Deep Singh

Building Muster: one place for the skills your coding agents use.

[Deep Singh @SinghRiat]

Most AI agent failures happen before any code: you hand it a vague idea and it confidently builds the wrong thing.

Matt Pocock's `/grill-me` skill interrogates your half-formed idea until the vagueness is gone.

https://t.co/Ilxktls6kp by @mattpocockuk

[Deep Singh @SinghRiat]

@RealNipuna Thanks for your advice! I'm trying to grow my profile too 🙏

[Deep Singh @SinghRiat]

Congratulations to @conductor_build and team! Conductor has been my daily driver. Excited for what comes next!

[Deep Singh @SinghRiat]

First Arc, and now Conductor. What's next? 😢

[Deep Singh @SinghRiat]

@Samaytwt Bro, boss is out here spying on his emails 💀

[Deep Singh @SinghRiat]

You installed 40 skills. Which ones actually work?

Muster lets you test them in a playground. Click Run, and your Claude Code creates a small example in an empty folder.

Muster serves it. Click Open.

Your project stays untouched. Nothing goes to Muster.

[ClaudeDevs @ClaudeDevs]

Users can now auto-connect plugins when they log in to Claude, with MCP servers that accept Claude as a trusted issuer.

If you’re building plugins for Claude, add support so users get started faster and stay connected without re-authorizing.

[Deep Singh @SinghRiat]

I stole this from @steipete’s “Just Talk To It”: spend about 20% of your time on refactoring, and let the agents do it.

So every few days I run /tidy in Claude Code: knip for dead code, jscpd for copies, and splits for big files.

https://t.co/hKJFxXXeHu

[Deep Singh @SinghRiat]

@zehf 🤯🤯

[Deep Singh @SinghRiat]

@mattpocockuk Awesome! Love your skills

[Deep Singh @SinghRiat]

“When a check goes green, ask what it proved. Then what it didn’t.”

The best line here. Agents write tests as fast as code, but tests can share the same blind spots as the code!

We now make agents explicitly call out what their tests might have missed

https://t.co/DvYQBu9S70

[Addy Osmani @addyosmani]

https://t.co/2gyp2Erpxx

[Deep Singh @SinghRiat]

@emilkowalski Ooo, this is cool!

[Deep Singh @SinghRiat]

@BryanJCollins How did you find the most used ones? 👀

[Deep Singh @SinghRiat]

@enzo66lin Nice clean measurement, 27k vs 11k makes the cost very concrete! For a per-project allowlist, is the plan to keep global skills minimal and move the rest into each repo's .claude/skills? 🤔

[Deep Singh @SinghRiat]

@usual_member @lydia_shann 🔥🔥

[Deep Singh @SinghRiat]

Yesterday we shipped 14 releases of Muster across 17 PRs.

Agents write code in parallel, fresh agents review every diff, and each PR brings its own changelog entry. Merging to main triggers a release.

One command turns the changelog into this deck, which we review together.

[Deep Singh @SinghRiat]

@davekiss Good find. Worth checking 'never used' against how rare the task is, though. A release or rollback skill can sit idle for weeks and still be worth its line in the list 👀

[Deep Singh @SinghRiat]

@addyosmani I 100 agree - it’s so powerful! However, the issue I’m seeing now is that we are installing heaps of skills and it’s becoming very cumbersome to manage. I’ve been working on a project to solve that (check out https://t.co/xuVX3MoJ8E) - let me know what you think 😁

[Deep Singh @SinghRiat]

@trq212 Awesome - thank you for sharing! At one point I had hundreds of skills to manage locally, some duplicate and some out of date. I built https://t.co/xuVX3MoJ8E - let me know what you think 😁