Version: 2026.04 | Last Updated: April 2026| Status: Draft

Action Approval

The Action Approval dashboard enables human-in-the-loop oversight for AI agent actions, providing approval workflows, policy enforcement, and automation playbooks.

Overview

When an agent action exceeds risk thresholds or triggers a policy, it enters the approval queue for human review.

Source: owkai-pilot-frontend/src/components/AgentAuthorizationDashboard.jsx

warning

Approval and denial decisions are irreversible once submitted. Review the action details, risk score, and policy context carefully before making a decision -- denied actions must be resubmitted by the agent.

Dashboard Tabs

Tab	Description
Pending	Actions awaiting approval
Workflows	Approval workflow configuration
Policies	Governance policies
Automation	Automation playbooks
Execution	Execution history

Pending Actions Queue

Queue Filters

Filter	Options
Source	All, Agent, MCP, High Risk
Status	Pending, Approved, Rejected
Risk Level	Low, Medium, High, Critical

Action Card Information

Each pending action displays:

┌─────────────────────────────────────────────────────────────┐
│ database_write                                  Risk: 75/100 │
├─────────────────────────────────────────────────────────────┤
│ Agent: financial-advisor-001                                │
│ Resource: production_db.customers                           │
│ Description: Update customer profile                        │
├─────────────────────────────────────────────────────────────┤
│ NIST: AC-3     MITRE: T1003     CVSS: 6.5                  │
├─────────────────────────────────────────────────────────────┤
│ [Approve]  [Deny]  [Details]                                │
└─────────────────────────────────────────────────────────────┘

Approval Actions

Approve Action

1. Review action details and risk score
2. Click Approve
3. Add optional comment
4. Confirm approval

Audit Trail: Records approver, timestamp, and justification.

Deny Action

1. Review action details
2. Click Deny
3. Provide reason (required)
4. Confirm denial

Agent Notification: Agent receives denial reason via webhook.

Emergency Override

For time-sensitive situations:

1. Click Emergency Override
2. Provide detailed justification (required)
3. Select override type:
   • Expedited Approval: Skip additional approvers
   • Risk Acknowledgment: Approve despite high risk

Security Alert: Emergency overrides trigger security team notification.

Approval Workflows

Configure multi-level approval chains:

Workflow Structure

{
  "name": "High-Value Transaction Approval",
  "description": "Multi-level approval for transactions over $10,000",
  "steps": [
    { "level": 1, "role": "team_lead", "timeout_hours": 4 },
    { "level": 2, "role": "manager", "timeout_hours": 8 },
    { "level": 3, "role": "director", "timeout_hours": 24 }
  ],
  "triggers": [
    { "field": "risk_score", "operator": ">=", "value": 70 },
    { "field": "action_type", "operator": "in", "value": ["funds_transfer"] }
  ],
  "approvers": ["team-leads", "finance-managers"]
}

Workflow Triggers

Trigger	Description	Example
risk_score	Risk threshold	>= 70
action_type	Specific actions	database_delete
resource	Resource patterns	*.pii
agent_id	Specific agents	autonomous-*

Automation Playbooks

Automate routine approval decisions:

Playbook Configuration

{
  "id": "auto-approve-low-risk",
  "name": "Auto-Approve Low Risk Reads",
  "status": "active",
  "risk_level": "low",
  "approval_required": false,
  "trigger_conditions": {
    "risk_score": { "min": 0, "max": 30 },
    "action_type": ["database_read", "api_read"],
    "environment": ["development", "staging"],
    "business_hours": true
  },
  "actions": [
    { "type": "auto_approve", "delay_seconds": 0 },
    { "type": "log", "message": "Auto-approved low-risk read" }
  ]
}

Playbook Features

Feature	Description
Template Library	Pre-built playbook templates
Dry-Run Testing	Test playbooks before activation
Version History	Track playbook changes
Analytics Dashboard	Monitor playbook performance

Playbook Actions

Action Type	Description
auto_approve	Automatically approve
auto_deny	Automatically deny
escalate	Send to next approval level
notify	Send notification
log	Add audit log entry
webhook	Call external webhook

Policy Enforcement

The Enhanced Policy Tab provides governance policy management:

Policy Structure

{
  "policy_name": "Block PII Access After Hours",
  "description": "Deny PII access outside business hours",
  "effect": "DENY",
  "conditions": {
    "resource_type": "pii",
    "time": { "outside": "09:00-17:00" }
  },
  "priority": 100
}

Policy Decision Badge

Actions display policy evaluation results:

Badge	Meaning
ALLOW	Policy permits action
DENY	Policy blocks action
REQUIRE_APPROVAL	Needs human review
CONFLICT	Multiple policies conflict

Execution History

Track all approval decisions:

Field	Description
Action ID	Unique action identifier
Decision	Approved/Denied/Auto-approved
Approver	Who made the decision
Timestamp	When decision was made
Duration	Time from submission to decision
Playbook	Automation playbook used (if any)

API Reference

Endpoint	Method	Description
/api/authorization/pending	GET	Get pending actions
/api/authorization/approve/{id}	POST	Approve action
/api/authorization/deny/{id}	POST	Deny action
/api/authorization/workflows	GET	List workflows
/api/automation/playbooks	GET	List playbooks

Source: ow-ai-backend/routes/authorization_routes.py

Best Practices

1. Set SLAs: Configure approval timeouts to prevent queue buildup
2. Use playbooks: Automate repetitive low-risk approvals
3. Review denials: Monitor denial patterns to improve policies
4. Escalate promptly: Don't let critical actions wait
5. Document decisions: Always add context to approvals/denials

Troubleshooting

Queue not refreshing

Solution: Check WebSocket connection; refresh page if needed.

Playbook not triggering

Solution: Verify trigger conditions match action attributes exactly.

Approval timeout

Solution: Check email notifications; escalate to backup approver.

---

Source: AgentAuthorizationDashboard.jsx, authorization_routes.py