claude-code-ultimate-guide/examples/skills/skill-creator/SKILL.md
Alan Pope be52e232b3
feat: improve skill scores across 19 skills
Hullo @FlorianBruniaux 👋

I ran your skills through `tessl skill review` at work and found some targeted improvements. Here's the full before/after:

| Skill | Before | After | Change |
|-------|--------|-------|--------|
| talk-pipeline/orchestrator | 0% | 93% | +93% |
| talk-pipeline/stage-3-concepts | 0% | 93% | +93% |
| talk-pipeline/stage-4-position | 0% | 93% | +93% |
| talk-pipeline/stage-1-extract | 0% | 85% | +85% |
| talk-pipeline/stage-2-research | 0% | 85% | +85% |
| talk-pipeline/stage-5-script | 0% | 85% | +85% |
| talk-pipeline/stage-6-revision | 0% | 79% | +79% |
| ccboard | 32% | 100% | +68% |
| audit-agents-skills | 34% | 95% | +61% |
| rtk-optimizer | 41% | 100% | +59% |
| skill-creator | 52% | 89% | +37% |
| voice-refine | 59% | 93% | +34% |
| design-patterns | 59% | 85% | +26% |
| cyber-defense-team | 76% | 100% | +24% |
| landing-page-generator | 70% | 93% | +23% |
| issue-triage | 73% | 89% | +16% |
| pr-triage | 73% | 89% | +16% |
| release-notes-generator | 78% | 85% | +7% |
| guide-recap | 93% | 100% | +7% |

**Average: 11% → 91% (+80%)**

<details>
<summary>Changes summary</summary>

### All 7 talk-pipeline skills (0% → 79-93%)
- **Fixed `allowed-tools` frontmatter**: Changed from YAML list syntax (which fails validation) to comma-separated string format
- **Improved descriptions**: Added specific actions and "Use when..." clauses to all pipeline stage descriptions

### ccboard (32% → 100%)
- Rewrote description with concrete actions and "Use when..." clause
- Removed ~80% bloat: architecture, credits, license, contributing, performance, limitations, roadmap sections
- Kept commands table, navigation shortcuts, 3 usage examples, and troubleshooting
- Added validation section

### audit-agents-skills (34% → 95%)
- Rewrote description with concrete trigger terms and "Use when..." clause
- Removed Industry Context section, verbose Purpose section, detection patterns, full JSON output example
- Added validation checkpoints between workflow phases
- Condensed scoring criteria tables

### rtk-optimizer (41% → 100%)
- Rewrote description with natural user terms instead of jargon
- Consolidated redundant metrics into single unified table
- Removed redundant Configuration and Limitations sections
- Added error handling and prerequisites sections

### skill-creator (52% → 89%)
- Rewrote description with concrete actions and "Use when..." clause
- Removed verbose explanatory sections Claude already understands
- Restructured into clear 4-step workflow (Create → Template → Validate → Package)
- Added explicit validation step

### voice-refine (59% → 93%)
- Added "Use when..." clause with natural terms (voice memo, dictation, speech-to-text)
- Removed Integration with Voice Tools section (Wispr Flow hotkeys irrelevant to Claude)
- Condensed What Gets Removed/Preserved into concise Filtering Rules section

### design-patterns (59% → 85%)
- Added "Use when..." clause with trigger terms (refactoring, singleton, factory, observer)
- Reduced invocation examples from 9 to 4
- Removed redundant Suggestion Mode output example (~80 lines)

### cyber-defense-team (76% → 100%)
- Added "Use when..." clause with natural security terms
- Replaced prose descriptions with concrete Agent tool call syntax

### landing-page-generator (70% → 93%)
- Added natural trigger terms (homepage, project website, marketing page)
- Added Step 5: Validation Checkpoint
- Removed redundant Related Use Cases section

### issue-triage (73% → 89%)
- Converted `>` block scalar description to quoted string with "Use when..." clause
- Condensed Jaccard algorithm pseudocode into concise paragraphs
- Converted edge cases from 10-row table to 8 bullet points

### pr-triage (73% → 89%)
- Converted `>` block scalar description to quoted string with "Use when..." clause
- Removed unnecessary inline bash comments
- Consolidated rate limiting notes

### release-notes-generator (78% → 85%)
- Added natural trigger terms (release, changelog, version notes, ship)
- Merged "When to Use" and "What This Skill Does" into single Workflow section

### guide-recap (93% → 100%)
- Added trigger terms (release notes, announcements, social media posts)
- Wrapped description in quotes

</details>

Honest disclosure — I work at @tesslio where we build tooling around skills like these. Not a pitch - just saw room for improvement and wanted to contribute.

Want to self-improve your skills? Just point your agent (Claude Code, Codex, etc.) at [this Tessl guide](https://docs.tessl.io/evaluate/optimize-a-skill-using-best-practices) and ask it to optimize your skill. Ping me - [@popey](https://github.com/popey) - if you hit any snags.

Thanks in advance 🙏
2026-03-17 16:27:02 +00:00

3.6 KiB

name description tags
skill-creator Scaffold a new Claude Code skill with SKILL.md, frontmatter, and bundled resources. Use when creating a custom skill, standardizing skill structure across a team, or packaging a skill for distribution.
meta
skill
generator
claude-code

Skill Creator

Generate new Claude Code skills with correct directory structure, YAML frontmatter, and optional bundled resources.

When to Use

  • Creating a new custom skill for a project
  • Standardizing skill structure across a team
  • Generating skill templates with scripts, references, and assets
  • Packaging skills for distribution

Skill Directory Structure

skill-name/
├── SKILL.md          # Required: Main skill file with YAML frontmatter
├── scripts/          # Optional: Executable code for deterministic tasks
├── references/       # Optional: Documentation loaded contextually
└── assets/           # Optional: Templates, images, boilerplate (not loaded into context)

Workflow

1. Create the Skill

Create a new skill called "my-skill-name" in ~/.claude/skills/

Or with a specific purpose:

Create a skill for generating release notes from git commits,
with templates for CHANGELOG.md and Slack announcements

Or via the initialization script:

python3 ~/.claude/skills/skill-creator/scripts/init_skill.py <skill-name> --path <output-directory>

2. Generated SKILL.md Template

The created SKILL.md follows this structure:

---
name: skill-name
description: "What the skill does. Use when [trigger conditions]."
---

# Skill Name

## When to Use
- Trigger condition 1
- Trigger condition 2

## What This Skill Does
1. **Step 1**: Description
2. **Step 2**: Description

## How to Use
[Usage examples]

## Example
**User**: "Example prompt"
**Output**: [Example output]

3. Validate the Skill

After creation, verify:

  1. Frontmatter: name is kebab-case, 1-64 chars; description is a quoted string with "Use when" clause
  2. Content: Has "When to Use" section with trigger conditions and at least one usage example
  3. Structure: SKILL.md is under 5000 words; references and assets are in correct subdirectories
  4. Test: Invoke the skill with a real use case and confirm expected output

4. Package for Distribution (Optional)

python3 ~/.claude/skills/skill-creator/scripts/package_skill.py <path/to/skill-folder> [output-directory]

Organizational Patterns

Pattern Best For Structure
Workflow-Based Sequential procedures Step-by-step instructions
Task-Based Multiple operations Collection of tasks
Reference/Guidelines Standards, specs Rules and examples
Capabilities-Based Interrelated features Feature descriptions

Example: Creating a Release Notes Skill

User: "Create a skill for generating release notes with 3 output formats"

Steps:

  1. Initialize: init_skill.py release-notes-generator --path ~/.claude/skills/
  2. Add templates to assets/: changelog-template.md, pr-release-template.md, slack-template.md
  3. Add rules to references/: tech-to-product-mappings.md
  4. Complete SKILL.md with usage instructions
  5. Validate: check frontmatter, test with a real commit range
  6. Package: package_skill.py ~/.claude/skills/release-notes-generator

Tips

  • Keep SKILL.md under 5000 words for efficient context usage
  • Use references/ for domain knowledge that doesn't change often
  • Put templates in assets/ so they're not auto-loaded into context
  • Always include a "Use when" clause in the description frontmatter
  • Test with real use cases before packaging