Skill Conductor

Plan twice, build once.

A Claude Code skill that orchestrates the complete lifecycle of skill development using a Conductor-inspired workflow: Context → Spec → Plan → Implement → Review → Publish.

What It Does

• Socratic questioning — Guides you through creating project context and skill specifications
• Structured workflow — Every change goes through spec → plan → implement → review → publish
• Automated validation — Checks SKILL.md format, trigger conflicts, and description quality
• Multi-project support — Each project maintains its own independent workspace
• Intent recognition — Understands what you mean, not just the words you use

Installation

git clone https://github.com/Yungho/skill-conductor.git ~/.claude/skills/skill-conductor

Then copy the .claude/CLAUDE.md content into your project's .claude/CLAUDE.md (merge with existing if you have one).

Usage

/skill-conductor setup          # Initialize project context (once per project)
/skill-conductor new <desc>     # Create a new development track
/skill-conductor implement      # Execute next pending task
/skill-conductor status         # Show progress overview
/skill-conductor review <id>    # Run 7-step review pipeline
/skill-conductor publish <id>   # Version + CHANGELOG + Git + GitHub
/skill-conductor validate <path> # Quick validate a SKILL.md

You don't need to use exact commands. Natural language works too:

• "I need a new skill for tracking my diet" → new
• "Can you check if my skill is formatted correctly?" → review
• "Let's ship this to GitHub" → publish

Project Structure

skill-conductor/
├── SKILL.md                    # Core instructions (83 lines, lean)
├── docs/commands/              # Detailed per-command instructions
│   ├── setup.md
│   ├── new.md
│   ├── implement.md
│   ├── status.md
│   ├── review.md
│   ├── publish.md
│   └── validate.md
├── templates/                  # File templates for workspace generation
├── scripts/                    # Validation and utility scripts
├── evals/                      # Test cases for trigger accuracy
├── README.md
└── LICENSE

Workspace

Each project maintains its own independent workspace. The workspace directory name is configurable (default: skill-conductor/).

your-project/                       another-project/
├── skill-conductor/                ├── conductor/          ← custom name
│   ├── index.md                    │   ├── index.md
│   ├── product.md                  │   ├── product.md
│   ├── guidelines.md               │   ├── guidelines.md
│   ├── workflow.md                 │   ├── workflow.md
│   ├── references.md               │   ├── references.md
│   ├── tracks.md                   │   ├── tracks.md
│   └── tracks/                     │   └── tracks/
│       └── <name>/                 │       └── <name>/
│           ├── index.md            │           ├── index.md
│           ├── spec.md             │           ├── spec.md
│           ├── plan.md             │           ├── plan.md
│           └── metadata.json       │           └── metadata.json
└── some-skill/                     └── src/
    └── SKILL.md

Set your workspace directory name in .claude/CLAUDE.md.

Design Principles

1. Intent Recognition — Semantic understanding, not keyword matching
2. Progressive Disclosure — Load context incrementally
3. Adaptive Checkpoints — Adjust verification frequency by complexity
4. External Verification — WebSearch to verify format rules (not cached memory)
5. Socratic Questioning — Ask before generating

Key Format Rules

• description must be single-line string (double quotes, no | or >)
• description ≤ 1024 characters
• name lowercase + hyphens, ≤ 64 characters
• SKILL.md under 500 lines

Inspired By

• Conductor — Gemini CLI extension for context-driven development
• Agent Skills — Open standard for AI agent skills

License

MIT — Author: Yungho