The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to handle untrusted data from Kafka streams and implements comprehensive defensive measures.
Ingestion points: Untrusted external data enters the system as raw bytes in references/consumer-patterns.py and via the Kafka engine table in references/clickhouse-schema.sql.
Boundary markers: Strict structural boundaries are defined using msgspec.Struct schemas (e.g., OrderMessage), which enforce type-safety and reject unexpected fields during deserialization.
Capability inventory: The patterns utilize clickhouse-driver for database operations and confluent-kafka for stream communication; no unsafe execution sinks like eval() or exec() are used.
Sanitization: The skill provides multiple layers of sanitization, including business rule validation (e.g., _validate_order_integrity), ISO timestamp parsing with format verification, and length constraints on string inputs.
[EXTERNAL_DOWNLOADS]: The skill references well-known, industry-standard Python libraries including msgspec, clickhouse-driver, confluent-kafka, and pydantic. These are established packages from official registries and do not pose an inherent risk.
[COMMAND_EXECUTION]: The skill uses Bash and Grep as allowed tools. The provided code examples and scripts use these tools for standard operational tasks and do not involve dynamic command assembly or unsafe execution of user-supplied strings.

validating-clickhouse-kafka-pipelines