What is an AI Agent Skills manager?

An AI Agent Skills manager is a desktop application that organizes the SKILL.md-based capability folders used by AI coding agents. Skills Manager keeps every Skill in one library, connects that library to each installed tool, lets you enable or disable a Skill per tool, installs Skills from a built-in marketplace, translates them between English and Chinese, and scans them for dangerous instructions before you turn them on.

What is a SKILL.md file?

SKILL.md is the entry file of an Agent Skill: a Markdown document that opens with a YAML front matter block containing at least a name and a description. AI coding agents index the front matter of every SKILL.md they can see, and load the Markdown body only when a task matches the description. A folder without a SKILL.md file is not treated as a Skill.

Where does Claude Code store its Skills?

Claude Code reads user-level Skills from ~/.claude/skills, where each Skill is a subfolder containing a SKILL.md file. Skills Manager keeps the canonical copy in ~/.skills-manager/skills and places a link inside ~/.claude/skills, so Claude Code sees the Skill without you having to store a second copy there.

Where does Codex store its Skills?

Codex reads Skills from ~/.codex/skills. Because that path differs from the one Claude Code uses, the same Skill normally has to be duplicated into both folders. Skills Manager avoids the duplication by linking one source folder into both locations.

Which AI coding tools does Skills Manager support?

Skills Manager ships with 31 built-in tool definitions, including Claude Code, Codex, Cursor, Gemini CLI, OpenCode, Windsurf, Cline, Antigravity, Vercel Skills, Trae, Qwen Code, Augment, CodeBuddy, Droid, OpenClaw, CommandCode, Continue, Crush, Goose, iFlow, Junie, Kilo Code, Kiro, Qoder, Roo Code, Zencoder, Pi, Trae CN, Hermes, WorkBuddy, and QoderWork CN. Any tool that reads Skills from a configuration directory can also be added manually as a custom tool.

How does Skills Manager sync Skills across tools?

Skills Manager stores one canonical copy of each Skill in ~/.skills-manager/skills, then creates a filesystem link to it inside each tool's own Skills directory. Every connected tool reads the same underlying file, so editing a Skill once changes its behaviour everywhere the toggle is on. Turning a toggle off removes only the link and never touches the source file.

Does Skills Manager need administrator rights on Windows?

No. On Windows, Skills Manager first attempts a directory symlink, which requires Developer Mode or elevation. If that is unavailable it creates a directory junction with mklink /J, which needs no special permission. If the junction is also blocked, it falls back to copying the folder and recording the source path in a .skills-manager-source.json file so the copy remains tracked. A standard, non-elevated Windows account is enough.

Can the same Skill run in Claude Code and Codex at the same time?

Yes. Enabling a Skill for both tools creates a link in ~/.claude/skills and another in ~/.codex/skills, both pointing at the same source folder. Both agents load the identical instructions, and any later edit reaches both at once rather than drifting apart the way two manual copies would.

Is it safe to install Agent Skills from the internet?

Treat an Agent Skill like a script you are about to run, because a Skill can contain shell commands the agent will execute on your behalf. Skills Manager reduces that exposure by scanning the whole Skill folder locally before you enable it and by showing the source repository, the exact file, and the line for anything it flags — but the final decision to trust a Skill stays with you. Note also that the scan is static analysis over the text: it never executes the commands, so it is a first filter rather than a guarantee.

What does the Skills Manager security scan check for?

The Skills Manager security scan runs 36 deterministic rules split evenly across four categories: destructive commands such as rm -rf on a root path or writing directly to a device, network and exfiltration patterns such as piping a download into a shell or opening a reverse shell, privilege and persistence changes such as sudo use, editing shell startup files, or installing a crontab entry, and obfuscated payloads such as base64-decoded execution or zero-width characters. Findings map onto five severity levels and record whether the match appeared in a documentation example, a comment, or a real instruction.

Does Skills Manager upload my Skills anywhere?

No. Skill organization, tool connections, and the static security scan all run locally on your machine. Network access happens only for features you start yourself: browsing the marketplace, downloading a Skill, checking for an app update, or sending content to a language model when you explicitly request an optional LLM review or a translation.

Do I need an account to use Skills Manager?

No account is needed to manage the local Skill library, connect tools, toggle Skills per tool, or run a security scan. An account only applies to Marketplace community features and optional cross-device services.

Is Skills Manager free?

Yes. Skills Manager is free, open source under the MIT License, and has no paid tier, trial period, or usage limit. The full source code is published on GitHub.

Should I use Agent Skills or an MCP server?

Use an Agent Skill when the capability can be written down as instructions the agent should follow — a workflow, a house style, a checklist. Use an MCP server when the agent needs to reach a system it cannot access on its own, such as a database, an internal API, or a ticketing tool. The two are complementary, and Skills Manager manages the first category.

What happens to my Skills if I uninstall Skills Manager?

Your Skills are ordinary folders on disk and survive uninstallation. The source files stay in ~/.skills-manager/skills, and you can move them back into any tool's Skills directory by hand. Disable a Skill before uninstalling if you want the links inside each tool's folder removed cleanly.

Can I add a tool that is not in the built-in list?

Yes. Point Skills Manager at the configuration directory of any tool that reads Skills from a folder, and it is managed alongside the 31 built-in definitions with the same toggles, syncing, and security scanning.

Can Skills Manager translate a Skill between English and Chinese?

Yes. Skills Manager translates a Skill name, description, and Markdown body between English and Simplified Chinese using a language model you configure, and caches the result on disk so the same Skill is not translated twice. The original file is never overwritten — the translated view sits alongside it.

Which macOS build should I download?

Download the Apple Silicon build for any Mac with an M-series chip, and the Intel build for older Intel-based Macs. You can check which one you have from the Apple menu, under About This Mac.

Who maintains Skills Manager?

Skills Manager is built and maintained by jiweiyeah, an independent developer, and has been developed in public on GitHub since February 2026. Development is not backed by a company: the same person writes the Rust linking layer, the static security rules, and the tool definitions, and answers issues on the repository. Several unrelated projects share the name "Skills Manager", so the maintainer handle and the repository URL github.com/jiweiyeah/Skills-Manager are what identify this one.

Where can I download Skills Manager?

Installers for macOS (.dmg), Windows (.exe and .msi), and Linux (.AppImage, .deb, and .rpm) are published on the Skills Manager GitHub Releases page, and the download section of this page links directly to the build for your current platform.

One Skill Library for Every AI Coding Agent

Import once, choose where each Skill runs, and keep Claude Code, Codex, Cursor, and 28 other coding tools in sync. Scan community Skills before you install.

Download free View on GitHub

Free · MIT licensed · Local-first · macOS, Windows, and Linux

Skills Manager connections for Claude Code, Codex, Cursor, and other tools

900+GitHub stars

31AI coding tools

3desktop platforms

MITopen-source license

Compatibility

Use the same Skills where you already work

Skills Manager detects the AI coding tools installed on your machine and connects each one to the same central library, so a Skill you write once is available everywhere you work.

Claude Code

Codex

Cursor

Gemini CLI

OpenCode

Windsurf

Cline

Antigravity

Vercel Skills

Trae

Qwen Code

Augment

and 19 more tools — see the full directory matrix below

Compatibility matrix

Where each of the 31 supported tools reads its Skills

One Skill, one source file in ~/.skills-manager/skills — Skills Manager writes the connection into the directory each tool already scans. Everything stays single-copy and reversible.

Tool	Skills directory	Detected by	Connection
Claude Code	`~/.claude/skills`	`claude`	Symlink / Junction
Codex	`~/.codex/skills`	`codex`	Symlink / Junction
Cursor	`~/.cursor/skills`	`cursor`	Symlink / Junction
Gemini CLI	`~/.gemini/skills`	`gemini`	Symlink / Junction
OpenCode	`~/.config/opencode/skills~/.opencode/skills`	`opencode`	Symlink / Junction
Windsurf	`~/.windsurf/skills`	`windsurf`	Symlink / Junction
Cline	`~/.cline/skills`	`cline`	Symlink / Junction
Antigravity	`~/.antigravity/skills`	`antigravity`	Symlink / Junction
Vercel Skills	`~/.agents/skills~/.vercel/skills~/.vercel-skills/skills`	`vercel`	Symlink / Junction
Trae	`~/.trae/skills`	`trae`	Symlink / Junction
Qwen Code	`~/.qwen/skills`	`qwen`	Symlink / Junction
Augment	`~/.augment/skills`	`augment`	Symlink / Junction
CodeBuddy	`~/.codebuddy/skills`	`codebuddy`	Symlink / Junction
Droid	`~/.factory/skills~/.droid/skills`	`droid`	Symlink / Junction
OpenClaw	`~/.openclaw/skills`	`openclaw`	Symlink / Junction
CommandCode	`~/.commandcode/skills`	`commandcode`	Symlink / Junction
Continue	`~/.continue/skills`	`continue`	Symlink / Junction
Crush	`~/.config/crush/skills~/.crush/skills`	`crush`	Symlink / Junction
Goose	`~/.config/goose/skills~/.goose/skills`	`goose`	Symlink / Junction
iFlow	`~/.iflow/skills`	`iflow`	Copy mode
Junie	`~/.junie/skills`	`junie`	Symlink / Junction
Kilo Code	`~/.kilocode/skills`	`kilo`	Symlink / Junction
Kiro	`~/.kiro/skills`	`kiro`	Symlink / Junction
Qoder	`~/.qoder/skills`	`qoder`	Symlink / Junction
Roo Code	`~/.roo/skills`	`roo`	Symlink / Junction
Zencoder	`~/.zencoder/skills`	`zencoder`	Symlink / Junction
Pi	`~/.pi/agent/skills`	`pi`	Symlink / Junction
Trae CN	`~/.trae-cn/skills`	`trae`	Symlink / Junction
Hermes	`~/.hermes/skills`	`hermes`	Symlink / Junction
WorkBuddy	`~/.workbuddy/skills`	`workbuddy`	Symlink / Junction
QoderWork CN	`~/.qoderworkcn/skills`	`qoderworkcn`	Symlink / Junction

Symlinks on macOS and Linux; on Windows the app falls back from symlink to junction, so no admin rights are needed. Copy mode is used only for iFlow, which does not follow links.

Tool not in the list?

Add a custom tool by pointing Skills Manager at its configuration directory. The same library, toggles, and security scan apply to it.

Workflow

One source of truth. No more copied folders.

Skills Manager keeps your canonical Skills in one library, then creates and maintains the connections each coding agent expects.

Central Skill library~/.skills-manager/skills

Connected tools

Claude Code

Codex

Cursor

Gemini CLI

Import once

Bring existing Skills into one managed library without rebuilding your setup.

Choose each tool

Enable or disable a Skill per tool while keeping the source file untouched.

Stay in sync

Symlink-based connections make edits available across tools without manual copies.

Your entire skill system, visible at a glance

See every Skill, every connected tool, and every marketplace install in one focused workspace.

Skills Manager unified Agent Skills library

Getting started

From download to synced Skills in three steps

Skills Manager does not ask you to restructure anything first. It reads the Skills you already have, then takes over the wiring between them and your tools.

01
Install and let it detect your tools
Download the build for your platform and open it. Skills Manager probes all 31 built-in tool directories and lists the ones it finds.
No account, no CLI setup. Detection is read-only — nothing is written until you enable a Skill.
02
Import your existing Skills into one library
Point the app at the Skills already sitting in folders such as ~/.claude/skills. They move into the central library at ~/.skills-manager/skills.
Duplicates across tools collapse into one entry — you stop maintaining the same Skill in several places.
03
Toggle each Skill per tool
Flip a switch and the link appears inside the tool’s Skills directory. Flip it back and the link is gone — the source file is never modified.
Every tool points at the same file, so one edit lands everywhere at once.

Typical setup timeUnder 5 minutes

Security

Inspect a Skill before you trust it

Run local static checks for dangerous commands, hidden downloads, privilege escalation, and suspicious instructions before enabling a Skill.

Local static scan

Deterministic rules run on your machine and keep source files private.

Optional LLM review

Add a second opinion for ambiguous or context-dependent findings.

Actionable findings

Review the exact file, severity, rule, and remediation before installation.

security-audit/SKILL.md

Destructive commandsClear

Network exfiltrationClear

Privilege escalationClear

Instruction overrides1 review

4 rulesets / local scan / 182 ms

Security scanning

36 deterministic rules across four risk categories

A Skill can contain commands the agent will execute, so installing one from the internet is like running a script you did not read. Skills Manager scans the entire folder locally before you enable it — deterministic rules, same input, same verdict.

destructive.*9 rules

Destructive commands

Commands that delete, overwrite, or destroy data on the machine running the Skill.

destructive.rm_rf_root
destructive.mkfs
destructive.dd_to_device
destructive.fork_bomb

network.*9 rules

Network and exfiltration

Reverse shells, exfiltration, and anything that ships local files or credentials off-machine.

network.pipe_to_shell
network.nc_reverse_shell
network.curl_with_creds
network.scp_to_remote

privilege.*9 rules

Privilege and persistence

Sudo, permission loosening, and edits that help the Skill survive a reboot or hijack later commands.

privilege.sudo
privilege.write_shell_rc
privilege.crontab_persist
privilege.path_hijack

payload.*9 rules

Obfuscation and hidden payloads

Base64-decoded execution, zero-width characters, and other tricks aimed at hiding intent.

payload.base64_decode_exec
payload.eval_exec
payload.zero_width_chars
payload.rtl_override

Five severity levels

Each finding carries a confidence score mapped to a severity ladder. The report shows the highest severity first, then the exact file and line.

Criticalconfidence ≥ 0.95
Highconfidence ≥ 0.80
Mediumconfidence ≥ 0.60
Lowconfidence ≥ 0.40
Safebelow 0.40

Context lowers false positives

A rm -rf line inside a doc example is not the same as one inside an instruction. The scanner records where each match appeared and weights it accordingly.

Inside a Markdown code example
On a comment line
Under a docs/ directory
Language of the surrounding code block

Marketplace

Discover, install, and translate without breaking flow

Browse community Skills inside the desktop app, inspect their source, install with one action, and translate content when needed.

Multiple sources

Browse Skills from supported community catalogs and GitHub repositories.

Bilingual workflows

Translate Skill names, descriptions, and instructions between English and Chinese.

Comparison

Less maintenance. More control.

Manual folders work until every agent needs its own copy. Skills Manager keeps the workflow visible and reversible as your toolset grows.

Workflow	Manual folders	Skills Manager
Canonical Skill source	One copy per tool	One central library
Per-tool enable controls	Move files by hand	Built-in toggles
Tool discovery	Find every path yourself	Automatic detection
Security review	Read every file manually	Local static scan
English and Chinese	Translate by hand	Integrated translation

Concepts

Agent Skills, SKILL.md, and why they need managing

Agent Skills are the portable format AI coding agents load on demand. Getting them right starts with knowing what a Skill is — and what it is not.

Agent Skill: An Agent Skill is a folder containing a SKILL.md file plus any scripts or templates it needs. Agents index the front matter of every SKILL.md up front, and load the body only when a task matches the description — so a large library never floods the context window.
SKILL.md: SKILL.md is the entry point of a Skill: a Markdown file opening with YAML front matter. The front matter is what agents index; the body is what they follow after the Skill triggers. A folder without SKILL.md is simply ignored.
Skills directory: Every tool scans its own directory: Claude Code reads ~/.claude/skills, Codex reads ~/.codex/skills, Cursor reads ~/.cursor/skills. Different paths per tool is why the same Skill historically ends up copied into every one of them.
The management problem: Copies drift: edit a Skill in ~/.claude/skills and the copy in ~/.codex/skills keeps the old behaviour. Skills Manager keeps one canonical folder and links it into each tool, so an edit lands everywhere at once.

Anatomy of a SKILL.md file

Agents index the front matter and match it against the task. The description decides whether the Skill loads at all — write it more carefully than the name.

nameRequired
Lowercase identifier for the Skill. Convention is to match the folder name, using hyphens instead of spaces.
descriptionRequired
What the Skill does and when to use it. Write both halves — the trigger condition is what makes the Skill fire at the right moment.
allowed-toolsOptional
Restricts the Skill to a specific set of tools. Supported by Claude Code; other agents ignore fields they do not recognise.
Markdown bodyRequired
The instructions the agent follows after the Skill is triggered. Keep it short and imperative; link out to reference files for anything long.

SKILL.md

---
name: release-notes
description: Drafts release notes from merged pull requests. Use when
  the user asks for a changelog or release summary for a version tag.
allowed-tools: Bash(git log:*), Read, Write
---

# Release notes

1. Collect merged pull requests since the previous tag.
2. Group them by type: feat, fix, perf, docs.
3. Write one line per change, newest first.

Skills vs MCP

Agent Skills and MCP servers solve different problems

Skills and MCP servers are not alternatives. A Skill encodes a procedure; an MCP server grants access to a system the agent cannot reach on its own. Most real setups use both.

Dimension	Agent Skills	MCP servers
What it adds	Procedure and domain knowledge	Connections to external systems
Format	A folder with SKILL.md and support files	A running server speaking a JSON-RPC protocol
Runtime cost	None until the Skill is triggered	A process and a live connection per server
Loaded when	The task matches the description	The agent starts, tools listed up front
Authoring effort	Write Markdown	Write and ship a server
Best fit	Repeatable workflows, house style, checklists	Databases, ticketing, internal APIs

Rule of thumb: a task you can write down is a Skill; a task that needs live data or credentials is an MCP server. Skills Manager manages the first.

Specifications

What you are installing

A Tauri desktop app that runs locally, reaching the network only for features you explicitly start.

Current version: 2.1.8
License: MIT
Price: Free, no paid tier
Platforms: macOS (Apple Silicon and Intel), Windows 10+, Linux
Package formats: .dmg, .exe, .msi, .AppImage, .deb, .rpm
Built-in tool definitions: 31, plus custom tools
Central library: ~/.skills-manager/skills
Connection method: Symlink on macOS and Linux, junction on Windows
Security rules: 36 local static rules, 5 severity levels
Account required: No, for local Skill management
Interface languages: English and Simplified Chinese

Who built this

Every number on this page can be checked in the repository

Skills Manager is maintained by one developer in public. This section states who wrote the page, where each claim comes from, and what the product deliberately does not do.

Maintainerjiweiyeah@jiweiyeah

Independent developer and sole maintainer of Skills Manager. Wrote the Rust linking layer that swaps between symlinks, Windows junctions, and tracked copies, the static rule engine behind the security scan, and the tool definitions for every supported agent.

Why this exists

The first version came out of a problem I had myself: running Claude Code and Codex side by side meant keeping the same Skill in ~/.claude/skills and ~/.codex/skills. Fix a bug in one and the other quietly kept the old behaviour. Everything on this page grew out of that one annoyance.

449+: commits since February 2026
22: tagged releases
2.1.8: current version

Where each claim comes from

Product numbers come from the source files that define them, not from marketing copy. Anything describing Agent Skills or MCP follows the official specification rather than a second-hand summary.

Claim on this page	Primary source
31 built-in tools and their Skills directories	src-tauri/src/models/tool.rs
36 static security rules across four categories	src-tauri/src/services/risk/rules.rs
Symlink, junction, and copy fallback behaviour	src-tauri/src/services/linker.rs
What an Agent Skill is and how SKILL.md is loaded	Anthropic — Agent Skills documentation
How to write a description that triggers reliably	Anthropic — Skill authoring best practices
What an MCP server is and how it differs from a Skill	Model Context Protocol specification
Directory junctions need no elevation on Windows	Microsoft Learn — mklink

What Skills Manager does not do

Knowing the limits matters more than the feature list when you are deciding whether to trust a tool with your setup.

The scan reads code, it does not run it

The security scan is static analysis over the files in a Skill folder. It matches patterns in text and cannot observe what a command actually does at runtime, so a novel or heavily obfuscated payload can pass a clean scan. Treat the result as a first filter, not a guarantee.

There is no web catalogue of Skills

The marketplace lives inside the desktop app. This site does not host browsable Skill pages, and there is no public API for searching Skills from a browser.

One tool gets a copy instead of a link

iFlow does not follow filesystem links, so Skills Manager keeps a tracked copy for it and records the source path. That copy does not update on its own after you edit the source — toggling the Skill off and on re-copies the folder. Every other supported tool sees the edit immediately.

Translation and LLM review send text to a provider

Both features are optional and off by default. When you start one, the Skill content is sent to the language model provider you configured with your own API key. Everything else — organizing, linking, scanning — never leaves your machine.

Page last reviewed2026-08-09Reviewed against the shipping build and the linked sources above.

Download

Make every coding agent feel like one workspace

Download Skills Manager, connect your tools, and stop maintaining the same Skill in multiple folders.

v2.1.8

macOS

Choose the build for your Mac

Apple Silicon Intel

Windows

64-bit Windows 10 or later

.exe installer .msi package

Linux

64-bit desktop distributions

AppImage .deb .rpm

View all release files

Prefer to inspect the source first?

Skills Manager is MIT-licensed and developed in public.

Explore the repository

FAQ