Skip to content
Agent Tree Knowledge Hub
Subscribe

OpenClaw Integration Architecture

Status: Production (Phases 2-6 complete)

Overview

Section titled “Overview”

OpenClaw is a separate agent orchestration platform. The integration work unified OpenClaw agents with the Agent System, eliminating duplication and establishing clear routing between the two systems.

Phase Summary

Section titled “Phase Summary”

Phase 2: Unified Router

Section titled “Phase 2: Unified Router”
  • Smart routing across 3 systems: Agent System tools, OpenClaw agents, Claude Code
  • Weighted keyword scoring with multi-word bonus
  • Health-aware OC routing (30s cache, demotion when gateway down)
  • Retired 6 OC agents → redirected to AS equivalents
  • JSONL routing log for audit trail
  • 50-case test suite: 100% accuracy, under 15ms latency

Phase 3A: AS Tools as OpenClaw Skills

Section titled “Phase 3A: AS Tools as OpenClaw Skills”

Created SKILL.md wrappers for cross-platform access:

  • research-agent
  • quanta-analytics
  • deck-architect
  • qa-agent

Phase 4: Shared Memory Protocol

Section titled “Phase 4: Shared Memory Protocol”
  • shared_memory.py — write, read, search with namespace ownership
  • Synced 803 chunks from 10 OC agents to vault
  • Shared YAML frontmatter schema
  • Cross-system search: vault markdown + SQLite FTS in one query
  • Shared rate-limit store with file locking

Phase 5: Agent Consolidation (3 merges)

Section titled “Phase 5: Agent Consolidation (3 merges)”
  • COO + task_manager: COO handles calendar/reminders, task_manager handles CRUD
  • Academic + canvas_sync: canvas_sync for data sync, academic for interactive queries
  • Tutor consolidation: conversational tutoring + deep analysis split

Phase 6: Production Hardening

Section titled “Phase 6: Production Hardening”
  • Per-request cost estimates and weekly reports
  • Vault export with OpenClaw health data
  • Runbook for operations: restart, add agents, debug routing, emergencies

Architecture Decision: Hybrid Tiered Model

Section titled “Architecture Decision: Hybrid Tiered Model”

After analyzing 4 options, we settled on “Option D” hybrid:

  • MCP for shared services (tools, data access)
  • OC skills for OC-native features (channels, task management)
  • AI-OS engines for heavy compute (research, deck building, RAG)

This eliminates service overlap while preserving each system’s strengths.

Unified Skill Registry

Section titled “Unified Skill Registry”
  • 67 skills registered
  • 4 tiers: worker, specialist, orchestrator, executive
  • 5 backend types: mcp, oc-skill, cli, http, claude-code
  • 16 categories
  • Auto-registration hooks in scaffold tools
  • Validation script for registry integrity