Patchright Skill

Security checks across malware telemetry and agentic risk

Overview

This skill needs careful review because it provides stealth browser automation for CAPTCHA/Cloudflare bypass and runs a persistent local browser-control server.

Install only if you specifically need authorized browser automation for systems you control. Avoid using it with personal or production logged-in accounts, do not use it to bypass third-party anti-bot protections, pin dependencies, and stop the background server after each session.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Risk analysis

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.

#
ASI02: Tool Misuse and Exploitation
High
What this means

The agent could be used to automate sites in ways intended to evade anti-bot protections, including actions on real accounts or public services.

Why it was flagged

The skill explicitly advertises bypassing bot-protection systems, which is not necessary for ordinary localhost/dev-server QA and can enable misuse on third-party services.

Skill content
Bot Detection Bypass: YES (Cloudflare, reCAPTCHA, etc.)
Recommendation

Use only on applications and environments you own or are authorized to test; remove or disable anti-bot bypass behavior for general browser automation.

#
ASI05: Unexpected Code Execution
Medium
What this means

A command sent to the server can read or manipulate the current page beyond normal click/type/screenshot operations.

Why it was flagged

The server exposes an evaluate tool that runs arbitrary caller-supplied JavaScript in the active browser page.

Skill content
async def evaluate(self, script):
        ...
        result = await self.page.evaluate(script)
Recommendation

Disable the evaluate tool by default, or require explicit user approval and restrict it to trusted local test pages.

#
ASI07: Insecure Inter-Agent Communication
Medium
What this means

Another local process could potentially issue browser commands, take screenshots, read page text, or manipulate the persistent browser session.

Why it was flagged

The browser-control server listens on localhost and the shown request handler executes JSON commands without an authentication token or origin check.

Skill content
self.server = await asyncio.start_server(self.handle_client, '127.0.0.1', PORT)
Recommendation

Use a random authenticated token, a permission-restricted Unix socket, or per-invocation execution instead of an unauthenticated fixed localhost port.

#
ASI10: Rogue Agents
Medium
What this means

Browser sessions and logged-in state may remain available after the immediate task unless the server is explicitly stopped.

Why it was flagged

The skill instructs use of a long-running background server that keeps browser state alive across commands.

Skill content
scripts/server.py runs background server -> session persists
Recommendation

Stop the server after use, avoid using sensitive logged-in accounts, and prefer a disposable browser profile or automatic timeout.

#
ASI09: Human-Agent Trust Exploitation
Medium
What this means

Users may believe the skill is limited to local testing when it also supports automating public web services.

Why it was flagged

The broader artifact set includes third-party Google search automation, while the main description emphasizes localhost/dev-server testing.

Skill content
This skill can be used to automate Google searches and extract search results using the Patchright browser automation library.
Recommendation

Clearly disclose third-party web automation behavior and restrict default triggers to owned/local environments.

#
ASI04: Agentic Supply Chain Vulnerabilities
Low
What this means

Installing later package versions could introduce behavior not represented by the reviewed artifacts.

Why it was flagged

The skill relies on an external package with only a lower-bound version, so future package changes could alter behavior.

Skill content
patchright>=0.0.1
Recommendation

Pin and verify the Patchright version before installation, and install in an isolated environment.