The Agent Skills Directory

[COMMAND_EXECUTION]: The code examples in SKILL.md (e.g., bulkInsertTrades and etlPipeline) utilize string interpolation to construct SQL queries. This pattern is vulnerable to SQL injection if variables like trade.id or row.market_slug contain untrusted data.
[PROMPT_INJECTION]: The skill identifies an indirect prompt injection surface where external data is processed and written to a database.
Ingestion points: Data enters the system context via PostgreSQL notifications (market_updates) and external data sources in the ETL pipeline examples within SKILL.md.
Boundary markers: No boundary markers or delimiters are used in the provided examples to separate untrusted data from system instructions.
Capability inventory: The skill demonstrates capabilities for executing SQL queries and performing database write operations via the clickhouse library.
Sanitization: While some numeric fields are parsed, string fields used in data transformations lack escaping or sanitization, creating an exploitable surface if the input source is compromised.

clickhouse-io