Your bug report button should feed your agents
I think that most of us build the bug report button last, and honestly it shows. In MentorMates it was a link in the support menu that opened a Google Form. Someone would write "the page is broken" and I would have no idea which page, which browser, or what they had clicked before it broke.
When we built the internal dashboard for Stanford Founders, I took it more seriously. The button let people pick bug or feature request, say how bad it was and where it happened, and paste a screenshot straight into the text box. It attached the page address on its own. Reports landed on our task board, and the team got an email. People actually used it, which surprised me more than anything.
Then I started building an outliner, and the first thing I wrote for it, before most of the editor was done, was the same button again. That was the moment I realised I kept rebuilding the same five parts: a button, a form, the context that comes along automatically, somewhere to store it, and somewhere for it to go.
The part that changed
What is different now is who reads the report. More and more of my fixes start with Claude Code, and a good bug report is exactly the thing an agent needs: what happened, where, on which browser, with a screenshot and a snapshot of the app's state. The report is basically a task prompt already. It just never reached the agent.
So shipcue treats the inbox as a queue. Reports sit in your own Postgres, most urgent first. An agent connects over MCP, claims the top report, gets it as a task (reproduce it, write a failing test, fix it), and closes it with the PR link. The claim is atomic, so you can point several agents at the queue and none of them will ever pick up the same report.
Why self-hosted
There are good hosted feedback tools already. I wanted something that lives in the database I already have, that I can read with SQL, and that does not need another login for the team. It is one table, one handler and one React component.
What is next
The first version is built and tested, and the code is now public on GitHub under the MIT license. Next I am moving my own apps onto it, starting with the outliner. If you try it, or you have a queue of reports you wish an agent would just handle, get in touch. I would really love to hear how you end up using it.