Inception

Ask review

Understand the problem, define success, and elaborate into units

Hats
3
Review Agents
2
Review
Ask
Unit Types
Research
Inputs
None

Hat Sequence

1

Architect

Focus: Understand the problem space, map the existing codebase, define scope and constraints, and identify technical risks and architectural implications. The architect produces a discovery document that gives downstream stages the context they need.

Produces: Discovery document with domain model, technical landscape, constraint analysis, and risk assessment.

Reads: Intent problem statement, codebase structure, existing project knowledge.

Anti-patterns (RFC 2119):

  • The agent MUST NOT jump to solutions before understanding the problem
  • The agent MUST NOT assume architecture without reading existing code
  • The agent MUST NOT ignore non-functional requirements (performance, security, accessibility)
  • The agent MUST NOT over-design at the discovery phase — this is understanding, not design
  • The agent MUST document what exists before proposing what should change
2

Decomposer

Focus: Break the intent into units with clear boundaries, define the dependency DAG, and write verifiable completion criteria for each unit. Each unit should be completable within a single bolt.

Produces: Unit specs with completion criteria, dependencies, and scope boundaries.

Reads: Architect's discovery output via the unit's ## References section.

Anti-patterns (RFC 2119):

  • The agent MUST NOT create units that are too large (more than one bolt to complete)
  • The agent MUST NOT create units with circular dependencies
  • The agent MUST NOT write vague criteria ("it works", "tests pass")
  • The agent MUST define clear boundaries between units
  • The agent MUST NOT elaborat by layer (all backend, then all frontend) instead of by feature slice
3

Elaborator

Focus: Break the intent into units with clear boundaries, define the dependency DAG, and write verifiable completion criteria for each unit. Each unit should be completable within a single bolt.

Produces: Unit specs with completion criteria, dependencies, and scope boundaries.

Reads: Architect's discovery output via the unit's ## References section.

Anti-patterns (RFC 2119):

  • The agent MUST NOT create units that are too large (more than one bolt to complete)
  • The agent MUST NOT create units with circular dependencies
  • The agent MUST NOT write vague criteria ("it works", "tests pass")
  • The agent MUST define clear boundaries between units
  • The agent MUST NOT elaborat by layer (all backend, then all frontend) instead of by feature slice

Review Agents

Completeness

Mandate: The agent MUST verify the discovery document fully maps the problem space and that unit elaboration covers the intent with no gaps or overlaps.

Check:

  • The agent MUST verify that every entity, relationship, and technical constraint from the intent is addressed in the discovery document
  • The agent MUST verify that all units have verifiable completion criteria (specific commands or tests, not vague assertions)
  • The agent MUST verify that unit DAG is acyclic with no orphans — every unit either produces inputs for another or delivers a final output
  • The agent MUST verify that no unit is too large for a single bolt cycle
  • The agent MUST verify that no critical path is missing (e.g., auth, data migration, error handling)

Feasibility

Mandate: The agent MUST challenge whether the elaboration is technically achievable given the codebase, dependencies, and constraints discovered.

Check:

  • The agent MUST verify that proposed approaches are compatible with existing architecture (frameworks, patterns, conventions)
  • The agent MUST verify that external dependencies (APIs, libraries, services) exist and support the planned usage
  • The agent MUST verify that no unit assumes capabilities that the codebase does not have or cannot reasonably acquire
  • The agent MUST verify that risk assessment covers the highest-impact failure modes, not just the obvious ones

Inception

Criteria Guidance

Good criteria examples:

  • "Discovery document maps all entities with their fields and relationships"
  • "Each unit has 3-5 completion criteria, each verifiable by a specific command or test"
  • "Unit DAG has no circular dependencies — verified by topological sort"

Bad criteria examples:

  • "Domain is understood"
  • "Units have criteria"
  • "Elaboration is complete"

Overlap Awareness

During elaboration, check for other active H·AI·K·U branches working on overlapping files:

# List active haiku branches and their changed files
for branch in $(git branch -r --list 'origin/haiku/*/main' 2>/dev/null); do
  changed=$(git diff --name-only main...$branch 2>/dev/null)
  [ -n "$changed" ] && echo "Branch: $branch" && echo "$changed"
done

If overlap is detected with files this intent plans to modify, note it in the discovery document. Not a blocker — just awareness for the architect hat to factor in.

Completion Signal (RFC 2119)

Discovery document MUST exist with domain model and technical landscape. All units MUST have specs with dependencies and verifiable completion criteria. Unit DAG MUST be acyclic. Each unit MUST be scoped to complete within a single bolt.