Guide

Governance Policies 101

Cadreen evaluates every tool call against governance policies before execution. This guide shows how to create policies, evaluate actions, and understand the approval flow.

How governance works

When the model proposes a tool call, Cadreen evaluates it. You don't need to build this. It happens automatically on every request.

Click a step for details

Create a policy

POST/api/v1/cadreen/policies

Request

curl -X POST https://accomplishanything.today/api/v1/cadreen/policies \
  -H "Authorization: Bearer sk_cadreen_..." \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Refund Policy",
    "rules": [
      {
        "condition": "Require manager approval for refunds over $100",
        "effect": "require_human_approval"
      },
      {
        "condition": "Auto-approve refunds of $100 or less",
        "effect": "auto_approve"
      }
    ]
  }'

Note

The policy is created in draft status. It must be confirmed before it goes live.

Policy fields and rule effects

EffectWhat happens

autoTool call executes immediately

handoffBlocked until you confirm

escalateSent to a higher authority

clarifySystem asks for more information

Evaluate a tool call

Test a policy against an action before it happens in production:

POST/api/v1/cadreen/policies/evaluate

Request

curl -X POST https://accomplishanything.today/api/v1/cadreen/policies/evaluate \
  -H "Authorization: Bearer sk_cadreen_..." \
  -H "Content-Type: application/json" \
  -d '{
    "action": "process_refund",
    "context": {
      "amount": 250,
      "customer_tier": "enterprise"
    }
  }'

Response

{
  "decision_id": "dec_01abc",
  "action": "approve",
  "policy_triggered": "Refund Policy",
  "checks": [
    {"rule": "Max auto-approval", "result": "exceeds", "detail": "$250 > $100 threshold"},
    {"rule": "Account age", "result": "pass", "detail": "Customer since 2023"}
  ],
  "reason": "Requires manual review due to amount threshold"
}

The approval flow

When governance blocks a tool call, you see it in the chat response:

Blocked response

I need your approval before I can process_refund.

Say "yes" to proceed or "no" to skip.

You can respond with:

yes / y / confirm / approveConfirm all pending

no / n / skip / cancelCancel all pending

1,3 / 1 3Confirm specific items by number

all / everythingConfirm all

Note

Or respond in natural language ("go ahead", "I need more details first") and the system interprets your intent.

What gets audited

Every governance decision is logged. These appear in intelligence traces and are available for compliance reporting.

Tool name — which tool was called

Decision — handled on its own, asked for permission, needs a human

Reason — why the decision was made

Confidence — governance confidence level

Domain — governance domain

Timestamp — when it happened

How governance works for all tools

Every tool call goes through governance — both Cadreen tools and client tools. The difference is where they execute: Cadreen tools run server-side after approval, client tools are returned to you for local execution. If governance blocks a client tool, it becomes a conversation just like a blocked Cadreen tool.

Note

Next: Tool Calling with /completions — hybrid execution, tool chaining, and self-healing.