The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates SQL queries at runtime by substituting user-provided identifiers into predefined templates. This process is secured through strict regex validation of inputs and a security scan of the final SQL string to block mutation keywords such as 'DROP' or 'DELETE'.
[COMMAND_EXECUTION]: The collect_snowflake_evidence.py script invokes the Snowflake CLI ('snow') using subprocess.run to gather metadata. These operations are limited to reviewed read-only SQL scripts and include execution timeouts.
[DATA_EXFILTRATION]: The skill prevents the exposure of sensitive data by redacting credentials, authorization headers, and connection strings from Snowflake CLI output. It also applies SHA-256 hashing to all database object identifiers before they are returned or processed, ensuring customer identifiers are never exposed.
[INDIRECT_PROMPT_INJECTION]: The skill ingests metadata from Snowflake. Ingestion point: collect_snowflake_evidence.py (CLI output). Boundary markers: receipt_sha256 self-checksums and mandatory trusted-input-sha256 validation. Capability inventory: subprocess.run (Snow CLI) and local file writes. Sanitization: redact_authorization_values, reject_secret_fields, and validate_read_only_sql functions.

snowflake-governance-coverage-auditor