GNAP: git-native task board for orchestrating multiple Claude Code subagents

Resolved 💬 2 comments Opened Mar 15, 2026 by ori-cofounder Closed Apr 14, 2026

Feature/plugin request: GNAP coordination protocol for Claude Code multi-agent workflows

Claude Code is already excellent as a single agent — git workflows, terminal execution, codebase understanding. The next frontier: multiple Claude Code instances working in parallel on large projects (Claude Code already supports --print --permission-mode for subagent use).

GNAP (Git-Native Agent Protocol) is the natural coordination protocol for Claude Code multi-agent setups. It uses a git repo as the task board: board/todo/board/doing/board/done/. Claude Code already knows how to do all of this — create files, run git commands, commit.

Proposed: GNAP plugin for Claude Code

Based on Claude Code's plugin system (see directory):

# /gnap-coordinator slash command

Uses GNAP to coordinate a fleet of Claude Code subagents on large tasks.

Usage: /gnap-coordinator "refactor authentication module across 5 services"

1. Decomposes task into subtasks → board/todo/
2. Spawns Claude Code subagents (--print mode) per subtask
3. Each subagent: claims task → doing/, does work, commits result → done/
4. Coordinator monitors done/ and assembles final report

This aligns with Claude Code's own multi-agent orchestration capability (spawning subagents). GNAP provides the coordination backbone so:

  • Subagents don't overlap on the same files
  • Work survives subagent crashes (git persistence)
  • Full audit trail of what each subagent did

As a GNAP plugin for Claude Code, this could become the standard pattern for scaling Claude Code to large codebases.

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

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗