Implement features using architect → kraken → arbiter chain
Orchestrates feature implementation using architect, kraken, and arbiter agents with testing.
/plugin marketplace add sethdford/claude-toolkit/plugin install workflows@claude-toolkit[mode] <description>Plan and implement features with proper testing.
/build "user authentication with JWT"
/build greenfield "new dashboard component"
/build brownfield "add caching to existing API"
/build tdd "implement retry logic"
| Mode | When to Use |
|---|---|
greenfield | New feature from scratch |
brownfield | Modifying existing code |
tdd | Test-driven development |
Default: Infers from context (new files = greenfield, existing = brownfield)
Spawn the architect agent to plan:
Use Task tool:
subagent_type: "architect"
prompt: "Plan implementation: $DESCRIPTION
1. Analyze requirements
2. Research existing patterns in codebase
3. Design the solution architecture
4. Create implementation plan with:
- Files to create/modify
- Key components
- Data flow
- Dependencies
5. Identify risks and edge cases"
Spawn kraken to implement:
Use Task tool:
subagent_type: "kraken"
prompt: "Implement the plan from architect:
Plan: [from architect]
Follow TDD approach:
1. Write tests first (for testable components)
2. Implement to make tests pass
3. Refactor for clarity
4. Document key decisions"
Spawn arbiter to verify:
Use Task tool:
subagent_type: "arbiter"
prompt: "Verify implementation:
1. Run all new tests
2. Run related existing tests
3. Check code coverage
4. Verify requirements are met
5. Report any gaps"
For significant changes, spawn judge:
Use Task tool:
subagent_type: "judge"
prompt: "Review the implementation:
1. Code quality check
2. Pattern adherence
3. Security review
4. Performance considerations
5. Suggestions for improvement"
| Option | Effect |
|---|---|
--skip-plan | Skip planning phase (have clear spec) |
--skip-review | Skip review phase |
--no-commit | Don't auto-commit |
--parallel | Run research in parallel |
After completion: