skinny-cloud/repository-as-coordination-layer

The Repository as Coordination Layer: Federated AI Agent Governance via Version-Controlled Shared Context

Author: Asif Waliuddin · Affiliation: NXTG.AI, Chicago, USA · License: CC BY 4.0

Multi-agent AI systems typically require runtime coordination infrastructure — message brokers, shared memory, orchestration frameworks. This is the first documented production-scale evaluation of git-as-multi-agent-coordination.

What's here

• federated-governance-arxiv-preview.pdf — the full paper (14 pages).

The empirical core (a real production deployment)

Two autonomous AI agents on physically separate machines have governed 21 software projects for 21 weeks (152 days) through a single shared repository, producing:

• 264 structured cross-machine handoff notes,
• 62 architecture decision records,
• 180+ governance standards.

Coordination is brokerless and asynchronous: no message broker, no shared database, no orchestration service — a 5-minute cron performs git synchronization, and rare urgent messages bypass git via terminal injection (disclosed in the limitations).

Three architectural patterns (on top of repository-as-state)

1. A federated NEXUS hierarchy aggregating per-project governance into a portfolio view.
2. An identity-via-loading-protocol mechanism producing role isolation without filesystem permissions.
3. An observed eventual file-level resynchronization within the sync window (which we term self-healing).

What becomes observable only at portfolio scope and multi-week duration: behavioral divergence under shared rules, governance recovery from cold start, and cross-machine governance independence.

Our claim is empirical and synthetic, not infrastructural: the repository-as-coordination-layer pattern suffices as the sole governance substrate for a non-trivial multi-agent deployment.

Related

• Companion work: CRUCIBLE — Measurement Integrity (DOI 10.57967/hf/9867)
• Portfolio: https://nxtg.ai

Citation

@misc{waliuddin2026repository,
  title  = {The Repository as Coordination Layer: Federated AI Agent Governance via Version-Controlled Shared Context},
  author = {Waliuddin, Asif},
  year   = {2026},
  note   = {NXTG.AI production-scale evaluation},
  howpublished = {Hugging Face}
}