An upgraded skill for building, testing, and iterating on Claude skills — based on Anthropic's Complete Guide to Building Skills for Claude.

better-skill-creator walks you through the entire skill development lifecycle: planning, writing, testing, iterating, validating, and packaging. Instead of re-reading documentation every time you build a skill, this skill embeds the best practices directly into the workflow.

It handles all three skill categories:

• Document & Asset Creation — skills that produce consistent outputs (docs, presentations, code)
• Workflow Automation — multi-step processes with validation gates
• MCP Enhancement — workflow guidance layered on top of MCP server integrations

This is a derivative of Anthropic's built-in skill-creator. Key upgrades:

1. Download the latest release or clone this repo
2. Zip the better-skill-creator/ folder
3. Go to Settings → Capabilities → Skills
4. Click Upload skill and select the zip

Place the better-skill-creator/ folder in your skills directory.

Start a conversation with any of these:

Help me build a skill for generating weekly reports from CSV data 

I have a Notion MCP connected — help me create a skill for project setup workflows 

Review this skill and suggest improvements 

Use the better-skill-creator to turn this conversation into a skill 

The skill will guide you through the appropriate phases based on where you are in the process. Building a skill from scratch? It starts with planning. Already have a draft? It jumps straight to testing.

better-skill-creator/ ├── SKILL.md # Main instructions (7-phase workflow) ├── LICENSE.md # Apache 2.0 with attribution ├── references/ │ ├── categories-and-patterns.md # 3 categories + 5 design patterns │ ├── troubleshooting.md # Diagnosis flowchart for common issues │ └── schemas.md # JSON schemas for evals, grading, benchmarks ├── scripts/ │ ├── quick_validate.py # Comprehensive skill validator │ ├── run_eval.py # Run trigger evaluations │ ├── run_loop.py # Description optimization loop │ ├── aggregate_benchmark.py # Aggregate benchmark results │ ├── generate_report.py # Generate reports │ ├── improve_description.py # Description improvement suggestions │ ├── package_skill.py # Package skill as .skill file │ └── utils.py # Shared utilities ├── agents/ │ ├── grader.md # Assertion evaluation agent │ ├── comparator.md # Blind A/B comparison agent │ └── analyzer.md # Benchmark analysis agent ├── assets/ │ └── eval_review.html # Trigger eval review template └── eval-viewer/ ├── generate_review.py # Launch interactive result viewer └── viewer.html # Viewer frontend 

1. Planning — Classify the skill category, define 2–3 use cases, set success criteria
2. Writing — Pick a design pattern, write frontmatter and instructions following progressive disclosure
3. Testing — Run test cases with and without the skill, draft assertions
4. Improving — Read feedback, generalize fixes, keep instructions lean
5. Validation — Run the pre-flight checklist and automated validator
6. Description optimization — Generate trigger evals, run the optimization loop
7. Packaging — Bundle and deliver the .skill file

The included validator catches issues the original missed:

python scripts/quick_validate.py path/to/your-skill/ # Example output: # ✅ Skill is valid! # # WARNINGS (1): # ⚠ Description doesn't include trigger conditions. # Add 'Use when...' phrases so Claude knows when to activate the skill.

Checks include: SKILL.md naming, YAML format, kebab-case enforcement, reserved name prefixes (claude/anthropic), angle bracket detection, description quality (trigger phrases, vagueness, length), folder-name consistency, README detection, and reference file TOC coverage.

Apache 2.0 — see LICENSE.md for details.