The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes natural language queries and displays database content, creating a surface for indirect prompt injection where malicious instructions could be embedded in data or queries.
Ingestion points: Ingests untrusted data via the query parameter in nl2sql (SKILL.md) and database row data returned in execute_sql (scripts/tunnel.py).
Boundary markers: The skill body in SKILL.md contains core principles instructing the agent to be cautious with data modification and to prioritize safety, but does not implement code-level sanitization of database outputs.
Capability inventory: The skill has network access capabilities via urllib.request to communicate with the Database Viking API Gateway (scripts/dbw_client.py).
Sanitization: The implementation relies on backend "security rules" enforced by the Volcengine API, as evidenced by specific error handling for blocked DML operations in _extract_error_info (scripts/tunnel.py).
[EXTERNAL_DOWNLOADS]: The skill requires the 'volcengine' Python package, which is the official SDK for the service provider (Bytedance). This is a standard dependency for interacting with Volcengine cloud resources.
[COMMAND_EXECUTION]: The skill allows for the execution of SQL commands. This is the intended primary function of the tool and is managed via an authenticated API gateway with stated security monitoring.

byted-viking-aisearch-database