All work · AI tool

HOMETOWN

AI Feedback Tool

I built and shipped an internal Chrome extension that turns a click on any page or component into an AI-drafted Jira ticket, written for both the developer and the coding agent that picks it up.

  • ClientHometown
  • RoleSenior User Experience Designer
  • TeamUsed by PMs and developers
01

The problem

Filing bugs by hand was tedious. The team didn’t like doing it, it was seen as too time-consuming, and some developers were frustrated by how much time went into writing up each bug manually.

02

One click to a complete ticket

Click any page or component, and the tool drafts a Jira ticket with everything a developer needs attached: a screenshot, DOM data, and environment details. It also sets a priority and routes the ticket to the right team.

Capture a page or a component

Capture the whole visible page, drag to select an area, or pick a single component. Hovering highlights each element with its selector and size, so you click exactly what’s wrong.

Side panel capture options: capture the full visible page, select a component, or drag to select an area, with recent tickets below.Selecting a component: the Meet Hometown heading is highlighted with its CSS selector and size while the panel waits for a click.

An AI draft, ready to review

The tool drafts the summary and description, and pre-fills the issue type, priority, and labels from what it found on the page. “Why these values?” explains each choice.

AI-drafted ticket: cropped component image, auto summary and description, pre-filled issue type, priority, and labels, with a Why these values link and the attachments list.
03

Written for two readers

Each ticket is written for two readers: the developer, and the coding agent that picks it up. The same ticket gives a person the context to understand the issue and gives an agent the structure to act on it.

The ticket in Jira

Detected issues are ranked by severity, and the steps to reproduce include the exact element selector. A developer can read it at a glance; a coding agent can follow it to the exact spot on the page.

The filed ticket in Jira with summary, detected issues ranked by severity, steps to reproduce with the exact element selector, expected and actual results, and labels.
04

A person approves every ticket

The AI drafts; people decide. Users review the draft, edit anything that’s off, and approve it before it submits, so nothing reaches Jira without a person signing off.

Review, then file

Before filing, users see every attachment and can uncheck anything unrelated to keep it out of the ticket. Nothing goes to Jira until they click File ticket.

Review step: attachments with checkboxes to leave out anything unrelated, and Discard and File ticket buttons.

Filed, with a link back

The panel confirms the ticket was created, shows exactly what was sent, and links straight to it in Jira.

Filed: the panel confirms BUG-9 was created in Jira, with the description that was sent and buttons to copy the link, open it, or start a new capture.
05

Built with Claude Code, shipped to the team

I built the tool myself as a Chrome extension using Claude Code. To roll it out, I shipped it through our internal GitHub so anyone on the team could install it, and ran a short training session to get people started. My team’s PMs and developers adopted it for filing feedback and bugs.

About these screens

The original runs internally at Hometown. The screens here come from a recreation I rebuilt for this portfolio, so no internal data is shown.

06

Outcome

The tool replaced a manual process the team had avoided. Filing a bug became a click and a quick review, and the team got better at filing and tracking bugs as a result.

  • Easier to file

    A click on the problem, a quick review, and the ticket is filed with everything attached.

  • Easier to track

    Consistent, complete tickets made bugs easier to follow from report to fix.

  • Adopted

    PMs and developers used it in production for filing feedback and bugs.