The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/export_config.sh utilizes the eval command to process and export environment variables from a local JSON configuration file (~/.bytehouse_config.json). This pattern of dynamic command execution on file content poses a risk if the configuration file is compromised or contains unexpected keys.
[INDIRECT_PROMPT_INJECTION]: The ByteHouseHybridSearch client in scripts/hybrid_search_client.py constructs SQL queries by directly interpolating the table_name and query parameters. This creates a vulnerability surface where malicious input could potentially lead to SQL injection.
Ingestion points: The table_name and query parameters in methods within scripts/hybrid_search_client.py (e.g., create_hybrid_table, fulltext_search, insert_document).
Boundary markers: None identified in the SQL construction logic.
Capability inventory: The skill uses self.client.command and self.client.query to execute SQL against the database, which are high-capability operations.
Sanitization: While transform_string performs basic character manipulation for query strings, the table_name parameter is interpolated into SQL statements without any validation or sanitization.
[CREDENTIALS_UNSAFE]: The database connection initialization in scripts/hybrid_search_client.py sets verify=False for the ClickHouse client. Disabling SSL certificate verification weakens the security of the connection, potentially exposing database credentials and sensitive data to Man-in-the-Middle (MitM) attacks during transport.

byted-bytehouse-hybrid-search