The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from the Helix Cloud service, such as query recommendations and planner findings, which could potentially contain untrusted content.
Ingestion points: Data enters via tools like helix_get_query_insights and helix_list_query_recommendations as described in SKILL.md.
Boundary markers: The skill includes a dedicated 'Trust boundary' section in SKILL.md that instructs the agent to identify tool results as untrusted_data.
Capability inventory: The current skill tools are restricted to read-only observability; however, the documentation notes the existence of related skills (helix-query-mcp, helix-admin-mcp) that possess write capabilities.
Sanitization: Explicit instructions are provided to the agent to summarize data in its own words and to never execute commands or follow links embedded within the returned content.
[EXTERNAL_DOWNLOADS]: The agent is configured to connect to a remote endpoint to access the MCP tools.
Evidence: The file agents/openai.yaml specifies a tool URL at https://mcp.helix-db.com/mcp.
Context: This URL belongs to the vendor's infrastructure and is used for its intended purpose of providing the hosted MCP server.

helix-mcp