The Agent Skills Directory

[COMMAND_EXECUTION]: The skill references the use of the montecarlo command-line interface, specifically for applying monitor configurations (montecarlo monitors apply).
[DYNAMIC_EXECUTION]: The skill dynamically generates content at runtime, including YAML monitor definitions (Workflow 2) and SQL validation queries (Workflow 5), which are saved to local project files.
[EXTERNAL_DOWNLOADS]: The documentation suggests a curl command to https://integrations.getmontecarlo.com/mcp/ as a troubleshooting step to verify connectivity to the official Monte Carlo integration endpoint.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes untrusted SQL files and dbt models to generate reports and queries.
Ingestion points: Reads .sql files and dbt models from the models/ directory as specified in SKILL.md and workflows.md.
Boundary markers: Absent. The skill does not define specific delimiters to isolate untrusted SQL content from the instructions.
Capability inventory: Performs file writes to the monitors/ and validation/ directories, executes shell commands via the montecarlo CLI, and initiates network requests through the monte-carlo MCP server (referenced in SKILL.md and workflows.md).
Sanitization: Absent. There is no evidence of sanitization or escaping of the ingested SQL code before it is used to construct impact reports or validation queries.

monte-carlo-prevent