The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes well-known, official Python libraries for interacting with data warehouses and the Monte Carlo API, including 'pycarlo', 'snowflake-connector-python', 'google-cloud-bigquery', 'databricks-sql-connector', 'psycopg2-binary', and 'pyhive'.
[COMMAND_EXECUTION]: The generated scripts execute SQL queries on the user's data warehouse to collect metadata. The templates incorporate proper identifier quoting and validation (e.g., '_quote_identifier', '_safe_hive_identifier') across all warehouse scripts to prevent SQL injection vulnerabilities.
[DATA_EXFILTRATION]: Warehouse metadata, lineage, and logs are transmitted to Monte Carlo's official ingestion gateway ('integrations.getmontecarlo.com'). This is the primary intended function of the skill and utilizes authenticated API calls with environment-provided credentials.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests data from warehouse system tables (e.g., Snowflake 'QUERY_HISTORY' in 'collect_lineage.py') and local server logs (e.g., HiveServer2 logs in 'collect_query_logs.py').
Boundary markers: Boundary markers are not present in the raw log data, but the scripts use rigid regex parsing instead of direct LLM interpolation of the content.
Capability inventory: The skill uses the 'pycarlo' SDK and 'requests' for network operations in 'push_metadata.py', and provide utilities for writing JSON manifests to the local file system.
Sanitization: Inputs are sanitized via SQL identifier quoting and regex-based identifier validation across all warehouse templates.
[SAFE]: The skill implements a security-first approach by including '_safe_paths.py' in every template directory, ensuring file operations are restricted to the current working directory unless explicitly overridden by an environment variable. It also performs host validation for Redshift endpoints and includes memory safety checks.

monte-carlo-push-ingestion