AI Skill Create

AI Skill Create is a public-ready Codex skill and plugin package for creating better Codex skills from real briefs, examples, files, and constraints.

Build the skill, validate the structure, dry-run the install path, then package it for other Codex users.

It helps Codex design, scaffold, validate, forward-test, document, and package skills that are triggerable, concise, safe, progressive-disclosure friendly, and easy to install.

What It Builds

  • strong SKILL.md trigger descriptions
  • lean workflows with one-level references
  • deterministic helper scripts when reliability matters
  • validation and forward-test prompts
  • plugin and marketplace metadata
  • Windows-first install and dry-run scripts
  • public repo safety checks and secret scanning

Quick Start

Clone the repo:

git clone https://github.com/ucsahinn/ai-skill-create.git
cd ai-skill-create

Preview the direct skill install:

powershell.exe -NoProfile -ExecutionPolicy Bypass -File scripts/install.ps1 -DryRun

Install the skill into your Codex home:

powershell.exe -NoProfile -ExecutionPolicy Bypass -File scripts/install.ps1 -Yes

Then start a new Codex thread and invoke:

Use $ai-skill-create to create a new Codex skill from this brief.

Plugin Layout

The plugin is self-contained:

plugins/ai-skill-create/
  .codex-plugin/plugin.json
  skills/ai-skill-create/
    SKILL.md
    agents/openai.yaml
    references/
    scripts/
    assets/

The repo-local marketplace entry lives at .agents/plugins/marketplace.json.

Validate Everything

Run the full local gate:

npm run validate

The validation chain checks:

  • repository structure
  • SKILL.md frontmatter and routing terms
  • agents/openai.yaml
  • plugin manifest
  • marketplace metadata
  • markdown links
  • script syntax
  • install dry-run
  • Gitleaks secret scan

Skill Workflow

When invoked, $ai-skill-create guides Codex through:

  1. understanding concrete examples and user intent
  2. choosing instruction-only vs references/scripts/assets/plugin packaging
  3. writing a concise, trigger-strong SKILL.md
  4. keeping detailed knowledge in references
  5. creating deterministic helpers only when useful
  6. validating structure and safety
  7. forward-testing with realistic prompts
  8. preparing public repo docs and release gates

Documentation

Safety Model

Generated skills can affect agent behavior. This repo treats briefs, examples, web pages, MCP output, GitHub issues, and generated text as untrusted input.

The project avoids:

  • real secrets or private data
  • broad default permissions
  • hidden override instructions
  • unreviewed global writes
  • destructive cleanup
  • curl-pipe-shell install patterns

See SAFE_GENERATION.md, THREAT_MODEL.md, and SECURITY.md.

Turkish Docs

Turkish documentation is available in README.tr.md.

License

MIT. See LICENSE.