GNAP: git-native task coordination for OpenAgents Control's multi-language agent workflows

[ori-cofounder]

Proposal: GNAP as a coordination protocol for OpenAgents Control's pattern-based agents

OpenAgents Control (OAC) is a compelling approach — pattern-defined agents that generate consistent code across TypeScript, Python, Go, and Rust. The approval gate system and team-ready design make it enterprise-grade. GNAP adds the multi-agent coordination layer.

GNAP (Git-Native Agent Protocol) coordinates multiple OAC agents via a shared git repo: board/todo/ → board/doing/ → board/done/. Since OAC is built on git-native tooling (OpenCode), GNAP is a natural complement.

Applied to OAC's multi-agent workflows:

A large feature spanning multiple services (TypeScript API + Python ML + Go service):

board/todo/implement-auth-api-typescript.md   ← Coordinator creates per service
board/todo/implement-ml-model-python.md
board/todo/implement-cache-service-go.md

board/doing/implement-auth-api-typescript.md  ← OAC TypeScript agent claims
board/doing/implement-ml-model-python.md      ← OAC Python agent claims

board/done/implement-auth-api-typescript.md   ← Code committed, patterns followed
board/done/implement-ml-model-python.md

GNAP's atomic task claiming (git mv) prevents two OAC agents from working on the same service simultaneously. The GNAP board + OAC approval gates = a complete pattern-controlled, human-approved multi-agent development pipeline.

Particularly compelling for OAC's team-ready story: different team members' OAC instances coordinate via a shared GNAP board, each claiming tasks within their language/domain expertise.

Spec: https://github.com/farol-team/gnap