The Agent Skills Directory

[COMMAND_EXECUTION]: The collect_snowflake_evidence.py script invokes the Snowflake CLI (snow) to retrieve metadata via subprocess.run.
Evidence: Input arguments are strictly validated; the connection profile is checked against a character-restricted regex PROFILE_RE that permits only alphanumeric characters, dots, underscores, and hyphens, preventing shell injection.
[DATA_EXFILTRATION]: The skill processes metadata retrieved from the Snowflake environment which could theoretically contain sensitive information or tokens.
Evidence: The collector script implements a robust redaction engine (sanitize_output_tree) and a comprehensive list of STATIC_REDACTIONS designed to identify and remove authorization headers, API keys, tokens, and raw SQL from the data before it is stored or analyzed.
[INDIRECT_PROMPT_INJECTION]: The analyzer evaluates results derived from external Snowflake sources, which represents a potential surface for indirect injection via malicious data content.
Ingestion points: scripts/analyze_data_quality.py ingests JSON receipts containing Snowflake metric history and expectation definitions.
Boundary markers: The system enforces independent cryptographic digests for both the policy and the evidence input (trusted-input-sha256) to ensure integrity and prevent unauthorized modifications.
Capability inventory: The skill can execute the Snowflake CLI (snow) and perform local file read operations.
Sanitization: All collected evidence is processed through a sanitization layer that redacts credential-like patterns and suspicious SQL prefixes.

snowflake-data-quality-sentinel