Human–AI Collaboration Protocol (HACP)

HACP (v0.1) is a minimal, vendor-neutral protocol and reference implementation for safe, auditable, and productive collaboration between humans and high-capability AI systems.

This repository contains:

• the normative spec (HACP.md)
• schemas and SHACL shapes for compliance testing
• a canonical reference mapping to the "Beast" implementation
• examples and a pilot plan for non-classified lab testing

Goal: make the protocol real and executable so that institutions and vendors can interoperate — without surrendering human authority.

Quick start

1. Read the normative spec: HACP.md
2. Inspect the schemas: schemas/task-manifest.yaml and schemas/culture-profile.yaml
3. Validate examples with SHACL (CI job included): See .github/workflows/validate-shacl.yml
4. Review reference fixtures:
   • Task fixtures: examples/rdf/irreversible_task.ttl, examples/rdf/low_confidence_escalation.ttl
   • Culture fixture: examples/rdf/culture_profile.ttl

License

Apache 2.0 — see LICENSE