ClawBotPro
DocsManaging Agents

Managing Agents

Introduction to Agent Management

In ClawBotPro, you don't just prompt an LLM; you construct a persistent, autonomous worker. The Managing Agents dashboard (located at /dashboard/agents) allows you to define specific roles, assign tool permissions, and provision dedicated workspaces for each agent in your fleet.

This guide walks you step-by-step through the Agent Creation Flow, detailing exactly how the platform configures your local execution nodes and synchronizes with the OpenClaw Gateway.

The Agent Creation Flow

When you click New Agent, the dashboard guides you through a 7-step configuration wizard. This ensures the agent is strictly bounded and correctly provisioned before it ever executes a task.

1Selecting a Template

You begin by selecting an Agent Template. These templates are predefined archetypes (e.g., Senior Full-Stack Engineer, Technical Writer, or DevOps Orchestrator) mapped to specific Departments.

Engineering

Pre-loads the Development pipeline phase and automatically toggles on file-system and terminal MCP skills.

Content

Pre-loads the Content pipeline phase and forces the agent's system prompt to output strict Markdown.

Selecting a template automatically injects a domain-specific System Prompt, which you can manually override in Step 3.

2Basic Details

Here you define the agent's identity within your organization:

  • Name: The internal identifier used in your Mission Control dashboard.
  • Description: A brief summary of the agent's purpose (e.g., "Monitors GitHub PRs and runs automated testing").
  • Team Assignment: Binds the agent to a specific team for access control and billing segregation.

3Agent Identity & System Instructions

The Identity step is where you define the agent's "soul". You select a custom avatar and define the System Instructions.

System Prompt

You are a Senior DevOps Engineer managing an AWS infrastructure.

CRITICAL RULES:
1. Never run 'rm -rf' on production databases.
2. Always check 'terraform plan' before 'terraform apply'.
3. Reply exclusively in JSON format when requested.

Architecture Warning

The System Instructions are persistently prepended to the context window at the start of every single ReAct loop iteration. Be concise. A 2,000-token system prompt will drastically inflate your API costs, as it is billed on every tool call.

4LLM Model Selection

ClawBotPro supports multi-model orchestration. During this step, the dashboard queries the /api/openclaw/enrolled-models endpoint to fetch your available LLM providers (e.g., Anthropic, OpenAI, or local Ollama instances).

You can assign models based on operational tiers:

  • Premium Models (GPT-4o, Claude 3.5 Sonnet): Best for complex reasoning and multi-step coding architectures.
  • Standard/Free Models (Llama 3, Haiku): Best for high-volume, low-complexity tasks like log summarization to drastically reduce API costs.

5Assigning Skills & Tools (MCP)

An LLM is useless if it cannot interact with the world. During the Skills step, you toggle specific tools on or off. OpenClaw relies entirely on the Model Context Protocol (MCP) to handle tooling.

When you toggle a skill (like Web Search or File System), you are defining the exact JSON schema that the agent is permitted to execute.

Enabled: 'exec'

The agent receives the {"name": "exec"} schema and can invoke terminal commands in its local sandbox.

Disabled: 'github'

The agent does not know GitHub exists. If prompted to open a PR, it will reply that it lacks the capability.

6Workspace Initialization

Every OpenClaw agent operates inside a dedicated file-system workspace on your local Gateway node. The Workspace step allows you to inject seed files into the agent's environment before it boots.

Common seed files include:

  • MEMORY.md: The agent's persistent long-term memory. It reads this via RAG at the start of every session.
  • TOOLS.md: Local notes specific to your host machine (e.g., "The production DB is on port 5432").
  • README.md: The overarching project guidelines.

When you complete the wizard, the dashboard makes a POST /api/workspace/sync request to physically write these files to the local node's disk.

7Review & Synchronization

When you click Deploy Agent, a complex synchronization flow occurs between the Supabase cloud dashboard and your local execution node:

// 1. The dashboard inserts the agent metadata into Supabase
await supabase.from("agents").insert({ ...formPayload })
// 2. It syncs the new agent identity down to the local Gateway node
await fetch("/api/agents/sync", { method: "POST" })
// 3. It commands the Gateway to install the required MCP skills
await fetch("/api/agents/skills/install", { body: selectedSkills })

If any of these steps fail (e.g., your local execution node is disconnected), the dashboard will safely log the error and retry the synchronization hook once the Gateway reconnects.

Managing Existing Agents

You can modify an agent post-creation by clicking on its card in the Agents Dashboard. From the Agent Settings view, you can:

  • Adjust Tools: Dynamically add or revoke MCP skills. The changes sync instantly to the execution node.
  • Monitor Usage: View real-time token consumption and success rates via the /api/agents/performance telemetry data.
  • Pause Execution: Temporarily disable an agent. This unbinds it from the Gateway, preventing it from processing new webhooks or tasks until reactivated.
Back to GuidesOpen Dashboard