The IssueFixer: from report to production
The problem
The Livalos Platform runs our customers' websites. Every bug, small or large, has to be found, described, fixed, tested and rolled out. That takes time, and without help reports pile up. A customer who reports an error and waits for weeks trusts you less.
The approach
We built the IssueFixer, a bot that works with Claude and GitLab.
Report: a customer or a system reports an error. Reports about the same error are grouped together.
Issue: the error becomes an issue in GitLab, in English, describing what goes wrong and where.
Label: an issue with the label `issuefixerbot` is picked up by the bot.
Fix: the bot fetches the code, lets Claude fix the error and creates a proposed change (a merge request).
Review: a reviewer checks the change and tests it on a real test site. If in doubt, it goes back to the bot with feedback.
Merge: an approved change joins the other changes in the development branch.
Release: the changes go out in a release to a test environment, then to a small group of sites and then to all sites.
The bot runs on the platform and on the dashboard, since the weekend of 16 August 2026. A human stays the gatekeeper. A bot change never goes live without review.
The result
| Number | Value |
|---|---|
| Merged changes on the platform in 2026 | 741 |
| Merged changes in September | 160 |
| Merged changes in October | 141 |
| Merged changes on the dashboard in 2026 | 263 |
| Platform releases in 2026 | 25 |
Source: Livalos GitLab. The 741 covers all merged changes on the platform, a large part of which comes from the IssueFixer.
What the customer gets
Errors disappear faster. A bug does not wait for a free hour.
Every change is checked. A human tests before it goes live.
Your platform improves every week. No big jumps once a year, but a stream of small improvements and 25 releases in 2026.
Want something like this in your own organisation? We teach you how to set up a process where an agent does the work and a human guards the gate.