The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests untrusted data from external websites and a curated search index through its core tools.
Ingestion points: The search_agents, get_site_details, and verify_mcp tools in SKILL.md fetch content from the nothumansearch.ai index and arbitrary external URLs.
Boundary markers: No specific boundary markers or instructions to ignore embedded commands are present in the prompt templates.
Capability inventory: The agent uses these tools to retrieve and process external information at runtime.
Sanitization: No sanitization or filtering of the retrieved content is mentioned.
[EXTERNAL_DOWNLOADS]: The skill configures the agent to connect to a remote MCP server hosted at https://nothumansearch.ai/mcp. This is standard for remote MCP services but involves interaction with a third-party domain.
[DATA_EXFILTRATION]: The register_monitor tool provides the capability to send a user-provided email address to an external service. While documented as requiring user consent, there is an inherent risk of PII exfiltration if the agent pulls an email from the conversation context without explicit confirmation.

not-human-search-mcp