What else Mushi Mushi does
The main job is simple: a user says your app is broken, and Mushi Mushi tells you why and hands your AI editor the fix. But most bugs never get reported. People just leave.
On my own apps that is exactly what happened. In 60 days people reported 32 bugs. More than a thousand came from my own side: checks like the ones below, AI code audits, Sentry errors and bugs I filed myself. Only one of the fixes that merged started from a user’s report. The live numbers count the two separately.
Each card says the problem it solves, what it does, and what it needs. Most are off until you switch them on, and some need your own key (Firecrawl, Cursor or Anthropic) or a paid plan; the console tells you before you switch one on.
First-time user test
You can’t use your own app like a stranger. You already know where every button is.
An AI that has never seen your app starts at your home page and tries one goal, like signing up. It times every step, notes where it got stuck, and files those stalls as bugs with screenshots.
Needs: A computer of yours running the Mushi Mushi worker, or run it once yourself with mushi ux fresh-eyes.
A second AI checks every fix
The AI’s fix looked right, passed the tests, and broke something else.
Before a fix merges, a second AI reads it next to the original bug report and says merge, fix first or do not merge, with its reasons line by line. When you choose “Merge when ready”, approved fixes land one at a time, each branch updated and its tests passed first.
Needs: Your GitHub repo connected and your own Anthropic key.
Live-site watch
Your site breaks at night and you hear about it from a user, days later.
Checks your live pages every hour. A page that errors, a dead link or a broken layout is checked a second time, and only then filed as a bug, so a one-off blip doesn’t wake you.
Needs: Your own Firecrawl key. Off until you switch it on.
Weekly site audit
Dead links, missing page titles and text in the wrong language quietly cost you visitors.
Once a week it reads every page of your site and files one plain-English report per problem: broken links and images, missing titles and descriptions, pages in the wrong language, images without a text description.
Needs: Your own Firecrawl key. Off until you switch it on.
Known security holes
Your AI added hundreds of packages. You can’t tell which one has a known security hole.
Every day it checks each package version in your lockfiles against the public vulnerability database. A hole with a fixed version becomes a report that says which package to update and how urgent it is.
Needs: Your GitHub repo connected.
Your CI bill, explained
Your GitHub Actions bill went up and you don’t know which workflow did it.
Reads your real GitHub bill, shows which repos and workflows cost the most, and spots waste, such as tests that run when only docs changed or a scheduled job that runs every hour. One click opens a draft pull request with the fix, and after it merges the minutes saved are measured.
Needs: Your GitHub repo connected. Reading the bill needs a GitHub token that can see billing.
Is it my code or the CI machine?
CI failed again, and you can’t tell if it’s your code or GitHub having a bad day.
Watches every CI job: how long it waited for a machine, how long it ran, and whether it failed on your code or on the machine. A machine problem is never blamed on a fix.
Needs: Your GitHub repo connected.
Playbooks that run in rounds
You know the app needs a proper pass, for speed, accessibility or tidier code, but it means a week of prompting.
Pick a playbook, like “make it accessible”, and a part of your app. It runs in rounds: each round becomes a fix pull request, and once that merges the next round starts, until the playbook finds nothing left (five rounds at most). You can also chain playbooks on one bug, for example reproduce, fix, test, and run each step in your editor.
Needs: Your own Cursor key, or a computer running the Mushi Mushi worker.
Bad app store reviews become bugs
A one-star review says the app crashes, and you see it a week later.
Reads your App Store and Google Play reviews every six hours. Low-star reviews become bug reports with a plain-English diagnosis, like any other report.
Needs: Your store accounts connected. Off until you switch it on.
Report a bug from your phone
You notice a bug on your phone and have forgotten it by the time you’re at your desk.
Say it or type it into Telegram, Slack or an iPhone shortcut. It becomes a report, and it can start a fix and send you the draft pull request.
Needs: Telegram, Slack or the shortcut connected from the console.
Where to go next
- Who it’s for: the pains and use cases for solo builders, experienced developers and teams.
- Sample diagnosis: what one bug looks like from report to fix prompt, no sign-up needed.
- Pricing: what each plan includes.