Phase 3 / Ep 13: Permission Model Overview —— tools.profile and Security Philosophy

⏱ Est. reading time: 4 min Updated on 4/13/2026

🎯 Learning Objective: Understand OpenClaw's tiered permission design and master the tools.profile configuration.

1. Security Philosophy: Principle of Least Privilege

OpenClaw Agents possess powerful tool capabilities. Power means danger—if you give an Agent full Shell access, it might accidentally delete your files.

Core Principle: Never give an Agent unnecessary capabilities.

2. Four-Tier Permission Model

graph LR
    Basic["🟢 basic\nRead-only · No Shell\nSecure Chat"] --> Standard["🟡 standard\nFile Read/Write\nLimited Shell"]
    Standard --> Coding["🔵 coding\nFull Shell\nCode Execution · Git"]
    Coding --> Full["🔴 full\nSystem-level Access\nNetwork · Process Management"]

    style Basic fill:#dcfce7,stroke:#16a34a
    style Standard fill:#fef9c3,stroke:#ca8a04
    style Coding fill:#dbeafe,stroke:#2563eb
    style Full fill:#fee2e2,stroke:#dc2626

3. Capability Comparison Table

Capability 🟢 basic 🟡 standard 🔵 coding 🔴 full
Chat Response ✅ ✅ ✅ ✅
File Read ✅ ✅ ✅ ✅
File Write ❌ ✅ ✅ ✅
Shell Commands ❌ Restricted Whitelist ✅ ✅
Git Operations ❌ ❌ ✅ ✅
Code Execution ❌ ❌ ✅ ✅
Network Access ❌ ❌ Restricted ✅
Process Management ❌ ❌ ❌ ✅
System Configuration ❌ ❌ ❌ ✅

4. Configuration Method

# Set permission level
openclaw config set tools.profile basic
openclaw config set tools.profile standard
openclaw config set tools.profile coding
openclaw config set tools.profile full

# Set for a specific Agent
openclaw agents config code-agent tools.profile coding
openclaw agents config ops-agent tools.profile full

5. Typical Configuration Scenarios

Agent Type Recommended Permission Runtime Environment Reason
Daily Assistant basic Any Only needs chat and reminders
Content Creation standard Any Needs to read and write files
Coding Assistant coding Native / Docker Needs Shell and Git
Ops & Monitoring full Docker/VM Only Needs system-level permissions

⚠️ Red Line: Never use the full permission in a native macOS environment. It must be run in a Docker container or UTM virtual machine.

Next Episode Preview: Ep 14, use the coding permission to make the Agent your programming assistant—clone repositories, write code, run tests, and submit PRs.