
- Dependency injection via FastAPI Depends() é limpo
- Cursor-based pagination é mais eficiente que offset

3. Failures

What didn't work and should be avoided:

- Não usar session.commit() em async — usar await session.flush()
- Não importar models diretamente nos routers — usar schemas
- Evitar nested queries — usar joinedload para N+1

4. Gotchas

Surprises and edge cases discovered:

- O endpoint /users tem rate limit de 100 req/min
- Redis cache TTL deve ser < session timeout
- Arquivos >10MB precisam de multipart upload

5. Commands

Useful commands discovered during implementation:

- Test runner: `pytest tests/ -v --cov=src`
- DB migration: `alembic upgrade head`
- Lint: `ruff check src/ --fix`

---

Storage

Learnings are stored in:

.pantheon/learnings/<feature>/learnings.md

Lifecycle:

• Created: When first wave completes
• Updated: After each subsequent wave
• Deleted: After feature is merged (temporary, not permanent memory)

---

Format