monte-carlo-remediation
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes shell access to discover and execute administrative CLI tools such as
gh,dbt, andairflow. It identifies available capabilities at runtime and generates commands to trigger pipeline runs or code changes. - [DATA_EXFILTRATION]: The agent retrieves technical metadata including table schemas, query history, and asset lineage via the Monte Carlo MCP server. It also facilitates communication with external service APIs using
curlfor remediation purposes. - [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted data from data quality alerts, incident summaries, and troubleshooting agent (TSA) results to decide on remediation actions.
- Ingestion points: Alert context (
get_alerts), Troubleshooting Agent outputs (run_troubleshooting_agent), and database metadata (get_table,get_queries_for_table). - Boundary markers: None identified in the instructional workflows.
- Capability inventory: Shell command execution, CLI tool access, network requests via
curl, and database metadata retrieval. - Sanitization: The skill enforces a human-in-the-loop pattern, requiring the agent to present a full remediation plan and rollback strategy for user approval before execution.
Audit Metadata