This file provides context for AI assistants (Claude, etc.) working with the HCA specification repository.
Holonic Compound Architecture (HCA) is a protocol specification for building autonomous, composable, and continuously-improving agentic systems.
Key concepts:
- Holons: Agents that are both autonomous wholes AND cooperative parts
- Holarchy: Recursive tree structure of holons (authority down, results up)
- Compound Loop: PLAN → WORK → REVIEW → REVISE with learning extraction
- Protocol Boundaries: A2A (coordination), MCP (tools), ACP (northbound)
hca/
├── README.md # Entry point, accessible overview
├── DOCUMENT-INDEX.md # Navigation, reading paths, classification
├── CLAUDE.md # This file - AI assistant context
├── .gitignore
│
├── spec/ # NORMATIVE - The Rules (MUST implement)
│ ├── SPEC.md # Core specification (~37KB)
│ ├── GLOSSARY.md # Term definitions (~20KB)
│ ├── HCA-EXTENSIONS.md # JSON-RPC method schemas (~37KB)
│ └── STATE-MACHINES.md # Formal state diagrams (~32KB)
│
├── reference/ # INFORMATIVE - Explains & Guides
│ ├── QUICK-REFERENCE.md # 1-page cheat sheet
│ ├── ARCHITECTURE-DIAGRAM.md # Visual Mermaid diagrams
│ ├── BUILDING-BLOCKS.md # Protocol rationale
│ └── CONFORMANCE-CRITERIA.md # Certification requirements
│
├── guidance/ # INFORMATIVE - Implementation Patterns
│ └── SECURITY-CONSIDERATIONS.md
│
└── working/ # PROJECT - Not Part of Spec
├── ANALYSIS-SUMMARY.md # Validation summary
├── PROTOCOL-GAP-ANALYSIS.md
└── CRITICAL-FIXES-REQUIRED.md
| Classification | Folder | Uses RFC 2119? | Creates Requirements? |
|---|---|---|---|
| Normative | spec/ |
Yes | Yes |
| Informative | reference/, guidance/ |
No | No |
| Working | working/ |
No | No |
- Root Holon: Top-level, no parent, mission control, external interface
- Domain Holon: Intermediate, coordinates specialty area, delegates
- Leaf Holon: No children, executes concrete work, invokes tools
- PLAN (~40% effort): Query knowledge base, create plan, allocate budget
- WORK: Execute plan, invoke tools via MCP, produce artifacts
- REVIEW (~40% effort): Separate evaluator holon, produce verdict
- REVISE: On failure with retries remaining, loop back to PLAN
- COMPOUND: Extract lessons for knowledge base (outer loop)
All HCA methods use JSON-RPC 2.0 with hca/ prefix:
Task Management:
hca/task/delegate,hca/task/negotiate,hca/task/statushca/task/complete,hca/task/escalatehca/task/checkpoint,hca/task/resume,hca/task/transfer
Review:
hca/review/request,hca/review/verdict
Knowledge:
hca/knowledge/contribute,hca/knowledge/queryhca/knowledge/deprecate,hca/knowledge/attest,hca/knowledge/conflict
Anti-Patterns:
hca/antipattern/register,hca/antipattern/check
Human Intervention:
hca/human/request-approval,hca/human/override,hca/human/release
Holon Lifecycle:
hca/holon/register,hca/holon/deregisterhca/holon/evolve,hca/holon/promote,hca/holon/reputation
Artifacts:
hca/artifact/store,hca/artifact/retrieve
- Level 1 (Minimal): Holons + delegation + artifacts + review
- Level 2 (Standard): + Full compound loop + iteration + negotiation
- Level 3 (Full): + Knowledge base + cross-task learning
- Use RFC 2119 keywords (MUST, SHOULD, MAY) only in
spec/documents - Keep cross-references up to date
- Update GLOSSARY.md when adding new terms
- Add state diagrams to STATE-MACHINES.md for new protocol flows
- Add JSON-RPC schemas to HCA-EXTENSIONS.md for new methods
- Update table of contents in the document
- Add to DOCUMENT-INDEX.md if significant
- Update related documents that reference it
- Consider if GLOSSARY.md needs new terms
- Within same folder:
[GLOSSARY](GLOSSARY.md) - To other folders:
[SPEC](../spec/SPEC.md)or[SPEC](spec/SPEC.md)from root
# HCA [Document Title]
**Holonic Compound Architecture (HCA) v0.1 - [Document Type]**
[Brief description]
------
*This document is part of the Holonic Compound Architecture (HCA) Specification v0.1*- Task is in exactly one state at any time
- Producer and reviewer MUST be different holons
- Iteration MUST be bounded (max retries + budget)
- Escalation MUST include evidence of what was attempted
- Query knowledge during PLAN phase
- Contribute lessons during COMPOUND phase
- Both success AND failure produce lessons
- Lessons have lifecycle (Active → Deprecated → Archived)
- Authority is constraint-based, not command-based
- Authority flows DOWN (constraints, goals, budgets)
- Results flow UP (artifacts, verdicts, status)
- Children MAY refuse or negotiate tasks
| Version | Date | Status |
|---|---|---|
| 0.1.0 | 2026-01-06 | Draft |
- A2A Protocol - Coordination boundary
- MCP Specification - Tool boundary
- ACP Specification - Northbound boundary
- OpenHands - Runtime substrate
- ADK - Agent framework option
- Define in
spec/HCA-EXTENSIONS.mdwith full JSON-RPC schema - Add state diagram to
spec/STATE-MACHINES.mdif new flow - Add terms to
spec/GLOSSARY.md - Update
reference/QUICK-REFERENCE.mdmethod list - Consider conformance level implications
- Define in
spec/SPEC.mdwith normative requirements - Add to
spec/GLOSSARY.md - Update
spec/STATE-MACHINES.mdif stateful - Update
DOCUMENT-INDEX.mdif major addition
- Check DOCUMENT-INDEX.md for authoritative paths
- Use relative paths from document location
- Verify link works (file exists at path)
This context file helps AI assistants work effectively with the HCA specification.