ClawTonomy

Task management, AI dispatch & self-improving autonomy for OpenClaw.
Create tasks. The AI sees them, works on them, dispatches sub-agents, and syncs everything back — with hard safety guards.

---

How It Works

The AI follows this loop:

1. REASON — Check what's happening (tasks, schedules, notifications)
2. DECIDE — Figure out what needs attention
3. BUILD — Create the solution
4. VERIFY — Test it and prove it works
5. DONE — Mark complete with evidence

Sync Bridge — The Core Innovation

OpenClaw has no native task system. Without ClawTonomy's sync bridge, the AI never sees your tasks. ClawTonomy solves this by injecting task data into three files the AI already reads:

File	When AI Reads	What It Contains
TASKS.md	Every conversation	Full task list with statuses, priorities, due dates
HEARTBEAT.md	Every 30 minutes	Overdue warnings, blocked tasks, dispatch status
AGENTS.md	Session start	Mandatory rules — the AI cannot ignore tasks

Sync happens automatically after every task mutation. No manual steps.

Task Lifecycle

Tasks flow through states: pending → in_progress → completed (with branches to blocked, deferred, cancelled). Every state change triggers a sync.

---

AI Dispatch

Dispatch tasks to parallel AI sub-agents:

# Dispatch via CLI
clawtonomy task dispatch <name>

# Dispatch via API
curl -X POST http://localhost:8767/api/dispatch \
  -H "Content-Type: application/json" \
  -d '{"task_name": "fix-auth", "mode": "agent"}'

Two dispatch modes:

• Agent mode — openclaw agent --local — Interactive sub-agent session
• Cron mode — openclaw cron add --at now — Isolated background session

Dispatched tasks auto-set to in_progress, track ai_dispatched = true, and record the session ID.

---

Quick Start

# Clone to your OpenClaw skills directory
cd "${OPENCLAW_HOME:-$HOME/.openclaw}/workspace/skills"
git clone https://github.com/rar-file/autonomy.git clawtonomy

# Run install script
cd clawtonomy
bash install.sh

# Create your first task
clawtonomy task create "my-first-task" "Explore the ClawTonomy system"
clawtonomy task list

# Start the web dashboard
python3 web_ui.py
# Open http://localhost:8767

---

Commands

Task Management

Command	Description
clawtonomy task create <name> [desc]	Create a new task
clawtonomy task list	Show all tasks with statuses
clawtonomy task work <name>	Mark as in-progress
clawtonomy task complete <name> "proof"	Mark complete with proof
clawtonomy task delete <name>	Delete a task
clawtonomy task sync	Force sync to TASKS.md + HEARTBEAT.md + AGENTS.md
clawtonomy task dispatch <name>	Dispatch to AI sub-agent
clawtonomy task status	Show dispatch/sync status

Natural Language Tasks

Create tasks in plain English — no flags needed:

clawtonomy task create "Fix the OAuth bug in auth.py by Friday, high priority"
# → name: fix-the-oauth-bug-in-auth-py
# → due: Friday
# → priority: high
# → tags: [bug, auth]

GitHub Integration

Command	Description
clawtonomy gh prs	Your open PRs
clawtonomy gh reviews	PRs waiting for review
clawtonomy gh ci-status	CI status on default branch
clawtonomy gh notifications	Unread notifications

System Monitoring

Command	Description
clawtonomy vm health	System health overview
clawtonomy vm process_list	List all processes
clawtonomy vm top_cpu	Top CPU consumers
clawtonomy vm docker_ps	Docker containers

---

Web Dashboard

Real-time dashboard at http://localhost:8767 with 36 API routes:

• Task Board — Live task list with status badges, priorities, due dates, subtask progress
• Quick Add Bar — Natural language task creation from the dashboard
• Dispatch Panel — One-click dispatch to AI sub-agents
• Personality Editor — Edit AI personality with A/B testing and version history
• Token Usage — OpenClaw API usage tracking
• System Health — CPU, memory, disk, load monitoring
• Task Templates — Pre-built task blueprints
• Dependency Graph — Visual task relationships
• Dark Theme — Custom SVG assets, auto-refresh every 30 seconds

Start it:

python3 web_ui.py

Key API Endpoints

Method	Endpoint	Purpose
GET/POST	/api/tasks	List / create tasks
PUT	/api/tasks/<name>	Update task
POST	/api/dispatch	Dispatch to AI
POST	/api/tasks/sync	Force sync
POST	/api/tasks/parse	NL → structured task
GET	/api/tasks/<name>/subtasks	Get subtasks
GET/PUT	/api/personality	Personality editor
POST	/api/personality/ab-test	A/B test personalities
GET	/api/personality/versions	Version history
POST	/api/digest	Generate daily digest
GET	/api/context-window	AI context usage

Full API reference: see SKILL.md

---

Safety Guards

Hard Limits

Limit	Value	Purpose
Max concurrent tasks	5	Prevent overload
Max sub-agents	3	Limit parallelism
Max schedules	5	Control recurring work
Daily token budget	50,000	Cost protection
Max iterations per task	5	Stop endless building

Anti-Hallucination

WRONG: "Task complete" (no proof)
RIGHT: "Task complete. Proof: Tested X, verified Y exists"

Guard	Description
Verification Required	Must prove work before marking complete
Attempt Tracking	Max 3 attempts before forced stop
File Verification	Files must exist (actually checked)
Command Testing	Commands must work (actually run)
Evidence Required	Hand-waving rejected

Approval Required

These actions need explicit approval:

• External API calls
• Sending messages
• File deletion
• Public posts
• Git push
• Installing packages

---

OpenClaw Integration

ClawTonomy is built specifically for OpenClaw:

• Sync Bridge — Injects tasks into TASKS.md, HEARTBEAT.md, AGENTS.md so the AI actually sees them
• Sub-agents — openclaw agent --local for parallel dispatch
• Cron — openclaw cron add for scheduled/background tasks
• Heartbeat — Respects your existing HEARTBEAT.md (injects between markers)
• Memory — Uses memory_search / memory_get for context
• Skills — Registered as an OpenClaw skill via SKILL.md

---

Configuration

Edit config.json:

{
  "version": "3.5.0",
  "limits": {
    "max_concurrent_tasks": 5,
    "daily_task_budget": 20
  },
  "github": {
    "default_repo": null,
    "notify_on_ci_fail": true
  },
  "web_ui": {
    "port": 8767,
    "auto_refresh": 30
  }
}

---

Example Session

# User creates a task (natural language)
$ clawtonomy task create "Fix OAuth login bug, high priority, due Friday"
Task 'fix-oauth-login-bug' created (priority: high, due: Friday)

# Check tasks
$ clawtonomy task list
  [pending] fix-oauth-login-bug: Fix OAuth login bug (priority: high, due: Fri)

# Dispatch to AI sub-agent
$ clawtonomy task dispatch fix-oauth-login-bug
Dispatched 'fix-oauth-login-bug' via agent mode
Session: sess_abc123

# AI works on it in parallel...
# Later, check status:
$ clawtonomy task list
  [completed] fix-oauth-login-bug: Fix OAuth login bug
    Proof: "Fixed token refresh in auth.py:47. Tested: login works with OAuth"

---

Assets

Visual assets included:

Asset	Description
assets/logo.svg	Main logo
assets/logo-banner.svg	Banner with badges
assets/diagram-workflow.svg	5-step agentic workflow
assets/diagram-sync.svg	3-file sync bridge flow
assets/diagram-dispatch.svg	AI dispatch architecture
assets/diagram-lifecycle.svg	Task status lifecycle
assets/diagram-architecture.svg	Full system architecture
assets/diagram-heartbeat.svg	Heartbeat injection flow
assets/diagram-limits.svg	Safety limits overview
assets/diagram-update.svg	Self-update mechanism

---

Documentation

Doc	Purpose
SKILL.md	OpenClaw skill manifest — full API reference, 12 mandatory rules
FULL_AUTONOMY.md	Complete autonomy capabilities reference
USAGE.md	Human + AI agent usage guide

---

License

MIT License — Built for OpenClaw

---

_{Built with 🔧 for the OpenClaw community — ClawTonomy v3.5.0}