Example prompts and workflows

You don't need special syntax. Name the project, say what you want and what the assistant may change. These examples use a project called Acme.

Find and summarize

What came in on the Acme project today?

Summarize this week's feedback on the Acme site and group it by page. Flag anything that's a change request, not a bug.

Show me open high-priority bugs on Acme that nobody is assigned to.

Find reports on Acme that mention "checkout" from the last 30 days.

Which Acme reports are assigned to me and still in progress?

Which open reports on Acme haven't been touched in the last 30 days?

Catch up on a report

Comments hold the history: what the client added later, what your team already tried, and earlier triage notes.

Summarize the discussion on https://app.ybug.io/reports/detail/hkz92hg6sf. What did the client add after the first report, and is anything still waiting on us?

Read the comments on the checkout reports from this week and tell me which ones we already answered.

Before a release or a client call

What's still open on Acme? Group it by priority and flag anything that blocks checkout or sign-up.

Write a short update for the Acme client: what we resolved in the last two weeks and what's still in progress. Plain language, no internal notes.

The assistant reads internal comments too, so if the update goes to a client, ask it to leave internal notes out, and read it before you send it.

Combine with your other tools

If your assistant is connected to other tools too, like GitHub, Linear or your docs, it can move between them in one conversation.

Read the coupon report on Acme and draft a GitHub issue for it with steps to reproduce, the browser and the page. Show me the draft before you create it.

Check whether any open Acme reports are about the pricing change we shipped on Monday, and link each one to the matching pull request in an internal note.

Ybug only provides the Ybug side. What the assistant can do in the other tool depends on that tool's own connector.

Investigate a bug with a coding agent

In a coding agent like Claude Code, Codex, Cursor or VS Code, the report becomes the agent's brief. It reads the page URL, browser, device, screenshot, console errors and failed requests from Ybug, then looks through your repo with its own tools.

Read the checkout report from yesterday on Acme and find the likely cause in this repo.

Investigate the coupon report on the Acme project and propose a fix in this repo.

You can also paste a report link from the dashboard:

Look at https://app.ybug.io/reports/detail/hkz92hg6sf and tell me what's wrong.

The assistant gets the report's text, page, browser, device and screenshot link. You don't need to ask for the logs: when a report has console errors or failed requests, the agent reads them as part of the investigation. Some details, like attachments, are only in the dashboard. See What stays in the dashboard.

Ybug doesn't change your code or open pull requests. The coding agent does that with its own tools, as it would for any other task.

Triage the new pile

Review the reports on Acme since Friday. Suggest priorities and likely duplicates, then apply the changes I approve with an internal note on each.

Tag every Acme report about the pricing page with "pricing". Create the tag if it doesn't exist.

Set a type on this week's Acme reports that don't have one yet. Show me your picks before you save them.

Assign the open checkout bugs on Acme to Jana, and take the pricing ones myself.

Ask the assistant to suggest changes first and apply them after you say yes. Most clients also ask you to confirm before a tool makes a change.

From feedback to fix

You decide what gets fixed, and you review what's done. The agent does the work in between.

1. A client reports a bug with the widget. It lands in Ybug with a screenshot, the page and the browser.
2. You pick the reports worth fixing and move them to Open.
3. Your coding agent works through what's Open: it sets each report to In progress, fixes it in your repo and leaves an internal note on what changed.
4. You review the fix, then move the report to Resolved.

The prompt for step 3:

Work through the Open reports on the Acme project. Set each one to In progress, fix it here and leave an internal note on what changed. Don't resolve anything until I've reviewed it.

Moving a report to Open is your instruction to the agent, not a lock. Ybug doesn't limit which reports the agent can touch, so say it in the prompt.

Auto-replies go out on Resolved and Closed

If you've enabled the corresponding auto-reply, moving a report into the Resolved or Closed category can email the reporter, even when the status has a custom name. That's true whether you or the agent makes the move, so keep that step for yourself, as in the prompt above.

Worked examples

These show what a good request looks like from start to finish. The reports and numbers are made up.

Weekly feedback digest

Review all Acme feedback created this week, including reports without a type. Read full descriptions before classifying them. Group recurring problems by page, include report numbers, and say how many reports you reviewed. Suggest changes, but don't save them.

What the assistant does: it lists the week's reports page by page, opens the ones with shortened descriptions, then groups them.

A good answer looks like this:

I reviewed 23 reports created Sep 14–18 (all 1 page).

/checkout (6 reports): coupon codes rejected in Safari (4, likely the same bug), total not updating after removing an item (2). /pricing (4 reports): two ask about annual billing, which is a question, not a bug. Two say the currency switcher resets. Other (13 reports): no pattern; listed below by number.

Suggested: set the 4 Safari coupon reports to High priority, tag the annual-billing ones "question".

Then you reply with what to apply, for example: "Apply the priorities. Skip the tags."

Duplicate review

Look for likely duplicates among the Open reports on Acme. For each pair, tell me why you think they match. Don't change anything yet.

A good answer gives pairs with reasons you can check:

1. #412 and #418. Same page (/checkout), same browser (Safari 17), both describe the coupon field rejecting a valid code. 2. #405 and #421. Both mention the menu overlapping the logo, but on different pages. Possibly the same CSS bug. Worth a look.

Then approve the ones you agree with. Saving changes needs STARTUP or higher:

Tag #418 with "duplicate" and add an internal note pointing to #412. Leave pair 2 alone.

Ybug has no merge over MCP. A tag plus an internal note keeps the link visible in the dashboard, and you can close the duplicate yourself.

Bug investigation

Investigate the coupon report on the Acme project in this repo. Separate what the report says from what you're guessing. Tell me what evidence is missing and how to verify the fix.

A good answer keeps facts, guesses and next steps apart:

Reported: coupon code "SPRING10" rejected on /checkout, Safari 17 on macOS, desktop, 1440×900. The screenshot shows the error under the coupon field. No console errors, but POST /api/coupons returned 422 invalid_code. Missing: whether it happens in other browsers. No steps to reproduce beyond the description. Likely cause: Safari autofill adds a trailing space, CouponField.tsx sends the raw value, and the API doesn't trim it. Verify: add a test for a code with a trailing space, then try it in Safari with autofill on.

If you're on STARTUP or higher, you can then ask it to leave that summary as an internal note on the report.

Tips

• Name the project. If you have several projects, say which one. The assistant can also list them first.
• Say what it may change. "Suggest" and "apply" lead to very different results.
• Ask for internal notes. A short comment on each change makes the assistant's work easy to review in the dashboard.
• Use your status names. The assistant reads your project's statuses, so "move it to QA" works if you have a QA status.
• Big lists take a few pages. The assistant fetches up to 100 reports at a time. For large projects, narrow it down with a status, a date or a keyword.
• Browser and device questions are slower. Lists don't include the browser or device, so "which reports came from Safari" makes the assistant open reports one by one. Narrow by date or status first.
• Paste a link to point at one report. The assistant finds a report by the id at the end of its link. A number like #412 works for reports it has already listed in the conversation, but it can't look a report up by number alone.