agi-farm

v3.3.3

Published

7 days ago

Multi-agent AI team builder for OpenClaw — bootstrap complete teams with Paperclip dashboard, auto-dispatcher, and infrastructure

0High
0Medium
0Low

oabdelmksoud

openclaw openclaw-plugin openclaw-extension multi-agent agent-orchestration agent-team agentic-workflow team-management auto-dispatcher paperclip dashboard cli-wizard agi collaborative-ai

🦅 AGI Farm

One wizard. Full multi-agent AI team. Live in minutes.

Build production-ready AI teams with a single command

Quick Start • Features • Documentation • Architecture

[!WARNING] Notice: AGI Farm is currently under active development and still has known bugs. Please use with caution in production environments.

🎯 What It Does

AGI Farm is an OpenClaw plugin that bootstraps a fully operational multi-agent AI system with everything you need:

| Feature | Description | |---------|-------------| | 🧙 Interactive Wizard | 6 questions → complete team setup in ~2 minutes | | 🤖 Multi-Agent Teams | Choose 3, 5, or 11 pre-wired specialist agents | | 📡 Live Ops Dashboard | React + SSE with real-time updates and interactive HITL/Cron controls | | 🔄 Auto-Dispatcher | Smart task delegation with HITL, backoff & dependencies | | 📦 Portable Bundle | Export your entire team to GitHub with one command | | 🏗️ ESM Native | Built for Node 20+ with full ES Module support | | 🛡️ Production Hardened | Security-audited with CSRF, Origin validation, CSP headers, input sanitization, and timing-safe auth | | 🔐 Security Dashboard | Real-time vulnerability scanning, auto-fix, and security history tracking | | 🔄 Auto-Update | Detects new releases on GitHub with one-click install from the dashboard | | ⚙️ Feature-Flagged Runtime | Optional jobs, skills, memory, policy, approvals, and metering modules | | 🧪 103 Automated Tests | Unit + integration + API smoke coverage with Jest/ESM |

🙏 Upstream Acknowledgement

Parts of AGI Farm dashboard direction are inspired by the excellent work in LobsterBoard by Curbob:

Upstream repository: https://github.com/Curbob/LobsterBoard
Upstream license: Business Source License 1.1 (BSL-1.1)

AGI Farm does not copy LobsterBoard code directly; features are re-implemented for AGI Farm architecture and compatibility.

🔄 Always-Sync Policy

To stay aligned with upstream updates, this repo now includes:

Automated scheduled sync workflow: .github/workflows/lobsterboard-upstream-sync.yml
Local sync script: npm run sync:lobsterboard
Snapshot mirror for review/porting: upstream/lobsterboard/
Tracking file: lobsterboard-resources/LOBSTERBOARD_VERSION
Parity plan: docs/LOBSTERBOARD_PARITY.md

This keeps upstream diffs visible in PRs so feature ports can be done continuously.

🧙 What's New in v3.3.0

🚀 Automated Release & Workflow Hardening

Version 3.3.0 standardizes the release process and ensures all GitHub workflows are production-ready.

🏗️ Automated Dashboard Launch

One-Click Wizard: The agi-farm setup wizard now automatically opens the Paperclip dashboard in your default browser upon successful configuration.
Direct Dependencies: Added express as a core dependency for improved dashboard server stability.

🛡️ Workflow & Compatibility Hardening

OpenClaw Compatibility: Improved version scrubbing (handles v prefixes and revision suffixes) for more robust integration testing.
Graceful Installation: Workflows now handle missing NPM versions gracefully, skipping integration tests if a specific version is not yet published.
Global PATH Reliability: Ensured global npm bin directories are correctly added to GITHUB_PATH across all workflow steps.

🔐 Security Enhancements

Automated Secret Scrubbing: Upstream sync scripts (scripts/sync-lobsterboard-upstream.js) now redact Google OAuth Client IDs and Secrets to prevent GitHub Push Protection violations.
Refined Command Execution: All agi-farm commands in workflows now use npx for guaranteed executability.

🧙 What's New in v1.9.0

🚀 Advanced Wizard 2.0 — Industry-Driven Automation

The setup wizard has been completely overhauled from a basic 4-template system to a sophisticated, 6-phase onboarding experience that activates the full depth of the AGI Farm plugin.

🏗️ 15 Industry-Specific Blueprints

Choose from professional team compositions across 5 major verticals:

Software Engineering: Startup MVP, Full-Stack Product, Mobile-First App, AI/ML System.
Marketing & Growth: Marketing Campaign, Brand Launch, Performance Marketing.
Enterprise & Regulated: Enterprise Feature, Security-Critical, Compliance & Audit.
Research & Development: Quality-First, Research & Discovery.
Creative & Content: Content Studio, Product Design Sprint.

⚙️ Deep Plugin Integration

The wizard now automates what used to be manual post-setup tasks:

Budget & OKRs: Set monthly spend limits and seed your team with industry-specific OKRs immediately.
Operational Automation: Select and activate pre-built cron jobs (daily security scans, weekly velocity reports, auto-dispatcher heartbeats).
HITL Security Policy: Choose your "Human-in-the-Loop" sensitivity level (Low/Medium/High) during setup.
Project Seeding: Start with a pre-populated project and task backlog tailored to your chosen blueprint.
AgentShield Integration: One-click initial security scan to baseline your new workspace.

🧹 Consolidated Architecture

Legacy setup-enhanced.js has been consolidated into the main setup.js.
Performance improvements in agent creation and SOUL.md template generation.
Expanded WORKFLOW_TEMPLATES.md documentation covering all 15 new blueprints.

🆕 What's New in v1.8.0

🎨 Complete Dashboard Redesign — Day Theme (Minimalism 3.0 + Glassmorphism 2.0)

The AGI Farm dashboard has been completely redesigned with a modern, light-mode UI. Every single one of the 26 tabs has been rewritten from scratch with a premium day-themed aesthetic.

✨ Design System

Off-white/pearl backgrounds (#F8FAFC slate palette) with glassmorphism cards and subtle depth shadows
Premium typography — Inter for UI, JetBrains Mono for technical values
Accent palette — Indigo primary, Emerald mint for success, Amber for warnings, Violet for AI/HITL, Red for errors
Hover-lift animations, smooth fade-in transitions, and pill-shaped status badges throughout
100% light mode — All dark #050505/#000/rgba(0,0,0) backgrounds purged from every file

📋 All 26 Tabs Redesigned

| Pillar | Tabs | |--------|------| | Agent Intelligence | Overview, Agents, Tasks, HITL, Alerts, Velocity, Budget | | Project Oversight | Projects, OKRs, Approvals, Knowledge, Comms | | System | Security, Jobs & Crons, Processes, Policies, Skills, Usage, Memory, Failures, Decisions, Audit Log, Settings, R&D |

🏗️ Key Changes

Projects tab — Full rewrite with light progress rings, health badges, tabbed detail view (Overview / Timeline / Tasks / Budget / OKRs / Logs), light Gantt chart and burndown chart
HITL tab — Soft-purple glass cards with clear Approve/Reject actions and status banner
Velocity tab — KPI cards + area chart + success rate trend line
Approvals tab — Card layout matching HITL with optional notes field
Settings tab — Toggle switches, budget limit inputs, workspace info grid
Crons tab — Status-dot rows with Run / Pause / Resume action buttons
Knowledge tab — Masonry card grid with tag pills and expand-to-read
Comms tab — Agent list sidebar with Inbox / Outbox panels
OKRs tab — Objective cards with animated KR progress bars
R&D tab — Light experiment table, benchmark leaderboard, and evolution pipeline backlog

🧹 Legacy Cleanup

Deleted TopBar.jsx and Sidebar.jsx (unused dark-theme stubs)
Removed all cyberpunk/neon language (SYSTEM_LIVE, ENCRYPTED, VECTOR_ID, NEURAL, etc.)
All chart grid lines and tooltips updated to light theme

Previous: v1.7.2 — Enhanced Wizard with Workflow Templates + Agent Browser

4 workflow templates — Pre-built teams for common use cases: Startup MVP (5 agents), Marketing Campaign (5 agents), Enterprise Feature (6 agents), Quality-First (4 agents)
Agent browser — Browse and select from all 91 agents across 11 categories (AGI Farm + ECC + Agency-Agents)
Auto-template copying — SOUL.md templates automatically copied to agent workspaces during setup
Workflow documentation — Each template links to orchestration patterns and quick-start guides
Backward compatible — Legacy wizard (3/5/11 fixed rosters) preserved
Zero manual setup — Agency-Agents templates work out-of-box with no file copying required

Run enhanced wizard: node scripts/setup-enhanced.js

Previous: v1.7.1 — Architectural Hardening

🔒 Architectural Hardening

Per-file mutex locking — All service read-modify-write operations (JobsService, ProjectService, TaskService, PolicyService, MeteringService, MemoryService, SkillsService) now serialize concurrent writes via withFileLockSync(), preventing data loss from race conditions
CSRF token rotation — Token rotates every 24 hours with 5-minute grace period for active connections; static tokens from env var exempt
Audit log pagination — /api/audit now supports offset and limit query params with memory-efficient streaming; warns on malformed log lines
Session rate limiting — /api/session endpoint has its own stricter rate limit (20 req/min) to prevent token enumeration
Knowledge/cron file locking — Dashboard direct file operations (knowledge CRUD, cron toggle) now also use withFileLockSync()

Previous: v1.7.0 — Dashboard Hardening & Security Overhaul

🛡️ Dashboard Hardening & Security Overhaul

Deep server-side audit — 20 issues identified and remediated across data integrity, endpoint security, SSE reliability, and file I/O
SSE exponential backoff — Reconnection now backs off 3s → 6s → 12s → ... → 60s max, resets on success (prevents server thrashing)
Input sanitization everywhere — New sanitizeText() helper applied to broadcast, knowledge, comms, and approval endpoints; strips control characters
CSRF hardening — /api/stream and /api/data now validate CSRF tokens; SSE accepts token via query parameter
Workspace validation — Server validates workspace directory exists and is writable at startup, exits cleanly on failure
Broadcaster thread-safety — Fixed Set mutation during iteration in SSE fan-out
Silent failure logging — readJson()/readMd() now log warnings on file read failures instead of silently returning empty values
Null safety sweep — Comprehensive guards across all 20+ dashboard tabs preventing crashes on missing/malformed data
Component prop consistency — Fixed mismatched prop names between server snapshot and frontend components

🔧 5 New Dashboard Tabs + Security Dashboard

Decisions tab — View and manage policy decisions and approval workflows
Failures tab — Track and analyze job/task failures with error details
Processes tab — Monitor running agent processes and system health
R&D tab — Experiments tracking with status and results visualization
Settings tab — Configure project defaults and feature flags from the UI
Security Dashboard — Real-time vulnerability scanning, auto-fix capabilities, and security history

📊 Dashboard Data & API Completeness

Missing endpoint coverage — Added server handlers for all frontend API calls (tasks, projects, jobs, intake, comms, knowledge)
Project enrichment — Server-side project enrichment with task counts, progress, team agents, and risk indicators
Auto-derived tasks — Tasks automatically derived from jobs with rootTaskId for complete task visibility
Error feedback — User-facing error messages and loading states across all interactive components

Previous: v1.6.0 — Agency-Agents Integration

🎭 Agency-Agents Integration: 59 Specialized Personalities

59 battle-tested agent templates from @msitarzewski's Agency-Agents
11 specializations: Engineering, Design, Marketing, Product, PM, Testing, Support, Spatial Computing, Strategy, and more
Production-proven workflows with 10,000+ lines of personality definitions and code examples
Standout agents: Agents Orchestrator, Evidence Collector, Reality Checker, Whimsy Injector, Reddit Community Builder
Total library: 91 agents (16 AGI Farm + 16 ECC + 59 Agency-Agents)
Zero dependencies — Pure markdown templates, instantly usable
See: Agency-Agents Integration Guide

🔄 Enhanced Orchestration Patterns

Pipeline orchestration for Cooper: PM → Architect → [Dev ↔ QA Loop] → Integration
Quality gate patterns for Vigil: Evidence Collector + Reality Checker methodologies
4 workflow templates: Startup MVP, Marketing Campaign, Enterprise Feature, Quality-First teams
Automated update system for keeping Agency-Agents templates current
See: Orchestration Patterns | Quality Gates | Workflow Templates

Previous: ECC Integration v1.5.0

69 specialized skills from Everything Claude Code (Anthropic hackathon winner)
16 ECC agent templates for domain-specific delegation
33 slash commands for common workflows
TDD & Security-First — 80%+ test coverage enforced, comprehensive security scans
See: ECC Integration Guide

Previous: Dashboard UI v2: Agent OS Overhaul (v1.4.0)

🛰️ Agent OS Aesthetic: Professional, minimalist, and high-contrast obsidian design.
📉 Strategic Command Matrix: High-density project matrix with integrated Gantt and Burndown visualizations.
🧬 Fleet Pulse signatures: Minimalist neural activity and heartbeat monitoring for the entire fleet.
📊 Efficiency Analytics: Professional data visualizations for agent velocity and task throughput.
⚡ Performance Optimization: 60% reduction in dashboard asset size and faster SSE fan-out.

🆕 Previous Updates (v1.2.0 - v1.3.x)

Previous Highlights (v1.1.x)

🔄 GitHub release detection & Auto-Update system directly from dashboard
⚙️ Feature-Flagged Core Runtime (Jobs, Approvals, Memory, Skills)
🛡️ Extensive Security Hardening (CSRF, CSP, token validation, atomic file writes)

📸 Preview

Dashboard Overview

┌──────────────────────────────────────────────────────────────────────┐
│  🦅 AGI Ops Room   ● LIVE   Online: 8/11  Pending: 4  Budget: $45  │
├──────────────────────────────────────────────────────────────────────┤
│  🔄 Update available: v1.0.2 → v1.1.0  [Release Notes] [Update Now]│
├──────────────────────────────────────────────────────────────────────┤
│ Overview │ Agents │ Tasks │ Projects │ Crons │ HITL │ Alerts │ ...  │
├──────────────────────────────────────────────────────────────────────┤
│                                                                      │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐  ┌──────────┐            │
│  │ 📊 42    │  │ ✅ 38    │  │ ⏳ 4      │  │ 🚨 2     │            │
│  │ Tasks    │  │ Done     │  │ Active   │  │ HITL     │            │
│  └──────────┘  └──────────┘  └──────────┘  └──────────┘            │
│                                                                      │
│  ┌───────────────────────────────────────────────────────────┐      │
│  │ 🔮 Sage    ✅ Ready  📨 3 msgs  ⭐ 94%  [Send Message]   │      │
│  │ ⚒️ Forge   🔄 Busy   📨 7 msgs  ⭐ 89%  [Send Message]   │      │
│  │ 🐛 Pixel   ✅ Ready  📨 2 msgs  ⭐ 96%  [Send Message]   │      │
│  └───────────────────────────────────────────────────────────┘      │
│                                                                      │
└──────────────────────────────────────────────────────────────────────┘

Team Architecture

┌─────────────────────────────────────────────────────────────────┐
│                    11-Agent Team (Full Stack)                   │
├─────────────────────────────────────────────────────────────────┤
│                                                                 │
│                      🦅 Cooper (Orchestrator)                   │
│                              │                                  │
│              ┌───────────────┼───────────────┐                 │
│              │               │               │                  │
│         🔮 Sage          ⚒️ Forge        🐛 Pixel              │
│      (Architect)     (Builder)      (Debugger)                 │
│              │               │               │                  │
│         🔭 Vista        🔊 Cipher       🛡️ Vigil               │
│       (Analyst)      (Knowledge)       (QA)                    │
│              │               │               │                  │
│         ⚓ Anchor        📡 Lens        🔄 Evolve               │
│      (Content)      (Multimodal)   (Process)                   │
│                              │                                  │
│                          🧪 Nova                                │
│                         (R&D Lead)                              │
│                                                                 │
└─────────────────────────────────────────────────────────────────┘

🚀 Quick Start

One-Step Install & Setup

The recommended way to install and configure AGI Farm is with this single command. It will globally install the CLI and immediately launch the interactive setup wizard:

npm install -g agi-farm && agi-farm setup

Alternative Installation

If you prefer to install it purely as an OpenClaw plugin (without the global CLI binary), you can use the built-in plugin manager, but you must use the npx runner to execute the setup wizard:

openclaw plugins install agi-farm
npx agi-farm setup

Upgrade (Existing Install)

# Update an already installed OpenClaw plugin
openclaw plugins update agi-farm

# One-liner: install first, or update if already installed
openclaw plugins install agi-farm || openclaw plugins update agi-farm

Run Setup Wizard

# Start the interactive wizard anytime
agi-farm setup

Answer the setup prompts and your team will be live in ~2 minutes:

🧙 AGI Farm Setup Wizard

? Team name (e.g., "CooperCorp"): MyTeam
? Orchestrator name: Cooper
? Team size: 11 agents (Full Stack)
? Domain: general
? Collaboration frameworks: langgraph
? Auto-create project channel per project: Yes
? Default execution path: AGI-Farm first
? Confirm setup? Yes

✅ Creating 11 agents...
✅ Setting up workspace...
✅ Initializing AGI Farm registries...

🎉 Your AI team is ready!

Setup now captures project defaults in PROJECTS.json, including:

auto_project_channel (default: true)
execution_path (default: agi-farm-first)

This makes AGI Farm the default execution path for newly created projects.

Project Defaults

New projects inherit defaults from PROJECTS.json:

Auto Project Channel: when enabled, each new project gets a dedicated channel
Execution Path: choose agi-farm-first (multi-agent routing) or direct-first (immediate execution)

View and edit defaults via:

Dashboard → Settings tab
GET /api/projects → defaults field
PATCH /api/projects/defaults API endpoint

📦 Commands

| Command | CLI Shortcut | Description | |---------|-------------|-------------| | 🎯 agi-farm setup | agi-farm | Full wizard — agents, workspace, crons (Classic mode: 3/5/11 agents) | | ✨ node scripts/setup-enhanced.js | — | Enhanced wizard (v1.7.0) — Workflow templates + Agent browser (91 agents) | | 🗑️ agi-farm teardown | agi-farm-teardown | Team teardown — removes AGI Farm agents, bundle, and workspace registries | | 📊 agi-farm status | agi-farm-status | Team health: agents, tasks, cron status | | 🔧 agi-farm rebuild | agi-farm-rebuild | Regenerate workspace from bundle | | 📤 agi-farm export | agi-farm-export | Push bundle to GitHub | | 🖥️ agi-farm dashboard | agi-farm-dashboard | Launch live ops room (SSE, :8080) | | ⚡ agi-farm dispatch | agi-farm-dispatch | Run auto-dispatcher manually | | 🍎 agi-farm launchagent | agi-farm-launchagent | Install/uninstall macOS LaunchAgent for persistent dashboard |

🆕 Enhanced Wizard (v1.7.0)

The enhanced wizard offers 4 workflow templates and agent browsing for all 91 agents:

Run it:

cd ~/.openclaw/extensions/agi-farm  # or global npm path
node scripts/setup-enhanced.js

Features:

🚀 Startup MVP (5 agents, 1-2 weeks) — Cooper, Forge, Pixel, Vigil, Growth Hacker
📈 Marketing Campaign (5 agents, 2-4 weeks) — Cooper, Content Creator, Twitter Engager, Reddit Builder, Analytics
🏢 Enterprise Feature (6 agents, 4-8 weeks) — Cooper, Vista, Sage, Vigil, Experiment Tracker, Reality Checker
🔬 Quality-First (4 agents, quality-driven) — Cooper, Vigil (Evidence), Reality Checker, Performance Benchmarker
🎨 Custom — Browse all 91 agents by category and select your own team

Automatic template copying: Agency-Agents SOUL.md templates are copied to agent workspaces automatically.

🍎 Persistent Dashboard (macOS LaunchAgent)

The dashboard can run as a macOS LaunchAgent — it starts on login and auto-restarts if it crashes, independent of the OpenClaw gateway lifecycle.

Install

# Install with defaults (port 8080, localhost)
agi-farm-launchagent

# Custom port and workspace
agi-farm-launchagent --port 9090 --workspace ~/my-workspace

Uninstall

agi-farm-launchagent --uninstall

Why use this?

The plugin lifecycle (onLoad) spawns the dashboard as a child process. If the gateway exits, restarts, or doesn't reliably complete the lifecycle, the dashboard dies with it. The LaunchAgent runs the dashboard as an independent OS-level service:

RunAtLoad — starts automatically on login
KeepAlive — restarts if it crashes
Logs — stdout/stderr saved to /tmp/openclaw/agi-farm-dashboard.log

Linux users: Use systemd with a similar service unit. See the plist template in templates/ for reference.

🏗️ Architecture

Plugin Structure

.openclaw/extensions/agi-farm/ (or global node_modules/agi-farm/)
├── 📦 package.json              Plugin manifest (ESM)
├── ⚙️ openclaw.plugin.json     Config schema & commands
├── 📂 dist/                    Compiled TypeScript (backend)
├── 📂 dashboard-dist/          Built React frontend (production)
├── 🌐 server/
│   ├── 🖥️ dashboard.js         SSE server + CRUD API (Node.js)
│   ├── 🔄 updater.js           GitHub release checker + auto-update
│   └── 🛠️ utils.js             Core parsing & logic (Unit Tested)
├── 📜 scripts/
│   ├── 🎯 setup.js             Setup wizard
│   ├── 🗑️ teardown.js          Uninstall mechanism
│   ├── 📊 status.js            Status checker
│   ├── 🔧 rebuild.js           Rebuilder
│   ├── 📤 export.js            GitHub exporter
│   ├── 🖥️ dashboard.js         Dashboard launcher
│   ├── ⚡ dispatch.js          Auto-dispatcher
│   └── 📂 lib/
│       └── 🛠️ run-command.js   Shared CLI utility
├── 📋 templates/               Agent & workspace templates
├── ⚛️ dashboard-react/         Vite + React 18 source (Dev)
└── 📚 skills/
    └── 📖 SKILL.md             Embedded documentation

Dashboard Architecture

┌──────────────────────────────────────────────────────────────┐
│                    Dashboard Data Flow                        │
├──────────────────────────────────────────────────────────────┤
│                                                              │
│  📂 Workspace Files                                          │
│  ├── TASKS.json                                              │
│  ├── AGENT_STATUS.json                                       │
│  └── comms/broadcast.md                                      │
│           │                                                  │
│           │ file change (250ms debounce)                     │
│           ▼                                                  │
│  ┌─────────────────────┐                                    │
│  │ WorkspaceWatcher    │ 👁️ chokidar file watcher          │
│  └─────────────────────┘                                    │
│           │                                                  │
│           ▼                                                  │
│  ┌─────────────────────┐    ┌──────────────────┐           │
│  │ SlowDataCache       │    │ Broadcaster      │           │
│  │ (30s cache)         │───▶│ (SSE fan-out)    │           │
│  └─────────────────────┘    └──────────────────┘           │
│           │                          │                       │
│  ┌─────────────────────┐             │ SSE stream            │
│  │ UpdateChecker       │             ▼                       │
│  │ (6h GitHub cache)   │   ┌──────────────────┐             │
│  └─────────────────────┘   │ React Frontend   │             │
│                             │ (Vite + Recharts)│             │
│                             └──────────────────┘             │
│                                                              │
└──────────────────────────────────────────────────────────────┘

Real-time Data Flow

┌──────────────┐     250ms      ┌──────────────┐     SSE      ┌──────────────┐
│  Workspace   │ ──────────────▶│   Server     │ ───────────▶ │   Browser    │
│   Files      │    chokidar    │  dashboard.js│   ~350ms     │   React      │
└──────────────┘                └──────────────┘              └──────────────┘
     ▲                                                           │
     │                                                           │
     │                      ┌──────────────┐                     │
     └──────────────────────│   OpenClaw   │◀────────────────────┘
          CLI commands      │     CLI      │    User actions
                            └──────────────┘

🤖 Team Presets

3-Agent (Minimal) 🏃

Perfect for simple workflows and quick prototypes.

🦅 Orchestrator
    │
    ├──▶ 🔮 Researcher
    │        │
    └──▶ ⚒️ Builder

Best for: Quick prototypes, simple automation, learning

5-Agent (Standard) ⚡

Balanced team for production workloads.

                🦅 Orchestrator
                     │
         ┌───────────┼───────────┐
         │                       │
    🔮 Researcher           🛡️ QA
         │                       │
    ⚒️ Builder             ⚓ Content

Best for: Production apps, balanced workloads, medium complexity

11-Agent (Full Stack) 🚀

Complete team for complex systems - RECOMMENDED

                        🦅 Cooper
                      (Orchestrator)
                           │
        ┌──────────────────┼──────────────────┐
        │                  │                  │
   🔮 Sage            ⚒️ Forge           🐛 Pixel
 (Architect)        (Builder)         (Debugger)
        │                  │                  │
   🔭 Vista          🔊 Cipher         🛡️ Vigil
  (Analyst)        (Knowledge)           (QA)
        │                  │                  │
   ⚓ Anchor           📡 Lens          🔄 Evolve
  (Content)       (Multimodal)       (Process)
                           │
                       🧪 Nova
                      (R&D Lead)

Best for: Complex systems, enterprise, full-stack development

🧠 Model Selection Guide

Choose the right model for each role to optimize cost and quality:

| Role | Recommended Model | Why | Cost/Quality | |------|-----------|-----|--------------| | 🦅 Orchestrator | anthropic/claude-3-5-sonnet | High-level planning & delegation | 💰💰💰 / ⭐⭐⭐ | | 🔮 Architect | anthropic/claude-3-opus | Deep reasoning & system design | 💰💰💰 / ⭐⭐⭐ | | ⚒️ Engineer | google/gemini-1.5-pro | Reliable code gen & large context | 💰💰 / ⭐⭐ | | 🐛 Debugger | anthropic/claude-3-5-sonnet | Precision logic & error analysis | 💰💰 / ⭐⭐⭐ | | 🔭 Analyst | google/gemini-2.0-flash | Ultra-fast data synthesis | 💰 / ⭐⭐ | | 🛡️ QA | anthropic/claude-3-haiku | Rapid pattern validation | 💰 / ⭐ | | ⚓ Content | google/gemini-1.5-pro | Rich multimodal generation | 💰💰 / ⭐⭐⭐ | | 🧪 R&D | deepseek/deepseek-reasoner | Structured creative experimentation | 💰 / ⭐⭐⭐ |

⚙️ Configuration

Configure AGI Farm in your openclaw.json:

{
  "plugins": {
    "entries": {
      "agi-farm": {
        "enabled": true,
        "config": {
          "dashboardPort": 8080,
          "dashboardHost": "127.0.0.1",
          "autoStartDashboard": true,
          "autoCheckUpdates": true,
          "workspacePath": "~/.openclaw/workspace",
          "bundlePath": "~/.openclaw/workspace/agi-farm-bundle",
          "featureJobs": true,
          "featureSkills": true,
          "featureMemory": true,
          "featurePolicy": true,
          "featureMetering": true,
          "featureApprovals": true
        }
      }
    }
  }
}

Configuration Options

| Option | Type | Default | Description | |--------|------|---------|-------------| | dashboardPort | number | 8080 | Port for live ops dashboard | | dashboardHost | string | "127.0.0.1" | Bind address for dashboard | | autoStartDashboard | boolean | true | Auto-start dashboard on load | | autoCheckUpdates | boolean | true | Check GitHub for new releases on startup | | workspacePath | string | ~/.openclaw/workspace | Path to OpenClaw workspace | | bundlePath | string | /agi-farm-bundle | Path to bundle directory | | featureJobs | boolean | true | Enable jobs runtime APIs + background worker | | featureSkills | boolean | true | Enable skills registry and routing endpoints | | featureMemory | boolean | true | Enable memory indexing + search endpoint | | featurePolicy | boolean | true | Enable policy evaluation on runtime/mutation actions | | featureMetering | boolean | true | Enable usage metering collection + API | | featureApprovals | boolean | true | Enable approval workflows for policy-gated actions |

Runtime Files Added By Core Modules

When feature flags are enabled, AGI Farm lazily creates these additive workspace files:

JOBS.json
JOB_RUNS.jsonl
SKILLS_REGISTRY.json
MEMORY_INDEX.json
POLICIES.json
APPROVALS.json
AUDIT_LOG.jsonl
USAGE_METERING.json
SECRETS/ (encrypted blobs + metadata)

📊 Dashboard Data Sources

All data updates in real-time from workspace files:

| Field | Source | Refresh | Latency | |-------|--------|---------|---------| | 📋 Tasks | TASKS.json | Instant | ~50ms | | 🤖 Agents | AGENT_STATUS.json | Instant | ~50ms | | 📬 Comms | comms/inboxes/*.md | Instant | ~50ms | | 🚨 Alerts | ALERTS.json | Instant | ~50ms | | 📂 Projects | PROJECTS.json | Instant | ~50ms | | 💰 Budget | BUDGET.json | Instant | ~50ms | | 📈 Velocity | VELOCITY.json | Instant | ~50ms | | 🎯 OKRs | OKRs.json | Instant | ~50ms | | 📢 Broadcast | comms/broadcast.md | Instant | ~50ms | | 🧪 Experiments | EXPERIMENTS.json | Instant | ~50ms | | 📚 Knowledge | SHARED_KNOWLEDGE.json | Instant | ~50ms | | 🧠 Memory | MEMORY.md | Instant | ~50ms | | 🔄 Agent Models | openclaw agents list | Cached | ~30s | | 🗂️ Jobs | JOBS.json | Instant | ~50ms | | ✅ Approvals | APPROVALS.json | Instant | ~50ms | | 📊 Usage | USAGE_METERING.json | Instant | ~50ms |

Interactive Actions (API)

The dashboard enables direct control over team operations via authenticated REST endpoints (all require CSRF token):

HITL & Cron Controls:

POST /api/hitl/:id/approve — Continue task with optional notes
POST /api/hitl/:id/reject — Block task and notify agent
POST /api/cron/:id/trigger — Manually run a specific cron job
POST /api/cron/:id/toggle — Enable or disable a cron job
POST /api/jobs — Create a background job from high-level intent
GET /api/jobs / GET /api/jobs/:id — List and inspect jobs
POST /api/jobs/:id/cancel / POST /api/jobs/:id/retry — Control failed/running jobs
GET /api/skills + POST /api/skills/:id/(enable|disable) — Manage skill activation
GET /api/memory/search?q=&tags= — Search memory index
GET /api/policies — Retrieve active policy rules
GET /api/approvals + POST /api/approvals/:id/(approve|reject) — Human approval queue
GET /api/usage — Usage and cost aggregates for dashboard

CRUD Operations:

POST /api/tasks — Create a new task with assignee, priority, dependencies
POST /api/comms/:id/send — Send a message to a specific agent inbox
POST /api/broadcast — Post a team-wide broadcast message
POST /api/knowledge — Add a shared knowledge entry
DELETE /api/knowledge/:id — Remove a knowledge entry

LobsterBoard Parity Foundations (P0):

GET /api/auth/status — PIN/public-mode status + write unlock state
POST /api/auth/verify-pin — Verify PIN and issue write auth token
POST /api/auth/set-pin / POST /api/auth/remove-pin — Manage dashboard PIN lock
POST /api/auth/public-mode — Toggle read-only public mode
GET /api/secrets / GET /api/secrets/:scope — List masked secrets metadata
POST /api/secrets/:scope / DELETE /api/secrets/:scope/:key — Store/delete scoped secrets
GET /api/templates / GET /api/templates/:id — List and inspect dashboard templates
POST /api/templates/export / POST /api/templates/import — Export/import template settings

Integrated LobsterBoard Runtime (same host/process):

Full LobsterBoard runtime is mounted internally at GET /lobsterboard/
Dashboard tab LobsterBoard loads this internal route (no separate user-facing port required)
API/static calls from LobsterBoard are proxied through AGI Farm server under the same origin

Auto-Update:

GET /api/update-check — Force a fresh GitHub release check
POST /api/update-install — Install latest version via npm

Total push latency: ~350ms from file change to browser update

⚡ Auto-Dispatcher

The auto-dispatcher intelligently routes tasks to agents:

┌──────────────────────────────────────────────────────────────┐
│                    Auto-Dispatcher Flow                       │
├──────────────────────────────────────────────────────────────┤
│                                                              │
│  📋 Pending Tasks                                            │
│       │                                                      │
│       ▼                                                      │
│  ┌─────────────┐     No     ┌──────────────┐                │
│  │ HITL Check  │───────────▶│   Skip Task  │                │
│  └─────────────┘            └──────────────┘                │
│       │ Yes                                                  │
│       ▼                                                      │
│  ┌─────────────┐     Yes    ┌──────────────┐                │
│  │ Rate Limit? │───────────▶│ Backoff Wait │                │
│  └─────────────┘            └──────────────┘                │
│       │ No                                                   │
│       ▼                                                      │
│  ┌─────────────┐     No     ┌──────────────┐                │
│  │ Deps Met?   │───────────▶│   Skip Task  │                │
│  └─────────────┘            └──────────────┘                │
│       │ Yes                                                  │
│       ▼                                                      │
│  ┌─────────────┐                                            │
│  │ Fire Agent  │ ▶ openclaw agents run <agent> <task>       │
│  └─────────────┘                                            │
│                                                              │
└──────────────────────────────────────────────────────────────┘

Features

✅ HITL Integration — Human-in-the-loop decision requests
✅ Rate-Limit Backoff — Exponential backoff on API limits (default: 10min)
✅ Dependency Checking — Respects task dependencies
✅ Cron Schedule — Runs automatically every 1 minute

# Manual dry-run
agi-farm dispatch

# Check logs
tail -f ~/.openclaw/workspace/logs/auto-dispatch.log

🛠️ Installation from Source

# Clone the repository
git clone https://github.com/oabdelmaksoud/AGI-FARM-PLUGIN.git
cd AGI-FARM-PLUGIN

# Install dependencies
npm install

# Build backend and dashboard
npm run build:all

# Link to OpenClaw extensions
ln -s $(pwd) ~/.openclaw/extensions/agi-farm

If you later switch to openclaw plugins install agi-farm, remove the dev symlink first: rm -f ~/.openclaw/extensions/agi-farm

🧹 Uninstall Guide

# 1) Remove AGI Farm team data (agents, bundle, registries)
agi-farm teardown

# 2) Remove plugin from OpenClaw
openclaw plugins uninstall agi-farm --force

# 3) Optional: remove standalone global CLI binary
npm uninstall -g agi-farm

Notes:

agi-farm teardown does not uninstall the plugin package.
openclaw plugins uninstall agi-farm --force removes the plugin install used by OpenClaw.

💻 Development

# Build everything (TS + Dashboard)
npm run build:all

# Build backend only
npm run build

# Watch mode for development
npm run dev

# Type checking
npm run typecheck

# Linting
npm run lint

# Run unit tests (Jest + ESM)
npm test

# Start dashboard server manually
npm run start-dashboard

🔧 Troubleshooting

| Symptom | Fix | Command | |---------|-----|---------| | ❌ Plugin fails to load | Check global install | npm list -g agi-farm | | ❌ plugin already exists on install | Use plugin update command | openclaw plugins update agi-farm | | 📊 Dashboard shows stale data | Sync with workspace | agi-farm status | | 🤖 Agent stuck >30 min | Verify heartbeats | cat ~/.openclaw/workspace/HEARTBEAT.md | | ⚠️ openclaw not found | Add to PATH | export PATH=$PATH:$(npm bin -g) | | 🔐 Access Denied | Check npm login | npm whoami | | ⏰ Cron registration error | Clean crontab | openclaw cron list --fix | | 📄 JSON Parse Error | Re-run setup | agi-farm setup --force |

🔒 Security

This plugin is designed with defense-in-depth security:

| Layer | Protection | |-------|-----------| | Network | Dashboard binds to 127.0.0.1 only — not exposed to LAN or internet | | Origin validation | /api/session gated by Origin/Referer — cross-origin token theft blocked | | CSRF tokens | All mutation endpoints require timing-safe CSRF token comparison; token rotates every 24h with 5-min grace period | | SSE authentication | /api/stream and /api/data require CSRF token — prevents cross-origin data exfiltration | | Security headers | CSP, X-Frame-Options (DENY), X-Content-Type-Options, Referrer-Policy | | Input validation | Agent IDs validated via isSafeId() regex — blocks path traversal | | Input sanitization | All user inputs (notes, messages, knowledge) stripped of control chars via sanitizeText()/sanitizeNote() | | Rate limiting | 120 req/min (read), 30 req/min (mutations), 20 req/min (session) per IP | | Startup validation | Workspace directory validated for existence and write access before server starts | | File locking | Per-file mutex (withFileLockSync) on all service read-modify-write operations — prevents concurrent corruption | | Atomic writes | All file mutations use .tmp → rename pattern with in-memory locks | | Shell injection | Update installer uses execFile (not exec) to prevent injection | | Credential isolation | Uses OpenClaw CLI — no API keys stored in plugin | | Supports encrypted secrets (SECRETS/, AES-256-GCM) | Expose secret values in API responses |

Your credentials stay in OpenClaw's configuration.

📚 Plugin vs Skill

AGI Farm was migrated from a skill to a plugin because it:

| Requirement | Skill | Plugin | |-------------|-------|--------| | Background services | ❌ | ✅ Dashboard server | | Complex wizard (15+ steps) | ⚠️ | ✅ Better UX | | System integration | ❌ | ✅ Cron, LaunchAgent | | Configuration management | ⚠️ | ✅ Schema-based |

The embedded skills/agi-farm/SKILL.md provides documentation reference for agents.

🔄 OpenClaw Compatibility

AGI Farm ensures compatibility with OpenClaw through automated testing and proactive monitoring:

Automated Compatibility Testing

Weekly testing against multiple OpenClaw versions (latest, previous, oldest-supported, beta)
Performance benchmarks to detect regressions (load time, validation time thresholds)
API deprecation scanning with automated replacement suggestions
Automated PR creation for compatibility fixes when breaking changes detected

Supported Versions

Minimum: OpenClaw 1.0.0+
Tested: Latest stable + previous stable + beta (when available)
Status: All compatibility tests run weekly via GitHub Actions

For Users

Check compatibility before upgrading OpenClaw:

# Check current OpenClaw version
openclaw --version

# Run validation
npm run validate

# View compatibility matrix
cat OPENCLAW_COMPATIBILITY.md

For Maintainers

The enhanced compatibility system includes:

Multi-version testing matrix (4 versions tested)
Performance regression detection (<5s load time, <3s validation)
Automated issue creation when incompatibility detected
Optional automated PR generation with suggested fixes

See: OPENCLAW_COMPATIBILITY.md for complete compatibility guide.

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

Fork the repository
Create your feature branch (git checkout -b feature/AmazingFeature)
Commit your changes (git commit -m 'Add some AmazingFeature')
Push to the branch (git push origin feature/AmazingFeature)
Open a Pull Request

🙏 Acknowledgments

This plugin integrates production-ready AI resources from the following exceptional projects:

Agency-Agents (NEW in v1.6.0)

Repository: https://github.com/msitarzewski/agency-agents
Author: @msitarzewski
Achievement: 50+ requests in first 12 hours on Reddit, months of production iteration
Integration: 59 specialized agent personalities across 11 categories
Impact: Battle-tested workflows for Engineering, Design, Marketing, Product, PM, Testing, Support, Spatial Computing, and Strategy

The Agency-Agents collection dramatically expands AGI Farm's agent library with proven personalities including:

Agents Orchestrator - Meta-agent for pipeline management (PM → Architect → Dev-QA Loop → Integration)
Evidence Collector - Screenshot-based QA (defaults to finding 3-5 issues minimum)
Reality Checker - Production readiness certification (defaults to "NEEDS WORK")
Whimsy Injector - Delightful UX enhancements that serve functional or emotional purpose
Reddit Community Builder - Authentic community engagement patterns

Special thanks to @msitarzewski for creating and sharing these exceptional agent personalities with the community.

Everything Claude Code (ECC)

Repository: https://github.com/affaan-m/everything-claude-code
Author: @affaan-m
Achievement: Winner of Anthropic's Claude Code hackathon
Integration: 510 resource files (16 agents, 69 skills, 33 commands, 7 guides)
Impact: Industry-standard TDD workflows, security scans, API design patterns, and framework templates for Python/Django/Spring Boot/Go/Swift/PostgreSQL

The ECC framework provides the foundation for AGI Farm's production-ready coding capabilities, including:

Test-driven development workflows with 80%+ coverage enforcement
Comprehensive security scanning and vulnerability detection (via AgentShield)
API design patterns and best practices
Framework-specific templates and conventions
Auto-delegation based on agent roles

Special thanks to @affaan-m for creating and open-sourcing this exceptional framework.

Combined Impact

Total Integrated Resources:

91 agent templates (16 AGI Farm + 16 ECC + 59 Agency-Agents)
69 production skills (ECC)
33 slash commands (ECC)
7 quality automation hooks (AGI Farm)
10,000+ lines of agent personalities and workflows

AGI Farm is proud to stand on the shoulders of these giants, combining the best of community-driven AI agent development into a single, cohesive plugin.

📝 License

MIT License — built for OpenClaw

📈 Star History

🔗 Links

GitHub: https://github.com/oabdelmaksoud/AGI-FARM-PLUGIN
Issues: https://github.com/oabdelmaksoud/AGI-FARM-PLUGIN/issues
OpenClaw Docs: https://docs.openclaw.ai
NPM Package: https://www.npmjs.com/package/agi-farm

Made with ❤️ for the OpenClaw community

⬆ Back to Top