abubakarsiddik31/skill-manager

---

Contents

• The problem
• What is a “skill”?
• Features
• Supported tools
• Download
• Development
• Project structure
• Roadmap
• Contributing
• FAQ
• License

The problem

If you use more than one AI coding agent, your skills are scattered across ~/.claude/skills, ~/.agents/skills, ~/.cursor/skills, ~/.gemini/skills, ~/.copilot/skills, ~/.roo/skills, and more — eleven folders, zero shared view.

Left unchecked, this turns into skill hell: every skill you’ve ever installed sits there active, and your agent has to disambiguate against all of them just to pick the right one for what you’re doing right now. A crowded skills directory doesn’t just look messy — the more skills compete for a match, the worse an agent gets at triggering the right one. Disabling what you don’t need for the current project is how you climb back out.

Skill Manager fixes all of it. One dashboard, every tool.

What is a “skill”?

Anthropic introduced Agent Skills — folders containing a SKILL.md file with instructions an AI coding agent can load on demand — and the idea has since been picked up across the ecosystem (Claude Code, Codex, Cursor, Gemini CLI, OpenCode, and others).

Features

• 📥 Browse & install from GitHub collections — curated skill collections (Anthropic’s skills, Superpowers, Matt’s skills) ship inside the app: browse and search them offline, then install into any tool’s user or project folder in one click. Add any owner/repo collection of your own, too — every install records where it came from.
• 🛡️ Safety scanning — every install is scanned by NVIDIA SkillSpector (static analysis, no LLM, no API keys) before anything touches disk: prompt injection, data exfiltration, dangerous code, and ~70 other patterns. A risky skill is blocked with its full report until you explicitly choose “install anyway”, and one click re-scans every installed skill. See Safety scanning.
• ✨ Create skills — start a new skill from a minimal template in any tool’s folder and write its instructions in the built-in editor.
• 🗂️ Unified view across every tool’s skills directory — no more digging through a dozen different config folders by hand.
• 🔀 Enable / disable any skill without deleting it (skills move to a sibling .disabled/ folder — fully reversible, and it never touches the tool’s own config).
• 📝 View & edit — a skill’s SKILL.md renders as formatted markdown by default, with a one-click switch to raw edit.
• 🗑️ Delete skills you no longer need.
• 📁 Per-project breakdown — track individual project folders and see exactly which skills are wired into each one, separate from your global skills.
• 🧭 Detected projects — the app finds folders you already work in (Claude Code history, editor recents, git repos) and offers them as one-click suggestions in the sidebar and in a searchable picker, with activity times and skill counts on every row.
• 📌 Pinned & most-used first — pin the tools and projects you touch daily; the sidebar keeps your top three most-used projects (30-day window) on top, and every project list can sort by use or by skill count.
• 🔍 Search across names and descriptions.

Supported tools

All eleven paths are verified against each tool’s own docs — not guessed. If a tool changes its convention, open an issue and we’ll update the adapter.

The sidebar itself is tool-level: each tool’s view shows every folder it reads, skills in the shared ~/.agents/skills folder carry a “seen by” chip for every tool that discovers them, and toggling one warns that it affects all of them — there’s a single copy on disk.

Download

Grab a build from the latest release:

[!NOTE] v0.4.2 is not code-signed or notarized by Apple, and its Windows installer is not signed by a verified publisher. We have not completed Apple or Microsoft publisher verification. Only continue if you downloaded the installer from this repository’s GitHub Releases page. Code-signed releases are on the Roadmap. We know the extra steps are inconvenient; optional sponsorship helps fund code-signing and notarization work.

macOS — Gatekeeper will say “Skill Manager” Not Opened / Apple could not verify “Skill Manager” is free of malware. Click Done, then either:

• Right-click (Control-click) the app in Finder → Open → Open in the dialog that appears, or
• Go to System Settings → Privacy & Security, scroll down to the blocked-app notice, and click Open Anyway → Open Anyway again to confirm.

You only need to do this once — it launches normally after that.

Windows — SmartScreen will show Windows protected your PC. Click More info, then Run anyway.

Safety scanning

Skill Manager integrates NVIDIA SkillSpector (Apache-2.0), a security scanner for agent skills, as an optional layer:

uv tool install git+https://github.com/NVIDIA/skillspector.git
# or, with pip: pipx install / pip install from the same repo

Once the skillspector CLI is on your machine (the app also checks ~/.local/bin, /opt/homebrew/bin, and /usr/local/bin, and honors a SKILL_MANAGER_SKILLSPECTOR env var pointing at the binary):

• Every install is scanned first. The scanner runs in static-only mode (--no-llm: no LLM calls, no API keys) against the exact files the install would write — 71 vulnerability patterns across 17 categories, including prompt injection, data exfiltration, privilege escalation, supply-chain tricks, and dangerous code.
• Risky skills are blocked. Following NVIDIA’s triage policy, a risk score above 50/100 or any high/critical finding stops the install before anything touches disk. The full report (score, findings, locations, fixes) opens in-app; installing anyway is an explicit button.
• Everything you already have can be audited. The safety scan button scans all installed skills (in parallel, one scan per on-disk folder) and every skill card gets a score chip; the last result survives restarts and individual skills can be re-scanned from the editor.

No scanner installed? Installs work exactly as before — the app just says so instead of pretending. Scanning is local: skill contents never leave your machine (except SkillSpector’s optional OSV.dev CVE lookups, which fall back to offline automatically).

Development

Requires Node.js and the Rust toolchain (Tauri’s prerequisites are documented here).

git clone https://github.com/abubakarsiddik31/skill-manager.git
cd skill-manager
npm install
npm run tauri dev

To build a release binary for your platform:

npm run tauri build

Project structure

src/
  components/           react components grouped by role (layout, skills,
                        browse, modals, ui)
  hooks/                data + mutations (useGlobalSkills, useProjects, useProjectSkills)
  api/                  the invoke client over the Rust commands
  utils/                pure helpers + tests
  types/                shared frontend types split by domain
src-tauri/src/
  skills/               one adapter per tool, all implementing SkillAdapter
  collections/          GitHub collections: catalog, bundled index, install
  skillspector/         optional NVIDIA SkillSpector safety-scan integration
  commands/             tauri commands exposed to the frontend
  projects.rs           persisted list of tracked project folders
docs/                   the landing page (GitHub Pages)

Each tool implements the same SkillAdapter trait (src-tauri/src/skills/mod.rs), so adding support for a new tool is just a new module pointing at its skills directory.

Roadmap

• [x] Skill browser & install — browse curated community collections and install them into any tool’s skills folder in one click
• [x] Safety scanning — NVIDIA SkillSpector integration: pre-install scans, risky-install blocking with the full report, and one-click audits of every installed skill
• [ ] Update support for installed collection skills (in-app update checks against the repo they came from)
• [ ] Auto-update support (in-app update checks, no manual reinstall)
• [ ] Code-signed builds (no more Gatekeeper/SmartScreen warnings)
• [ ] Keep adding agents as they adopt SKILL.md — Windsurf and Trae are on the radar, but their conventions aren’t verified yet

See CONTRIBUTING.md for how to pick any of these up, or add support for a tool not listed above.

Contributing

This project gets meaningfully better with a handful of small, specific contributions — see CONTRIBUTING.md for exactly how to:

• Pin down a tool’s real skills directory (the single biggest gap right now)
• Add support for an entirely new coding agent
• Report a bug or open a fix

Every one of these is scoped to be doable in one sitting. Issues and PRs welcome. Please also read our Code of Conduct.

FAQ

Does this modify how my tools load skills? No. Skill Manager only moves folders between a skills directory and a sibling .disabled/ folder, and edits SKILL.md files directly on disk. It never touches a tool’s own config or settings files.

Is it safe to disable a skill? Yes — disabling moves the skill folder to .disabled/ next to it. Re-enabling moves it back. Nothing is deleted until you explicitly delete it.

Does it phone home or collect telemetry? No accounts, no analytics, no telemetry. The app reaches the network only to fetch the public collections catalog and to download a skill when you explicitly click install — everything else stays on your machine.

Why isn’t `` supported? Open an issue with a link to that tool’s own skills documentation and we’ll add an adapter — see Contributing.

License

MIT