Agent Governance

Control what agents can discover, execute and modify.

Policy version 4.2

Permission Matrix

CapabilityAccessPolicy note
  • Customer LookupREADAuto-executed for all agents
  • Ticket SearchREADAuto-executed for all agents
  • Knowledge SearchREADAuto-executed for all agents
  • Incident AnalysisEXECUTEAnalysis-only, no writes
  • Ticket UpdateAPPROVAL REQUIREDHuman approval before any write
  • Customer MessageAPPROVAL REQUIREDOutbound comms reviewed by an operator
  • Ticket DeleteBLOCKEDDestructive action, never delegated

Approval Queue

Write actions paused for a human decision.

  • Ticket #48291

    Awaiting approval

    Update ticket · Medium risk · requested by Investigation Agent

  • Contact #9921

    Awaiting approval

    Send customer message · High risk · requested by Support Resolution Agent

  • Ticket #48122

    Awaiting approval

    Reassign ticket · Low risk · requested by Support Resolution Agent

Governance Principles

  • Read is autonomous

    Read-only capabilities execute without interruption so investigation stays fast.

  • Write is reviewed

    Any capability that mutates enterprise state pauses for an explicit human decision.

  • Everything is traced

    Permission checks, approvals and tool calls are recorded in the execution trace.