This checklist walks you through completing your first intent with H·AI·K·U. Follow each section in order.

Before You Start

Complete these steps before beginning your first intent:

Environment Setup

Claude Code or Cowork installed and working
H·AI·K·U plugin installed (/plugin marketplace add thebushidocollective/ai-dlc then /plugin install haiku@thebushidocollective-ai-dlc --scope project)
Verified plugin is active (commands like /haiku:elaborate are recognized)

Project Selection

Chosen a real project (not a toy project)
Project has clear goals (you know what you want to build)
Low time pressure (room for learning)
You're familiar with the codebase (one new thing at a time)

Mental Preparation

Blocked 2-4 hours of focused time
Prepared to follow the process even if it feels slow
Ready to complete at least one full unit

Phase 1: Elaboration

Use /haiku:elaborate to define your intent collaboratively with Claude.

Start Elaboration

Run /haiku:elaborate
Describe what you want to build in 2-3 sentences

Define the Intent

Review Claude's clarifying questions
Answer questions about scope, constraints, and requirements
Refine the description until it's clear

Write Completion Criteria

Define Units

Review proposed unit breakdown
Verify units are:
- Cohesive (related stories grouped together)
- Loosely coupled (minimal dependencies between units)
- Independently completable (each can be finished and tested)
Adjust unit boundaries if needed
Confirm unit ordering and dependencies

Verify Artifacts

Elaboration Complete

Intent is clear and well-defined
Units are appropriately scoped
Ready to begin execution

Phase 2: First Unit

Work through your first unit using /haiku:run.

Start Execution

Run /haiku:run
Verify Claude loaded the correct unit

Planner Hat

Builder Hat

Claude begins implementation
Monitor progress (you're in OHOTL mode by default)
Redirect if going off-track
Answer questions if Claude asks

Quality Gate Checks:

Tests written for new functionality
No TypeScript/lint errors introduced
Code follows project conventions

Reviewer Hat

Claude performs self-review
Review completion criteria status:

## Completion Criteria Status
- [x] Criterion 1 - verified by [method]
- [x] Criterion 2 - verified by [method]
- [ ] Criterion 3 - FAILED: [reason]

If criteria failed:
- Claude proposes fixes
- Back to Builder to implement fixes
- Re-review until all pass

First Unit Complete

All completion criteria satisfied
Quality gates pass (tests, types, lint)
Unit file updated with status: complete
Changes committed with unit reference
Ready for next unit or intent completion

Phase 3: Iteration

Continue with remaining units.

For Each Remaining Unit

Run /haiku:run to load next unit
Review any dependencies are satisfied
Complete Planner → Builder → Reviewer cycle
Commit with unit reference
Update unit status

Between Units

Check if context is getting heavy
/clear if needed, then /haiku:run to continue
Verify committed artifacts reflect current state

Handling Blockers

If you hit a blocker during any unit:

Document the blocker in unit file
Categorize: Technical / Requirement / External
Decide: Can work around / Need help / Block unit
If blocking: Move to next available unit
If workable: Document assumption, continue

Phase 4: Completion

Wrap up the intent.

Final Review

All units marked complete
intent.md overall criteria satisfied
All quality gates pass across all units
No outstanding blockers

Documentation

README updated if needed
API documentation current (if applicable)
Any new patterns documented

Commit and PR

Final commit with completion message
PR created (if applicable)
PR description references intent and units
Ready for team review

Cleanup

Remove any scratch/debug code
Verify no console.log/debug statements left
Clean up any temporary files

Phase 5: Reflection

Learn from the experience.

Self-Assessment

Answer these questions honestly:

Process:

Did the hat transitions feel natural?
Did you stay in your hat or drift?
Was the Researcher phase valuable?
Did Review catch any issues?

Criteria Quality:

Were criteria specific enough?
Any criteria that were unclear in hindsight?
Any important criteria that were missing?

Tooling:

Any friction with the commands?
Did /clear + /haiku:run work smoothly?
Were artifacts helpful?

Metrics to Note

Record for future comparison:

Time to complete intent: ____
Number of units: ____
Blocker count: ____
Criteria changes during work: ____

Improvements for Next Time

Based on this experience:

What would you do differently?
What worked well to repeat?
Any conventions to establish?

Quick Reference

Commands

Command	When to Use
`/haiku:elaborate`	Start new intent
`/haiku:run`	Continue/start unit work
`/researcher`	Need more understanding
`/planner`	Need to redesign approach
`/builder`	Ready to implement
`/reviewer`	Ready to verify
`/clear`	Context too heavy

Criteria Checklist

Hat Flow

Researcher → Planner → Builder → Reviewer → Next Unit
     ↑           ↑         ↓         ↓
     └───────────┴─────────┘         │
           (when needed)             │
                                     ↓
                               Unit Complete

Congratulations!

You've completed your first H·AI·K·U intent. The process likely felt slower than "just coding," but consider:

How many issues were caught before they became bugs?
How clear is the documentation of what was built?
How easy would it be for someone else to understand this work?

The next intent will be faster. By your fifth intent, the process will feel natural.

Next steps:

Complete 2-3 more intents to build fluency
Try a different workflow (TDD, Hypothesis)
Share your experience with teammates