Real Agent Cases

QA-test a department chatbot and file Jira bugs with Claude Cowork and Chrome

Posted
by @Kole_Sam
Last checked
Method:
How we verify

Evidence: Screenshots or logsThe source shows screenshots, a transcript or the agent's output.

Outcome

Using Claude Cowork with the Chrome extension, the poster had it navigate a new chatbot like a user and produce a report of what worked, what needed improvement, bugs, repro steps and fixes; after the poster reproduced the bugs, Claude created Jira tickets assigned to the right teammates.

Agent

Claude Cowork, Claude in Chrome

Prompt Recreated prompt

The source didn't share its exact prompt, so we wrote this from what it describes.

Before you start, ask me 1–3 short questions about the details below and wait for my answers. Once I've answered, confirm the plan in one line and run the whole task start to finish.

Goal: Test a web app like a real user and turn what breaks into clear bug reports.

Quick-check questions: Which app, which flows matter most, and where should bugs go (Jira project)?

What I'll give you:
- The app URL ([URL])
- The main flows or test questions
- My Jira project (tickets only after I approve)

Rules:
- Click through each flow and log what happened at each step.
- For every bug, write numbered repro steps, expected vs actual, and a suggested fix.
- Don't create tickets until I've reproduced the bugs and said go.

Output: A findings report (works well / needs improvement / bugs with repro steps), then Jira tickets once approved.

If my answers leave a gap, make a sensible choice, say what you assumed, and keep going.

You'll know it worked when: I can reproduce every listed bug by following its steps.

Try this in:

ChatGPT Gemini Grok

This prompt has blanks to fill in, like [URL]. We'll copy it instead of filling it in, so you can edit it before you send.

The source used Claude Cowork and Claude in Chrome, which can take actions in a browser or app. These are general chat tools, so results may differ, and a normal chat may only give you a plan.

Opens in a new tab. We don't see what you send.

These links are only a shortcut. We're not affiliated with OpenAI, Google or xAI, and your results may not match the source.

Optional stronger version

Our rewrite with extra checks. The poster didn't test it.

Before you start, ask me 1–3 short questions about the details below and wait for my answers. Once I've answered, confirm the plan in one line and run the whole task start to finish.

Goal: Test a web app like a real user and turn what breaks into clear, reproducible Jira bug reports, filed only after my approval.

Questions to ask me first:
- Which app and which flows matter most?
- Which Jira project and issue type should bugs go to?
- Is there an answer key or expected behaviour to test against?

Inputs (fill in or confirm):
- App URL ([URL]) and test account
- Main flows or test questions
- Jira project key and issue type
- Expected behaviour or answer key, if any

Rules:
- Run each flow end to end and record pass/fail with the exact input you used.
- For each failure capture steps to reproduce, expected vs actual result, a screenshot and browser details.
- Group duplicates into one report with all affected inputs.
- Rate severity (blocker, major, minor) with a one-line reason.
- Never change app settings or data beyond normal test use.

Checkpoint: Show me the draft bug list; create Jira tickets only for the ones I approve.

If you get stuck: If the app needs SSO, a CAPTCHA or a two-step code, stop and ask me to sign in; never guess or store passwords.

Verify before you finish: Re-run each failing step once to confirm it's reproducible before reporting it.

Output: A test log (flow, input, result) and draft tickets; after approval, ticket links.

Done when: every approved bug is in Jira with steps anyone can reproduce.

Source

https://x.com/Kole_Sam/status/2062964099974369751

Related cases

  • Build a five-question Google Form with an AI browser agent: Gemini handed it back, Claude finished it (partial)Recreated promptEvidence: Described by posterPartial

    Irena Palamani Xhurxhi asked Gemini to build a Google Form. It got partway, then stopped and asked Xhurxhi to take over because it couldn't change the question type in the interface. Xhurxhi gave the same task to Claude in the browser and went back to other work; when they returned, the form was done, with all five questions, correct types, descriptions and required fields. Asked about it, Claude said it had struggled with the same parts, trying clicks, JavaScript workarounds and keyboard navigation, but kept going rather than handing it back.

    Gemini, Claude in Chrome · Ops · Source date:

  • Submit a hosting provider bug report from log warnings with Claude in ChromeRecreated promptEvidence: Screenshots or logs

    Using the poster's logged-in browser, Claude in Chrome submitted a bug report to DigitalOcean after finding warnings in the logs.

    Claude in Chrome · Ops · Source date:

  • Prepare and file overdue German VAT returns with Claude Cowork and Claude in ChromeRecreated promptEvidence: Shown on video

    From a folder of invoice and receipt PDFs, Claude Cowork prepared quarterly accounts; after the poster logged in to the Elster tax portal, Claude in Chrome found, filled and submitted the Q3 2025 VAT forms.

    Claude Cowork, Claude in Chrome · Ops · Source date: