ForClaude CodeandCodex

100 skills.
.
Time to muster them.

Select the highlighted words to see what else goes wrong.

Muster finds every skill you've installed and sorts them into folders, from your terminal, your browser or your menu bar. It flags the ones that clash, repeat or never run, and suggests fixes and improvements.Muster finds every skill you've installed, sorts them into folders and flags the ones that clash, repeat or never run.

Get your invite See it in action
  • Checked every day
  • Runs on your Mac
  • No account
  • Nothing changes without you

You saw a skill on X and installed it. Then a few more. Then a plugin brought its friends. Now there are 98, in no order, and you could name maybe twenty. Some have newer versions now. Nobody told you. All 98 load into every session, even the ones you never use. Two of them tell your agent opposite things. And you wonder why it’s been acting strange lately…

How it works

Muster gets them in line.

One line to install. It finds every skill in Claude Code and Codex, sorts them into folders, checks them every day, and lets you fix a clash in one click.One line to install. Then it finds, sorts, checks and fixes.

  1. 1

    You install it with one line

    Paste one command. Muster is now a command in your terminal, an app in your browser and an icon in your menu bar. No account, nothing to configure. It starts on the skills you already have.Paste one command, and Muster is in your terminal, your browser and your menu bar.

  2. 2

    It finds every skill you have

    Claude Code, Codex, your projects, your plugins: one scan finds every skill, wherever it lives. It only reads. Nothing moves.One scan finds every skill in Claude Code, Codex, your projects and plugins. It only reads.

  3. 3

    It sorts them into folders

    No more one long list. Your skills get folders and subfolders, and every new skill gets a suggested home. The files stay where they are, so your agent still finds them.Folders and subfolders, and a suggested home for every new skill. The files stay put.

  4. 4

    It checks them every day

    It flags what needs a look: skills that contradict each other, skills that do the same job, skills you never use, and descriptions too vague for your agent to pick. It also checks as soon as you add or change a skill, and whenever you ask.It flags clashes, duplicates, skills you never use and descriptions too vague to pick.

  5. 5

    You fix them in one click

    Every flag comes with a ready fix: merge the duplicates, settle the conflicts, switch off what you never use. Nothing changes until you say so.Every flag comes with a ready fix. Nothing changes until you say so.

Installing skills is easy now. Keeping track of what you installed is the hard part.

We built Muster for our own Macs first. This is what the first cleanup did to one of them. Now we’re letting a few people in at a time.

The Muster team

229 to 75

skills on one of our Macs, before and after the first cleanup

114
switched off. All from one plugin, installed for one job.
23
duplicates merged. 8 were animation skills under different names.
17
archived after 30 days unused. Nothing deleted.
9
conflicts fixed in the 75 that stayed

Every skill rides along on every turn. Muster saves you tokens, money and sanity.

Skill text in each turn of one bad session on that Mac, in tokens

Turns in one session. A turn is one call to the model: your message, or a step your agent takes on its own.

Tokens the skills put into each turn of one session
TurnBeforeAfter Muster
110,0007,500
210,0007,500
310,0007,500
415,00010,000
515,00010,000
615,00010,000
715,00010,000
820,00012,500
920,00012,500
1020,00012,500
1120,00012,500
1220,00012,500
  • What stays The list of the 75 kept, 7,500 tokens a turn, and each skill that runs because you asked for it.
  • Skills you don’t need On a 1M context window, Claude Code caps the skill list at about 10,000 tokens. The 154 that went filled it: 2,500 more, every turn.
  • A wrong pick A duplicate loads instead of the skill you meant, and stays for the rest of the session.
  • A clash Two of your skills with opposite rules both load, and your agent follows one. Muster narrows one so only the right one loads.
Try itTake a look

Browser, menu bar or terminal.

The full app in your browser, today’s summary in your menu bar, every command in your terminal. Same skills and counts, so a fix in one shows in all three.

Safari
127.0.0.1:55350/

  • In your browserThe full app, on a page your Mac serves. Folders, usage and every fix.

  • In your menu barToday’s runs and what needs a look, one click away all day.

  • In your terminalEvery command, from the keyboard or a script. Scan, check, fix and sort.

Privacy

Your skills stay on your Mac. The daily check is a normal session of your own Claude Code or Codex. Nothing goes to us.

Your Mac

Your skills~/.claude, ~/.agents, projects
reads
MusterMenu bar, browser, terminal
daily check
Your Claude Code or CodexTalks to Anthropic or OpenAI
  • Reads your skillsAnd your Claude Code and Codex history, to count which skill ran when. All on your Mac.
  • Changes nothing on its ownIt writes only to its own folder. Your skills stay as they are until you apply a fix.
  • Checks through your agentYour Claude Code or Codex runs the daily check on your plan, like any session. Nothing goes to us.
Roadmap

Ready for you.
Now for your team.

The personal version is done: one person, one Mac, free. Now we’re building team plans, and what people ask for when they sign up decides the order.

  1. Ready

    Your skills, on your Mac

    • Find, check and fix, every day
    • Folders, subfolders and usage, in your menu bar
    • Claude Code and Codex on macOS

  2. Building now

    Your team

    • A shared team space next to your private one
    • Suggest a skill, an admin approves, it lands on every teammate’s Mac
    • Connectors like Linear, GitHub and Notion, set up once for everyone

  3. Later

    More agents

    • Cursor, Gemini CLI and other agents
    • Windows and Linux
    • And a few bigger ideas we’re not ready to show yet
Questions

Short answers.
No sales call.

What people ask before they point something at their ~/.claude or ~/.agents folder.

Something we missed?

What does Muster work with?

Claude Code and Codex on macOS. It reads your global skills in ~/.claude/skills and ~/.agents/skills (plus Codex’s older ~/.codex/skills), the skill folders inside your projects, and skills that came with plugins, including ones you installed from skills.sh or GitHub. Other agents come later, then Windows and Linux.

I use both Claude Code and Codex. Does that help?

That’s where the mess grows fastest: you copy a skill over for Codex, then edit only one copy. Muster reads both, spots skills that exist in both places, shows where the copies drifted apart, and brings them back in line.

Will it change or delete my skills?

Not without asking. The scan only reads. Every fix is a suggestion: apply it, copy it and do it yourself, or skip it. Unused skills you wrote get archived, not deleted.

Does anything leave my Mac?

Nothing goes to us. There’s no Muster account and no Muster cloud. The daily check asks your own Claude Code or Codex to read your skills, so their contents go only where that agent already sends them.

Do folders change anything for my agent?

No. Folders and subfolders live in Muster, so you can sort a hundred skills however you like while every file stays exactly where Claude Code or Codex looks for it. Moving a skill between folders never touches the skill itself.

Does it use my Claude or ChatGPT plan?

Yes, if you sign in to Claude Code or Codex with your Claude or ChatGPT account: the check runs through the one you pick and counts toward that plan like any other session. With an API key, it’s billed to that key. By default it runs once a day and when you add or remove a skill; you can switch it to only when you ask.

Will a cleanup make my agent cheaper to run?

A little on every turn, and turns add up. Every skill’s description, about 100 tokens, goes with every turn, used or not, and a skill that loads by mistake keeps its full instructions in the session. With fewer skills, they all stay in view: past a limit, Claude Code drops descriptions, least-used first. On a Claude or ChatGPT plan, your limits last longer. On an API key, the bill is smaller, though most of these tokens are cached and cheap. Run /context in Claude Code before and after a cleanup to see it.

How does it spot conflicts and duplicates?

It reads every skill’s description, which is what your agent matches your request against, and then the full instructions. Two skills only count as a conflict if both would load for the same request and it can quote the lines that contradict; similar but compatible skills are duplicates. The fix depends on who owns the skill: yours get edited or merged, installed ones get a tiebreak line in ~/.claude/CLAUDE.md or ~/.codex/AGENTS.md, and unused ones get archived or switched off.

How do I get it, and how do I remove it?

The install comes with your invite. Uninstalling removes the app, its folder and its daily schedule; Muster stores nothing anywhere else. Your skills stay as they are, apart from fixes you applied.

What if it doesn’t work for me?

It’s free for personal use, so all it costs you is the uninstall. But please first and tell us what went wrong. That’s how it gets fixed for the next person.

What about my team?

That’s next. You’ll be able to suggest a skill from your Mac to a team space; an admin approves it, and Muster puts it on every teammate’s Mac. Connectors like Linear, GitHub and Notion come with it. Personal use stays free; team plans will be paid. If that’s what you need most, pick it when you sign up.

Early access

Muster your skills.

Answer three questions, add your email, and we’ll send your invite. Your answers decide what we build next.

Send us a message

Questions, ideas, a skill setup that breaks things. We read every message.