Skill Manager app icon
New from Vibe Slop · Version 1.0

Skill Manager

Switch skills on and off. Keep context lean.

Your agent loads the name and description of every skill it can find, whether the job needs it or not. Skill Manager switches skills on and off, one at a time or a whole themed group at once, so Claude Code, Codex, Cursor and friends only see the skills that fit the task.

Version 1.0 · Requires macOS 26 or later

Music: “Voxel Revolution” by Kevin MacLeod (incompetech.com), licensed under CC BY 4.0.

Every skill, one switch

See every skill installed across your agents and flip any of them on or off. Smart tiles show what's on, what's off and what needs a look.

Skill Manager's library: a sidebar of coloured smart tiles and groups, a list of 143 skills, and the details of one skill with its findings and install locations.

Switch skills by theme

Group skills by theme, like Unity, SwiftUI or writing, and turn on the group for the task in front of you. With one group at a time, switching one on switches the last one off.

The Groups view with 38 colour-coded groups, each with an on/off switch, and one group open showing its four skills.

Catch problems early

Deterministic checks flag broken metadata on every scan, offline. An optional AI review goes deeper, explains what's wrong and suggests a fix.

The Findings view listing warnings from an AI review, with one finding open showing a suggested change and the evidence behind it.

Every review, remembered

Past reviews are kept on your Mac with counts, failures and suggestions, so you can retry failed requests or look back later.

The review History view: a list of past reviews and one review open with totals for skills, suggestions and failed requests.
Why switch skills off?

Fewer skills on.
Sharper agents.

Agents like Claude Code and Codex don't load whole skills up front. They load every skill's name and description, then read the full skill once one fits the task. That list is always there, and the longer it gets, the more it costs and the harder the right pick becomes.

Everything on143 skills

A Unity skill, a Django reviewer and a slide-deck helper all compete for attention while you're fixing a SwiftUI layout.

Unity 2D group on6 skills
2d-pixel-perfect manage-sprite-atlas sprite-editor sprite-segment-3x3grid tilemap-palette-create tilemap-ruletile-createempty

Only the skills for today's job are on. The agent's choice is obvious, and there's more room for your code and conversation.

Leaner contextEvery skill that's on keeps its name and description in context all session. Skills you switch off aren't listed at all.
Nothing cut to fitAgents cap the skill list. Past the cap, they trim descriptions or leave skills out, so the one you need can get missed.
Fewer wrong picksAgents choose skills by their descriptions. With only relevant ones on, a Unity skill can't win when you're writing SwiftUI.
No clashing skillsTwo skills that cover the same ground can give opposite advice, like rival commit-message styles. Keep one on and the agent gets one clear answer.
Plays well with

Your whole agent crew.

Claude Code Codex Cursor GitHub Copilot Antigravity OpenCode Cline Junie Kilo Code Devin

Plus shared skills in ~/.agents/skills, your own project folders, and read-only Codex and Claude plugin catalogs.

What's inside

Small app.
Lots of tricks.

Everything you need to keep dozens (or hundreds) of skills under control, and nothing you don't.

Switch any skill

Turn one copy of a skill, every copy, or a whole selection on or off. Off moves it out of your agents' folders; on puts it back.

Groups by theme

Bundle skills by project or task and switch the whole group at once. Drag and drop to regroup, or let AI suggest a starting set.

Menu bar switching

Flip groups and favourite skills from the menu bar, with a one-group-at-a-time mode for quick swaps.

Preview first

See every folder and link a change will touch before a single file moves.

Compare versions

Put two SKILL.md files side by side with line numbers and a changes-only view.

Deterministic checks

Rule-based checks run on every scan, fully offline. The same skill always gets the same result.

Expanded AI review

An opt-in, deeper read of each skill, run through Claude Code or Codex on your existing subscription, or with your own API key.

Properly native

Built with SwiftUI for macOS 26. Follows Light and Dark mode, and respects Reduce Motion.

How switching works

Flip it off. Flip it back. Nothing deleted.

Agents pick up whatever sits in their skill folders. Skill Manager moves skills in and out of those folders carefully, so you can change your setup without fear.

  1. Pick

    Choose one copy of a skill, every copy of it, or a whole group.

  2. Preview

    See exactly which folders and links will move, and where they'll go.

  3. Switch

    Off parks the skill in a safe folder on the same volume. On moves it back. Each batch is journaled first, and symlinks stay intact.

Findings

Two ways to spot trouble.

Deterministic checks catch broken skills on every scan. AI review goes further, reading each skill the way an agent would.

Local checks

Deterministic, always on

Every scan runs the same rule-based checks on each skill's format and YAML front matter, right on your Mac. Same skill, same result, every time. No account, no network, no waiting.

  • Missing or malformed metadata
  • Descriptions that break the Agent Skills spec
  • Different versions of the same skill
AI review

Expanded, on request

Goes beyond the rules. A model reads each skill against twelve checks, quotes the exact lines behind every finding and suggests a fix.

  • Vague triggers and overbroad descriptions
  • Contradictory or repeated instructions
  • Unsafe actions and needless context loading

Runs on the Claude Code or Codex you already use, with your existing subscription, or with an Anthropic or OpenAI API key. You pick the model and can preview the request first.

Claude CodeCodexAnthropic APIOpenAI API
Questions

Asked, answered.

Why would I switch skills off?

Every skill that's on costs context, and it's one more thing the agent might pick by mistake, or one more set of instructions that might clash with another skill's. Switching off the skills a task doesn't need keeps context lean and the right choice obvious.

Groups make it quick: switch on the set for today's work from the menu bar, and the rest stay off.

Don't agents only load a skill when they use it?

Partly. Claude Code, Codex and other agents that follow the Agent Skills standard load skills in stages: the name and description of every skill up front, the full SKILL.md once a skill fits the task, and bundled files only when they're needed.

So an unused skill costs little, but not nothing. Its name and description stay in context all session, one entry for every skill you have on. Agents also cap that list, so with lots of skills they start trimming descriptions or leaving skills out.

Does Skill Manager delete my skills?

No. Turning a skill off moves its folder (or link) to ~/Library/Application Support/SkillManager/Disabled, or to a storage folder you pick on the same volume. Turning it on moves it back to exactly where it was, and it never replaces a file that's already there.

Does it send my skills anywhere?

Only if you ask it to. Scanning and the deterministic checks run entirely on your Mac.

AI review is opt-in: you pick the reviewer and model, you can preview the request first, and background review stays off until you turn it on. It runs through the Claude Code or Codex you already have, on your existing subscription, or through your own Anthropic or OpenAI API key, which is kept in your Mac's Keychain.

Crash reports are opt-in too. Skill Manager asks once, and if you say yes it sends a report to Sentry after a crash or freeze. Reports never include your skills, your name is removed from file paths, and nothing tracks how you use the app. Change it anytime in Settings › Privacy.

Which agents does it support?

GitHub Copilot, Claude Code, Cursor, Codex, Antigravity, OpenCode, Cline, Junie, Kilo Code and Devin, plus shared skills in ~/.agents/skills. You can add your own search folders and project locations in Settings.

Will my agent notice the change straight away?

Switching a skill off moves its folder out of your agents' skills folder, and switching it on moves it back. Claude Code and Codex notice the change in a running session, though Codex may need a restart if a change doesn't show up. Other agents may need a new session.

A skill your agent has already used in a conversation stays in that conversation until you start a new one.

What do I need to run it?

A Mac running macOS 26 or later. AI reviews need Claude Code or Codex signed in, or an Anthropic or OpenAI API key. Everything else works without one.

How do I install and update it?

Download the disk image, open it and drag Skill Manager into Applications. It's signed and notarized by Apple.

It keeps itself up to date: choose Check for Updates… in the Skill Manager menu, or turn on automatic updates in Settings › Updates.

Only the skills you need, switched on.

Version 1.0 is out now for macOS 26 and later.