The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The TypeScript code examples for bulkInsertTrades and the CDC listener demonstrate unsafe interpolation of external data into SQL queries. This pattern presents a vulnerability surface where malicious data from a source database or stream could execute unintended SQL commands.
Ingestion points: The trades parameter in bulkInsertTrades and the msg.payload received from the PostgreSQL LISTEN notification in SKILL.md.
Boundary markers: No boundary markers or 'ignore' instructions are present to prevent the agent from treating data as instructions.
Capability inventory: The skill utilizes the clickhouse library to execute queries and the pg library for database notifications.
Sanitization: The examples use template literals (e.g., '${trade.id}') for SQL construction instead of parameterized queries or dedicated escaping functions.
[EXTERNAL_DOWNLOADS]: The skill references external Node.js dependencies for database connectivity.
Dependencies: clickhouse and pg (PostgreSQL client) are suggested for installation and usage in the provided integration patterns.

clickhouse-io