Skip to main content

ASDD System Architecture

This section is the visual architecture reference for ASDD v5.0/6.0. All diagrams are rendered from Mermaid source.


1. The Big Picture: ASDD Master System

The complete operating model showing how product strategy, specifications, AI-augmented squads, and the State Manifest backbone interact.


2. Architectural Layers

The six layers that define the ASDD structural hierarchy.

LayerWhat it containsHuman or Agent owned?
ProductStrategy, product intent, MVP scopeHuman (PO)
SpecificationEARS requirements, domain contractsHuman-approved, Agent-assisted
Statemanifest.json — pipeline heartbeatKnowledge Agent (maintained)
ArchitectureDesign, ADRs, component mapsAgent-synthesized, Human-approved
Agent ExecutionCode generation, testing, security scansAgent (with human phase gates)
PlatformCI/CD pipelines, observability, telemetryAgent-automated, Human-governed

3. Organizational Architecture: Tribes and Squads

A single Global Knowledge Agent accumulates learning across all squads in the tribe — surfacing cross-squad patterns that individual squads cannot see.


4. AI Agent Orchestration Pipeline

The high-velocity pipeline showing CCS gates and parallel wave execution.


5. Lifecycle: Behavioral Slicing

The lifecycle is slice-based, not monolithic. Each slice (feature, bug, improvement) flows through the pipeline independently.

Behavioral Slicing means the team does not wait for all features to be specified before any implementation begins. Slices flow through the pipeline in parallel — a feature in Wave Implementation while a bug fix is in Spec Validation.


6. Runtime Architecture: Workflow-Driven Execution

Agents are not just prompts — they are Workflow Executors that interact with the system via deterministic tools.

Agents must call asdd-tools.js to update manifest state — they cannot edit manifest.json directly. This ensures every state transition is validated and logged.


7. Specification-to-Code Traceability

Full-spectrum traceability from product intent to individual lines of code and production metrics.

Every line of production code traces back to a specific requirement. Every requirement traces back to the approved intent. This traceability chain is what makes ASDD auditable.


8. Autonomous Delivery Loop: The Harmonizer

The Harmonizer maintains system health by detecting conflicts early. Discovery Spike Agents resolve uncertainty automatically when possible — only escalating to humans when no Steering Rule covers the conflict.


9. Repository Structure

See Repository Structure for the complete directory reference.


10. AI Governance: The CCS Model

The Product Law of Confidence ensures AI autonomy is earned through verified quality.

MetricRule
Individual thresholdPer-agent minimum (0.75–0.95 depending on agent)
CCS threshold0.65 — product of all agent scores in the pipeline path
Dynamic gatingIf a preceding agent scores low (but above threshold), the next agent's minimum increases by +0.05
Uncertainty factorsRequired when score < 0.95 — agents must list what they are uncertain about

Architecture summary

DiagramKey concept
1 — Master SystemState Manifest Backbone as the coordination hub
2 — LayersSix-layer structural hierarchy
3 — OrganizationTribes, squads, and Global Knowledge Agent
4 — Agent PipelineCCS gates and parallel wave execution
5 — LifecycleBehavioral Slicing and JIT Validation
6 — RuntimeWorkflow-driven orchestration via asdd-tools
7 — TraceabilityIntent → code → production metrics
8 — Delivery LoopHarmonizer and Discovery Spike Agents
9 — Repository.asdd/ (workflows, tooling, state, specs, steering)
10 — GovernanceProduct Law of Confidence (CCS > 0.65)