← Back to Skills Marketplace
47
Downloads
0
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install worker-safety
Description
Operations Safety Reference - Hard Limits apply unconditionally for OpenClaw agent operations.
Usage Guidance
This skill is an instruction-only safety policy that will tell an OpenClaw agent to refuse a range of dangerous or risky operations (upgrades, writing outside workspace, fetching-and-executing unknown URLs, bulk deletions, etc.). It asks for no credentials and does not install anything, so technical risk is low. Before installing: be aware that the skill will refuse some legitimate actions you might ask the agent to perform (for example, upgrading OpenClaw via package managers or writing files outside the workspace). Because it is instruction-only, enforcement relies on the agent following the skill—if you need enforceable, system-level controls, implement them at the platform/runtime level rather than relying solely on a skill. If you plan to allow certain actions this skill forbids, you will need an owner-level override outside the skill.
Capability Analysis
Type: OpenClaw Skill
Name: worker-safety
Version: 1.0.0
The bundle contains a safety configuration (SKILL.md) designed to harden the OpenClaw agent against common attacks and misuse. It establishes 'Hard Limits' that instruct the agent to refuse high-risk operations such as system-level configuration changes, writing outside the designated workspace, exfiltrating sensitive identity/config files, and executing instructions from external URLs. The instructions are purely defensive and align with the stated purpose of providing an operations safety reference.
Capability Assessment
Purpose & Capability
The skill name/description (worker-safety / Operations Safety Reference) matches the SKILL.md content: a list of hard limits and refusal behaviors. It does not declare unrelated env vars, binaries, or install steps, so requested capabilities are proportionate to the stated safety purpose.
Instruction Scope
SKILL.md contains policy-like runtime instructions (what to refuse, how to warn, alternatives). It references OpenClaw-specific files/paths (openclaw.json, workspace files) only to restrict access, and explicitly forbids reading/disclosing protected files. It does not instruct the agent to read or exfiltrate external secrets or system files. The refusal rules are specific rather than open-ended.
Install Mechanism
There is no install spec and no code files. Being instruction-only means nothing is downloaded or written to disk by the skill itself.
Credentials
The skill requires no environment variables, credentials, or config paths. It references system paths (e.g., /root/.openclaw/workspace/) only to prohibit writing outside them; no unrelated credentials are requested.
Persistence & Privilege
The skill is not marked 'always: true' and uses normal autonomous invocation defaults. It contains self-protection guidance (do not delete this safety skill) which is reasonable for a safety/reference skill. There is no attempt to modify other skills or system-wide configs in the instructions.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install worker-safety - After installation, invoke the skill by name or use
/worker-safety - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
Version 1.0.0 of worker-safety skill
- Introduces strict Hard Limits to enforce safety for OpenClaw agent operations, covering system integrity, network exposure, external execution, and file access.
- Implements clear refusal and warning procedures when unsafe or forbidden requests are received, including never providing step-by-step instructions for restricted actions.
- For group chat scenarios, adds additional rules to protect sensitive files, prevent behavioral overrides, and block exposure of system topology.
- Offers guidance to always state the risk, present safe alternatives, and notify both users and coordinators when actions are refused.
Metadata
Frequently Asked Questions
What is Worker Safety?
Operations Safety Reference - Hard Limits apply unconditionally for OpenClaw agent operations. It is an AI Agent Skill for Claude Code / OpenClaw, with 47 downloads so far.
How do I install Worker Safety?
Run "/install worker-safety" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Worker Safety free?
Yes, Worker Safety is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Worker Safety support?
Worker Safety is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Worker Safety?
It is built and maintained by forest (@yurken); the current version is v1.0.0.
More Skills