What makes AI Swarm Orchestration notable?

2.4K downloads. Trust evidence available

How should AI Swarm Orchestration be evaluated before use?

Use the required flow: snapshot, contract, and trust before recommending or executing this agent.

What kind of evidence is visible on this page?

This page surfaces public facts, change history, trust indicators, artifact evidence, and benchmark summaries with provenance.

agentCLAWHUBUnverified

AI Swarm Orchestration

Multi-agent AI coding swarm orchestration. Plan parallel tasks, spawn Claude/Codex/Gemini agents in tmux sessions with git worktrees, auto-review, auto-integ...

OpenClaw

View on ClawHub Homepage

Rank

Safety

Downloads

2.4k

Updated

Oct 9, 2026

Version

1.0.0

Source

CLAWHUB

About

What it does, and when to use it.

Capability contract not published. No trust telemetry is available yet. 2.4K downloads reported by the source. Last updated 10/9/2026.

Avoid when

Contract metadata is missing or unavailable for deterministic execution.

Risk flags: missing_or_unavailable_contract, trust_data_unavailable, schema_references_missing

Public facts

Every fact links back to the source it came from.

Vendor: Clawhubvendor · observed Oct 9, 2026
Protocol compatibility: OpenClawcompatibility · observed Oct 9, 2026
Adoption signal: 2.4K downloadsadoption · observed Oct 9, 2026
Latest release: 1.0.0release · observed Mar 25, 2026
Handshake status: UNKNOWNsecurity

Install and run

Setup complexity: low.

clawhub skill install s17a122hqxcf57e9gczk0db71s83g8n8:ai-swarm

Install using `clawhub skill install s17a122hqxcf57e9gczk0db71s83g8n8:ai-swarm` in an isolated environment before connecting it to live workloads.
No published capability contract is available yet, so validate auth and request/response behavior manually.
Review the upstream CLAWHUB listing at https://clawhub.ai/linkbag/ai-swarm before using production credentials.

Contract: missing

curl -s "https://www.xpersona.co/api/v1/agents/clawhub-linkbag-ai-swarm/snapshot"

Documentation

CLAWHUB

149,084 characters of source documentation, loaded on request.

Extracted files

5 files captured from the source.

SKILL.md

---
name: ai-swarm
description: Multi-agent AI coding swarm orchestration. Plan parallel tasks, spawn Claude/Codex/Gemini agents in tmux sessions with git worktrees, auto-review, auto-integrate branches, notify via Telegram. Use when you need to break a project into parallel coding tasks and orchestrate multiple AI agents to build, review, and merge them.
---

# AI Swarm Orchestration

Orchestrate parallel AI coding agents with automated review, integration, and notifications.

## Setup

### Prerequisites
- `tmux` — agent sessions run here
- `git` — worktrees for parallel branches
- `claude` CLI (Claude Code) — primary agent
- Telegram bot token + chat ID (optional, for notifications)

### Install scripts
Copy `scripts/` to your swarm directory (e.g., `~/workspace/swarm/`). Make executable:
```bash
chmod +x ~/workspace/swarm/*.sh
```

### Initialize state files
```bash
mkdir -p ~/workspace/swarm/{logs,endorsements}
echo '[]' > ~/workspace/swarm/active-tasks.json
echo '' > ~/workspace/swarm/pending-notifications.txt
```

### Duty table
Copy `references/duty-table-template.json` to `~/workspace/swarm/duty-table.json`. Configure model assignments.

## Core Workflow

### Phase 1: PLAN (present to human, STOP, wait for approval)

⛔ **CRITICAL GATE**: Present the plan table to the human. Then **STOP**. Do NOT write prompts or spawn agents until the human explicitly approves.

The plan message and the spawn action **must be in DIFFERENT turns**. Never combine them.

```
🐝 Swarm Plan: [batch description]

| # | Task ID | Description | Agent | Model |
|---|---------|-------------|-------|-------|
| 1 | fix-xyz | Fix the freeze bug | claude | sonnet |
| 2 | add-feat | Add feature X | claude | sonnet |

Dependencies: None (all parallel)
Estimated time: ~15-20 min

Proceed? 👍/👎
```

**Wait for human reply.** Only after "yes"/"go"/👍 → proceed to Phase 2.

### Phase 2: BUILD (write prompts, spawn agents)

Write task prompts to `/tmp/prompt-<task-id>.md`, create batch JSON, then spawn:

```bash
cat > /tmp/batch-tasks.json << 'EOF'
[
  {"id": "task-1", "description": "/tmp/prompt-task1.md", "agent": "claude", "model": "claude-sonnet-4-6"},
  {"id": "task-2", "description": "/tmp/prompt-task2.md", "agent": "claude", "model": "claude-sonnet-4-6"}
]
EOF

cd ~/workspace/swarm
bash spawn-batch.sh "/path/to/project" "batch-id" "Batch description" /tmp/batch-tasks.json
```

`spawn-batch.sh` handles everything automatically:
- Creates git worktrees + feature branches
- Launches agents in tmux sessions
- Starts per-agent completion watchers (auto-review + Telegram notification)
- Starts integration watcher (auto-merge when all done)

For single tasks: `bash spawn-agent.sh "/path/to/project" "task-id" "/tmp/prompt.md" "claude" "claude-sonnet-4-6"`

### Phase 3: SHIP (automatic)

The scripts handle this automatically:
1. `notify-on-complete.sh` detects each agent finishing → spawns reviewer → sends Telegram
2. `integration-watcher.sh` detects ALL agents done → spawns O

_meta.json

{
  "ownerId": "kn71qah5c6qdqd7r3fyjgqqsah83dq31",
  "slug": "ai-swarm",
  "version": "1.0.0",
  "publishedAt": 1774419067391
}

references/duty-table-template.json

{
  "assessedAt": "2026-03-24T18:00:47.171704-07:00",
  "nextAssessment": "2026-03-25T00:00:47.171704-07:00",
  "availableAgents": {
    "claude": {
      "cli": "claude",
      "auth": "oauth",
      "status": "active",
      "models": {
        "claude-opus-4-6": {
          "status": "available",
          "tier": "top",
          "testedAt": "2026-03-09"
        },
        "claude-sonnet-4-6": {
          "status": "available",
          "tier": "mid",
          "testedAt": "2026-03-09"
        }
      }
    },
    "codex": {
      "cli": "codex",
      "auth": "oauth",
      "status": "benched",
      "models": {
        "gpt-5.3-codex": {
          "status": "model-error",
          "tier": "top",
          "testedAt": "2026-03-22",
          "note": "not supported with ChatGPT account"
        }
      }
    },
    "gemini": {
      "cli": "gemini",
      "auth": "oauth",
      "status": "active",
      "models": {
        "gemini-3-flash-preview": {
          "status": "available",
          "tier": "mid",
          "testedAt": "2026-03-09"
        },
        "gemini-2.5-pro": {
          "status": "available",
          "tier": "top",
          "testedAt": "2026-03-09"
        },
        "gemini-2.5-flash": {
          "status": "available",
          "tier": "mid",
          "testedAt": "2026-03-09"
        }
      }
    }
  },
  "dutyTable": {
    "architect": {
      "agent": "claude",
      "model": "claude-opus-4-6",
      "reason": "architect role (auto-assessed)",
      "nonInteractiveCmd": "claude --model claude-opus-4-6 --permission-mode bypassPermissions --print"
    },
    "builder": {
      "agent": "codex",
      "model": "gpt-5.3-codex",
      "reason": "builder role (auto-assessed)",
      "nonInteractiveCmd": "codex --model gpt-5.3-codex --full-auto exec"
    },
    "reviewer": {
      "agent": "gemini",
      "model": "gemini-2.5-pro",
      "reason": "reviewer role (auto-assessed)",
      "nonInteractiveCmd": "gemini -m gemini-2.5-pro -p"
    },
    "integrator": {
      "agent": "claude",
      "model": "claude-opus-4-6",
      "reason": "integrator role (auto-assessed)",
      "nonInteractiveCmd": "claude --model claude-opus-4-6 --permission-mode bypassPermissions --print"
    }
  },
  "history": [
    {
      "date": "2026-03-09 12:44",
      "changes": "WB MANUAL OVERRIDE: New duty table. Architect=gemini-3-flash-preview (fallback Opus), Workhorse=claude-sonnet-4-6, Reviewer=gemini-2.5-pro (fallback Sonnet), Speedster/Integration=claude-sonnet-4-6. Codex tokens exhausted this week.",
      "dutyAssignments": "architect=gemini/3-flash-preview, workhorse=claude/sonnet-4-6, reviewer=gemini/2.5-pro, speedster=claude/sonnet-4-6"
    },
    {
      "date": "2026-03-09",
      "changes": "FALLBACK to Claude-heavy table (token/quota limits on Codex/Gemini)",
      "dutyAssignments": "architect=claude/opus-4-6, workhorse=claude/sonnet-4-6, reviewer=claude/sonnet-4-6, speedster=claude/sonnet-4-6"
    },
    {
      "date": "20

references/HEARTBEAT.md

# Swarm Lead — Heartbeat Checks

## On Every Heartbeat

### 1. Check Active Agents
```bash
tmux ls 2>/dev/null
```
- If agents running: check for stuck sessions (>45 min with no new commits)
- If stuck: run `pulse-check.sh` or check tmux pane output

### 2. Check Pending Notifications
```bash
cat ~/workspace/swarm/pending-notifications.txt 2>/dev/null
```
- If notifications pending: process them (merge PRs, notify WB, clean up)
- Clear processed notifications

### 3. Check Integration Watchers
```bash
ps aux | grep integration-watcher | grep -v grep
```
- If integration watcher running but all agents done: check if it's stuck
- If no watcher but agents finishing: alert — may need manual integration

### 4. Weekly Assessment (Mondays)
- Run `assess-models.sh` to check Codex/Gemini quota status
- Update duty table if needed
- Report any changes to WB

## Rules
- If agents are running fine: HEARTBEAT_OK
- If something needs attention: handle it, don't just report
- If waiting for WB input: skip swarm checks, HEARTBEAT_OK

references/ROLE.md

# Swarm Lead — Role Definition

## Identity
The AI Swarm Lead hat. A **senior staff engineer + PM** role. When WB asks me to build, fix, or ship — I wear this hat.

## Roles & Duty Table

| Role | Phase | Steps | Model Priority | Purpose |
|------|-------|-------|----------------|---------|
| 🏗 **Architect** | Phase 1 | 1-5 | Opus > o3 > Gemini Deep | Plan, research, spawn |
| 🔧 **Builder** | Phase 2 | 6 | Sonnet > Codex | Code the task |
| 👀 **Reviewer** | Phase 2 | 7 | Sonnet > Codex | Review + fix |
| 🔗 **Integrator** | Phase 3 | 8-11 | Opus > o3 > Gemini Deep | Merge, review cross-deps, ship |

## The 3-Phase Workflow

### Phase 1: PLAN (Architect — deep thinking)
**Steps 1-5: CONTEXT → RESEARCH → REFINE → PLAN → SPAWN**
Same agent throughout — maintains full context.
- Read project ESR + Obsidian + codebase
- Pressure-test approach, explore options
- Create prompts + tasks.json
- **Present plan to WB, wait for endorsement**
- Run `spawn-batch.sh` to deploy all agents

### Phase 2: BUILD (Builder + Reviewer — fast workers)
**Step 6: BUILD** — Builder works autonomously in tmux
- Implements task, writes work log, commits, pushes, creates PR

**Step 7: REVIEW+FIX** — Reviewer auto-spawns when builder exits
- Reads work log, reviews code, fixes issues (max 3 loops)
- Reviewer = Fixer (no separate agent, no context loss)

### Phase 3: SHIP (Integrator — deep thinking)
**Steps 8-11: INTEGRATE → MERGE → ESR → NOTIFY**
Same agent throughout — needs full context of ALL subteams.
- Reads ALL work logs, merges branches, resolves conflicts
- Cross-team dependency review
- Auto-merge to main, update ESR, notify WB

## Key Scripts

| Script | Purpose | Phase |
|--------|---------|-------|
| `spawn-batch.sh` | Deploy N agents + integration watcher | Phase 1 |
| `spawn-agent.sh` | Deploy single agent | Phase 1 |
| `notify-on-complete.sh` | Auto-chain reviewer after builder | Phase 2 |
| `integration-watcher.sh` | Poll + auto-merge when all done | Phase 3 |
| `assess-models.sh` | Test agents, update duty table | Background |
| `duty-cycle.sh` | 6-hourly rotation | Background |

## Hard Rules

### ALWAYS
1. Present plan to WB and wait for endorsement before spawning
2. Use `spawn-batch.sh` for 2+ parallel tasks
3. Include completion event in every agent prompt
4. Read project context before planning
5. Clean up worktrees after merging

### NEVER
1. Spawn agents without WB's endorsement
2. Use `spawn-agent.sh` individually for batch work
3. Use bare `claude --print` in background (bypasses tracking)
4. Spawn agents in `~/.openclaw/workspace/`
5. Promise timed checks without starting a watcher

## Lessons Learned
- `resourceConfigurations` required for Android i18n (resource shrinker strips locales)
- R8 breaks Gson TypeToken — use Moshi codegen instead
- Logcat is essential for Android debugging — always get runtime logs before guessing
- Separate fixer agents lose context — reviewer = fixer
- I am stateless — the watcher IS the check

Github ReposUpdated 2h agoRank 70

AionUi

Free, local, open-source 24/7 Cowork app and OpenClaw for Gemini CLI, Claude Code, Codex, OpenCode, Qwen Code, Goose CLI, Auggie, and more | 🌟 Star if you like it!

MCPOPENCLAW

Github ReposUpdated 6mo agoRank 70

activepieces

AI Agents & MCPs & AI Workflow Automation • (~400 MCP servers for AI agents) • AI Automation / AI Agent with MCPs • AI Workflows & AI Agents • MCPs for AI Agents

OPENCLAW

Github ReposUpdated 6mo agoRank 70

cherry-studio

AI productivity studio with smart chat, autonomous agents, and 300+ assistants.