The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill automatically installs the official snowflake-cli tool using uv if it is not present on the system. This ensures the environment has the necessary tooling from a reputable source.
[COMMAND_EXECUTION]: The wrapper script executes snow CLI commands to interact with Snowflake databases. It includes extensive safety mechanisms:
Detects mutations (e.g., INSERT, UPDATE, DROP) and requires an explicit --execute flag and user confirmation before running them.
Enforces restricted file permissions (0600) on all generated output files and configuration directories.
Prevents secrets (passwords, tokens) from being passed as command-line arguments, instead handling them via protected files or environment variables.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it executes SQL queries and object definitions that may contain instructions from untrusted data.
Ingestion points: SQL content is ingested through the --sql and --file parameters, and object definitions via --definition-file in scripts/snowflake.sh.
Boundary markers: Instructions mandate the use of --preview for inspecting SQL and object changes, followed by manual user approval.
Capability inventory: The skill utilizes snow sql for SQL execution and snow object for management, providing full access to Snowflake database resources within the connection's role scope.
Sanitization: The wrapper implements a local-only SQL configuration to block remote file inclusion (!source), performs regex-based mutation detection to force previews, and validates JSON definitions to ensure they do not exceed size limits or contain sensitive credential fields that would be exposed in process arguments.

snowflake