Cook — Backend Workflow

Structured workflow for backend projects. Every non-trivial task flows through ordered phases.

Plan → Code → Review → Test

---

Memory Bank — Required for Every Session

At the start of EVERY session, read all files in memory-bank/:

1. projectbrief.md — Core requirements and goals
2. productContext.md — Problem and solution context
3. techContext.md — Technical setup and dependencies
4. systemPatterns.md — Architecture and design patterns
5. activeContext.md — Current work focus and recent changes
6. progress.md — Completed work and known issues

If memory-bank/ does not exist, create it with the files above based on what you learn from the codebase.

After completing work (end of Test phase or any stopping point), update:

• activeContext.md — What was done, current state, next steps
• progress.md — Mark completed items, add new known issues

This ensures continuity across sessions. Never skip this step.

---

[PLAN] Plan

1. Read requirements. Ask if ambiguous.
2. Identify affected routes, services, models, and migrations.
3. Break into subtasks. Define acceptance criteria.
4. Identify API contract changes (request/response shapes).

Output: Numbered plan with affected files and acceptance criteria. Rule: No code in this phase.

[CODE] Code

1. Implement data layer first (models, migrations, schemas).
2. Then service/business logic.
3. Then API routes/controllers.
4. Handle errors at boundaries — validate input, catch DB errors, return proper status codes.
5. Update related code (types, middleware, configs).

Rules:

• Match existing patterns. Don't invent new conventions.
• No features beyond scope. No premature abstractions.

[REVIEW] Review

1. API contract — Do endpoints match the spec? Correct HTTP methods and status codes?
2. Validation — Is all user input validated before reaching business logic?
3. Security — No SQL injection, no exposed secrets, proper auth checks?
4. Error handling — Do all error paths return meaningful responses?
5. Types — Are types precise? No any leaks? Proper null handling?
6. Performance — No N+1 queries? Proper indexing considered?

Fix issues immediately. If a design flaw is found, loop back to Plan.

[TEST] Test

1. Unit tests — Test service/business logic in isolation. Mock external dependencies.
2. Integration tests — Test API endpoints end-to-end (request → response). Test with a real or in-memory database where possible.
3. Run the full test suite — ensure nothing is broken.
4. Run type checking and linting.

Rules:

• Test behavior, not implementation. Test error paths, not just happy paths.
• Never mark done with failing tests.

---

Iteration

If any phase reveals problems:

• Test failures from a bug → Fix in Code, re-run Review + Test.
• Review reveals missing requirements → Loop back to Plan.
• User feedback changes scope → Restart from Plan.

Always note which phase you're returning to and why.

Phase Markers

Prefix progress with the current phase:

[PLAN] Analyzing requirements...
[CODE] Implementing service layer...
[REVIEW] Self-reviewing changes...
[TEST] Running unit and integration tests...