Design, scaffold, build, review, evolve, and publish production-grade Claude Code skills following the Agent Skills open standard.
- Plan — Analyze use cases, select complexity tier (1-4), design architecture with sub-skill decomposition
- Build — Scaffold complete skill file trees with SKILL.md, sub-skills, scripts, references, and agents
- Review — Audit any skill with a 0-100 health score across 6 quality categories
- Evolve — Fix triggering issues, improve instructions, refine architecture based on feedback
- Publish — Package as
.skillfiles, generate install scripts, prepare for GitHub distribution - Convert — Port skills to OpenAI Codex, Google Gemini CLI, Google Antigravity, and Cursor
- Eval — Run evaluation pipelines with assertions, grading, and multi-agent execution
- Benchmark — Measure performance with variance analysis, multiple trials, and threshold gating
git clone https://github.com/AgriciDaniel/skill-forge.git
cd skill-forge
bash install.shbash install.sh --uninstall| Command | Description |
|---|---|
/skill-forge |
Interactive skill creation wizard |
/skill-forge plan <domain> |
Architecture and design planning |
/skill-forge build <name> |
Scaffold and build a skill |
/skill-forge review <path> |
Audit an existing skill (0-100 score) |
/skill-forge evolve <path> |
Improve a skill from feedback |
/skill-forge publish <path> |
Package for distribution |
/skill-forge eval <path> |
Run eval pipeline to test skill quality |
/skill-forge benchmark <path> |
Benchmark skill with variance analysis |
/skill-forge convert <path> |
Convert to Codex, Gemini, Antigravity, or Cursor |
Create a simple skill:
/skill-forge build my-tool
Design a complex skill ecosystem:
/skill-forge plan "DevOps toolkit for Docker and Kubernetes management"
Review an existing skill:
/skill-forge review ~/.claude/skills/my-skill
Convert a skill for other platforms:
/skill-forge convert ~/.claude/skills/my-skill
Quick scaffold with the CLI script:
python skill-forge/scripts/init_skill.py devops-toolkit --tier 3 --sub docker,k8s,monitor| Tier | Name | Structure | Best For |
|---|---|---|---|
| 1 | Minimal | Single SKILL.md | Simple workflows, document generation |
| 2 | Workflow | SKILL.md + scripts | Tasks needing deterministic validation |
| 3 | Multi-Skill | Orchestrator + sub-skills | Complex domains with multiple workflows |
| 4 | Ecosystem | Full system with agents | Enterprise-grade parallel analysis |
skill-forge/ # Main orchestrator (Tier 4)
SKILL.md # Entry point and routing
references/ # On-demand knowledge (10 files)
scripts/ # Execution scripts (8 files)
assets/templates/ # Skill templates (4 tiers)
skills/
skill-forge-plan/ # Architecture planning
skill-forge-build/ # Scaffolding and generation
skill-forge-review/ # Quality auditing
skill-forge-evolve/ # Improvement and iteration
skill-forge-eval/ # Evaluation pipeline
skill-forge-benchmark/ # Performance benchmarking
skill-forge-publish/ # Distribution and packaging
skill-forge-convert/ # Multi-platform conversion
agents/
skill-forge-architect.md # Architecture design agent
skill-forge-writer.md # Content writing agent
skill-forge-validator.md # Validation agent
skill-forge-converter.md # Platform conversion agent
skill-forge-executor.md # Eval execution agent
skill-forge-grader.md # Eval grading agent
skill-forge-analyzer.md # Benchmark analysis agent
skill-forge-comparator.md # Blind A/B comparison agent
- Python 3.10+ — Required for scaffolding, validation, packaging, conversion, eval, and benchmarking scripts
- Claude Code — The CLI tool these skills are built for
- No external Python packages required (stdlib only)
- Agent Skills Standard — Open standard for AI agent skills
- 3-Layer Architecture — Directive + Orchestration + Execution
- Inspired by claude-seo and Anthropic Skills
- Issues: GitHub Issues
- Discussions: GitHub Discussions