Transform Agent Zero into a proactive, identity-aware autonomous assistant
A0-Claw brings enterprise-grade agent identity management to Agent Zero, implementing the OpenClaw protocol for persistent, proactive AI operations. This system creates a structured identity layer that survives session boundaries, enabling truly autonomous behavior with appropriate guardrails.
graph TB
subgraph "Identity Layer"
A[IDENTITY.md] --> B[SOUL.md]
B --> C[USER.md]
C --> D[AGENTS.md]
end
subgraph "Operational Layer"
E[HEARTBEAT.md] --> F[TODO.md]
F --> G[MISSIONS.md]
G --> H[MEMORY.md]
end
subgraph "Execution Layer"
I[Scheduler Tasks] --> J[Proactive Checks]
J --> K[Maintenance Actions]
end
D --> E
H --> I
flowchart LR
A[User Input] --> B{Bootstrap Stage}
B -->|Stage 1| C[Identity Definition]
B -->|Stage 2| D[Operational Contract]
C --> E[Core Files Generated]
D --> F[Behavioral Parameters]
E --> G[Heartbeat Activation]
F --> G
G --> H[Autonomous Operation]
| Component | Purpose | Lifecycle |
|---|---|---|
| AGENTS.md | Workspace configuration & startup routine | Persistent |
| IDENTITY.md | Core identity & behavioral defaults | Post-bootstrap |
| SOUL.md | Values, preferences & anti-patterns | Post-bootstrap |
| USER.md | User context & working preferences | Post-bootstrap |
| HEARTBEAT.md | Proactive check protocol | Persistent |
| MEMORY.md | Curated long-term memory | Evolving |
| TODO.md | Active task queue | Dynamic |
| MISSIONS.md | Strategic priorities | Dynamic |
| TOOLS.md | Operational reference | Static |
- Agent Zero installed and operational
- Project workspace access
# Clone or copy skill to Agent Zero skills directory
cp -r a0-claw /a0/skills/
# Initialize your agent
/a0/skills/a0-claw/scripts/init-agent.sh <agent-name> <timezone>
# Example:
/a0/skills/a0-claw/scripts/init-agent.sh aurora America/New_York- Create Project: In Agent Zero UI, create project matching your agent name
- Set Instructions: Configure project instructions to reference workspace files
- Bootstrap: Send
BOOTSTRAP.mdcontent as first message - Extended Setup: Upon completion, agent offers deep operational configuration
- Activate: Configure heartbeat task for autonomous operation
Objective: Establish core identity parameters
Process:
- Send
BOOTSTRAP.mdas initial message - Agent initiates conversational discovery:
- Name selection
- Nature definition
- Communication vibe
- Signature emoji
- Agent populates:
IDENTITY.mdUSER.md(basic)SOUL.md(basic)
Completion Criteria: Agent confirms identity establishment, deletes BOOTSTRAP.md
Objective: Define working relationship parameters
Process:
- Agent offers: "Bootstrap complete. Continue to extended setup?"
- If accepted, deep diagnostic conversation covering:
- Current project state & blockers
- Communication preferences
- Tool authorization matrix
- Escalation protocols
- Memory management strategy
Completion Criteria: Agent updates operational files with contextual parameters
Project Name: [agent-name]
Workspace: /a0/usr/projects/[agent-name]/
Working Directory: [configured in Agent Settings]
Primary Location: Project root (/a0/usr/projects/[agent-name]/)
Runtime Injection: Mirror to .a0proj/instructions/ for automatic context loading
# Hybrid mirror deployment
cp /a0/usr/projects/[agent-name]/*.md /a0/usr/projects/[agent-name]/.a0proj/instructions/| Parameter | Value |
|---|---|
| Task Type | Scheduled |
| Initial Schedule | 0 */6 * * * (6-hour intervals) |
| Context | Dedicated |
| Attachments | HEARTBEAT.md, project root, instructions/ |
System Prompt:
Wake up [agent]. Read HEARTBEAT.md, MISSIONS.md, and TODO.md.
Perform small maintenance/progress actions.
If nothing needed, reply exactly: HEARTBEAT_OK
- Heartbeat Protocol: Scheduled wellness checks
- Stale Item Detection: Automatic identification of aging tasks
- Maintenance Rotation: Cyclic review patterns (TODO, MISSIONS, MEMORY)
- External Action Gates: Configurable authorization requirements
- Destructive Operation Warnings: Prevention protocols
- Privacy Protection: Data exposure prevention
- Daily Capture: Automatic session logging
- Curated Long-term: Manual MEMORY.md management
- Continuity Assurance: File-based persistence
sequenceDiagram
participant U as User
participant A as Agent
participant F as Identity Files
participant H as Heartbeat
U->>A: Send BOOTSTRAP.md
A->>F: Create IDENTITY, USER, SOUL
A->>U: Request identity parameters
U->>A: Provide preferences
A->>F: Update files
A->>U: Offer extended setup
opt Extended Accepted
U->>A: Accept
A->>F: Deep operational config
A->>U: Diagnostic questions
U->>A: Detailed responses
A->>F: Update AGENTS, TODO, MISSIONS
end
A->>H: Activate heartbeat
loop Every 6 hours
H->>A: Wake and check
A->>F: Read state
alt Action needed
A->>U: Report/suggest
else Nothing needed
A->>H: HEARTBEAT_OK
end
end
- Maintain 100-300 line sweet spot per file
- Prefer updating existing files over fragmentation
- Use absolute paths in task attachments
- Version control your workspace
- Be specific in identity definition
- Surface real blockers in extended setup
- Define clear escalation boundaries
- Establish trust through demonstrated accuracy
- Start with daily heartbeat, scale frequency based on need
- Scope tasks narrowly for reliability
- Clear task chat context periodically
- Review MEMORY.md monthly for consolidation
Protocol: OpenClaw
Original Implementation: @iamcapote
Platform: Agent Zero
Reference Article: A0-Claw on BIThub
MIT License - See LICENSE file for details.
"Identity is not what you are given. It is what you become through action."
A0-Claw — Enterprise Agent Orchestration