Skip to main content

OpenClaw Agent Setup Guide

How to configure OpenClaw agents for maximum productivity — whether you're building software or managing marketing campaigns.

Two Approaches, One Platform

OpenClaw supports two agent configurations. Choosing the right one depends on what the agent does.

1. Project Agents — For Development Work

When agents work on codebases (implementing features, fixing bugs, reviewing code), you don't need identity files or persona configurations. Instead, use a thread-based setup:

Conversation thread  →  agent
├── working folder: /path/to/project
├── skills: global + project-level
└── tools: browser profile, credentials

How it works:

  1. Each conversation thread (e.g., in Slack) connects to an agent instance
  2. The agent is bound to a working folder — the project it operates on
  3. Skills load automatically — shared skills (available globally) and project-specific skills (from the project's configuration)
  4. You give the agent a task: a description, a GitHub issue link, a Jira ticket, or an ADO work item
  5. The agent works autonomously using all available skills

Why this is better than per-person configuration:

  • No need to create identity files for each developer
  • Skills are shared — one "backend development" skill works for everyone
  • Switch projects by starting a new thread pointed at a different folder
  • Same person, multiple projects, zero config duplication

Example — multi-project team:

ThreadAgentWorking FolderSkills
#1Backend agent/repos/my-api/Global + project skills
#2Frontend agent/repos/my-app/Global + project skills
#3DevOps agent/repos/infra/Global + project skills

All three agents use shared skills. No identity setup needed.

2. Role Agents — For Marketing, Content, Social Media

When the agent's voice and personality matter — writing copy, managing social accounts, creating content — use OpenClaw's built-in identity system:

  • Soul files define the agent's personality and communication style
  • Identity configuration maintains consistent brand voice
  • Persona settings shape how the agent interacts

These agents don't need a fixed working folder. Their value comes from maintaining a consistent identity across interactions.

Example — marketing team:

AgentIdentitySkills
Brand copywriterCreative, on-brand voiceMarketing, copywriting
Social media managerCasual, engaging toneSocial media, scheduling
Content strategistAnalytical, data-drivenAnalytics, content planning

Setting Up Project Agents

Step 1: Create a Connection

Set up your messaging integration (Slack, etc.) so each thread maps to an agent session.

Step 2: Configure the Working Folder

Point the agent to the project directory. The agent starts here and has access to:

  • Project source code
  • Configuration files (including SpecWeave config if present)
  • Project-specific skills that load automatically

Step 3: Assign Tools

Tools handle environment-specific concerns:

  • Browser profiles — pre-configured with saved passwords for relevant services
  • Service accounts — API keys and tokens
  • External integrations — Jira, GitHub, ADO connections

The agent invokes tools as needed. Credentials stay in the tool layer — the agent never sees raw passwords.

Step 4: Give It Work

Drop a task into the thread:

  • Plain text description
  • Link to a GitHub issue
  • Link to a Jira or ADO work item
  • Reference to an existing SpecWeave increment

The agent picks it up and starts working.

Skills: Global vs. Project-Level

Global skills are available to all agents across all projects:

  • Marketing, analytics, social media
  • General development patterns
  • Common workflows

Project-level skills load from the project's configuration:

  • Domain-specific workflows
  • Project conventions and patterns
  • Custom automation

This means a marketer working on three different projects uses the same global marketing skill — no need to train separate agents per project. Project-specific context comes from the working folder.

When to Use Which

ScenarioAgent TypeWhy
Feature implementationProject agentCode context matters, identity doesn't
Bug fixingProject agentNeeds codebase access, not personality
Code reviewProject agentSkills-based, folder-bound
Marketing copyRole agentBrand voice and tone matter
Social media postsRole agentConsistent personality across posts
Content strategyRole agentAnalytical persona adds value
Customer communicationRole agentConsistent, empathetic voice

Key Takeaway

Project agents  = working_folder + skills + tools        → no identity needed
Role agents = identity + personality + skills + tools → identity IS the value

Keep them separate. Project agents should be lightweight and interchangeable. Role agents should be carefully crafted personas that maintain consistency.