The Agent Skills Directory

[SAFE]: The skill focuses on providing legitimate technical guidance and enforces security best practices for database interactions. It specifically includes a safety rule (agent-query-safety.md) to prevent resource exhaustion and unbounded queries.
[EXTERNAL_DOWNLOADS]: The skill references official ClickHouse resources and tools. It suggests the installation of mcp-clickhouse (a Python package) and connects to mcp.clickhouse.cloud and other vendor-owned subdomains. These resources are scoped to the vendor's own infrastructure.
[COMMAND_EXECUTION]: The instructions provide templates for executing clickhouse-client and curl to interact with ClickHouse clusters. These are the intended primary functions of the skill and include recommended safety settings like max_execution_time and max_rows_to_read.
[INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface for potentially untrusted data through database metadata and query results.
Ingestion points: Data enters the agent context via system.columns, system.tables, and the results of arbitrary SELECT queries.
Boundary markers: The skill uses standard markdown formatting but does not implement specific adversarial boundary markers for database content.
Capability inventory: The agent can perform network operations and execute database commands, including write operations if explicitly enabled by the user via environment variables.
Sanitization: No explicit sanitization or filtering of database output is mandated, though the query safety rules mitigate some volumetric risks.

clickhouse-best-practices