Requirement to Implementation

Structured workflow that turns a requirement into verified, documented code. Every task (feature, bug fix, refactor, chore, or performance improvement) flows through the same phases.

Phase 0: Intake

Identify the requirement source and extract key information.

Produce a one-paragraph requirement summary confirming your understanding. Ask the user about literally anything:technical implementation, UI & UX, concerns, tradeoffs, etc. but make sure the questions are not obvious, be very in-depth and continue interviewing me continually until it's complete, then write the plan to the file.

Phase 1: Context Loading

Read the Memory Bank to ground yourself in project state:

1. .cursor/memory-bank/projectbrief.md: scope and goals
2. .cursor/memory-bank/productContext.md: product context
3. .cursor/memory-bank/techContext.md: tech stack and constraints
4. .cursor/memory-bank/systemPatterns.md: architecture and patterns
5. .cursor/user-memory-bank/activeContext.md: current focus
6. .cursor/user-memory-bank/progress.md: what's done, what's pending

If any file is missing or stale, note it; you'll update them in Phase 5.

Phase 2: Exploration & Analysis

Explore the codebase to map the blast radius of the change:

1. Identify affected files: entry points, services, APIs, jobs, UI, tests
2. Read existing patterns: find similar implementations to follow
3. Spot risks: PII/sensitive data exposure, authorization gaps, integration side effects, migration needs
4. Note test coverage: existing tests that need updating, gaps that need filling

Use exploratory subagents for broad investigations. Use targeted searches for specific lookups.

Phase 3: Plan & Approve

Present a structured plan to the user. Do not implement until approved.

Plan Template

## Requirement
[One-sentence summary]

## Approach
[High-level description of the solution]

## Changes
1. [File/area]: [what changes and why]
2. [File/area]: [what changes and why]
...

## Verification Strategy
- [ ] [How you'll verify correctness (tests, manual checks, etc.)]
- [ ] [Specific test files to create or update]

## Risks & Considerations
- [Anything the user should be aware of]

## Questions (if any)
- [Clarifying questions that surfaced during exploration]

Ask 4-6 clarifying questions if ambiguities remain. Wait for answers before finalizing the plan.

Once the user approves, proceed to implementation.

Phase 4: Implement

1. Create a TodoWrite task list mirroring the approved plan's changes
2. Mark each todo in_progress as you begin it, completed when done
3. Work step by step: one todo at a time, verifying as you go

Verification (flexible per task)

Choose the verification approach that fits:

After each meaningful change, run the project's test command for affected paths (for example npm test, pytest, go test ./..., bundle exec rspec; use what the host repo documents). Fix failures before moving on.

Implementation Principles

• Follow existing patterns found in Phase 2
• One concern per change: don't mix unrelated modifications
• Check lints after substantive edits
• If a step reveals the plan was wrong, stop and re-plan with the user

Phase 5: Document

After implementation is complete and verified:

1. Update .cursor/user-memory-bank/activeContext.md: reflect current focus and recent changes
2. Update .cursor/user-memory-bank/progress.md: mark what's done, note what's next
3. Update .cursor/memory-bank/systemPatterns.md: if new patterns were introduced
4. Update .cursor/memory-bank/techContext.md: if tech stack or constraints changed

Quick Reference: Phase Flow

Intake → Context → Explore → Plan → [Approve] → Implement → Document
  ↑                            ↓
  └──── Re-plan if needed ─────┘

Cardinal rule: No implementation without an approved plan. Plans are cheap; rework is expensive.

Guardrails

• No implementation without an approved plan: present the plan and wait for explicit approval before writing production code.
• Do not skip intake or exploration on an ambiguous requirement; map the blast radius first.
• Commit or push only on explicit user request; never print or commit secrets.