Openclaw Web Search Mcp

Provides AI agents with web search, page and PDF extraction, YouTube transcripts, summarization, semantic search, and automated research workflows.

This skill appears to implement the advertised browsing and research features but has several red flags you should resolve before installing: - The YouTube transcript code runs a shell command (yt-dlp) via execSync and calls a hardcoded path (/home/nishu/.local/bin/yt-dlp). Ask the author to (a) remove the hardcoded path, (b) document that yt-dlp is a required external binary (and how to install it), or (c) use a Node library instead. Running unverified shell commands is a risk. - SKILL.md says 'Google search' but the code scrapes DuckDuckGo. Confirm which search provider you expect and whether scraping is acceptable for your use case. - mcp.json lists a tool ('crawl_site') that is not implemented; verify the tool list and behavior to ensure there is no hidden functionality. - Because the code uses execSync and will write temporary files, run this skill in a sandboxed environment (or on an isolated agent) until you validate it. Review/modify the youtubeTranscript code to avoid arbitrary shell execution if you cannot fully trust the package. If you are not comfortable auditing or modifying the code, do not install this skill in a production or high-privilege agent. Asking the author for corrected packaging and explicit runtime requirements would materially reduce the risk.

Type: OpenClaw Skill Name: openclaw-web-search-mcp Version: 1.0.0 The skill bundle contains a critical command injection vulnerability in 'extraction/youtubeTranscript.js' where the 'url' input is directly interpolated into an 'execSync' shell command. Additionally, this file uses a hardcoded absolute path to a specific local user's directory ('/home/nishu/.local/bin/yt-dlp'), which is highly irregular for a portable bundle and suggests poor development practices or an environment-specific exploit. While these flaws present a significant RCE risk, they appear to be unintentional vulnerabilities rather than purposefully designed malware, as no exfiltration or persistence logic was identified.