The Agent Skills Directory

[COMMAND_EXECUTION]: The skill enables the execution of SQL commands on ByteHouse instances. It implements a safety check within scripts/execute_sql.py and scripts/search_client.py that identifies non-query commands (DML/DDL such as INSERT or DROP) and requires either interactive user confirmation or an explicit --force flag before execution.
[INDIRECT_PROMPT_INJECTION]: The skill utilizes a Text2SQL model to translate natural language user input into SQL commands. This architecture presents a surface where malicious instructions could potentially influence the generated database commands.
Ingestion points: User-provided natural language queries in scripts/text2sql.py and database metadata retrieved via scripts/list_tables.py.
Boundary markers: The is_dql validation logic in scripts/execute_sql.py serves as a guardrail to prevent the automatic execution of non-query commands.
Capability inventory: The skill can execute SQL (queries and modifications) and perform network requests to Volcengine's AI services.
Sanitization: SQL command prefixes are checked against an allow-list, and basic character escaping is applied to parameters in the search_client.py module.
[EXTERNAL_DOWNLOADS]: The skill communicates with Volcengine's Ark and ByteHouse API endpoints for Text2SQL conversion and multi-modal embedding generation. These are official services operated by the vendor (ByteDance).
[SAFE]: The database connection client in scripts/client.py is configured to disable SSL certificate verification (verify=False). This is a common configuration for internal or development database access, though it represents a security trade-off in untrusted network environments.

byted-bytehouse-ai-query