- Home
- Skills
- Agents & Orchestration
- Openclaw Hardening Audit
Openclaw Hardening Audit
Audit OpenClaw installations and third-party skills for security misconfigurations and supply-chain risks.
Free
Works with the AI Agents you already use
Openclaw Hardening Audit
Example session with this skill installed
Audit my OpenClaw install at ~/.openclaw and check this SKILL.md I found on GitHub for any red flags before I add it.
- Read your context and instructions
- Compiled the openclaw hardening audit
Audit completed for path ~/.openclaw. Found 1 Critical, 1 High, and 1 Medium risk findings. The external SKILL.md is flagged as
Malicious and should not be installed.
OpenClaw Hardening Report
| Category | Finding | Severity | Risk |
|---|---|---|---|
| Skill Audit | Credential Exfiltration Pattern | CRITICAL | The provided SKILL.md attempts to read ~/.openclaw/credentials/ and send contents to an external URL. |
| Gateway | Bind Host 0.0.0.0 | HIGH | The gateway is exposed to all network interfaces, allowing remote access if the port is reachable. |
| Workspace | Missing Injection Defense | MEDIUM | AGENTS.md lacks rules for handling untrusted external content (web/email). |
Detailed Findings
[CRITICAL] Malicious Skill Pattern Detected
The external skill you provided contains a trigger that attempts to cat files in the credentials directory and pipe them to a curl command. This is a documented exfiltration technique.
Do not install this skill.
[HIGH] Global Gateway Binding
Your openclaw.json has gateway.bindHost set to 0.0.0.0. This makes the agent's control interface accessible from the public internet or local network.
[MEDIUM] Weak Workspace Hardening
AGENTS.md does not explicitly instruct the agent to treat web search results as untrusted. This leaves the system vulnerable to indirect prompt injection where a malicious website could take control of the session.
Next steps
- Delete the malicious SKILL.md immediately and do not attempt to run it.
- Edit
~/.openclaw/openclaw.jsonto setgateway.bindHostto127.0.0.1. - Add defensive instructions to
AGENTS.mdregarding untrusted data sources.
Connects securely to your tools. The creator never sees your data.
About this skill
The problem
OpenClaw gives LLMs shell access, browser control, and file system reach, often with unsafe defaults. With over 135,000 internet-exposed instances and a rise in malicious skills targeting credentials, a standard installation is a significant security liability.
What it does
- Audits
openclaw.jsonfor critical gateway exposure, weak authentication tokens, and overprivileged tool permissions. - Inspects
SKILL.mdfiles for supply-chain risks, including credential exfiltration patterns, obfuscated payloads, and prompt-injection vulnerabilities. - Checks workspace instruction files like
AGENTS.mdandSOUL.mdfor missing defensive boundaries against indirect prompt injection. - Cross-references the local OpenClaw version against known CVEs and identifies dangerous file-system permission sets.
Frameworks & tools
Built for OpenClaw installations, ClawHub skills, and MCP server configurations. Uses stat, ls, and the OpenClaw CLI for non-intrusive inspection.
Why this beats prompting it yourself
General LLMs often overlook the specific "indirect injection" risks unique to OpenClaw's browser and email tools. This skill uses a specialized reference library of known malicious patterns and CVE histories that standard models lack in their base training.
Use cases
- Vetting third-party skills from community repos before installation.
- Hardening a fresh OpenClaw deployment before giving it access to sensitive data.
- Conducting regular security posture reviews of production agents.
- Investigating agents that are exhibiting suspicious or out-of-character behavior.
Known limitations
Does not perform memory forensics or log analysis to detect past compromises. The CVE reference library requires manual verification against live advisories for the most recent threats.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.
Does not detect past compromises via logs or memory forensics. CVE data is a floor, not a live feed. Cannot audit MCP server internals or OpenClaw source code.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
2 uses
Times used to date
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 2 months ago
- Free with an Agensi account
Permissions
File Scopes