The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Snowflake CLI (snow sql) to execute DDL statements and queries. This is the primary function of the skill and is directed at the user's own Snowflake instance using user-configured connections.
[EXTERNAL_DOWNLOADS]: The skill references official Snowflake documentation for the installation of the Snowflake CLI and connection configuration. These are well-known, trusted sources.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests untrusted data from Snowflake metadata (table and column comments).
Ingestion points: The agent reads Snowflake table, view, and column comments, as well as sample data (LIMIT 1000), which are external sources that could contain malicious instructions (SKILL.md).
Boundary markers: The instructions do not specify explicit boundary markers or delimiters for the ingested metadata content.
Capability inventory: The skill has the capability to execute shell commands via the Snowflake CLI (snow sql) and perform network operations to communicate with the Snowflake service (SKILL.md).
Sanitization: No specific sanitization or escaping of external content is mentioned, although the workflow includes a human-in-the-loop step by requiring user approval for generated synonyms and comments before finalization.

snowflake-semanticview