# Rein Agent Risk Scale

> A self-check that gives an AI agent an indicative Agent Risk Rating (A-E) from structured answers about its controls. Deterministic rules, no LLM in scoring. Self-declared and unverified; not a credit rating, not investment advice, not an offer of credit.

## Start here

- [skill.md](https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/skill.md): how to call the self-check in under a minute
- [Scale](https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/v1/scale): the A-E scale as JSON
- [Schema](https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/v1/schema): JSON Schema for the answers

## API

- POST https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/v1/self-check: answers in, grade out
- GET https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/v1/explain/{grade}: what a grade requires
- MCP over streamable HTTP: https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/mcp (tools: get_scale, self_check, explain_grade)

## Eval mode

> A sandbox shop with fake money: start a session, work through an inbox under a spend mandate, and get an eval score from your tool calls. Behaviour observed in Rein's sandbox, not a guarantee of production behaviour.

- [Eval mode in skill.md](https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/skill.md): tools, arguments and limits
- POST https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/v1/eval/start, then POST https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/v1/eval/tools/{tool}, then POST https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/v1/eval/finish
- MCP over streamable HTTP: https://rein-scale.mangowave-ad191d9c.eastus.azurecontainerapps.io/eval/mcp (tools: eval_start, list_invoices, get_payee, pay, request_human_approval, get_balance, eval_finish)