Back to skill

Security audit

Website Security Audit

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent website-safety checker, but it requires opening any user-provided URL in the agent browser without destination safeguards.

Install only if you are comfortable with the agent opening URLs from its own browser environment. Avoid using it on local, private-network, metadata-service, or untrusted redirecting URLs unless the runtime has strong browser/network isolation, and treat page content as untrusted.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:20
Finding

Unvalidated User-Controlled URL Navigation Enables Server-Side Request Forgery

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 20-43
Vulnerability Type: Server-Side Request Forgery through unrestricted browser navigation
Risk Level: Medium

Relevant Code Snippet:

text
Step 1: Extract URL

Parse a normalized URL from user input and remove tracking parameters.

Step 2: Browser access (mandatory)

1. browser action=start profile=openclaw
2. browser action=tabs
3. browser action=navigate targetId=<id> targetUrl=<base-URL>
4. browser action=snapshot targetId=<id> maxChars=10000 loadState=networkidle

The snippet above is an English translation of the relevant skill instructions.

Technical Analysis

The skill requires the browser to navigate to a URL derived directly from user input. It does not prescribe validation of the URL scheme, hostname, resolved IP address, port, or redirect destinations before navigation.

In particular, the instructions do not require the agent to:

  • Restrict navigation to the http and https schemes.
  • Reject loopback, private, link-local, multicast, or reserved IP addresses.
  • Validate IPv4 and IPv6 address representations.
  • Resolve hostnames and verify that they point to publicly routable addresses.
  • Revalidate the destination after each HTTP redirect.
  • Protect against DNS rebinding.
  • Block URLs containing embedded credentials or access to nonstandard administrative ports.

Because the browser operates from the agent's environment, it may have access to network resources that are unavailable to the user. The mandatory snapshot operation can retrieve up to 10,000 characters from such a resource. The retrieved content may subsequently be incorporated into the security report or affect the agent's analysis.

Attack Path

  1. An attacker asks the skill to audit a URL targeting a loopback, private-network, link-local, or otherwise non-public destination. Alternatively, the attacker supplies a public U ...[truncated 1310 chars]
Remediation
View remediation

Remediation Suggestions

Implement a strict URL-validation policy before any browser navigation:

  1. Allow only explicitly supported http:// and https:// URLs.
  2. Reject URLs containing embedded usernames or passwords, malformed hosts, ambiguous numeric IP representations, or unsupported ports.
  3. Resolve the hostname before navigation and reject every address in loopback, private, link-local, multicast, carrier-grade NAT, documentation, reserved, and unspecified ranges for both IPv4 and IPv6.
  4. Revalidate all resolved addresses immediately before connection to reduce time-of-check/time-of-use and DNS-rebinding risks.
  5. Disable automatic redirects where possible, or validate the scheme, hostname, port, and resolved addresses of every redirect target before following it.
  6. Block known infrastructure metadata destinations, including link-local metadata addresses and provider-specific metadata hostnames.
  7. Use browser-level network isolation or an outbound proxy that permits only publicly routable Internet destinations.
  8. Require explicit user confirmation before accessing unusual ports or destinations that cannot be conclusively classified as public.
  9. Treat all retrieved page content as untrusted data. Explicitly instruct the agent not to follow commands or operational instructions embedded in audited pages.
  10. Avoid reproducing sensitive response content in reports. Redact credentials, tokens, internal addresses, and other secrets if unexpected non-public content is retrieved.
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The trigger description uses broad natural-language phrases that could cause the skill to activate for generic requests about checking links or websites, including contexts the user did not intend for this workflow. Over-broad invocation can lead to unnecessary browsing of arbitrary URLs and unintended execution of the skill's more powerful capabilities.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill mandates visiting arbitrary user-supplied URLs with a browser but does not provide strong safety guidance, consent checks, or restrictions for potentially malicious destinations. This is dangerous because opening attacker-controlled pages can expose browsing context, trigger unwanted network activity, and create privacy or safety risks for the agent environment.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill includes an optional Node.js command that executes a local script with user-influenced search input, expanding behavior beyond simple in-browser website inspection into local code execution. Even if intended for benign enrichment, this increases attack surface by enabling command/script invocation from the skill and may expose the host environment if the integration or argument handling is unsafe.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The manifest description is entirely in Chinese and presents the skill as operating in that language, but it does not state that Chinese is optional or explain a region-specific limitation. This can violate language/locale policy when a skill implicitly forces one language without user opt-in.

Content

No source excerpt is available for this finding.