haiku

H·AI·K·U

Human + AI Knowledge Unification — A lifecycle orchestration system for any structured work.

Studios, stages, and quality gates — disciplined form, reliable results.

What It Is

H·AI·K·U is a Claude Code plugin that structures AI-assisted work into focused, iterative cycles with quality enforcement. It works for any domain — software development, sales, marketing, compliance, incident response, and more.

Four-layer hierarchy:

Studio > Stage > Unit > Bolt

• Studio — A lifecycle template (software, sales, compliance, etc.)
• Stage — A phase within the studio with its own hats and review gate
• Unit — A discrete piece of work with completion criteria
• Bolt — One iteration cycle through the hat sequence for a unit

Installation

/install gigsmart/haiku-method

No external dependencies. The plugin ships a single binary (plugin/bin/haiku) that handles MCP tools, hooks, and state management.

Quick Start

/haiku:new              # Create a new intent — describe what you want to do
/haiku:resume              # The orchestrator tells you what to do next

That's it. The orchestrator (haiku_run_next) drives the stage loop. You follow the actions it returns.

12 Built-In Studios

Engineering

Studio	Stages
Software	inception → design → product → development → operations → security
Data Pipeline	discovery → extraction → transformation → validation → deployment
Migration	assessment → mapping → migrate → validation → cutover
Incident Response	triage → investigate → mitigate → resolve → postmortem
Compliance	scope → assess → remediate → document → certify
Security Assessment	reconnaissance → enumeration → exploitation → post-exploitation → reporting

Go-to-Market

Studio	Stages
Sales	research → qualification → proposal → negotiation → close
Marketing	research → strategy → content → launch → measure
Customer Success	onboarding → adoption → health-check → expansion → renewal
Product Strategy	discovery → user-research → prioritization → roadmap → stakeholder-review

General Purpose

Studio	Stages
Ideation	research → create → review → deliver
Documentation	audit → outline → draft → review → publish

The Stage Loop

Every stage runs the same five-step cycle:

1. Decompose — Break the stage into units with completion criteria and a dependency DAG
2. Execute — For each unit, run the bolt loop through the stage's hat sequence
3. Adversarial Review — Spawn review agents (per-stage + cross-stage) to verify the work
4. Persist — Save stage outputs to their scoped locations
5. Gate — Evaluate the review gate: auto, ask, external, or await

Commands

Command	Purpose
/haiku:new	Create a new intent
/haiku:resume	Advance through stages (orchestrator-driven)
/haiku:composite	Create multi-studio intent with sync points
/haiku:refine	Amend specs mid-execution or refine upstream stages
/haiku:review	Pre-delivery code review
/haiku:reflect	Post-completion analysis
/haiku:operate	Post-delivery operational tasks
/haiku:capacity	Historical throughput analysis
/haiku:triggers	Poll providers for events
/haiku:setup	Configure providers and quality gates
/haiku:new --template <name>	Create from a template (10 templates across 7 studios)

Architecture

One binary: plugin/bin/haiku

Mode	Usage	Purpose
haiku mcp	MCP server (stdio)	16 state tools + orchestrator + review tools
haiku hook <name>	Hook execution	10 hooks (quality gates, context injection, etc.)
haiku migrate	CLI command	Migrate .ai-dlc/ intents to .haiku/

Source in packages/haiku/ (TypeScript), compiled to a single bundled binary.

State Model

State lives in committed artifacts, not ephemeral files:

Level	Location	Key Fields
Intent	intent.md frontmatter	studio, active_stage, status, started_at, completed_at
Stage	stages/{stage}/state.json	phase, status, started_at, gate_outcome
Unit	unit-*.md frontmatter	bolt, hat, status, started_at, completed_at

MCP tools (haiku_intent_*, haiku_stage_*, haiku_unit_*) are the primary interface.

Providers

Six bidirectional provider categories for external tool integration: