A specialized framework for defining and enforcing role-based tool access for AI subagents to ensure safe and efficient task execution.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install openclaw-subagent-toolset
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install openclaw-subagent-toolset using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Subagent Toolset?
The Subagent Toolset is a robust framework designed to manage and constrain the capabilities of AI subagents within Openclaw Skills environments. By categorizing tasks into specific roles—such as Explore, Plan, Verification, Coding, and Secretary—developers can prevent AI agents from performing unauthorized or dangerous operations while ensuring they have the precise tools needed for the job.
This skill provides a structured methodology for role-based access control (RBAC) in AI agent swarms. It utilizes soft constraints within task prompts to guide LLM behavior, reducing the risk of unintended file modifications or unauthorized command execution during complex automated workflows. It is an essential component for developers building complex, multi-step AI agents that require strict operational boundaries.
Subagent Toolset Use Cases
- Restricting information-gathering agents to read-only tools to prevent data corruption.
- Creating specialized documentation assistants (Secretary) that only summarize without altering source code.
- Managing high-privilege coding agents for focused debugging and feature implementation.
- Validating system states with Verification agents using read-only terminal commands.
How Subagent Toolset Works
- Identify the specific task type required for a new subagent (e.g., Planning vs. Coding).
- Reference the corresponding tool subset definition provided by the Subagent Toolset skill.
- Prepend the selected tool permissions and explicit prohibitions to the subagent's task prompt to establish behavioral boundaries.
- Spawn the subagent with the constrained prompt to guide its behavior throughout the session.
- The subagent operates within its defined scope, prioritizing the allowed tools for its designated role while avoiding forbidden actions.
Subagent Toolset Setup
To integrate the Subagent Toolset into your Openclaw Skills configuration, follow these steps:
- Access your subagent management interface or agent orchestration logic.
- Define the tool subsets within your system prompt library.
- When initiating a subagent session, inject the constraint prompt dynamically:
# Example of a constraint injection for an Explore agent
PROMPT="You are an Explore-type sub-agent. You may only use the following tools: read_file, glob_search, grep_search, WebFetch, WebSearch, Skill. Forbidden: exec, sessions_spawn, edit, write."
- Monitor subagent logs to ensure compliance with the provided tool constraints.
Subagent Toolset Data Schema & Taxonomy
The Subagent Toolset organizes permissions into distinct profiles. Below is the mapping used within Openclaw Skills:
| Profile | Primary Tools | Forbidden Tools |
|---|---|---|
| Explore | read_file, glob_search, WebFetch, WebSearch | exec, sessions_spawn, edit, write |
| Plan | read_file, TodoWrite, WebFetch, StructuredOutput | exec, sessions_spawn, edit, write |
| Verification | bash (read-only), read_file, PowerShell | sessions_spawn, edit, write |
| Coding | write_file, edit_file, bash, TodoWrite | None (Full Access) |
| Secretary | read_file, TodoWrite, SendUserMessage | exec, write, glob_search, grep_search |
Subagent Toolset Advanced Features
- Role-based prompt templating for rapid subagent deployment across various task types.
- Multi-agent swarm orchestration with specialized permission layers for enhanced security.
- Integration with StructuredOutput to ensure consistent data delivery from constrained subagents.
- Custom tool subset definitions allowing developers to tailor permissions for proprietary enterprise workflows within Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Agent Sorting Hat
A structural diagnostic skill used to identify and fix AI agent role misalignment, team gaps, and boundary issues before persona refinement.
Research Synthesis
A specialized skill for AI agents to perform systematic literature reviews and complex multi-step research synthesis.
Audit Log Hook
A robust logging utility for Openclaw Skills that records tool call lifecycles to ensure security, accountability, and performance monitoring.
Physics Simulation Skill
A specialized AI agent skill for solving complex physical equations and performing high-fidelity numerical simulations for scientific research.
OpenClaw Memory Management Format
A multi-layered memory architecture that combines semantic search with structured Markdown storage to maintain persistent AI agent context.
Aliyun Qwen VL
A professional-grade multimodal skill that enables AI agents to analyze images, read charts, and perform structured visual data extraction using Alibaba Cloud Qwen VL models.
Featured*
