Approval Workflows

Configure how approval requests are routed, who can approve, and what happens when approvals aren't handled.

Routing

By Trust Tier

Route approvals based on agent trust:

Trust Tier	Default Routing
Tier 1	Auto-approve most operations
Tier 2	Team lead approval
Tier 3	Security team + Team lead
Tier 4	Security team required
Untrusted	Admin approval required

By Operation Type

Route based on what the agent is attempting:

Operation Type	Example Routing
DATABASE_WRITE	Data team
EXTERNAL_API_CALL	Security team
FILE_WRITE	Team lead
Financial operations	Finance + Security

By Risk Level

OpenBox calculates a risk level for each request:

Risk Level	Calculation	Routing
Low	Low-risk operation + high trust tier	Team member
Medium	Standard operation + standard tier	Team lead
High	Sensitive operation or lower tier	Security team
Critical	High-risk operation + low tier	Admin required

Escalation

Time-Based Escalation

If not handled within threshold, escalate:

0-30 min:    Primary approver (Team lead)
30-60 min:   Secondary approver (Security team)
60+ min:     Admin + Alert

Configure in Organization → Settings → Approval Workflows.

Automatic Escalation Triggers

• Approval approaches timeout
• Multiple pending approvals from same agent
• Agent trust tier drops while approval pending
• High-risk operation from flagged agent

Notification Channels

Configure how approvers are notified:

Email

• Immediate notification on new request
• Reminder at 50% of timeout
• Escalation notification

Slack

#approvals channel:

🔔 Approval Required
Agent: Customer Support Agent (Tier 3)
Operation: EXTERNAL_API_CALL → stripe.com
Risk: Medium
Timeout: 23h remaining

[Approve] [Reject] [View Details]

Webhook

Send events to your systems for custom integrations:

{
  "event": "approval_requested",
  "approval_id": "apr_abc123",
  "agent_id": "agt_xyz789",
  "operation": {
    "type": "EXTERNAL_API_CALL",
    "target": "stripe.com/customers"
  },
  "risk_level": "medium",
  "timeout_at": "2024-01-16T09:14:32Z"
}

Approval Policies

Auto-Approve Rules

For trusted scenarios, skip the queue:

auto_approve:
  - condition:
      trust_tier: 1
      operation_type: DATABASE_READ
    reason: "Tier 1 agents can read databases"

  - condition:
      trust_tier: [1, 2]
      operation_type: LLM_CALL
    reason: "LLM calls allowed for trusted agents"

Auto-Reject Rules

Block without human review:

auto_reject:
  - condition:
      trust_tier: 5  # Untrusted
      operation_type: DATABASE_WRITE
    reason: "Untrusted agents cannot write to database"

Require Multiple Approvers

For high-risk operations:

multi_approval:
  - condition:
      risk_level: critical
    required_approvers: 2
    must_include: ["security_team"]

Approval Audit Trail

Every approval decision is logged:

Approval ID: apr_abc123
Agent: Customer Support Agent
Operation: EXTERNAL_API_CALL
Requested: 2024-01-15 09:14:32 UTC
Decision: APPROVED
Decided by: john@company.com
Decided at: 2024-01-15 09:18:45 UTC
Comment: "Verified customer requested refund"
Trust Impact: None

Access in:

• Organization → Audit Log (all approvals)
• Agent Detail → Adapt (per-agent)
• Compliance → Attestation (for auditors)

Next Steps

1. View Audit Log - See all approval decisions and other organization activity
2. Review Agent Approvals - See approvals for a specific agent
3. Compliance - Use approval evidence alongside attestation and audit logs