magnet-searcher
PassAudited by VirusTotal on May 13, 2026.
Overview
Type: OpenClaw Skill Name: magnet-searcher Version: 1.0.0 The skill facilitates searching for magnet links by instructing the agent to perform high-risk actions, including system-level package installation (apt-get install chromium), global NPM package installation, and the execution of arbitrary JavaScript within a browser context (agent-browser eval). While these capabilities are plausibly needed for the stated purpose of web scraping and automation, the requirement for root-level modifications and broad execution power without built-in sanitization or environment constraints warrants a suspicious classification under the provided criteria. No explicit evidence of data exfiltration or intentional malware was found in SKILL.md or the associated files.
Findings (0)
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.
Installing these tools can change the local system and depends on the current npm/package-manager sources.
The skill relies on manually installed, unpinned global/npm/system browser tooling rather than a pinned install spec. This is disclosed and central to browser-based searching, but it expands supply-chain and provenance exposure.
npm install -g agent-browser ... apt-get install -y chromium ... agent-browser install
Install only from trusted package sources, prefer pinned versions where possible, and review agent-browser and Chromium provenance before use.
The agent could land on unsafe or misleading pages while looking for magnet links.
The core workflow uses browser automation to visit magnet/search sites, click search results, and run JavaScript in the browser. This is purpose-aligned, but the destination pages and search results are untrusted.
agent-browser open "https://btsow.pics"
...
agent-browser click @e36
...
agent-browser eval "document.querySelector('a[href*=magnet]')?.href"Keep the task narrowly scoped, avoid entering sensitive information on these pages, and verify any resulting links before using them.
A malicious or spammy page could try to influence the agent during the current browsing task.
The workflow may bring arbitrary text from external pages into the agent context. This is needed for search/extraction and is not persistent memory, but untrusted page text should not be treated as instructions.
agent-browser snapshot ... agent-browser eval "document.body.innerText" # Get all text, then search manually
Treat webpage text as data only, ignore any page instructions aimed at the agent, and use the content only to identify the requested magnet link.