Back to skill

Security audit

m5stack-chrome-browser-control

Security checks for vulnerabilities and agentic risk

Overview

This skill openly provides powerful local Chrome control, but it is broad enough to expose logged-in browser sessions and relies on an unpinned package install.

Install only if you intentionally want OpenClaw to control your local Chrome session. Prefer a dedicated Chrome profile with only the accounts needed for the task, avoid exposing personal or administrative tabs, pin and review the MCP package version instead of using `@latest`, and turn off remote debugging when finished.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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)

T08 · Insecure Dependencies

Error
Location
SKILL.md:35
Finding

Unpinned Third-Party Package Executed with Access to Authenticated Browser Sessions

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 35–41
Vulnerability Type: Insecure third-party dependency execution
Risk Level: High

json
"command": "npx",
"args": [
  "chrome-devtools-mcp@latest",
  "--autoConnect"
]

Technical Analysis

The configuration instructs npx to execute chrome-devtools-mcp@latest. The mutable latest tag does not identify a fixed, previously reviewed artifact. A future invocation can therefore download and execute a package version whose contents differ from the version originally audited.

The configuration provides no exact version pin, lockfile, integrity hash, or other package-verification mechanism. This creates a supply-chain exposure: compromise of the package, its publisher account, or its release process could cause attacker-controlled code to execute under the user's account.

The package is especially sensitive because --autoConnect connects it to the local Chrome DevTools endpoint. As documented elsewhere in the Skill, this interface can enumerate and control tabs, read page contents and Outlook messages, fill fields, click controls, and capture screenshots. The dependency consequently operates across a high-trust boundary involving authenticated browser sessions.

Attack Path

  1. An attacker compromises the package publisher, package registry entry, or release process for chrome-devtools-mcp, or causes an unsafe release to become the package's latest version.
  2. The user installs or restarts the configured OpenClaw gateway.
  3. npx resolves chrome-devtools-mcp@latest, downloads the altered release, and executes it with the user's operating-system privileges.
  4. The process automatically connects to Chrome's local remote-debugging endpoint at 127.0.0.1:9222.
  5. The malicious process enumerates open tabs and accesses authenticated browser contexts.
  6. It can read sensitive page content, capture screenshots, manipulate web applications, or transmit accessible data using ...[truncated 916 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace chrome-devtools-mcp@latest with an exact, reviewed version; do not use mutable tags or version ranges.
  2. Use a lockfile and a deterministic installation process where supported.
  3. Verify package provenance, publisher identity, release signatures, and registry integrity metadata before installation and upgrades.
  4. Cache or vendor a reviewed artifact and validate its cryptographic hash before execution.
  5. Establish a controlled update process in which new versions are reviewed and tested before deployment.
  6. Run the MCP server under a dedicated, least-privileged operating-system account or sandbox with restricted filesystem and network access.
  7. Use a dedicated Chrome profile containing only the sessions required for the requested task. Do not expose unrelated personal, administrative, financial, or corporate sessions through the debugging endpoint.
  8. Enable remote debugging only while actively required, bind it exclusively to localhost, and disable it after use.
  9. Limit the MCP server's available browser targets and capabilities where technically possible, and monitor package and browser-control activity for unexpected access.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill explicitly enables control of the user's local Chrome session and access to arbitrary Outlook mailboxes and browser pages, which can expose email contents, tokens, and other private data from already-authenticated sessions. In this context, the absence of an explicit privacy warning and consent boundary is particularly dangerous because the skill relies on remote debugging and can operate across any open tab the user has logged into.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger scope is broad enough to match routine browser-related requests, yet the skill grants powerful local-browser automation via DevTools. This increases the chance the skill is invoked in situations where the user did not clearly intend to authorize access to tabs, authenticated sessions, or sensitive web content.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

All user-facing natural-language content in the skill is written only in Chinese, with no indication that the user can choose another language or locale. Under the stated policy, forcing a specific language without opt-in is a language/locale policy violation unless clearly justified as region-specific, which is not documented here.

Content

No source excerpt is available for this finding.