Repository navigation
FAQ
| 🇬🇧 English | 🇹🇷 Türkçe |
English
The first run clones 34–48 repositories (depends on profile). This typically takes 2–5 minutes on a broadband connection. Each repo is cloned sequentially; the bottleneck is network speed and the number of repos in the chosen profile.
All skills are installed to ~/.agents/skills/. Each skill gets its own subdirectory named after the skill (e.g., ~/.agents/skills/academic-paper-review/). This location is auto-discovered by:
-
Claude Code — reads
~/.agents/skills/by default - OpenCode — reads the same directory
- Cursor — supports the same convention
- Any agent that uses the standard
SKILL.mddiscovery path
Simply delete the directory:
# macOS / Linux
rm -rf ~/.agents/skills
# Windows
rmdir /s %USERPROFILE%\.agents\skillsOr remove individual subdirectories for skills you no longer need.
Yes. Use --prefix /path/to/custom/dir to install to a custom directory instead of ~/.agents/skills/. Without it, the installer writes to ~/.agents/skills/ (the default expected by Claude Code, OpenCode, and Cursor).
Git is required to clone repositories. Install it from git-scm.com, restart your terminal, and try again.
To verify Git is installed:
git --versionThe --gui flag requires tkinter, which is Python's standard GUI toolkit.
| Platform | Solution |
|---|---|
| Windows | tkinter is bundled with the official Python installer. Reinstall Python from python.org and ensure "tcl/tk and IDLE" is checked. |
| macOS | tkinter is bundled with the official Python installer from python.org. If using Homebrew's Python, install with: brew install python-tk
|
| Linux (Debian/Ubuntu) | sudo apt install python3-tk |
| Linux (Fedora) | sudo dnf install python3-tkinter |
| Linux (Arch) | sudo pacman -S tk |
- Verify the skills are installed:
ls ~/.agents/skills/(should show subdirectories) - Check that each subdirectory contains a
SKILL.mdfile - Restart your editor / agent after installation
- Make sure your agent is configured to look in
~/.agents/skills/(this is the default for Claude Code, OpenCode, and Cursor)
The first run clones every repository from scratch. This is network-bound. Possible improvements:
- Use a faster internet connection
- Install fewer categories (e.g.,
python skills.py K2instead of the full profile) - After the first install, subsequent runs are faster (repos are cached)
Some repositories may be large or slow to clone. The installer retries each clone up to 3 times, with a 300-second timeout per clone attempt (sparse-checkout uses a 120-second timeout). If a repo consistently times out:
- Check your internet connection
- The repo may be temporarily unavailable (try again later)
- Skip that repo by choosing a more specific category
Some repos store their skills in a subdirectory rather than the root. The installer looks for the specified subpath. If the upstream repo changes its structure, this error may appear. Please open an issue if you encounter this.
These categories contain:
- K9 (Community): Repositories with lower GitHub star counts, lesser-known authors, or unverified quality
- K10 (Risk): Experimental, potentially unmaintained, or high-risk repositories
They are excluded from all profiles except all to maintain a high quality bar for the default experience. You can include them with:
python skills.py all
# or individually:
python skills.py K9 K10Each repository is placed into a category based on its primary function. Criteria include:
- The repository's description and README
- The content of its SKILL.md files
- The domain of the parent project (e.g., Supabase → Cloud, Expo → Mobile)
Python 3.8 or later. The installer uses from __future__ import annotations and pathlib.Path, which are stable in 3.8+.
Yes. The installer is tested on:
- Windows 10/11 (PowerShell 5.1+)
- macOS 12+ (Intel and Apple Silicon)
- Linux (Ubuntu 22.04+, Debian, Fedora, Arch)
- Reads the embedded
REPOSregistry (48 repos across 10 categories) - Filters repos based on the requested profile or category
- For each repo: clones it to a temp directory, finds
SKILL.mdfiles, copies them to~/.agents/skills/<name>/, cleans up temp files - Fixes invalid skill names (spaces and underscores become hyphens; other invalid characters are removed)
- Deduplicates — if two repos ship the same skill name, the higher-category version wins
- Saves a timestamped log and directory tree snapshot
When two repos contain a SKILL.md file with the same skill name, the one from the higher category number wins. For example, if both K2 (AI & LLM) and K4 (Frontend & UI) ship a skill named react-skills, the K4 version is kept because 4 > 2.
In your current working directory (where you run python skills.py), two timestamped files appear (exact paths per .gitignore): the detailed installation log, and the directory tree of ~/.agents/.
These are timestamped to the second, so every run produces unique files.
İlk çalıştırma 34–48 depoyu klonlar (profile bağlı). Geniş bant bağlantıda tipik olarak 2–5 dakika sürer. Her depo sırayla klonlanır; darboğaz ağ hızı ve seçilen profildeki depo sayısıdır.
Tüm beceriler ~/.agents/skills/ klasörüne yüklenir. Her beceri, beceri adıyla kendi alt klasörünü alır (örn. ~/.agents/skills/academic-paper-review/). Bu konum aşağıdakiler tarafından otomatik olarak keşfedilir:
-
Claude Code — varsayılan olarak
~/.agents/skills/okur - OpenCode — aynı dizini okur
- Cursor — aynı kuralı destekler
- Standart
SKILL.mdkeşif yolunu kullanan her ajan
Dizini silmeniz yeterlidir:
# macOS / Linux
rm -rf ~/.agents/skills
# Windows
rmdir /s %USERPROFILE%\.agents\skillsVeya artık ihtiyacınız olmayan bireysel beceri klasörlerini kaldırın.
Evet. ~/.agents/skills/ yerine özel bir dizine yüklemek için --prefix /ozel/dizin/yolu kullanın. Bu bayrak olmadan yükleyici, Claude Code, OpenCode ve Cursor'ın beklediği varsayılan konum olan ~/.agents/skills/ klasörüne yazar.
Depoları klonlamak için Git gereklidir. git-scm.com adresinden yükleyin, terminalinizi yeniden başlatın ve tekrar deneyin.
Git'in yüklü olduğunu doğrulamak için:
git --version--gui flag'i, Python'un standart GUI araç seti olan tkinter'ı gerektirir.
| Platform | Çözüm |
|---|---|
| Windows | tkinter, resmi Python yükleyicisiyle birlikte gelir. Python'u python.org adresinden yeniden yükleyin ve "tcl/tk and IDLE" seçeneğinin işaretli olduğundan emin olun. |
| macOS | tkinter, python.org adresindeki resmi Python yükleyicisiyle birlikte gelir. Homebrew Python kullanıyorsanız: brew install python-tk
|
| Linux (Debian/Ubuntu) | sudo apt install python3-tk |
| Linux (Fedora) | sudo dnf install python3-tkinter |
| Linux (Arch) | sudo pacman -S tk |
- Becerilerin yüklü olduğunu doğrulayın:
ls ~/.agents/skills/(alt klasörleri göstermeli) - Her alt klasörün bir
SKILL.mddosyası içerdiğini kontrol edin - Kurulumdan sonra düzenleyicinizi / ajanınızı yeniden başlatın
- Ajanınızın
~/.agents/skills/konumuna bakacak şekilde yapılandırıldığından emin olun (Claude Code, OpenCode ve Cursor için varsayılan budur)
İlk çalıştırma her depoyu sıfırdan klonlar. Bu ağ bağımlıdır. Olası iyileştirmeler:
- Daha hızlı bir internet bağlantısı kullanın
- Daha az kategori yükleyin (tam profil yerine
python skills.py K2) - İlk kurulumdan sonraki çalıştırmalar daha hızlıdır (depolar önbelleğe alınır)
Bazı depolar büyük veya klonlaması yavaş olabilir. Yükleyici, klon başına 300 saniyelik zaman aşımı ile her depoyu en fazla 3 kez dener (sparse-checkout 120 saniyelik zaman aşımı kullanır). Bir depo sürekli zaman aşımına uğrarsa:
- İnternet bağlantınızı kontrol edin
- Depo geçici olarak kullanılamıyor olabilir (daha sonra tekrar deneyin)
- Daha spesifik bir kategori seçerek bu depoyu atlayın
Bazı depolar becerilerini kök dizin yerine bir alt dizinde saklar. Yükleyici belirtilen alt yolu arar. Yukarı akış deposu yapısını değiştirirse bu hata görünebilir. Bununla karşılaşırsanız lütfen bir issue açın.
Bu kategoriler şunları içerir:
- K9 (Topluluk): Düşük GitHub yıldız sayılarına sahip, az bilinen yazarlar veya doğrulanmamış kalite
- K10 (Risk): Deneysel, potansiyel olarak bakımı yapılmayan veya yüksek riskli depolar
Varsayılan deneyimde yüksek kalite çıtasını korumak için tumu dışındaki tüm profillerden hariç tutulurlar. Bunları şu şekilde dahil edebilirsiniz:
python skills.py tumu
# veya tek tek:
python skills.py K9 K10Her depo, birincil işlevine göre bir kategoriye yerleştirilir. Kriterler şunları içerir:
- Deponun açıklaması ve README'si
- SKILL.md dosyalarının içeriği
- Üst projenin alanı (örn. Supabase → Bulut, Expo → Mobil)
Python 3.8 veya üstü. Yükleyici from __future__ import annotations ve pathlib.Path kullanır; bunlar 3.8+'da kararlıdır.
Evet. Yükleyici aşağıdakilerde test edilmiştir:
- Windows 10/11 (PowerShell 5.1+)
- macOS 12+ (Intel ve Apple Silicon)
- Linux (Ubuntu 22.04+, Debian, Fedora, Arch)
- Gömülü
REPOSkaydını okur (10 kategoride 48 depo) - İstenen profile veya kategoriye göre depoları filtreler
- Her depo için: geçici bir dizine klonlar,
SKILL.mddosyalarını bulur,~/.agents/skills/<adi>/konumuna kopyalar, geçici dosyaları temizler - Geçersiz beceri adlarını düzeltir (boşluklar ve alt çizgiler tirelere dönüşür; diğer geçersiz karakterler kaldırılır)
- Tekilleştirir — iki depo aynı beceri adını gönderirse, yüksek kategori numarasına sahip olan kazanır
- Zaman damgalı bir günlük ve dizin ağacı görüntüsü kaydeder
İki depo aynı beceri adına sahip bir SKILL.md dosyası içerdiğinde, daha yüksek kategori numarasına sahip olan kazanır. Örneğin, hem K2 (AI & LLM) hem de K4 (Ön Yüz & Arayüz) react-skills adında bir beceri gönderirse, 4 > 2 olduğu için K4 sürümü korunur.
Geçerli çalışma dizininizde (python skills.py çalıştırdığınız yerde) iki zaman damgalı dosya belirir (tam yollar .gitignore dosyasına göre): ayrıntılı kurulum günlüğü ve ~/.agents/ dizin ağacı.
Bunlar saniyeye kadar zaman damgalıdır, böylece her çalıştırma benzersiz dosyalar üretir.