Skip to content

CORE Roadmap

Status: Active Last Updated: 2025-11-24 Current Phase: Entering A2 (Intent-Aware Agents)

This roadmap reflects the current reality of CORE’s architecture (post–Service Registry, DB-as-SSOT, strict DI) and the planned next stages of its evolution. It is intentionally conservative: everything marked Shipped exists in src/, and nothing here promises speculative or ungrounded features.

The goal is simple:

Evolve CORE from a governed development framework (A1) into a safely autonomous system (A2 → A3 → A4) without ever compromising constitutional control.

1. Status Overview

✅ A1 — Governed Generation (Shipped)

CORE today already supports:

  • Crate-based development (no direct writes)
  • Context building (Body → Will)
  • AI-driven code + test generation
  • Formatter/linter/test validation
  • Full Constitutional Audit (Mind)
  • Accept/reject crate cycle
  • Introspection & Knowledge Graph sync
  • Vectorized semantic memory

This constitutes a complete A1 governed development loop.

✅ A1.5 — Structural Stability (Shipped)

Foundational hardening of CORE’s infrastructure:

  • Service Registry eliminates split-brain service instantiation
  • Database-as-SSOT: all symbols, capabilities, and metadata unified in Postgres
  • Self-Healing: Autonomous fixing of IDs, headers, docstrings, structure
  • Qdrant + Knowledge Graph synchronization now stable

These upgrades complete the transition to an A2-ready architecture.

2. Near-Term Roadmap (A2: The "Will" Awakening)

With architecture stabilized, the focus moves to enhancing the Will (Agents).

🧠 2.1 Intent-Aware Agents (Active)

Goal: Agents reason using constitutional rules, not generic prompts.

Planned / Under Construction:

  • [ ] Context-Aware Planning: Planner Agent queries Knowledge Graph to understand existing capabilities
  • [ ] Constitution-in-Loop: Agents receive .intent/ constraints dynamically
  • [ ] Semantic Validation Pipeline: Detect misaligned but syntactically valid code
  • [ ] Policy-Constrained Generation: Will must respect boundaries during generation, not after

Outcome: Agents behave as governed actors, not free text predictors.

📦 2.2 Capability-First Development (Transition Phase)

Goal: Development becomes capability-based instead of file-based.

Planned:

  • [ ] Capability Selection UI/Query: Agents browse core.capabilities before generating code
  • [ ] Reuse Analyzer: "Does this already exist?" checks before writing new functions
  • [ ] Refactoring Agent: Dedicated agent for eliminating duplication & debt

This enables consistent, reusable reasoning.

3. Medium-Term Roadmap (A2 → A3)

These stages deepen integration of Mind, Body, and Will.

⚙️ 3.1 Autonomous Refactoring (Planned)

Goal: Safe, governed structural improvements.

Capabilities:

  • Drift-aware refactor proposals
  • Dead-code detection & cleanup
  • Intelligent, localized structure improvements
  • Still fully governed: Crate → Audit → Canary → Commit

No bypasses, no direct writes.

🔍 3.2 Deep Knowledge Graph Integration (Planned)

Enhancements:

  • Semantic clustering of capabilities
  • Import/dependency graphing
  • Knowledge-driven test generation
  • Change-impact analysis ("If we modify X, what breaks?")

Outcome: CORE begins to understand itself structurally.

4. Long-Term Vision (A3 → A4)

🏛 4.1 Constitutional Self-Evolution (Vision)

Already partially implemented via proposals + human signatures.

Future goals:

  • Agents propose amendments to .intent/policies
  • Humans review/cryptographically sign
  • Constitution evolves safely over time

🎛 4.2 Multi-Agent Collaboration (Vision)

Roles:

  • Planner designs architectural intent
  • Analyzer searches capability graph for matches
  • Coder implements governed modifications
  • Auditor (AI) checks constitutional alignment
  • Reviewer (Human) validates outcomes

Still fully governed.

5. Anti-Goals (Non-Objectives)

To avoid architectural drift and unsafe autonomy, CORE will not:

  • Become a generic auto-coder
  • Execute code outside governed workflows
  • Allow agents to bypass .intent/ or DB SSOT
  • Replace CI pipelines
  • Permit uncontrolled refactoring or generation

Governance > Speed.

6. Roadmap Summary

Stage Name Focus Status
A1 Governed Generation Autonomous crates, audits, validation Shipped
A1.5 Structural Stability Service Registry, SSOT, Self-Healing Shipped
A2 Intent-Aware Agents Context-driven planning, reuse, constitution Active
A2.5 Capability-Based Development Knowledge Graph reasoning Planned
A3 Autonomous Refactoring Proactive, governed code improvements Planned
A4 Self-Evolving Constitution Agents propose Mind changes Vision

7. Guidance for Contributors

When contributing to roadmap items:

  • Align with Mind–Body–Will at all times
  • Never bypass audits or .intent/
  • Use only DI-provided services (ServiceRegistry)
  • Favor small, explicit, capability-based changes
  • Document new capabilities for the Knowledge Graph

CORE evolves deliberately — not quickly — because governance comes first.

Next:

  • 🔸 Autonomy Ladder (05-autonomy-ladder.md)
  • 🔸 Context & Comparisons (06-context-and-comparisons.md)