Contracts that make ownership explicit.

DH-ADOM Technical Specification v1.0

Status: Normative

Version: 1.0.0

Date: 2026-10-06

1. Scope

DH-ADOM specifies a repository-level operating model for AI-native software engineering. It defines governance hierarchy, agent roles, domain ownership, bounded delegation, lifecycle integration, evidence, state, handoff, and conformance requirements.

2. Normative language

The terms MUST, MUST NOT, SHOULD, SHOULD NOT, MAY, and RECOMMENDED are normative.

3. Governance hierarchy

1. Engineering Constitution

2. AI-SDLC

3. ADLC

4. Root Agent

5. Feature Agent

6. Specialist Agent

7. Implementation

4. Agent roles

4.1 Root Agent

The Root Agent MUST own global orchestration and MUST NOT require itself to implement every feature.

4.2 Feature Agent

Every meaningful bounded domain SHOULD have one persistent Feature Agent. A Feature Agent MUST declare intent, ownership, capabilities, tools, permissions, autonomy, risks, state, and handoff.

4.3 Specialist Agent

A Specialist Agent MUST be created under a parent Feature Agent or explicitly authorized Root Agent. It MUST have a bounded purpose and task-scoped permissions.

5. Delegation Contract

Every delegation MUST identify parent, child, purpose, task, scope, permissions, budget, lifetime, evidence requirements, and acceptance criteria.

6. Delegation invariants

• Child scope MUST be a subset of authorized parent scope unless explicitly widened through governance.

• Delegation depth MUST be bounded.

• Parallel child execution MUST be bounded.

• Child permissions MUST NOT exceed delegated authority.

• Completion MUST return evidence and handoff.

• Uncontrolled recursive spawning MUST be blocked.

7. Feature ownership

Feature Agent ownership is an engineering boundary. Cross-feature changes require coordination through the Root Agent and MUST be traceable.

8. Identity and contract

Agent identity is distinct from model identity. Each agent MUST have an identity record and SHOULD have a machine-readable passport and contract.

9. AI-SDLC integration

All material work MUST map to intent, requirements, architecture, implementation, validation, evidence, release, and feedback.

10. ADLC integration

All agents MUST follow an agent lifecycle that includes intent, contract, identity, capabilities, permissions, autonomy, validation, promotion, monitoring, revalidation, and retirement.

11. Evidence

Meaningful tasks MUST emit auditable evidence of implementation and validation.

12. Handoff

A completed task MUST update state and create a machine-readable handoff sufficient for an independent AI session to continue work.

13. Security

Agents MUST follow least privilege. Production actions MUST be policy-controlled. Permission or autonomy escalation MUST NOT be self-authorized.

14. Conformance

An implementation conforms to DH-ADOM v1.0 only when all mandatory requirements in CONFORMANCE-REQUIREMENTS.md pass or have a formally accepted exception.

DH-ADOM v1.0 Conformance Requirements

An implementation MUST satisfy all CRITICAL requirements and provide evidence for each requirement.