Chrome DevTools Web Research

SuspiciousAudited by ClawScan on May 10, 2026.

Overview

The skill’s web-research purpose is coherent, but it asks for broad access to a logged-in Chrome session and proactive browser control without clear site or action boundaries.

Use this only if you are comfortable giving the agent control over a Chrome profile. For safety, use a dedicated browser profile with minimal logins, close unrelated private tabs, restrict the allowed sites in your request, require confirmation before any non-read-only action, pin/review the MCP tool version, and turn off remote debugging when finished.

Findings (5)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

The agent may be able to read or use logged-in sessions and private tabs, not just public webpages.

Why it was flagged

This explicitly grants access to the user’s authenticated browser state, which can include active accounts and private pages, without a visible per-site or per-profile boundary.

Skill content
the agent can now see the user’s tabs, cookies, logins, and page state through Chrome DevTools MCP
Recommendation

Use a separate Chrome profile with only the accounts needed for the task, close private tabs, and disable remote debugging after use.

ConcernMedium Confidence
ASI02: Tool Misuse and Exploitation
What this means

A browsing task could accidentally interact with authenticated sites or submit information if the agent uses these controls too broadly.

Why it was flagged

Clicking and filling forms in a live logged-in browser can change account state or submit data, and the provided instructions do not show a clear require-confirmation rule for sensitive actions.

Skill content
Navigate, click, fill, or switch tabs as needed.
Recommendation

Require explicit user confirmation before form submissions, purchases, posts, deletes, account changes, or other non-read-only actions.

What this means

Private page contents, account context, or session-derived information could enter the agent’s context through the browser bridge.

Why it was flagged

Sensitive browser state is routed through an MCP bridge, but the visible artifacts do not define data boundaries such as which tabs are accessible, what is logged, or how credentials/page state are isolated.

Skill content
through Chrome DevTools MCP
Recommendation

Limit use to a dedicated browser profile and document clear tab, logging, retention, and credential-handling boundaries.

What this means

A future package version could behave differently from the version the skill author tested.

Why it was flagged

The external MCP package is unpinned and central to the skill, so the runtime behavior may change as the latest package changes.

Skill content
Use `chrome-devtools-mcp@latest` through `mcporter`
Recommendation

Prefer a pinned, reviewed version of chrome-devtools-mcp and verify the mcporter/MCP configuration before granting browser access.

NoteHigh Confidence
ASI01: Agent Goal Hijack
What this means

Search queries may be sent to multiple platforms, potentially under the user’s logged-in browser session.

Why it was flagged

The skill expands ambiguous search requests into a default Google, X, and Reddit workflow. This is disclosed and research-oriented, but users should know it broadens the task unless they restrict it.

Skill content
If the user only says “search this”, still run the chain unless they explicitly restrict scope.
Recommendation

Specify allowed sites or platforms when you want a narrower search.