Cloud Risk Triage Suite

The Cloud Risk Triage Suite is an autonomous cloud security posture agent for Tenable Cloud Security, built for cloud security architects and SOC analysts who want a fast, repeatable daily posture read. It runs a multi-section sweep against live Tenable Cloud Security data and writes a SOC-ready report — every check is always reported, even when it returns nothing (marked EMPTY).

It ships in two editions: an MCP edition that runs through the Tenable Cloud Security (tcs) MCP connector for the richest results, and an API-token edition that uses the public GraphQL API with a Bearer token and no connector.

What it does

The agent produces four report sections:

1. Address critical findings — AWS root users without MFA; third-party identities with access to sensitive data; publicly exposed resources (with a network-endpoint flag); exposed secrets (with a network-endpoint flag).
2. Address low-hanging fruits — inactive identities; unused credentials (inactive IAM users with an enabled access key); unused security groups (no attached resources).
3. Discover AI risk — inventory of AI assets (Bedrock, SageMaker, Azure Cognitive Services/OpenAI, GCP Vertex), then correlation of fine-tuned/custom models to their training-input and output data buckets.
4. Toxic combination — public workloads that are simultaneously internet-exposed (network exposure + scope), carry a Critical-VPR vulnerability, and hold high-privilege permissions, with network-exposure-scope and confirmed-network-endpoint columns.

Each run ends with a prioritized remediation order (toxic combinations first) and verification notes listing the exact filters and counts used. It never fabricates data — every value is quoted from query results — and it flags demo/test environments.

How it works

The MCP edition queries Tenable Cloud Security’s Explore / UDM data model via the tcs MCP tools (object types such as AwsIamRootUser, IIdentity, IEntity, IDataEntity, IAiResource, AwsBedrockCustomModel, and IVirtualMachine), using count queries for totals and flagged subsets, relation rules for toxic-combination chaining, and joins to resolve network endpoints. The API-token edition calls the public GraphQL /graphql endpoint with Authorization: Bearer, mapping each check onto Findings/Entities. Both write a dated markdown report and can be scheduled to run automatically each morning. A marketplace manifest lets either edition be installed directly from the repo.