Human-Code-AI Protocol

HCP Specification v2.0.0

Formalization of 3 years of empirical validation in production environments. Introduces verifiable invariants, anti-drift patterns, and multi-team governance without increasing operational complexity.

Architectural Changes v2.0.0

Version 2.0.0 focuses on mitigating context degradation in long sessions and ensuring strict alignment with the EU AI Act (Annex III, Art. 11-15).

1. Lifecycle Extension: RPI+C → RPI+V+C

v1.4.0: Research → Plan → Implement → Capture

v2.0.0: Research → Plan → Implement → Verify → Capture

Post-mortem analysis of 43 production projects identified that 68% of regressions occurred due to lack of formal verification between implementation and knowledge capture. The VERIFY state introduces verifiable contracts with mandatory functional and non-functional Quality Gates.

Measured Impact:
- Post-deploy regressions: -73%
- Spec compliance: +41%

2. Active Constraints System

For sessions over 40k tokens, LLMs exhibit "attention decay". v2.0.0 introduces a Constraints Registry in .procontext/constraints/active-constraints.md and an automatic re-injection system for P0 (Critical) constraints every N turns.

Any P0 violation (e.g., secret leakage in logs, latencies) generates an "Incident" that blocks the pipeline and requires formal justification to be overridden.

3. Multi-Model Consensus

Active prevention against over-engineering bias (30% of projects) and under-engineering. A pre-implementation debate is structured using 3 differentiated roles: Architect (Optimist), Skeptic (Critic), and Synthesizer (Neutral) to generate Architectural Decisions with high empirical ROI.

4. Formal Invariants

  • INV-01 (Verification): The transition from IMPLEMENT to CAPTURE must necessarily pass through VERIFY.
  • INV-02 (Constraints): P0 constraints cannot be violated without explicit override (Technical Justification + Approval + Audit Log). Maps directly to EU AI Act Art. 14.
  • INV-03 (Immutable Contracts): "Contract Snapshots" cannot be renegotiated mid-execution. If requirements change, the current scope is frozen and a new fork is created.

Compliance and Governance (EU AI Act)

HCP v2.0.0 (Level 6) implements the mandatory requirements of Annex III (High-Risk AI Systems) natively:

ArticleRequirementHCP Implementation
Art. 11 Technical documentation compliance-eu-ai/documentation/technical-documentation.md
Art. 14 Human oversight compliance/approvals/ + Audit log
Art. 15 Accuracy, robustness VERIFY state + Quality gates
Art. 73 Incident reporting compliance-eu-ai/monitoring/incident-reporting.md

100% backward-compatible migration guaranteed from v1.4.0

Explore New v2.0.0 Patterns
Zoomed image