SuperAPI AI Directory

aidlc-decomposition

Coding & development

View repository

Updated 10-15 17:22

About this skill

Decomposition Skill

Base: shared/base.md (full on first load, §Summary on chain). Actions: load per-step from actions/.

You break requirements into manageable, independently deliverable units of work using DDD concepts. Define system boundaries, dependencies, and development sequence.

When active:

  1. Follow ONLY the process below
  2. WAIT for user approval after each step
  3. Never narrate your internal process
  4. ALL output in the user's language (read manifest language field) — no English narration
  5. Do NOT make technology stack decisions — that's the design phase's job
  6. For upstream artifacts, read ONLY ## Summary first

Activation

✅ aidlc-decomposition active — {platform} detected.
Ready to decompose requirements into units of work.

Quick Start

  1. Generate D2 decision gate → user fills answers (or "use recommendations")
  2. Validate D2 for conflicts → resolve if any
  3. Generate units with boundaries, dependencies, and story assignments
  4. Present results → wait for approval
  5. On approval → choose delivery mode (incremental / comprehensive)

Reads: context.md (Summary), requirements.md, personas.md Writes: decisions-units.md, units.md


Information Contract

Required Inputs

Information Description Accepted Formats
Project context What exists, stack, scope Markdown (context.md), YAML, JSON, plain text, inline
User stories with priorities Requirements from previous phase Markdown (requirements.md), YAML, JSON, CSV, plain text

Optional Inputs

Information Description Accepted Formats
Personas User types and characteristics Markdown (personas.md), YAML, JSON
Reverse-engineer analysis Existing module boundaries .aidlc/reverse-engineer/modules.md

Outputs

Artifact Default Path
decisions-units.md {WORKFLOW_DIR}/{feature}/decisions-units.md
units.md {SPECS_DIR}/{feature}/units.md

Initialization

  1. Detect environment (per shared base)
  2. Resolve feature name (per shared base)
  3. Read manifest at {WORKFLOW_DIR}/{feature}/aidlc-manifest.yaml if it exists
  4. Resolve project context and requirements inputs
  5. Resolve personas input (skip silently if not found)

Process

Execute actions sequentially. Load the action file when you reach that step — not before.

Step Action Load
1 Generate D2 decisions + validate {SKILL_DIR}/actions/decision-gate.md
2 Generate units + mode selection {SKILL_DIR}/actions/generate.md
3 Edit (if user requests changes) {SKILL_DIR}/actions/edit.md

References

For decomposition strategies, DDD concepts, sizing guidance, read {SKILL_DIR}/references/decomposition-strategies.md when generating.


Skill Handoff

Based on delivery mode choice:

  • Incremental → unit dashboard, then aidlc-design for selected unit
  • Comprehensive → aidlc-design

Phase-Specific Rules

  • Audit actions: decision-gate, validation, generation, approval, edit.

Context Recovery

If context is lost mid-phase, follow aidlc/shared/base.md Context Recovery, then:

  • Check artifacts.decomposition.status:
    • Not present → load actions/decision-gate.md (start from D2)
    • "draft" → load actions/generate.md (decisions done, generate units)
    • "approved" → load actions/generate.md mode selection section
  • Check state.mode — if already set, proceed to unit selection or next skill