question: where is one browser/computer action authorized before execution?

### What I am trying to do

Understand where authorization lives at the public Cindy client-to-executor boundary before one browser, computer, phone, IM, or scheduled action executes.

I maintain [Vizier](https://github.com/vassiliylakhonin/vizier), an experimental deterministic authorization service. I am running a five-participant, one-week learning test—not a sales campaign—and want to start from a past action and Cindy's current control rather than assume a missing feature.

Would a maintainer be open to a 20-minute technical conversation about the last real reversible browser or computer action, what could have gone wrong, and what permission or confirmation ran immediately before execution? If a concrete seam is missing, I can do the integration work for one synthetic action. Timeout, REVIEW, and BLOCK remain stop conditions; no credentials or private data need to be shared.

Pilot scope and kill criteria: https://github.com/vassiliylakhonin/vizier/blob/main/docs/PILOT.md

### What I have tried

I reviewed the public README, support guidance, and public client boundary. The backend authorization path is not evident from those sources, so I am asking rather than inferring it.

### Environment

- Cindy version or commit: current public `main`, checked 2026-08-26
- Platform and OS version: not yet running; discovery conversation first

Start by reviewing the public README, support guidance, and public client boundary on the current main branch, which the report identifies as the available sources. The issue does not name a file, test, or implementation change; a maintainer conversation must first identify a real reversible action and an authorization seam. Done would mean agreeing on that seam and defining the synthetic integration scope.

Written by the indexing model from the issue text.