Security audit

CDP浏览器大师

Security checks across malware telemetry and agentic risk

Overview

This skill enables powerful logged-in browser control and raw session cookie extraction while under-disclosing and contradicting those risks.

Only install this after careful review. Use a dedicated temporary browser profile, restrict remote debugging to local trusted use, avoid exporting raw cookies, and do not use it on accounts or sites where session disclosure could cause financial, business, or privacy harm.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands

Findings (8)

Description-Behavior Mismatch

Medium

Confidence: 92% confidence
Finding: The manifest presents the skill as broad coding/development assistance, while the body actually enables logged-in browser automation, scraping, and credential-adjacent data access. This mismatch can cause the skill to be invoked in contexts where users and reviewers do not expect account/session-sensitive behavior, weakening informed consent and oversight.

Intent-Code Divergence

High

Confidence: 98% confidence
Finding: The security section claims returned content will not include sensitive credentials, but the documented functionality explicitly extracts and returns cookies, including HttpOnly cookies. This contradiction can mislead operators into exposing live session material while believing the skill is designed not to return secrets.

Context-Inappropriate Capability

High

Confidence: 99% confidence
Finding: The skill documents extraction of HttpOnly cookies from an already logged-in browser session, despite the manifest framing the tool as general programming help. HttpOnly cookies are specifically protected from normal script access because they often authenticate accounts, so exposing them through the skill enables session theft and account takeover.

Vague Triggers

Medium

Confidence: 90% confidence
Finding: The activation text is overly broad, covering generic programming, debugging, and deployment help, while the skill performs sensitive browser automation against logged-in sessions. Overbroad routing increases the chance the skill is auto-selected in inappropriate contexts, exposing browser control and session data beyond user expectations.

Missing User Warnings

High

Confidence: 98% confidence
Finding: The skill prominently advertises reuse of logged-in browser sessions and HttpOnly cookie access without a strong privacy/account-sensitivity warning. In context, this is especially dangerous because the capability targets authenticated state and can expose tokens that permit impersonation on third-party services.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The examples instruct users to start browsers with remote debugging enabled and then automate authenticated sessions, but do not clearly warn that remote debugging exposes broad browser control and access to session data. If that port is reachable by other local or network actors, it can enable takeover of open tabs, data extraction, and credential abuse.

Ssd 3

High

Confidence: 99% confidence
Finding: The skill directly instructs retrieval and reuse of authenticated cookies, including HttpOnly cookies, from a logged-in session. This operationalizes extraction of live authentication artifacts, making compromise of user accounts or bypass of normal login controls straightforward if the output is mishandled or abused.

Ssd 3

High

Confidence: 99% confidence
Finding: The example converts complete browser cookies into a reusable HTTP Cookie header for subsequent requests, which is effectively a recipe for replaying authenticated sessions outside the browser. This materially lowers the barrier to session hijacking, lateral reuse of account access, and stealthy scraping under a victim's authenticated identity.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.