Model or dataset
xingkongliang/skills-manager avatar
xingkongliang/skills-manager

Skills Manager keeps one skill library for 50+ agent tools

A lightweight desktop app to manage, sync, and organize AI agent skills across 50+ coding tools — Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.

7,526 stars615 forksRustMIT

At a glance

What is it?
A Tauri desktop app written in Rust that collects AI agent skills into one repo at ~/.skills-manager, then hands them to Claude Code, Codex, Cursor, Copilot, Gemini CLI and the rest by symlink or copy. The parts worth knowing before you trust it are the ones where it deliberately does nothing on its own.
Who is it for?
This fits people who run Claude Code, Codex, Cursor, Copilot or Gemini CLI side by side and want one place that knows where every skill came from and which agents hold it. It does not fit a single-tool setup, and it does not fit you if a preset applied yesterday has to reflect today's edits without another action.
Can I use it commercially?
Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository last received commits 1 day ago.
What is it written in?
Mainly Rust, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • xingkongliang/skills-manager7,526 ★ · +2,863
Star history
Repository2026-09-112026-11-15
xingkongliang/skills-manager4.7k7.5k

2,863 Stars in the last 65 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

One repo at ~/.skills-manager, four ways in

Every skill ends up in the same place. Sources come from Git repositories, local folders, `.zip` and `.skill` archives, or the skills.sh marketplace, and all of them land in one central repo that defaults to `~/.skills-manager` and is repointed in Settings. The 50+ supported tools no longer need hand-made copies, so when an agent reports a missing skill there is one directory to inspect and one list to filter.

That central copy carries bookkeeping. Each entry remembers where it came from, Git-based entries can be checked against upstream for new versions, and anything imported as a plain folder or an archive has no upstream to poll, which makes refreshing it a manual re-import. Inside the app you can open `SKILL.md` or `README.md` for an entry, read its source metadata, and set local content beside the upstream version before deciding to take the update.

Tags do the grouping instead of folders. You label entries, filter the library by source or by tag, and an Untagged pill collects the rows nobody has labelled yet, which is the quickest way to find the parts of your library the app will not group for you.

Homebrew cask on a Mac, and what the other installers carry

On macOS the app is a Homebrew cask, so the install is one line:

bash
brew install --cask skills-manager

A `.dmg` from the latest release does the same job if you would rather Homebrew not own it. Windows users pick `.exe` or `.msi`. Linux users pick `.AppImage`, `.deb`, or `.rpm`, built for x64 and arm64. Every one of those installers ships the CLI inside the app, so the command line entry point arrives with the window instead of as a separate install. From a checkout, the same Rust CLI is reached through three npm scripts that go by way of `scripts/run-rust-cli.mjs`:

json
"cli": "node scripts/run-rust-cli.mjs cli",
"cli:build": "node scripts/run-rust-cli.mjs build",
"cli:install": "node scripts/run-rust-cli.mjs install"

The manifest carries version 1.43.2, the same number as the 2026-10-31 release tag, and it is marked private, so the npm tree describes a build rather than a package you install from a registry. The shell is Tauri with the Rust half in `src-tauri/` and the interface in `src/`, React 19 with Vite, Tailwind, and i18next behind the language setting, next to a `skills/` directory and a `docs/` folder in the repo itself.

The first run is a fixed sequence. Install skills from a local folder, a Git repository, an archive, or the marketplace. Open Global Workspace from the sidebar and pick an agent, Claude Code for instance. Click a preset pill to activate that preset's skills for that agent, or use + Add Skills to pick from the library and toggle target agents inline; an active preset shows a check mark, a partial install shows a count badge. Project-local skills go through a Project Workspace with the same pills, or the same picker with its multi-agent target selector. Paths, custom tools, theme, language, proxy, and Git preferences all sit in Settings.

Presets copy once and do not follow later edits

Presets are named groups of skills. Clicking a preset pill in any workspace activates or deactivates all of its skills for the current agent scope, the sidebar keeps every preset within reach, and the All Agents overview manages every installed agent at once.

What a preset is not matters more than what it does. Applying one is a one-time copy, not a live sync. Edit a skill in the central library after the preset has gone out and the copy the agent is using keeps the older content, with nothing rewriting it behind your back. You re-apply the preset, or push that one skill from its card, to move the change across. Plan on a preset as a batch you run, not as a channel that stays connected.

The interface makes partial states visible rather than silent. An active preset shows a check mark, and a preset that reached only some of its skills shows a count badge, so a batch that stopped halfway reads as unfinished work instead of a finished one. Linked Workspaces are the exception that proves the rule: pointing at any directory as a skills root gives you a workspace managed standalone, deliberately outside global preset sync, which is what you want for skills that live somewhere unusual and for skills you would rather activate by hand.

Symlink or copy per agent, and the skills the app did not install

Sync to any supported tool is a single click and it works two ways, symlink or copy. Every skill card carries an agent icon badge per enabled agent, the badge reflects live sync state, and clicking it installs or removes that skill for that agent without leaving the card. Batch operations sit on the same model: multi-select for bulk enable, disable, export, or delete, and Project Workspaces add bulk enable and disable for project-local skills.

Global Workspace is the honest view rather than the tidy one. Each agent page lists every skill in that agent's global folder, including ones that arrived outside Skills Manager, because the point is to show what the agent actually sees. The price of that honesty shows up in the details. A skill the app never installed has no entry in the central library, so no source, no tags, and no preset sit behind it, and adding it yourself is what turns a stray file into something you can manage.

Project Workspaces carry the same model down into a repository. They show project-local skill folders for supported agents, compare them against the central library, sync changes in either direction, and handle nested skill directories plus per-agent assignment when exporting. The + Add Skills sheet is the shared entry point for both levels: search the central library, toggle target agents on always-visible chips with select-all and clear, and batch-add in one click.

Skill-aware merges and conflicts that wait for your decision

Backup and multi-device sync run through Git. You connect a private GitHub repository with one sign-in, or any other Git remote, and the app backs the library up automatically and keeps every connected device on the same set. Snapshot versions stay restorable at any time.

The merge knows what a skill is rather than treating the library as an opaque folder. A rename on one machine combines cleanly with an edit on another instead of colliding as two competing files. A true conflict does not block anything: your local version stays where it is until you pick keep mine, use remote, or keep both.

Install, remove, update, and sync operations are recorded locally. Settings has an Export Logs entry that bundles recent logs and the activity history into a single zip, which is the file to attach when a sync behaves differently than you expected. The activity history is what tells you which operation ran and when, so read it before you change sync mode.

What it will not do for you

Start with updates, because that is where people get surprised. In-app updates cover macOS and Windows. Nothing downloads or installs on its own: checking only notifies you, and installing and restarting each take a click. A Linux install, which arrives as an `.AppImage`, `.deb`, or `.rpm` for x64 and arm64, has no in-app updater in that list, so new versions land when you download the installer yourself.

Update tracking is narrower than the feature list suggests. It checks upstream for Git-based skills. A skill imported from a local folder or from a `.zip` or `.skill` archive has no remote to compare against, so keeping it current means re-importing it.

The last limit is about who owns the folder. The design has agents install, deploy, and report skills by driving Skills Manager rather than writing into an agent directory behind it, which is what keeps sources, presets, update tracking, and per-agent state intact, and the Dashboard sets that up in one click. Skip that step and an agent keeps writing into its own folder, where Skills Manager can only list the file on the agent's Global Workspace page, with no library entry behind it. Batch operations and the Add from Library sheet work on what the central library knows, so anything installed outside it stays outside your presets, your tags, and your batch exports.

Editorial conclusion

This fits people who run Claude Code, Codex, Cursor, Copilot or Gemini CLI side by side and want one place that knows where every skill came from and which agents hold it. It does not fit a single-tool setup, and it does not fit you if a preset applied yesterday has to reflect today's edits without another action. Before moving an existing library in, check the repo path in Settings, confirm each agent can reach that path, and export the logs after the first sync.

Frequently asked questions

What should Skills Manager have in place before an agent can use a skill?

A source and a home. Anything installed from a Git repo, a local folder, a `.zip` or `.skill` archive, or the skills.sh marketplace lands in the central repo at `~/.skills-manager` by default, and you pick target agents from the + Add Skills sheet or from a preset pill.

What does Skills Manager keep for every skill it tracks?

The source and its metadata, your tags, whether an upstream update exists for Git-based skills, and the per-agent install state shown as icon badges on the card. For any entry you can also open `SKILL.md` or `README.md` and set local content beside the upstream version.

How does Skills Manager apply one group of skills to several agents at once?

Group skills into a named preset, then click the preset pill in the workspace, which activates or deactivates all of its skills for the current agent scope. Applying a preset is a one-time copy rather than a live sync, and the All Agents overview handles every installed agent in one place.

Official sources

  1. License: MIT
  2. Project website
  3. README
  4. Releases
  5. xingkongliang/skills-manager on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/xingkongliang-skills-manager.svg)](https://hysenlabs.com/projects/xingkongliang-skills-manager)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/xingkongliang-skills-manager.svg)](https://hysenlabs.com/projects/xingkongliang-skills-manager)

Was this useful?