Use case

Approval Inbox

A reviewable inbox of pending agent actions — context-rich, SLA-bound, routable — where high-impact workflows go for human signoff before execution.

Overview

An approval mechanism that lives in email or Slack is a queue that decays. A dedicated inbox with context and SLAs is the difference between human-in-the-loop and human-as-fig-leaf.

What it solves

Makes human review of agent actions consistent, observable, and routable. Surfaces the rate of approvals as a signal about which workflows have earned more autonomy.

How we build it

Each pending approval carries the trigger event, the proposed action, the agent's reasoning, the trace ID, and the SLA. Reviewers approve, reject with reason, or modify. Reject and modify events feed the eval set. Expired requests follow a documented default.

  • Full context per approval
  • Approve, reject, or modify
  • SLAs with expiry defaults
  • Rejections feed eval set

What changes when it is in place

Review becomes consistent and tracked. The team can see which workflows need looser gates and which need tighter ones.