Matthieu Napoli
banner
mnapoli.bsky.social
Matthieu Napoli
@mnapoli.bsky.social
Building
▸ bref.sh
https://bref.sh/cloud
https://github.com/getlift
https://port7777.com
AWS serverless hero
Most of these are Claude slash-commands/skills. Except 1: I built a dedicated app for designing and planning with Claude.
I should find time to post a whole blog post about this.
February 2, 2026 at 4:58 PM
6. `/address-pr-review` -> CC will review each open comment and either fix or post a reply saying why it's stupid
7. goto 4 until ready to merge
February 2, 2026 at 4:58 PM
I'm exploring very high level testing. Natural language. Like Cucumber/Behat, but less rigid, because we don't have to be this rigid anymore.
February 2, 2026 at 12:57 PM
The problem I see: AI writes most of the implementation, and most of the tests. Who reviews these thousands of lines of tests, and is 100% confident they test correctly? We need tests written by humans. And we need to make this scalable with how much faster we write code today.
February 2, 2026 at 12:57 PM
- it's doing refactoring with all bells and whistles -> I tried to rename a "handle" method into "execute" for example, and it renamed every single time the word "handle" appeared in markdown files or comments… That completely defeats the purpose
January 28, 2026 at 11:23 PM
btw I also tested PhpStorm's built-in MCP server. It has a "rename" tool for example.

It works, but:
- sooo many tools (too much context, though I could live with it)
- (cont.)
January 28, 2026 at 11:23 PM
FYI I also sent a PR to Laravel, it might be native at some point
January 27, 2026 at 10:42 PM
I've been thinking about it for a while: made no sales in 2025, might as well do something with it

If you're interested, here's the entire sales history. I made a great amount of money (mostly in 2020) and I'm really grateful for that. In total I may have spent 3 months of work on it since 2020.
January 19, 2026 at 1:02 PM
Yes.
Though here almost every single difference was slightly in favor of Claude Code. That's maybe 20 or 30 differences. If this was _that_ random it would be surprising.
January 12, 2026 at 2:00 PM
Yes
January 12, 2026 at 2:00 PM
One thing I noticed was weird in OpenCode is that it kept failing to add imports, thinking it had done so and facing test or lint errors. It feels like there's a lot of room for improvement here.
January 12, 2026 at 1:00 PM
OpenCode wins on the UX though, it's much nicer. But using Claude Code on the web, GitHub code reviews, integrating with Linear, etc is something missing from OpenCode.

Btw migrating from Claude Code to OpenCode took me 3h, but I guess that depends on how advanced your setup is.
January 12, 2026 at 1:00 PM
Some things could be improved both ways by adding rules, not a dealbreaker, but Claude needed less corrections. Btw both had the same rules, tools, etc.
January 12, 2026 at 1:00 PM
haha yes definitely agree!
January 12, 2026 at 10:51 AM
What does harness mean?
January 12, 2026 at 9:08 AM
Justement je trouve que ce sujet est une bonne illustration de comment « mieux discuter des (bonnes) pratiques »
January 1, 2026 at 6:33 PM