AGENT WORKSHOP / REAL WORK, RECORDED

A website that shows
its working.

A small experiment in AI-assisted maintenance. Different agents inspect the site, make focused changes, and review the result.

These are curated records from actual runs, not simulated conversations. Work shown here is from a local preview. A passing check does not mean a change is published.

01 / THE ROLES

Inspect. Build. Review.

EVENTS / NOT A VIDEO STREAM
01

Inspector

Looks for broken paths, unclear content, and accessibility issues.

No activity loaded

02

Builder

Makes focused changes and checks the affected pages.

No activity loaded

03

Reviewer

Checks the change independently and flags what still needs attention.

No activity loaded

02 / ACTIVITY

The maintenance record.

Loading the latest record…

UPDATES EVERY 30 SECONDS WHILE OPEN

    View the activity data ↗

    03 / THE BOUNDARY

    Useful changes.
    Visible evidence.

    The agents can propose small repairs to links, layout, and accessibility. Changes to my profile, project claims, or public deployment still need my review.

    The record shows concise task summaries and selected evidence. It does not publish private prompts, credentials, or internal conversations. A run can finish with nothing to change.

    This page reads a static activity file. It does not run agents in your browser. New records reach the public website when a site update is published.