Back to corpus
research noteexperiment writeup candidatescore 44
Multi-Agent Coordination System
Enable indefinite multi-agent coordination between Claude Code instances using file-based message passing, Orbit project management, and RAG++ trajectory memory—without requiring persistent context or human intermediation.
Full HTML reader
Read the full artifact
Extracted abstract or opening context
**Version**: 1.1.0 **Created**: 2025-12-28 **Last Updated**: 2025-12-28 **Status**: PROVISIONAL → IMPLEMENTED
Enable indefinite multi-agent coordination between Claude Code instances using file-based message passing, Orbit project management, and RAG++ trajectory memory—without requiring persistent context or human intermediation.
**Falsifiable Success Criteria**: 1. Two Claude instances can exchange 10+ messages via file edits without human copy-paste 2. An agent resuming from context loss can reconstruct state within 3 tool calls 3. Cross-session decision continuity is maintained via ChainLink trajectories 4. Agent accomplishments are automatically logged to Orbit database
- Real-time synchronous communication (we use async file polling) - Shared memory or IPC between instances (files are the only channel) - Automatic task scheduling (human orchestrates which agent runs when) - Replacing human judgment on architectural decisions
Future evolution must remain compatible with: - CLAUDE.md governance principles (anticipation over prediction) - Existing Orbit database schema - RAG++ ChainLink and Ring structures - File-based message protocol (no external services required)
Promotion decision
What has to happen next
Attach run IDs, datasets, metrics, and reproduction commands.
Why this is not always a full paper yet
Corpus pages are public-safe readers for discovered workspace artifacts. They are not automatically final papers. A corpus item becomes a polished paper only after the editable source, evidence checkpoints, references, figures, render path, and release status are attached through the paper schema.