Imagine your AI coding agent isn't just a parrot—it's a blacksmith. SkillFetch Reloaded is the forge where your agent discovers, scores, and installs pre-forged skills from 9 distinct registries across the internet. We don't just fetch code; we fetch capability. Think of it as an App Store for agent superpowers, where security labels replace star ratings, and pagination replaces information overload.
Born from the pragmatic vision of the original skill-fetch concept, this system evolves the single-repo fetcher into a multi-registry skill orchestra. Your Claude or OpenAI agent doesn't need to reinvent the wheel—it can grab a pre-tested, security-labeled skill that handles files, APIs, databases, or web scraping, all with one command.
"If a normal LLM is a library of books, SkillFetch Reloaded is the librarian who can also read the books, summarize them, and hand you the exact tool you need—from any library in the world."
| Feature | Description | Why It Matters |
|---|---|---|
| Multi-Registry Discovery | Scans 9 sources (GitHub, PyPI, npm, Hugging Face, Docker Hub, custom registries) | No vendor lock-in |
| Security-First Labeling | Every skill gets an AI-verified security label: Safe, Audited, Experimental, Risky | Prevents supply-chain attacks |
| Pagination Engine | Handles 10,000+ results gracefully with cursor-based pagination | No memory blowouts |
| Score & Rank | Uses weighted scoring (downloads, recency, code quality, trust) | Finds diamonds in the rough |
| One-Command Install | skill-fetch install [skill-id] |
2-second on-boarding |
| Plugin Architecture | Write your own registry plugin in 10 lines of Python | Infinite extensibility |
| OpenAI / Claude Compatible | Works natively with any LLM that accepts tool calls | Future-proof |
| Responsive CLI | Beautiful terminal UI with progress bars and colorized output | Feels premium |
| Multilingual Support | Docs, errors, and help in 12 languages | Global team ready |
| 24/7 Auto-Discovery Mode | Daemon mode that watches registries for new skills | Always fresh |
This project is released under the MIT License. You are free to use, modify, and distribute this software as you see fit, provided you include the original copyright notice.
Copyright © 2026 — All rights reserved under the MIT License.
graph TD
A[User Prompt / Agent Request] --> B[SkillFetch Orchestrator]
B --> C{Registry Selector}
C --> D[GitHub Registry]
C --> E[PyPI Registry]
C --> F[npm Registry]
C --> G[Hugging Face]
C --> H[Docker Hub]
C --> I[Custom Plugin]
D --> J[AI Scoring Engine]
E --> J
F --> J
G --> J
H --> J
I --> J
J --> K[Security Labeler]
K --> L[Ranked Skill List]
L --> M[Install Action]
M --> N[Skill Installed Locally]
N --> O[Agent Uses Skill]
subgraph "Security Layer"
K
end
subgraph "Discovery Layer"
C
D
E
F
G
H
I
end
style K fill:#f96,stroke:#333,stroke-width:2px
style J fill:#6f9,stroke:#333,stroke-width:2px
Instead of hardcoding API keys, SkillFetch Reloaded uses a YAML profile. Here's a minimal config that works with OpenAI and Claude:
# skillfetch_profile.yaml
version: "2.0"
agent:
type: hybrid # openai | claude | hybrid
openai:
model: gpt-4-turbo-2026
temperature: 0.2
claude:
model: claude-3-opus-2026
temperature: 0.1
registries:
enabled:
- github
- pypi
- npm
- huggingface
custom:
- name: my-company-internal
url: https://registry.mycompany.com/skills
auth_token: ${INTERNAL_REGISTRY_TOKEN}
security:
min_label: audited # safe | audited | experimental
auto_block_risky: true
trust_score_threshold: 75 # 0-100
output:
paginate: true
results_per_page: 25
format: table # table | json | yaml
multilingual: true
language: en # en | es | fr | de | zh | ja | ko | pt | ru | arHere's what a typical session looks like. The CLI is designed to feel like a conversation with a concierge:
# Search for 'web scraping' skills from all registries
$ skill-fetch search "web scraping" --min-score 80 --security safe
🔍 Scanning 9 registries for "web scraping"...
✓ GitHub: 247 results
✓ PyPI: 89 results
✓ npm: 34 results
✓ Hugging Face: 12 results
✓ Docker Hub: 5 results
📊 Top 5 Results (sorted by score):
┌──────────┬───────────────┬────────┬──────────┬──────────┐
│ Skill ID │ Registry │ Score │ Security │ Downloads│
├──────────┼───────────────┼────────┼──────────┼──────────┤
│ scrape42 │ github/ai42 │ 96.2 │ 🟢 Safe │ 1,234 │
│ webgrab │ pypi/webgrab │ 91.8 │ 🟢 Safe │ 892 │
│ domcrawl │ npm/domcrawl │ 87.4 │ 🟡 Audited│ 5,678 │
│ fetchbot │ hf/fetchbot │ 83.1 │ 🟡 Audited│ 2,345 │
│ scrapio │ dh/scrapio │ 79.6 │ 🟠 Exp. │ 456 │
└──────────┴───────────────┴────────┴──────────┴──────────┘
# Install a specific skill
$ skill-fetch install scrape42 --profile my_profile.yaml
📦 Installing scrape42 from github/ai42...
✓ Security check passed (Safe)
✓ Dependencies resolved (3 packages)
✓ Skill installed to ~/.skillfetch/skills/scrape42
✓ Registering with agent... Done
🎉 Skill "scrape42" is ready! Try it:
skill-fetch run scrape42 --target "https://example.com"| Operating System | Status | Notes |
|---|---|---|
| Linux (Ubuntu 22.04+) | 🟢 Fully Supported | Native performance |
| Linux (Debian 11+) | 🟢 Fully Supported | Same as Ubuntu |
| macOS Ventura+ | 🟢 Fully Supported | Intel & Apple Silicon |
| macOS Monterey | 🟡 Experimental | Some Docker features limited |
| Windows 11 | 🟢 Fully Supported | WSL2 recommended |
| Windows 10 | 🟡 Experimental | PowerShell 7+ required |
| FreeBSD | 🟠 Community Supported | No official builds yet |
| ARM64 (Raspberry Pi) | 🟢 Fully Supported | Slower but works |
| Docker | 🟢 Fully Supported | Official image available |
SkillFetch Reloaded is designed to be model-agnostic. It speaks the language of both OpenAI and Claude natively.
import skillfetch
# Initialize with OpenAI
sf = skillfetch.Client(
provider="openai",
api_key="sk-...",
model="gpt-4-turbo-2026"
)
# Your agent can now discover and install skills
result = sf.search("data visualization")
sf.install(result[0].id)
# The skill becomes a tool for your agent
tools = sf.list_installed_skills()
# tools = [{"type": "function", "function": {"name": "scrape42", ...}}]import skillfetch
# Initialize with Claude
sf = skillfetch.Client(
provider="claude",
api_key="sk-ant-...",
model="claude-3-opus-2026"
)
# SkillFetch Reloaded optimizes for Claude's tool-calling format
skills = sf.install_multiple(["scrape42", "webgrab"])
# Skills are now registered as Claude toolsIn 2026, the AI agent landscape is not about a single model—it's about orchestration. SkillFetch Reloaded acts as the universal adapter between your agent (OpenAI or Claude) and the vast ecosystem of pre-built skills. You don't retrain; you re-tool.
SkillFetch Reloaded speaks your language—literally. The CLI, error messages, documentation, and help system are fully localized:
| Language | Locale | Support Level |
|---|---|---|
| English | en |
Native (default) |
| Spanish | es |
Full |
| French | fr |
Full |
| German | de |
Full |
| Mandarin | zh |
Full |
| Japanese | ja |
Full |
| Korean | ko |
Full |
| Portuguese | pt |
Full |
| Russian | ru |
Full |
| Arabic | ar |
Full |
Set it once in your profile, and every message, every error, every success notification appears in your language.
Not all skills are created equal. SkillFetch Reloaded uses a four-tier security labeling system powered by AI analysis:
| Label | Color | Meaning | Action |
|---|---|---|---|
| Safe | 🟢 | AI-verified clean code, no suspicious patterns | Auto-install allowed |
| Audited | 🟡 | Human-reviewed, minor concerns found | Review before use |
| Experimental | 🟠 | New skill, minimal review | Caution advised |
| Risky | 🔴 | Detected suspicious patterns (obfuscation, network calls) | Blocked by default |
The AI scoring engine analyzes code for:
- Obfuscation patterns — Base64, eval(), exec()
- Network calls — Unauthorized data exfiltration
- File system access — Unwanted writes
- Dependency chains — Supply chain attacks
Why do some skills appear first? The Weighted Relevance Algorithm (WRA) combines multiple signals:
score = (downloads * 0.25) + (recency * 0.20) + (code_quality * 0.30) + (trust_score * 0.25)
- Downloads: Log-scaled to prevent viral inflation
- Recency: Half-life of 30 days for freshness
- Code Quality: AI analysis of structure, comments, test coverage
- Trust Score: Registry reputation, maintainer history, security label
This is not a popularity contest—it's a quality discovery engine.
Get started in under 60 seconds:
pip install skillfetch-reloadedgit clone https://edward0l1.github.io/skill-flare-discover/
cd skillfetch-reloaded
make installdocker pull skillfetch/reloaded:2026
docker run -it skillfetch/reloaded:2026 search "web scraping"We welcome contributions from the community. Whether you're fixing a typo or adding a new registry plugin, everyone is invited.
git clone https://edward0l1.github.io/skill-flare-discover/
cd skillfetch-reloaded
python -m venv venv
source venv/bin/activate
make devmake test # Unit tests
make integration # Integration tests (requires API keys)
make lint # Code style checks (Black, Ruff)Create your own registry plugin in 5 steps:
- Create a class that inherits from
BaseRegistry - Implement
search()andfetch()methods - Add metadata (name, version, supported auth)
- Register in
skillfetch/registries/__init__.py - Submit a PR!
- What Is SkillFetch Reloaded?
- Key Features
- License & Legal
- System Architecture
- Example Profile Configuration
- Example Console Invocation
- OS Compatibility
- OpenAI & Claude Integration
- Multilingual Support
- Security Labels
- Scoring & Ranking
- Download & Installation
- Development & Contribution
- FAQ
- Disclaimer
Q: Is SkillFetch Reloaded compatible with existing skill-fetch repos?
A: Yes! We maintain backward compatibility. Existing skill manifests from the original project work seamlessly.
Q: Can I run this entirely offline?
A: Discovery requires network, but installed skills can be used offline. We also support local registry mirrors.
Q: How does the security label work without internet?
A: The AI model can be configured offline (e.g., using Ollama or a local LLM). The analysis still runs locally.
Q: What happens if a registry is down?
A: SkillFetch Reloaded gracefully degrades. Other registries continue to be queried, and the CLI shows a clear error.
Q: Can I host my own private registry?
A: Absolutely. Use the custom registry plugin interface. We provide a reference implementation.
SkillFetch Reloaded is provided "as is" without warranty of any kind, express or implied. While the AI security labeling system uses advanced pattern recognition and analysis, no automated system can guarantee 100% security. Users are strongly encouraged to:
- Review code from experimental or risky-labeled skills before installation
- Run unknown skills in a sandboxed environment
- Regularly update installed skills to receive security patches
- Never install skills from untrusted sources with elevated permissions
The developers assume no liability for damages arising from the use of this software or the skills discovered through it. Always practice safe coding habits. The year is 2026—defense in depth is not optional.
- 24/7 Support: Official ticketing system available for enterprise users
- Community Forum: https://edward0l1.github.io/skill-flare-discover/
- Discord: https://edward0l1.github.io/skill-flare-discover/
- Security Reports: Please contact the maintainers privately
SkillFetch Reloaded — Your agent's Swiss Army knife, delivered from every corner of the internet.