auditing-mcp-servers-for-tool-poisoning

SUSPICIOUS/medium risk by design rather than clearly malicious: the skill is coherent with its stated MCP security-audit purpose, but it installs and runs external tooling via official yet weakly pinned paths, may route scanned tool metadata to invariantlabs.ai by default, and includes active SSRF/network probing capabilities. No clear credential theft or deceptive exfiltration is present, but the combination of live probing plus third-party scanner data flow makes it riskier than a passive documentation skill.