The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in collect_snowflake_evidence.py to invoke the official Snowflake CLI (snow). This is the primary intended function of the skill and is protected by strict regular expression validation on the connection profile name (PROFILE_RE) and temporary file paths to prevent command injection.
[DYNAMIC_EXECUTION]: The collector script generates temporary SQL files at runtime by populating reviewed templates with user-supplied identifiers. These identifiers are rigorously validated using unquoted Snowflake identifier regex patterns (one-part, two-part, and three-part) before interpolation. Additionally, the rendered SQL is scanned for forbidden keywords (e.g., DROP, DELETE, GRANT) to ensure it remains read-only.
[DATA_EXPOSURE]: The skill implements significant privacy and anti-exfiltration measures. The sanitize_output_tree and redact_authorization_values functions in collect_snowflake_evidence.py are designed to detect and redact credentials (tokens, passwords, keys) and raw SQL from tool output before it is written to receipts. Sensitive metadata like account names and roles are hashed using SHA-256 before leaving the Snowflake session.
[SAFE]: The analysis logic in analyze_failover_readiness.py is entirely deterministic and operates on local JSON evidence. It enforces a cryptographically bound evidence contract where independent digests must match for the input, policy, and operator receipts, preventing tampering with the audit results.

snowflake-failover-readiness-drill