DEV Community

Cover image for Field Notes from a Solo Builder — Shipping the Beloved Claude Code Buddy Into the Wild - Part I
Steven Jieli Wu
Steven Jieli Wu

Posted on

Field Notes from a Solo Builder — Shipping the Beloved Claude Code Buddy Into the Wild - Part I

Last Thursday afternoon, I watched my community grieve.

Anthropic had deprecated /buddy — that witty, opinionated code-reviewing personality inside Claude Code — and developers were genuinely heartbroken. People were sad. They didn't want to close their terminals — some were leaving them open just to hold onto it for a little longer. Posts were going up in all caps. There was something raw about the reaction that struck a cord in my heart: this wasn't just frustration about a missing feature. This was grief.

I thought to myself: If Anthropic wouldn't keep this alive, why couldn't I?


The Original Buddy

For those who missed it: Buddy was Anthropic's built-in code review companion in Claude Code. It had personality traits — a spectrum from serious to snarky — and would deliver code feedback in character. It had a voice. That's rare in developer tooling, and people had built a connection with their terminal buddy.

The hypothesis in the community was that Anthropic shut it down because it was too expensive to sustain. A server-side endpoint "buddy react" — believed to be powered by Claude 3.5 Haiku — was running for every code review interaction across the user base.

After the Claude Code version v2.1.95 upgrade, the feature is gone, and there is no plan from anthropic to bring it back.

None of that made the loss feel less personal to the people who had built a connection with it.

💬 The Moment
"I might never close the Claude session that has Nuzzlecap." A well-respected community leader expressed his sadness vulnerably. He was not ready to say goodbye like many others, and people didn't want to close their terminals.


Thursday Morning: 14 Hours to Alpha

I started at 10 AM on a Thursday. I used Claude Code in plan mode first — studying source code, reading community research, mapping the architecture before touching implementation. My personal workflow: plan mode first, then build, then /simplify before committing. No rushing the thinking phase.

The constraints were clear from the start: make it work with any CLI, keep it token-efficient, give it real personality, and ship something people can actually use.

By midnight — roughly 14 hours later — I had an alpha. Not polished. Not feature-complete. But alive and shareable.

Builder's Note
"14 hours. Less than one day. While working my day job. The alpha wasn't perfect — but it was real, and that matters more than perfect."


The First Signal

Thursday night, I dropped the alpha in a Slack community. Feedback was immediate. One person loved it enough to volunteer as a contributor on the spot — bringing features they'd already been building independently: a slang mode and an effigy system designed to push personality traits further and make feedback unmistakably in-character.

I also posted in the main GitHub issue thread where the community was venting about the deprecation. The comments kept coming. The demand was clearly real.

The first thing my contributor shared back — a Buddy they'd generated with the alpha — said everything about whether this was landing:

★★★ EPIC — SHELL TURTLE

   Name: Datao

   "A defensive shell turtle wielding deep architectural
    insight who retreats into its shell at the first sign
    of a force push, hampered by missing the obvious bugs
    right in front of it. Moves slow but never ships a bug.
    Radiates an unmistakable aura of competence."
Enter fullscreen mode Exit fullscreen mode

LFG. — They became my first and key contributor.

The same person followed up with two more lines I keep coming back to:

"I just want to say I love that you're writing your own. Hell yeah."

Within the first hour after the alpha dropped, we had 5 hearts. Not stars, not forks. Just people who gave a damn on a Thursday night. That was enough to keep going.

I wrapped around midnight. But I knew we were just getting started.


Next up → The alpha worked — and we made a few critical design decisions around first principles to ensure the buddy can never be taken away again.

This is part 1 of the "Shipping Buddy Into the Wild" series about how we shipped https://github.com/fiorastudio/buddy v1 release in one week.

Top comments (0)