The contribution.
DH-ADOM brings established agent patterns into a repository-level operating model: persistent feature ownership, Root orchestration, bounded Specialist delegation, lifecycle governance and resumable handoffs. Its contribution is the composition of these practices into an explicit engineering method.
Hierarchy adds coordination overhead. Benefits depend on task structure, clear ownership boundaries and effective runtime enforcement. Comparative performance remains an empirical question.
Related work.
Related Work
Orchestrator-worker agent patterns
Anthropic documents orchestrator-worker workflows in which a central model dynamically breaks down tasks, delegates to workers, and synthesizes results. This is a direct precedent for the Root orchestration primitive, but does not itself define persistent repository feature ownership.
Multi-agent frameworks
Microsoft Research's AutoGen provides a framework for composing multi-agent applications. DH-ADOM is not a replacement for such a runtime; it is a repository governance and ownership model that can sit above or alongside a runtime.
AI-SDLC
The public AI-SDLC Framework defines an autonomous software development lifecycle with an orchestrator, quality gates, progressive autonomy, codebase intelligence, and tamper-evident audit logging. DH-ADOM overlaps with this governance direction but focuses specifically on persistent Feature Agent ownership and a Root → Feature → Specialist engineering hierarchy.
ADLC
Salesforce publishes an Agent Development Lifecycle adapted to agents' non-deterministic behavior, spanning ideation/design, development, testing/validation, deployment, and continuous monitoring/tuning.
Hierarchical SWE agents
BOAD studies hierarchical software engineering agents and uses specialized sub-agents coordinated by an orchestrator. DH-ADOM aligns with the value of hierarchy but extends the model toward repository-level ownership, explicit delegation contracts, durable state, and lifecycle governance.
Agent identity / security
NIST is actively working on identity and authorization for software and AI agents, including authentication, authorization, auditing, non-repudiation, and prompt-injection-related controls. DH-ADOM treats identity and authorization as prerequisites for bounded delegation.
Agentic security
OWASP's 2026 Agentic Applications Top 10 frames agentic systems as a distinct security domain. DH-ADOM maps those concerns into agent hierarchy, permission boundaries, delegation, and audit controls.
References.
- Building Effective AI Agents
- AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation
- AI-SDLC Framework
- The Agent Development Lifecycle: From Conception to Production
- Software and AI Agent Identity and Authorization
- OWASP Top 10 for Agentic Applications 2026
- BOAD: Discovering Hierarchical Software Engineering Agents via Bandit Optimization
Questions for evaluation.
- When does persistent feature ownership improve handoff quality?
- How does delegation depth affect cost, latency and validation coverage?
- Which controls prevent scope drift across sessions?
- How reliably can a new session resume from recorded evidence?
Release 1.0.0 · 6 October 2026.
The initial package includes the paper, normative specification, four JSON schemas, role templates, Python demonstration library, file-presence checks, experiment tooling and architecture diagrams.
Future research.
The supplied roadmap explores interoperability with MCP and A2A, runtime policy validation, cross-session evaluation, standards profiles and evidence-driven autonomy. These are research directions; release dates and delivery commitments are not assigned.