Claude Lead System

One control room for all your Claude Code terminals.

Detect file conflicts before they collide. Spawn, redirect, and message workers. Run tracked multi-step pipelines. All from a single local coordinator — no API tokens spent on coordination.

The Problem

Two Claude terminals. Same file. Neither knows the other is there.

Terminal A: editing src/auth/login.ts
Terminal B: editing src/auth/login.ts  ← collision in 3... 2... 1...

One overwrites the other. You find out when the build breaks.

With Lead:

Lead: "check conflicts"
→ CONFLICT: login.ts — terminal-a (editing) ↔ terminal-b (editing)
→ Both sessions notified. Collision blocked.

Quick Start

Quick (npm)

npm install -g claude-lead-system && claudex

Standard (git)

git clone https://github.com/DrewDawson2027/claude-lead-system.git
cd claude-lead-system && bash install.sh
claudex

Try without installing (Docker)

docker run -it ghcr.io/drewdawson2027/claude-lead-system

Then in any install method, enter coordinator mode:

/lead

You'll see a live dashboard of every active Claude terminal — what it's working on, what files it's touching, and whether any two sessions are about to collide.

Why you'd use it:

• You run 3+ Claude terminals and need one place to see what each is touching
• You've had two sessions clobber the same file — and want to prevent it
• You want to redirect, wake, or spawn workers without opening a new window

Common MCP tools (usable from any Claude session):

coord_spawn_worker    # Start a new Claude worker on a task
coord_watch_output    # Stream a worker's live output
coord_send_message    # Send instructions to a named worker

What It Does

| Feature | What It Does | | -------------------------------- | ------------------------------------------------------------------------------ | | Pre-edit conflict detection | Flags when two sessions are about to edit the same file — before the collision | | Conflict lifecycle tracking | Track, resolve, and recheck conflicts across sessions | | Worker spawn + kill + resume | Start, stop, and re-enter workers from one control point | | P2P worker messaging | Workers send messages directly to named peers via inbox files | | Broadcast | Send one message to all active workers simultaneously | | Operator dashboard | Live table: session, status, branch, files touched, last active | | Plan approval protocol | Workers pause in plan mode; lead approves before execution | | Pipeline orchestration | Run lint → test → build tracked from one place | | Budget gating | Cap tokens or turns per worker before spawning | | Zero API-token coordination | Filesystem carries coordination — no token cost for inter-worker comms | | Session resumption | Re-enter a prior Claude conversation, not a fresh start | | Context store | Shared key/value store accessible to all workers in a session | | Worktree isolation | Each worker can run in its own git branch — no conflicts | | Live output streaming | Worker output streamed via SSE — <1ms teammate switching in TUI | | Activity audit log | Append-only record of every coordination event | | 81 MCP tools | Full tool surface for coordination, governance, and orchestration |

Lead vs. Native Agent Teams

| Capability | Native Agent Teams | Lead System | | --------------------------------------- | ------------------ | ----------- | | Pre-edit conflict detection | — | ✅ verified | | Operator dashboard (all sessions) | — | ✅ verified | | Zero API-token coordination path | — | ✅ verified | | Budget/spawn/approval governance | — | ✅ verified | | In-process worker output streaming | — | ✅ verified | | In-context teammate lifecycle | ✅ verified | ✅ verified | | First-party cross-platform UX | ✅ verified | partial | | Minimum-setup (no external coordinator) | ✅ verified | — |

Use Lead when you need conflict detection, observability, and governance across 4+ terminals. Use Native Agent Teams for 1-2 collaborators where first-party UX matters most.

By the Numbers

• 847 automated tests (480 coordinator + 308 sidecar + 59 Python hooks)
• 85% line coverage (coordinator + sidecar)
• 81 MCP coordination tools
• 24 lib modules
• Zero API tokens spent on coordination (filesystem path)
• 8/8 macOS platform proofs passing · 8/8 Linux platform proofs passing

The Story

"I'm a Philosophy, Politics & Economics student at USC. I've never written a line of code in my life. I built this entire system — 81 tools, 847 tests, SLSA supply chain security — entirely through Claude Code."

Platform Support

| Platform | Status | Notes | | -------- | ------------ | --------------------------------------------------------------------------------------------------- | | macOS | Verified | Full support. iTerm2 or tmux recommended for split-pane. 8/8 proofs passing. | | Linux | Verified | 8/8 proofs passing. Requires bash + tmux. See compatibility matrix. | | Windows | Experimental | Advisory CI. Some features may not work. |

Requirements: Node.js 18+, bash (macOS/Linux), Python 3.10+ (for hooks)

Requirements

• Claude Code installed and authenticated
• Node.js ≥ 18
• jq (brew install jq / apt install jq)
• bash, python3

Docker users: all dependencies are bundled in the image.

More

• Getting Started — step-by-step first use walkthrough
• Docs — architecture, API contract, MCP tool reference
• Troubleshooting
• Security
• Contributing

Author: Drew Dawson — @DrewDawson2027

MIT License