Skill Framework

A comprehensive governance framework for AI Agent Skills — from classification to quality assurance to tooling.

Why This Matters

Building AI agents is easy. Governing 200+ skills across a team of agents is hard. This framework provides:

• L0-L4 Skill Classification — Every skill categorized by abstraction level
• 208-Skill Inventory — A fully cataloged, searchable skill database
• Quality Gates — Automated compliance checks before any skill goes live
• YAML Templates — Copy-paste starters for L1 Base, L2 Gateway, and L3 Ceiling skills
• Python Tooling — Lint, scan, and backfill tools for skill lifecycle management
• 4 Industry Verticals — Financial, telecom, healthcare, and government blueprints

Architecture

L0: Foundation Layer     →  Shared utilities, DB connectors, API clients
L1: Base Skill Layer     →  Single-function atomic skills (info-extractor, data-analyst)
L2: Gateway/Router Layer →  Intent routing, NL2SQL, cross-channel dispatching
L3: Scenario Layer       →  Business-scenario composites (fraud detection, SLA monitoring)
L4: Multi-Agent Layer    →  Agent-team orchestration, multi-phase pipelines

Quick Start

1. Use the Templates

# Create a new L1 Base skill
cp templates/skill-template-l1-base.yaml my-skill.yaml

# Create a new L2 Gateway skill
cp templates/skill-template-l2-gateway.yaml my-gateway.yaml

# Create a new L3 Ceiling skill
cp templates/skill-template-l3-ceiling.yaml my-scenario.yaml

2. Run the Tools

# Scan your skill directory and build inventory
python skill-architecture/scripts/inventory-scan.py --dir ./skills

# Lint skill YAML files for compliance
python skill-architecture/scripts/skill-lint.py --dir ./skills

# Backfill missing frontmatter in skill files
python skill-architecture/scripts/backfill-frontmatter.py --dir ./skills

3. Quality Gate Checklist

Use templates/audit-checklist.md to verify your skills meet governance standards before deployment.

Directory Structure

skill-framework/
├── skill-architecture/
│   ├── SKILL-ARCHITECTURE-v1.0.md      # Full architecture specification
│   ├── unified_skill_inventory.json     # 208-skill master inventory (JSON)
│   ├── unified_skill_inventory.csv      # 208-skill master inventory (CSV)
│   ├── skills_inventory.json            # Detailed skill catalog
│   ├── skills_detailed.json             # Per-skill metadata
│   ├── skills_dependencies.json         # Inter-skill dependency graph
│   ├── governance/                      # Quality gates, IM specs, remediation plans
│   ├── l2-patterns/                     # L2 pattern references (alert engine, routing, etc.)
│   ├── l3-scenarios/                    # L3 scenario skill blueprints
│   ├── l4-multi-agent/                  # L4 multi-agent orchestration patterns
│   ├── verticals/                       # Industry vertical blueprints (4 industries)
│   ├── roadmap/                         # Skill development roadmap
│   └── scripts/                         # Python tools (lint, scan, backfill)
├── templates/
│   ├── skill-template-l1-base.yaml      # L1 Base skill template
│   ├── skill-template-l2-gateway.yaml   # L2 Gateway skill template
│   ├── skill-template-l3-ceiling.yaml   # L3 Ceiling skill template
│   └── audit-checklist.md              # Quality gate checklist
└── README.md

Skill Classification Model

Level	Name	Scope	Example
L0	Foundation	Shared infra	DB connector, API client
L1	Base Skill	Single function	Info-extractor, data-analyst
L2	Gateway	Routing/orchestration	NL2SQL gateway, cross-channel router
L3	Scenario	Business composite	Fraud detection pipeline
L4	Multi-Agent	Agent team	5-phase scoring engine

Ecosystem

This framework is part of the Agent Skill Ecosystem:

Repo	Description	Skills
financial-ai-skills	104 financial AI skills	104
teleagent-skills	General business skills	5
agent-cluster-comm	Agent cluster communication	5
skill-framework (this repo)	Governance framework + templates	208 cataloged

Contributing

1. Fork this repository
2. Create your feature branch
3. Follow the L0-L4 classification when adding new skills
4. Use the provided YAML templates
5. Run skill-lint.py before submitting
6. Submit a pull request

License

MIT License — use freely in personal and commercial projects.