Installation & Setup¶
Prerequisites¶
- Python 3.11+ (tested with 3.11 through 3.14)
- git (to clone the repositories)
Quick Install¶
# 1. Clone both repositories side by side
git clone https://github.com/gfernandf/agent-skills.git
git clone https://github.com/gfernandf/agent-skill-registry.git
# 2. Install the runtime (with all optional extras)
cd agent-skills
pip install -e ".[all]"
That's it. The agent-skills CLI is now available.
Verify Installation¶
# Check CLI
agent-skills --help
# Run health check (uses Python baselines, no API key needed)
agent-skills run text.detect-language-and-classify \
--input '{"text": "Hello world"}'
Optional: LLM Provider¶
Some capabilities use OpenAI as a binding backend. To enable them:
# Copy the example env file
cp .env.example .env
# Edit .env and set your API key
# OPENAI_API_KEY=sk-...
# Then export it (Linux/Mac)
export OPENAI_API_KEY=sk-...
# Or on Windows PowerShell
$env:OPENAI_API_KEY = "sk-..."
Without an API key, the runtime falls back to deterministic Python baselines for all capabilities. This is fully functional for development and testing.
Install Options¶
# Minimal (only core dependencies: pyyaml, requests)
pip install -e .
# With PDF support
pip install -e ".[pdf]"
# With web scraping support
pip install -e ".[web]"
# With all extras (pdf + web)
pip install -e ".[all]"
# With development tools (extras + pytest)
pip install -e ".[dev]"
Registry Tooling¶
The registry repo has its own lightweight dependencies:
cd ../agent-skill-registry
pip install -r requirements.txt
Validate the registry:
python tools/validate_registry.py
python tools/generate_catalog.py
python tools/registry_stats.py
Project Layout¶
your-workspace/
├── agent-skills/ ← Runtime (this repo)
│ ├── cli/ CLI entry point
│ ├── runtime/ Execution engine, scheduler, binding system
│ ├── gateway/ Skill discovery and attach
│ ├── customer_facing/ HTTP API and MCP bridge
│ ├── official_services/ Python baseline implementations
│ ├── bindings/official/ Binding definitions per capability
│ ├── services/official/ Service descriptors
│ ├── policies/ Default binding selection
│ └── docs/ Architecture documentation
│
└── agent-skill-registry/ ← Contracts (companion repo)
├── capabilities/ Capability YAML definitions
├── skills/ Skill YAML workflows
├── vocabulary/ Controlled naming vocabulary
├── catalog/ Generated machine-readable catalogs
└── tools/ Validation and governance tooling
Both repos must be siblings in the same parent directory for the runtime to locate registry definitions automatically.
Environment Variables¶
| Variable | Default | Description |
|---|---|---|
OPENAI_API_KEY |
(none) | OpenAI API key for LLM-backed bindings |
AGENT_SKILLS_FS_ROOT |
cwd() |
Root directory for fs.file.read sandbox |
AGENT_SKILLS_DEBUG |
(unset) | Enable verbose debug logging |
AGENT_SKILLS_AUDIT_DEFAULT_MODE |
standard |
Audit mode: off, standard, full |
See .env.example for the full list.
Next Steps¶
- 10-minute onboarding — mental model and first skill run
- Runner guide — architecture deep dive
- Project status — current milestone