Huashu Agent Swarm

Infinite Agent Loop — Multi-Agent Swarm Mode

Inspired by Nicholas Carlini's experiment of using 16 Claude instances to autonomously build a C compiler. No master agent. Pure git self-organisation. Each agent independently claims tasks, writes code, and pushes.

Trigger Conditions

Use this skill when the user mentions "swarm mode", "multi-agent parallel", "infinite loop", "agent swarm", or "launch swarm".

Prerequisites

• tmux (brew install tmux)
• Claude CLI (already installed)
• A git repository (existing or new)

Usage Workflow

Step 1: Describe the Project

The user tells me:

• Project directory path (must be a git repository)
• Project goal and overall description
• Initial task list (or let the agents break it down themselves)
• Number of agents (default: 8)
• Code standards and test commands

Step 2: Initialise the Project

bash SKILL_DIR/scripts/setup_project.sh <project-directory>

This creates the following inside the project:

• AGENT_PROMPT.md — Generated from a template; I customise it based on user requirements
• TASKS.md — Initial task checklist
• current_tasks/ — Task claim directory
• agent_logs/ — Logs directory

I then customise AGENT_PROMPT.md using references/agent-prompt-template.md, filling in project-specific details.

Step 3: Launch the Swarm

bash SKILL_DIR/scripts/start_swarm.sh <number-of-agents> <project-directory>

This will:

1. Create a git worktree for each agent (shared .git object store — no disk waste)
2. Create a tmux session with one pane per agent
3. Each agent enters an infinite loop: pull → claim task → execute → push → next task

Step 4: Open the Dashboard

python3 SKILL_DIR/scripts/dashboard.py <project-directory> 8420

Open http://localhost:8420 in your browser to:

• View all agent statuses, git log, and task progress in real time
• View the latest logs for each agent
• Send instructions directly to agents via the input box (written to HUMAN_INPUT.md)
• Stop all agents with one click

You can also monitor via the command line:

# Terminal status
bash SKILL_DIR/scripts/status.sh <project-directory>

# Send instructions
bash SKILL_DIR/scripts/send_input.sh <project-directory> "your instruction"

# Attach to tmux directly to observe
tmux attach -t swarm-<project-name>

Step 5: Stop the Swarm

bash SKILL_DIR/scripts/stop_swarm.sh <project-directory>

Automatically stops all agents, merges branches, and cleans up worktrees.

Core Mechanisms

Git Self-Organisation Coordination

• Each agent claims tasks via current_tasks/*.lock files
• Agents track global progress through TASKS.md
• Agents understand other agents' work via git log
• Conflicts are resolved by agents themselves

Git Worktree Isolation

• No need for multiple clones — uses git worktree for isolation
• All worktrees share the same .git object store
• Each agent works independently in its own worktree

Infinite Loop

• Each agent automatically begins the next session after completing one
• Uses git pull to fetch the latest work from other agents
• Sleep intervals between sessions prevent API rate limiting

Key Configuration

| Parameter | Default | Description |

param($m)
$inner = $m.Groups[1].Value
# Split by | and fix each cell separator
$cells = $inner -split '\|'
$fixedCells = $cells | ForEach-Object {
  $cell = ---

name: huashu-agent-swarm description: Multi-agent swarm parallel collaboration using pure git self-organisation, ideal for large-scale project development. Use when the user mentions "swarm mode", "multi-agent", "parallel development", or "agent swarm".

Infinite Agent Loop — Multi-Agent Swarm Mode

Inspired by Nicholas Carlini's experiment of using 16 Claude instances to autonomously build a C compiler. No master agent. Pure git self-organisation. Each agent independently claims tasks, writes code, and pushes.

Trigger Conditions

Use this skill when the user mentions "swarm mode", "multi-agent parallel", "infinite loop", "agent swarm", or "launch swarm".

Prerequisites

• tmux (brew install tmux)
• Claude CLI (already installed)
• A git repository (existing or new)

Usage Workflow

Step 1: Describe the Project

The user tells me:

• Project directory path (must be a git repository)
• Project goal and overall description
• Initial task list (or let the agents break it down themselves)
• Number of agents (default: 8)
• Code standards and test commands

Step 2: Initialise the Project

bash SKILL_DIR/scripts/setup_project.sh <project-directory>

This creates the following inside the project:

• AGENT_PROMPT.md — Generated from a template; I customise it based on user requirements
• TASKS.md — Initial task checklist
• current_tasks/ — Task claim directory
• agent_logs/ — Logs directory

I then customise AGENT_PROMPT.md using references/agent-prompt-template.md, filling in project-specific details.

Step 3: Launch the Swarm

bash SKILL_DIR/scripts/start_swarm.sh <number-of-agents> <project-directory>

This will:

1. Create a git worktree for each agent (shared .git object store — no disk waste)
2. Create a tmux session with one pane per agent
3. Each agent enters an infinite loop: pull → claim task → execute → push → next task

Step 4: Open the Dashboard

python3 SKILL_DIR/scripts/dashboard.py <project-directory> 8420

Open http://localhost:8420 in your browser to:

• View all agent statuses, git log, and task progress in real time
• View the latest logs for each agent
• Send instructions directly to agents via the input box (written to HUMAN_INPUT.md)
• Stop all agents with one click

You can also monitor via the command line:

# Terminal status
bash SKILL_DIR/scripts/status.sh <project-directory>

# Send instructions
bash SKILL_DIR/scripts/send_input.sh <project-directory> "your instruction"

# Attach to tmux directly to observe
tmux attach -t swarm-<project-name>

Step 5: Stop the Swarm

bash SKILL_DIR/scripts/stop_swarm.sh <project-directory>

Automatically stops all agents, merges branches, and cleans up worktrees.

Core Mechanisms

Git Self-Organisation Coordination

• Each agent claims tasks via current_tasks/*.lock files
• Agents track global progress through TASKS.md
• Agents understand other agents' work via git log
• Conflicts are resolved by agents themselves

Git Worktree Isolation

• No need for multiple clones — uses git worktree for isolation
• All worktrees share the same .git object store
• Each agent works independently in its own worktree

Infinite Loop

• Each agent automatically begins the next session after completing one
• Uses git pull to fetch the latest work from other agents
• Sleep intervals between sessions prevent API rate limiting

Key Configuration

Risks and Mitigations

By Huashu | AI Native Coder · Independent Developer WeChat Official Account "Huashu" | 300k+ followers | AI Tools & Productivity Flagship products: Kitty Catchlight (AppStore Paid Rankings Top 1) · Play DeepSeek with One Book .Trim() if ($cell -match '^:?-+:? | Number of agents | 8 | Can be specified at launch | | Sleep interval | 5 seconds | Adjustable in agent_loop.sh | | Model | claude-opus-4-6 | Adjustable in agent_loop.sh |

Risks and Mitigations

By Huashu | AI Native Coder · Independent Developer WeChat Official Account "Huashu" | 300k+ followers | AI Tools & Productivity Flagship products: Kitty Catchlight (AppStore Paid Rankings Top 1) · Play DeepSeek with One Book ) { " $cell " } else { " $cell " } } '|' + ($fixedCells -join '|') + '|'

| Number of agents | 8 | Can be specified at launch | | Sleep interval | 5 seconds | Adjustable in agent_loop.sh | | Model | claude-opus-4-6 | Adjustable in agent_loop.sh |

Risks and Mitigations

| Risk | Mitigation |

param($m)
$inner = $m.Groups[1].Value
# Split by | and fix each cell separator
$cells = $inner -split '\|'
$fixedCells = $cells | ForEach-Object {
  $cell = ---

name: huashu-agent-swarm description: Multi-agent swarm parallel collaboration using pure git self-organisation, ideal for large-scale project development. Use when the user mentions "swarm mode", "multi-agent", "parallel development", or "agent swarm".

Infinite Agent Loop — Multi-Agent Swarm Mode

Inspired by Nicholas Carlini's experiment of using 16 Claude instances to autonomously build a C compiler. No master agent. Pure git self-organisation. Each agent independently claims tasks, writes code, and pushes.

Trigger Conditions

Use this skill when the user mentions "swarm mode", "multi-agent parallel", "infinite loop", "agent swarm", or "launch swarm".

Prerequisites

• tmux (brew install tmux)
• Claude CLI (already installed)
• A git repository (existing or new)

Usage Workflow

Step 1: Describe the Project

The user tells me:

• Project directory path (must be a git repository)
• Project goal and overall description
• Initial task list (or let the agents break it down themselves)
• Number of agents (default: 8)
• Code standards and test commands

Step 2: Initialise the Project

bash SKILL_DIR/scripts/setup_project.sh <project-directory>

This creates the following inside the project:

• AGENT_PROMPT.md — Generated from a template; I customise it based on user requirements
• TASKS.md — Initial task checklist
• current_tasks/ — Task claim directory
• agent_logs/ — Logs directory

I then customise AGENT_PROMPT.md using references/agent-prompt-template.md, filling in project-specific details.

Step 3: Launch the Swarm

bash SKILL_DIR/scripts/start_swarm.sh <number-of-agents> <project-directory>

This will:

1. Create a git worktree for each agent (shared .git object store — no disk waste)
2. Create a tmux session with one pane per agent
3. Each agent enters an infinite loop: pull → claim task → execute → push → next task

Step 4: Open the Dashboard

python3 SKILL_DIR/scripts/dashboard.py <project-directory> 8420

Open http://localhost:8420 in your browser to:

• View all agent statuses, git log, and task progress in real time
• View the latest logs for each agent
• Send instructions directly to agents via the input box (written to HUMAN_INPUT.md)
• Stop all agents with one click

You can also monitor via the command line:

# Terminal status
bash SKILL_DIR/scripts/status.sh <project-directory>

# Send instructions
bash SKILL_DIR/scripts/send_input.sh <project-directory> "your instruction"

# Attach to tmux directly to observe
tmux attach -t swarm-<project-name>

Step 5: Stop the Swarm

bash SKILL_DIR/scripts/stop_swarm.sh <project-directory>

Automatically stops all agents, merges branches, and cleans up worktrees.

Core Mechanisms

Git Self-Organisation Coordination

• Each agent claims tasks via current_tasks/*.lock files
• Agents track global progress through TASKS.md
• Agents understand other agents' work via git log
• Conflicts are resolved by agents themselves

Git Worktree Isolation

• No need for multiple clones — uses git worktree for isolation
• All worktrees share the same .git object store
• Each agent works independently in its own worktree

Infinite Loop

• Each agent automatically begins the next session after completing one
• Uses git pull to fetch the latest work from other agents
• Sleep intervals between sessions prevent API rate limiting

Key Configuration

Risks and Mitigations

By Huashu | AI Native Coder · Independent Developer WeChat Official Account "Huashu" | 300k+ followers | AI Tools & Productivity Flagship products: Kitty Catchlight (AppStore Paid Rankings Top 1) · Play DeepSeek with One Book .Trim() if ($cell -match '^:?-+:? | API rate limiting | Sleep intervals + adjustable agent count | | Merge conflicts | AGENT_PROMPT guides small, granular commits | | Infinite loop doing useless work | Log monitoring + stop conditions | | Disk space | stop_swarm.sh auto-cleans | | Cost spiral | Limit session count in AGENT_PROMPT |

By Huashu | AI Native Coder · Independent Developer WeChat Official Account "Huashu" | 300k+ followers | AI Tools & Productivity Flagship products: Kitty Catchlight (AppStore Paid Rankings Top 1) · Play DeepSeek with One Book ) { " $cell " } else { " $cell " } } '|' + ($fixedCells -join '|') + '|'

| API rate limiting | Sleep intervals + adjustable agent count | | Merge conflicts | AGENT_PROMPT guides small, granular commits | | Infinite loop doing useless work | Log monitoring + stop conditions | | Disk space | stop_swarm.sh auto-cleans | | Cost spiral | Limit session count in AGENT_PROMPT |

By Huashu | AI Native Coder · Independent Developer WeChat Official Account "Huashu" | 300k+ followers | AI Tools & Productivity Flagship products: Kitty Catchlight (AppStore Paid Rankings Top 1) · Play DeepSeek with One Book