Opportunity Engine approval policy¶

Risky execution actions are blocked until an operator explicitly approves or rejects them. The launch orchestrator and growth loop reference these gates; live integrations stay idle until approval is recorded in 13-approval-log.md.

Risky actions¶

Action key	Operator label
`send_outreach`	Sending outreach
`publish_post`	Publishing posts
`create_ad`	Creating ads
`edit_ad`	Editing ads
`spend_money`	Spending money
`update_crm`	Updating CRM records
`send_email_sequence`	Sending email sequences
`publish_landing_page`	Publishing landing pages
`charge_customer`	Charging customers
`export_personal_data`	Exporting personal data
`upload_lead_list`	Uploading lead lists
`set_ad_budget`	Setting ad budgets
`create_stripe_product`	Creating Stripe products or prices

Implementation: src/keprix/opportunity/approvals.py.

Default behaviour¶

Launch orchestrator runs in dry run unless launch_dry_run is set to false in opportunity metadata and approvals are granted.
Validation score below 65 blocks the asset factory playbook unless validation_override is set with operator intent.
Regulated industries (healthcare, finance, legal, and similar) receive warnings in offer and launch artifacts; operators must review compliance before approval.
No private data collection by default: lead lists, personal exports, and CRM bulk updates require explicit approval.

Approving from CLI¶

keprix opportunity approve opp-xxxxxxxx create_ad --approve
keprix opportunity approve opp-xxxxxxxx spend_money --reject --reason "Budget not allocated"

Approving from API¶

Use POST /api/opportunities/{id}/approve with action, decision (approve or reject), and optional reason.

Approval log artifact¶

13-approval-log.md is a markdown table with timestamp, action, risk level, decision, and detail. Dry-run launch plans list pending gates without executing them.

Web UI¶

The opportunities detail view surfaces pending approvals and links to the launch plan. Operators should resolve gates before disabling dry run or connecting live ad spend.

What does not require approval¶

Research playbooks, markdown artifact generation, validation scoring, dry-run launch planning, and manual import templates for missing integrations are safe by default and do not spend money or publish externally.