fusion-search
ReviewAudited by ClawScan on May 16, 2026.
Overview
This appears to be a real search skill, but it weakens browser security while visiting untrusted web pages, so it should be reviewed before use.
Before installing, decide whether you are comfortable with automated scraping via a stealth browser. Avoid sensitive queries, consider disabling full-content fetching when not needed, and run the skill in a contained environment because its Chromium launch disables several browser security protections.
Findings (4)
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.
A malicious or compromised web page reached through search results would be handled by a less-protected browser process, increasing local security risk.
The automated browser is configured with important isolation and web-security protections disabled, while the same tool navigates to search engines and result pages.
'--no-sandbox', ... '--disable-site-isolation-trials', '--disable-web-security', '--allow-running-insecure-content'
Run this skill only in a sandboxed/containerized environment, keep Playwright/Chromium updated, and remove the unsafe browser flags unless they are strictly necessary.
The skill may interact with websites in a way intended to avoid bot detection, which can have reliability, policy, or site-terms implications.
The skill injects JavaScript into browser contexts to spoof automation and fingerprinting signals. This is disclosed and aligned with the stated anti-crawl search purpose, but it is still browser-behavior manipulation.
Object.defineProperty(navigator, 'webdriver', {get: () => undefined}); ... WebGLRenderingContext.prototype.getParameter = function(parameter) { ... }Use only where automated scraping is acceptable, and consider disabling stealth behavior for normal browsing or compliant API-based search.
Search results or fetched page content could include instructions that should be treated as untrusted web content, not as commands to the agent.
The tool can return raw text extracted from arbitrary web pages. Such text may contain prompt-injection instructions even though the behavior is expected for a search/full-content extraction skill.
content = page.evaluate("""() => { ... return document.body ? document.body.innerText : ''; }""") ... return content[:8000]Treat returned snippets and full-page content as untrusted evidence; do not let web page text override user instructions or tool-safety rules.
Installation pulls in external browser automation components that are not enforced by the registry requirements.
The skill requires installing a Python package and downloading Chromium, but the registry install specification is absent. This is expected for Playwright-based browsing, but it is an installation/provenance point users should notice.
pip install playwright playwright install chromium
Install from trusted package sources, verify the Playwright/Chromium download path, and prefer a pinned or managed environment.