The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from external ClickHouse databases into the agent context, which presents an attack surface for indirect prompt injection. \n
Ingestion points: Results from client.query, rs.json(), and rs.stream() in references/client-ts.md. \n
Boundary markers: The skill includes clear instructions to use query_params and avoid string interpolation for security (SKILL.md, references/usage-guide.md). \n
Capability inventory: The skill provides tools for DDL execution (client.command), data insertion (client.insert), and raw HTTP execution (client.exec) (references/client-ts.md). \n
Sanitization: Input is sanitized using server-side parameterized binding as documented in the client API sections (references/client-ts.md). \n
[EXTERNAL_DOWNLOADS]: The skill instructs the user to install official packages from npm and utilizes testcontainers to fetch ClickHouse Docker images (references/testing-migrations.md, references/usage-guide.md). \n
[COMMAND_EXECUTION]: The skill provides commands for verifying package versions and environment status using bun pm ls and npm view (references/source-map.md). \n
[SAFE]: The skill demonstrates safe handling of sensitive data by recommending environment variables for passwords and tokens and following official ClickHouse integration patterns.

clickhouse