Skip to content

Latest commit

Β 

History

39 Commits

Folders and files

NameName
Last commit message
Last commit date
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 
Β 

Repository files navigation

🧠 AI Skill Maker

Build reliable AI skills from rough intent, repo evidence, and repeatable workflows.

License: Apache-2.0 Runtime: Node.js 18+ Skill: Codex Validation: self-check

ζ—₯本θͺž


AI Skill Maker is a meta-skill for designing, rendering, validating, and refreshing reusable AI skills and assistant instruction bundles.

It turns rough intent, a repeatable workflow, or an existing repository into a structured skill folder with evidence labels, validation scripts, release gates, and refresh-safe user rule blocks.

node scripts/self-check.mjs
node scripts/install-local-skill.mjs --dry-run

πŸ“– Contents

# Section What You Get
1 Why It Exists The problem this project solves
2 What It Builds Supported skill and adapter outputs
3 Quick Start Copy-paste commands to validate and render
4 Workflow The generation pipeline
5 Generated Layout What rendered skills contain
6 Command Reference Script-by-script usage
7 Validation & Release Required checks before shipping
8 Safety Model Trust gates and protected principles

✨ Why It Exists

Reusable AI skills are easy to draft and hard to keep reliable. A useful skill needs a narrow trigger, clear boundaries, runnable workflow, verifiable output, and enough structure that future agents can refresh it without erasing maintainer intent.

AI Skill Maker provides that structure. It helps an agent settle the root problem, collect evidence, choose the right output mode, render the skill from a strict config, and verify the result before it is treated as ready.

The goal is not a longer prompt. The goal is a smaller, sharper capability that can be reused, tested, refreshed, and trusted.


🧩 What It Builds

Output Best For Primary Renderer
πŸ› οΈ Functional skill Reports, PDFs, spreadsheets, API tasks, browser workflows, data processing scripts/render-skill.mjs
πŸ“„ Document/template skill Artifact structure, style, reusable assets, rendered fidelity scripts/render-skill.mjs
πŸ” Workflow automation skill Repeatable procedures through tools, CLIs, APIs, or scripts scripts/render-skill.mjs
🧭 Project maintainer skill Long-lived AI guidance for a specific repository scripts/render-project-skill.mjs
πŸ”Œ Adapter instruction bundle AGENTS.md, CLAUDE.md, Cursor rules, Copilot instructions scripts/render-adapter.mjs

πŸ›οΈ Design Pillars

Pillar What It Means
🎯 Root problem first Design around the recurring need, not the user's first phrasing
🧾 Evidence discipline Label claims as observed_fact, declared_intent, recommended_standard, or inferred_assumption
πŸ›‘οΈ Trust Gate Require explicit PASS/BLOCK review for permissions, sensitive data, dependencies, environment, external actions, and rollback
πŸ§ͺ Evaluation skeleton Ship trigger tests, output assertions, and BLOCK/ALLOW release gates that must be filled with real evidence
πŸ”’ Refresh safety Preserve user-authored rule blocks and protected core blocks
πŸ“ File budget Automatically check active Markdown instruction files against the 9,000-token ceiling

πŸš€ Quick Start

Requires Node.js 18+.

git clone https://github.com/sscodeai/ai-skill-maker.git
cd ai-skill-maker
node scripts/self-check.mjs

⚑ Create A Functional Skill

node scripts/render-skill.mjs --init-config functional > config.json
node scripts/validate-skill-config.mjs --input config.json --mode functional --strict
node scripts/render-skill.mjs --input config.json --output ./generated-skill --mode functional --strict
node scripts/validate-skill-output.mjs ./generated-skill
node scripts/file-budget.mjs ./generated-skill

🧭 Create A Project Maintainer Skill

node scripts/draft-project-config.mjs --repo . > project-config.json
node scripts/validate-config.mjs --input project-config.json --mode repo --strict
node scripts/render-project-skill.mjs --input project-config.json --output ./project-maintainer --mode repo --strict
node scripts/validate-project-skill.mjs ./project-maintainer

🧠 Install As A Codex Skill

By default, installation syncs the runtime payload into the local ai-skill-maker personal skill directory. If that directory already exists and is not this source checkout, it is replaced with the current payload: SKILL.md, agents/, references/, assets/, and scripts/.

Preview the target and replacement behavior without writing files:

node scripts/install-local-skill.mjs --dry-run

Install or sync the local skill:

node scripts/install-local-skill.mjs

🧭 Workflow

flowchart LR
  A["Root Problem<br/><small>smallest useful capability</small>"]
  B{"Trust Gate<br/><small>PASS / BLOCK</small>"}
  C["Evidence & Intent<br/><small>facts, intent, standards, assumptions</small>"]
  D["Strict Config<br/><small>schema-validated JSON</small>"]
  E["Render<br/><small>skill or instruction bundle</small>"]
  F["Validate<br/><small>structure, metadata, budget</small>"]
  G{"Release Gate<br/><small>ALLOW / BLOCK</small>"}

  A --> B
  B -->|PASS| C
  B -.->|BLOCK| H["Resolve risk<br/><small>permissions, secrets, dependencies</small>"]
  H -.-> B
  C --> D --> E --> F --> G
  G -->|ALLOW| I["Ready to use"]
  G -.->|BLOCK| J["Add evidence<br/><small>tests, assertions, license, trust</small>"]
  J -.-> F
Loading
Step Gate Output
1 Identify the smallest useful capability One-sentence root problem
2 Check permissions, secrets, dependencies, environment, external actions, rollback Trust Gate PASS/BLOCK
3 Separate facts from intent and assumptions Evidence-labeled config
4 Select mode and adapter Renderer choice
5 Validate before rendering Strict config check
6 Render or refresh Skill folder or instruction bundle
7 Validate output Structure, metadata, evidence labels, ledgers, user-rule markers, file budget
8 Release only with recorded evidence Trigger tests, output assertions, structure validation, file budget, trust, license attribution

πŸ—οΈ Generated Layout

General Skill

generated-skill/
  SKILL.md
  agents/
    openai.yaml
  references/
    skill-intent.md
    workflows.md
    resources.md
    verification.md
    generated-files.md
    evals/
      trigger-tests.md
      output-assertions.md
      release-gate.md
  scripts/
    health-check.mjs

Project Maintainer Skill

project-maintainer/
  SKILL.md
  agents/
    openai.yaml
  references/
    project-intent.md
    project-map.md
    architecture.md
    coding-standards.md
    content-style.md
    workflows.md
    verification.md
    release.md
    generated-files.md
    evals/
      trigger-tests.md
      output-assertions.md
      release-gate.md
  scripts/
    health-check.mjs

πŸ—‚οΈ Repository Layout

SKILL.md                         # maker runtime instructions
agents/openai.yaml               # Codex/OpenAI skill metadata
assets/examples/                 # starter configs for every render mode
assets/templates/skill/          # general skill template
assets/templates/project-skill/  # project maintainer skill template
references/                      # modes, schemas, adapters, checklists, rules, evals
scripts/                         # renderers, validators, repo scanners, guardrails

πŸ› οΈ Command Reference

Command Purpose
node scripts/render-skill.mjs --init-config <mode> Print a starter config for functional, document, workflow, or refresh mode
node scripts/validate-skill-config.mjs --input config.json --strict Validate a general skill config before rendering
node scripts/render-skill.mjs --input config.json --output <dir> --strict Render or refresh a general skill
node scripts/validate-skill-output.mjs <dir> Validate a rendered general skill folder
node scripts/draft-project-config.mjs --repo <repo> Draft a project maintainer config from repository signals
node scripts/validate-config.mjs --input config.json --mode genesis|repo --strict Validate a project maintainer config
node scripts/render-project-skill.mjs --input config.json --output <dir> --strict Render a project maintainer skill
node scripts/validate-project-skill.mjs <dir> Validate a rendered project maintainer skill
node scripts/render-adapter.mjs --input config.json --adapter agents|claude|cursor|copilot --output <path> Render assistant instruction files while preserving generated blocks
node scripts/check-core-principles.mjs Verify the protected core principle fingerprint
node scripts/file-budget.mjs [skill-dir] Enforce the 9,000-token ceiling for active Markdown instruction files
node scripts/check-release-gate.mjs <skill-dir> Check whether release-gate evidence has been recorded for required gates
node scripts/self-check.mjs Run the full repository health check
node scripts/install-local-skill.mjs [--dry-run] Preview or sync this maker into the local Codex personal skills directory

βœ… Validation & Release

Run these checks before merging, publishing, or installing a changed maker:

node scripts/self-check.mjs
node scripts/check-core-principles.mjs
node scripts/file-budget.mjs
git diff --check

Run this when you also want to compare the repository payload with the installed local skill:

node scripts/self-check.mjs --check-installed

Generated skills include references/evals/release-gate.md. Treat the gate as BLOCK/ALLOW only: trigger tests, output assertions, structure validation, file budget, trust, and license attribution must all have recorded evidence before a skill is described as release-ready.

The validators and guardrail scripts are intentionally scoped. They can check required files, YAML metadata, evidence labels, release-gate evidence cells, file budget, and user-rule preservation markers. They do not prove that a skill is absolutely safe, complete, or ready to publish; that still requires real trigger-test results, output assertions, Trust Gate review, and license evidence recorded in the release gate.


πŸ›‘οΈ Safety Model

AI Skill Maker is designed to produce instruction assets, not to perform unbounded external actions. Generated skills should require explicit user permission before destructive operations, credential-sensitive work, remote publishing, force-pushing, visibility changes, or other high-risk actions.

The maker's protected core principles live in references/rules/protected-core-principles.md and are locked by references/rules/core-principles.lock.json. Regenerate the lock only when the user explicitly approves changing a protected core principle.

Guardrail Enforced By
Protected core fingerprint scripts/check-core-principles.mjs
Instruction file budget scripts/file-budget.mjs
Config shape and evidence labels validate-skill-config.mjs, validate-config.mjs
Rendered output structure validate-skill-output.mjs, validate-project-skill.mjs
Recorded release-gate evidence scripts/check-release-gate.mjs
End-to-end repository health scripts/self-check.mjs

🌱 Inspiration

AI Skill Maker incorporates ideas from several open meta-skill projects:

Source Adopted Idea Implementation
CheshireMew/meta-skills Behavior constitution Protected core principles with a SHA-256 fingerprint lock
yaojingang/yao-meta-skill Evaluation and release gates references/evals/ skeleton for generated skills
gnipbao/dao-skill Root-problem thinking and Trust Gate Root-problem intake plus hard PASS/BLOCK trust checks

🀝 Contributing

Keep changes evidence-backed and easy to verify:

  • Update schemas, templates, and validators together when output shape changes.
  • Preserve <!-- BEGIN USER RULES --> blocks during refresh-related edits.
  • Keep active Markdown instruction files below the file budget.
  • Add or update scripts/self-check.mjs coverage for behavioral changes.
  • Run the validation commands in this README before opening a pull request.

πŸ“„ License

Apache-2.0

About

AI Skill Maker is a meta-skill for creating and refreshing reusable AI skills and assistant instruction bundles across platforms.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages