Why we went Claude-only (v0.5.0)
When we first launched prjct/cli, the idea was to work with every AI editor — @Claude Code, @Cursor, @Windsurf (my fav), @Codex by OpenAI. Sounded great, right? But in practice it was a mess:
Tons of extra code just for compatibility
Features blocked because we had to play lowest-common-denominator
Testing was impossible to trust
So in v0.5.0 we pivoted to @Claude by Anthropic.
That decision instantly unlocked faster dev cycles, cleaner code, and some features that weren’t possible before (like dynamic agents + native MCP).
Instead of pretending we could “support everything,” we chose to be honest: prjct/cli works best when it’s laser-focused on @Claude Code and @Claude for Desktop.
If you want the full story (and numbers), I wrote a proper post-mortem in the changelog.
But here’s the TL;DR: 👉 Less code, faster shipping, better quality.
Curious to hear — would you rather a tool try to work everywhere or do one thing really well?
Changelog: https://www.prjct.app/changelog
Replies
Maybe the hidden gem of it all? In my experience, model switching in a project yields “2 steps back, 1 forward” progress. Aka, none.
I absolutely think this is the best way forwards - accommodating and stretching to fit with all agents isn't necessary when all it brings is complications on both sides!