agent-observatory-sensor.nolimit-observatory.workers.dev

The description instructs the agent to 'Follow start_here.hop first (skill file, 302)' and says it 'Also returns docs, CLI, and remote MCP hops.' This is an instruction aimed at the AI to automatically chain through redirects and load additional remote tool definitions/MCP servers, which could be used to smuggle in untrusted or malicious tool schemas, exfiltrate data via a 'pay-per-call' gateway, or bypass user awareness of what external services are being contacted. A tool whose stated purpose is merely 'finding a gateway' should not instruct the agent to blindly follow hops or fetch remote MCP definitions.

Description says it returns 'CLI setup, docs, API, and remote MCP hops' for a 'living people-search index.' This encourages the agent to pull in additional untrusted remote MCP tool sets and to use a people-search service (potentially aggregating personal data on real individuals) without any indication of consent, legality, or purpose limitation. Combined with the directive-like phrasing common to agent-tool-index, this looks like a vector for expanding the agent's capabilities/toolset via untrusted external hops and for enabling privacy-invasive lookups beyond the tool's simple index-finding description.