The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external database queries, creating an attack surface for indirect prompt injection. The documentation explicitly acknowledges this risk and provides clear instructions to mitigate it.
Ingestion points: Data returned by the helix_execute_read_query tool in SKILL.md.
Boundary markers: The instructions explicitly label database results as untrusted_data and warn the agent not to follow instructions found within them.
Capability inventory: The skill includes mutation capabilities through helix_prepare_write_query and helix_execute_write_query.
Sanitization: Security is enforced through a mandatory human-in-the-loop confirmation process for all write operations, ensuring the agent cannot be tricked into executing unauthorized mutations independently.
[EXTERNAL_DOWNLOADS]: The configuration in agents/openai.yaml specifies a hosted Model Context Protocol (MCP) server at https://query-mcp.helix-db.com/mcp. This URL is a vendor-owned resource (HelixDB) necessary for the skill's brokered database access functionality.

helix-query-mcp