Skill Fleet

A modular AI capability platform that creates, manages, and deploys agent skills as reusable, standards-compliant components.

Skill Fleet transforms how AI agents learn by organizing capabilities in a hierarchical taxonomy. Instead of bloated monolithic prompts, skills are modular, versioned, and discoverable—loaded on-demand when agents need them.

Perfect for: AI teams building agent systems, platform engineers managing capability libraries, and organizations standardizing AI knowledge management.

---

Why Skill Fleet?

For Technical Teams

• DSPy-Powered Workflows: Built on DSPy with task-based orchestrators for reliable skill generation
• FastAPI Architecture: Clean API layer with background tasks, dependency injection, and async support
• agentskills.io Compliant: Standard YAML frontmatter ensures skills work across different agent frameworks
• Hierarchical MLflow Tracking: Parent runs for workflows, child runs for each phase

For Decision Makers

• Modular & Maintainable: Skills are versioned, tracked, and independently testable
• Standards-Based: Open specification compliance prevents vendor lock-in
• Scalable: Hierarchical taxonomy supports hundreds of skills with organized growth

For Everyone

• Easy to Use: Chat interface for creating skills without coding
• Validated: Automated compliance checking ensures quality
• Observable: Built-in analytics and usage tracking

---

Quick Start

Create your first skill in under 2 minutes:

# 1. Install dependencies
uv sync --group dev
cp .env.example .env
# Edit .env and add your GOOGLE_API_KEY

# 2. Start the API server
uv run skill-fleet serve

# 3. Create a skill via chat (in a new terminal)
uv run skill-fleet chat "Create a Python decorators skill"

The skill is created as a draft. After reviewing it in drafts/<job_id>/, promote it:

uv run skill-fleet promote <job_id>

---

Prerequisites

• Python: 3.12+
• Package Manager: uv
• API Keys: GOOGLE_API_KEY (Gemini is the default model)

---

Installation

# Clone the repository
git clone https://github.com/Qredence/skill-fleet.git
cd skill-fleet

# Install dependencies
uv sync

# Setup environment
cp .env.example .env
# Edit .env and add your GOOGLE_API_KEY

---

Interactive Workflows

💬 Real-Time Chat

Build skills with streaming updates and live reasoning display:

uv run skill-fleet chat "Create a Redis caching skill"

Features:

• Real-time progress updates (100ms polling)
• Live thinking/reasoning display
• Arrow key navigation for multi-choice questions
• HITL (Human-in-the-Loop) integration

📊 Validation & Quality

Validate skills against agentskills.io standards:

# Validate a skill
uv run skill-fleet validate skills/python/decorators

# Check compliance
uv run skill-fleet validate --strict skills/general/testing

🧠 DSPy Optimization

Tune prompts using MIPROv2 or GEPA optimizers:

uv run skill-fleet optimize --optimizer miprov2

---

API Reference

v1 API (Current)

The v1 API provides comprehensive skill management:

Endpoint	Description
POST /api/v1/skills	Create skill (starts background job)
GET /api/v1/skills/{id}	Get skill details
PUT /api/v1/skills/{id}	Update skill
POST /api/v1/skills/{id}/validate	Validate skill
POST /api/v1/skills/{id}/refine	Refine with feedback
GET /api/v1/jobs/{id}	Check job status
GET /api/v1/hitl/{job_id}	Poll for HITL prompts
POST /api/v1/hitl/{job_id}	Submit HITL response

CLI Commands

Command	Description
skill-fleet serve	Start FastAPI server
skill-fleet chat	Interactive skill creation
skill-fleet list	List all skills
skill-fleet promote <id>	Promote draft to taxonomy
skill-fleet validate <path>	Validate skill
skill-fleet terminal	Python-only CLI interface

---

Project Structure

skill-fleet/
├── src/skill_fleet/
│   ├── api/                 # FastAPI application
│   │   ├── v1/             # API v1 routes
│   │   ├── services/       # Business logic layer
│   │   └── schemas/        # Pydantic models
│   ├── cli/                # Typer-based CLI
│   │   └── commands/
│   │       ├── chat.py     # Interactive chat
│   │       └── terminal.py # Python-only interface
│   ├── core/               # Domain logic
│   │   ├── workflows/      # Task-based orchestrators
│   │   ├── modules/        # DSPy modules
│   │   └── signatures/     # DSPy signatures
│   ├── dspy/               # Centralized DSPy config
│   ├── infrastructure/     # Technical infrastructure
│   │   ├── db/            # Database layer
│   │   └── tracing/       # MLflow integration
│   ├── taxonomy/          # Skill taxonomy management
│   └── validators/        # Skill validation
├── skills/                # Skill taxonomy
│   ├── _core/            # Always-loaded skills
│   ├── python/           # Python skills
│   ├── devops/           # DevOps skills
│   └── ...
├── tests/
│   ├── api/              # API tests
│   ├── unit/             # Unit tests
│   └── integration/      # Integration tests
├── config/               # Configuration
└── docs/                 # Documentation

---

Architecture

FastAPI-Centric Design

The v0.3.5 restructure introduces a clean, FastAPI-centric architecture:

1. API Layer (api/): Routes, schemas, and dependency injection
2. Service Layer (api/services/): Business logic bridging API to workflows
3. Workflow Layer (core/workflows/): Task-based DSPy orchestrators
4. Infrastructure (infrastructure/): Database, tracing, monitoring

Workflow Orchestrators

Three-phase skill creation with HITL support:

1. Understanding Workflow: Analyzes requirements, generates plan
2. Generation Workflow: Creates skill content
3. Validation Workflow: Checks compliance, refines content

Each phase runs in a child MLflow run under a parent workflow run.

---

Configuration

Environment Variables

Variable	Required	Description
GOOGLE_API_KEY	Yes	Gemini API key
SKILL_FLEET_ENV	No	production or development
SKILL_FLEET_CORS_ORIGINS	Prod	Allowed origins (comma-separated)
DSPY_CACHEDIR	No	DSPy cache directory
DSPY_TEMPERATURE	No	Override LLM temperature

Config File

Edit config/config.yaml for:

• Model settings (default: gemini/gemini-2-flash)
• Optimizer configurations (MIPROv2, GEPA)
• Task-specific model assignments

---

Development

Running Tests

# Run all tests
uv run pytest

# Run unit tests only
uv run pytest tests/unit/

# Run with coverage
uv run pytest --cov=skill_fleet

Linting & Formatting

# Check and fix
uv run ruff check --fix .

# Format code
uv run ruff format .

Project Standards

• Python: 3.12+ with modern type hints (str | None)
• Line Length: 100 characters
• Quotes: Double quotes
• Docstrings: Google style
• Imports: Absolute only, no relative imports

---

Documentation

Getting Started

• Getting Started Guide - Installation and first steps
• Quick Start - Streaming chat guide

Core Concepts

• System Overview - Architecture and concepts
• AGENTS.md - Comprehensive working guide
• Developer Reference

API & Technical

• API v1 Documentation - REST API reference
• API Migration - v1 to v2 migration guide
• DSPy Framework - Workflow documentation
• Import Path Guide - Import conventions
• Service Extension - Adding services

Advanced

• HITL System - Human-in-the-Loop
• agentskills.io Compliance - Standards
• Background Jobs - Async job processing

---

Migration from v0.2.x

See docs/api/MIGRATION_V1_TO_V2.md for:

• Import path changes
• API endpoint updates
• Architecture changes
• Breaking changes and deprecations

---

License

Apache License 2.0. See LICENSE.

Contributing

See CONTRIBUTING.md for development workflow.

---

Version: 0.3.5
Status: Production Ready
Last Updated: January 2026