AI SDLC: The Asset Graph Model

A formal system for AI-augmented software development — 4 primitives, 1 operation, valid projections at every scale.

Foundation: Constraint-Emergence Ontology

What Is This?

A methodology that reduces all software development to:

Primitive	What it is
Graph	Topology of typed assets with admissible transitions
Iterate	Convergence engine — the only operation
Evaluators	Convergence test — {Human, Agent, Deterministic Tests}
Spec + Context	Constraint surface — what bounds construction

Everything else — stages, agents, TDD, BDD, projections — is parameterisation of these four primitives.

The Core Operation

iterate(Asset, Context[], Evaluators) → Asset'

while not stable(candidate):
    candidate = iterate(candidate, context, evaluators)
return promote(candidate)    // candidate becomes stable Markov object

The SDLC Graph (one projection)

Intent → Spec → Design → Code ↔ Tests → UAT → CI/CD → Telemetry
           ↑       ↑                                        │
           │       └── feedback (tech ambiguity) ───────────┤
           └──────── feedback (business ambiguity) ─────────┘

This graph is not universal — it is one domain-specific instantiation. The four primitives are universal; the graph is parameterised.

Feature Lineage

Every asset carries REQ keys from spec to runtime:

Spec:       REQ-F-AUTH-001 defined
Design:     Implements: REQ-F-AUTH-001
Code:       # Implements: REQ-F-AUTH-001
Tests:      # Validates: REQ-F-AUTH-001
Telemetry:  logger.info("login", req="REQ-F-AUTH-001", latency_ms=42)

One identifier, end to end. Feature views are generated at every stage by grepping the REQ key across artifacts.

Key Concepts

Projections

The formal system is a generator of valid methodologies. Each valid instance preserves the 4 invariants while varying:

Graph — from 2 nodes / 1 edge to the full SDLC
Evaluators — which types are active per edge
Convergence — what "done" means (all checks pass, question answered, time box expired)
Context — how many constraints bound construction

A 10-minute spike and a regulated medical device both use the same four primitives with different parameterisation.

Vector Types

Type	Purpose	Convergence
Feature	Deliver a capability	All required checks pass
Discovery	Answer a question	Question answered or time box expired
Spike	Assess technical risk	Risk assessed or time box expired
PoC	Validate feasibility	Hypothesis confirmed/rejected
Hotfix	Fix production issue	Fix verified in production

Spec / Design Separation

Spec — WHAT the system does, tech-agnostic. One spec, many designs.
Design — HOW architecturally, bound to technology. ADRs, ecosystem binding.
Code — HOW concretely, eventually fully automated. Disambiguation feeds back to Spec (business gap) or Design (tech gap).

Getting Started

1. Install (one command)

# From your project directory
curl -sL https://raw.githubusercontent.com/foolishimp/ai_sdlc_method/main/imp_claude/code/installers/gen-setup.py | python3 -

This creates the plugin config, hook scripts, workspace, graph topology, project constraints, and Genesis Bootloader. Restart Claude Code to load the plugin.

# Verify
curl -sL https://raw.githubusercontent.com/foolishimp/ai_sdlc_method/main/imp_claude/code/installers/gen-setup.py | python3 - verify

2. Start working (two commands)

/gen-start            # Detects state, selects feature/edge, iterates — "Go."
/gen-status           # Project-wide state, "you are here", signals — "Where am I?"

/gen-start handles everything: init, feature creation, edge selection, iteration. It detects your project state and routes to the right action automatically.

3. Power-user commands

/gen-iterate          # Advance an asset along a specific edge
/gen-spawn            # Spawn a new feature/spike/hotfix vector
/gen-trace            # Trace REQ keys across artifacts
/gen-gaps             # Find traceability gaps
/gen-checkpoint       # Save current progress
/gen-review           # Review an asset for promotion
/gen-spec-review      # Gradient check at spec boundaries
/gen-escalate         # View/process escalation queue
/gen-zoom             # Zoom into/out of graph edges
/gen-release          # Prepare a release

Reading the spec (no tooling needed)

If you just want to understand the methodology, start with:

INTENT.md — why this exists
AI_SDLC_ASSET_GRAPH_MODEL.md — the formal system
PROJECTIONS_AND_INVARIANTS.md — how it adapts to different scales

Repository Structure

ai_sdlc_method/
│
├── specification/                          # SHARED — the formal system (tech-agnostic)
│   ├── INTENT.md
│   ├── AI_SDLC_ASSET_GRAPH_MODEL.md
│   ├── PROJECTIONS_AND_INVARIANTS.md
│   ├── AISDLC_IMPLEMENTATION_REQUIREMENTS.md
│   ├── FEATURE_VECTORS.md
│   └── presentations/
│
├── imp_claude/                             # Claude Code implementation
│   ├── design/                             #   AISDLC_V2_DESIGN.md + ADRs 008-017
│   ├── code/                               #   Plugin: 4 agents, 13 commands, 4 hooks
│   │   └── installers/gen-setup.py         #   One-command installer (curl | python3)
│   └── tests/                              #   521+ tests (spec + implementation + installer)
│
├── imp_gemini/                             # Gemini Genesis implementation
│   ├── design/                             #   GEMINI_GENESIS_DESIGN.md + ADRs GG-001-008
│   ├── code/                               #   (future)
│   └── tests/                              #   (future)
│
├── imp_codex/                              # Codex Genesis implementation
│   ├── design/                             #   CODEX_GENESIS_DESIGN.md + ADR-CG-001+
│   ├── code/                               #   (future)
│   └── tests/                              #   (future)
│
├── imp_bedrock/                            # AWS Bedrock Genesis implementation
│   ├── design/                             #   BEDROCK_GENESIS_DESIGN.md + ADRs AB-001-008
│   ├── code/                               #   (future)
│   └── tests/                              #   (future)
│
├── docs/analysis/                          # Cross-cutting analysis
├── .ai-workspace/                          # Runtime workspace state
├── CLAUDE.md                               # Project guide
└── README.md                               # This file

Documentation

Document	What it covers
INTENT.md	Business intent and motivation
AI_SDLC_ASSET_GRAPH_MODEL.md	Formal system — 4 primitives, 1 operation, Hilbert space structure
PROJECTIONS_AND_INVARIANTS.md	Projections, vector types, spawning, fold-back, time-boxing
AISDLC_IMPLEMENTATION_REQUIREMENTS.md	Platform-agnostic implementation requirements
FEATURE_VECTORS.md	Feature decomposition for building the methodology itself
AISDLC_V2_DESIGN.md	Claude Code implementation design
GEMINI_GENESIS_DESIGN.md	Gemini Genesis implementation design
CODEX_GENESIS_DESIGN.md	Codex Genesis implementation design
BEDROCK_GENESIS_DESIGN.md	AWS Bedrock Genesis implementation design

Status

Version: 2.8.0 (Asset Graph Model — Multi-Tenant)

Spec: Complete (formal system, projections, invariants, consciousness loop, processing phases, sensory systems, UX)
Design: Complete (Claude ADRs 008-017, Gemini ADRs GG-001-008, Codex ADR-CG-001+, Bedrock ADRs AB-001-008)
Code: Phase 1a (Claude: configs, 4 agents, 13 commands, 4 hooks, installer)
Tests: 521+ tests (spec + implementation + installer), 34 E2E convergence tests
UAT / CI/CD / Telemetry: Not started

License

MIT

Acknowledgments

Foundation: Constraint-Emergence Ontology
ai_init (Key Principles, integrated as Context[])
Built with: Claude Code, Gemini CLI, Codex

Name	design-with-traceability
Description	Create technical solution architecture from requirements with REQ-* traceability. Designs components, APIs, data models, and interactions. Tags all design artifacts with requirement keys. Use after requirements are validated, before coding starts.

design-with-traceability

SKILL.md