Agent Skills: Harness Init Skill

プロジェクトをHarness仕様に初期化。開発の準備体操はこちらで。Use when user mentions '/harness-init', new project setup, launch a project, or introduce harness to existing project. Do NOT load for: implementation, reviews, or planning.

UncategorizedID: Chachamaru127/claude-code-harness/harness-init

Install this agent skill to your local

pnpm dlx add-skill https://github.com/Chachamaru127/claude-code-harness/tree/HEAD/skills/harness-init

Skill Files

Browse the full folder contents for harness-init.

Download Skill

Loading file tree…

skills/harness-init/SKILL.md

Skill Metadata

Name
harness-init
Description
"プロジェクトをHarness仕様に初期化。開発の準備体操はこちらで。Use when user mentions '/harness-init', new project setup, launch a project, or introduce harness to existing project. Do NOT load for: implementation, reviews, or planning."

Harness Init Skill

Sets up a project so VibeCoder can start development with natural language only. Completes with minimum 1 question, ready to start development immediately.

Quick Reference

  • "Want to launch a new project fastest" → this skill
  • "Leave it to you" / "Quickly" → proceed with defaults, no questions
  • "With Next.js + Supabase" → technology specification
  • "Introduce harness to existing project" → analyze existing code

Deliverables

  • Real project generation (e.g., create-next-app) + initial setup
  • Prepare Plans.md / AGENTS.md / CLAUDE.md / .claude/
  • Optional: Codex CLI compatibility (.codex/ + AGENTS.md)
  • Environment diagnosisSSOT initializationFinal validation
  • Ready to run Plan→Work→Review immediately

Usage

/harness-init                              # Interactive setup (min 1 question)
/harness-init blog --mode=solo             # With partial specification
/harness-init --stack=next-supabase        # Specify tech stack

Feature Details

| Feature | Reference | |---------|-----------| | Execution Flow | See references/execution-flow.md | | Smart Defaults | See references/smart-defaults.md | | Generated Files | See references/generated-files.md |

Optimized Flow

Before: Up to 11 dialogue rounds After: Minimum 1, maximum 2 dialogue rounds

Step 1: Unified question (1 time)
  ├─ What are you building?
  ├─ Who will use it?
  └─ Leave it to me or detailed settings?

Step 2: Confirmation (skip if "leave it to me")
  └─ Tech stack + project name

→ Execute setup (includes background analysis)

Step 3: Completion report

Arguments

| Argument | Description | Example | |----------|-------------|---------| | [project description] | What to build | "EC site" | | --mode | solo / 2agent | --mode=solo | | --stack | Tech stack | --stack=next-supabase | | --name | Project name | --name=my-app | | --lang | Language | --lang=en |

Smart Defaults

| Item | Default | Auto-determination | |------|---------|-------------------| | Language | ja | Config file | | Mode | Solo | 2-Agent if .cursor/ exists | | Tech stack | next-supabase | In auto mode | | Skills Gate | Auto-configured | Adjust with /skills-update |

Next Steps

After setup:

  • "/plan-with-agent I want to create XXX" → Create plan
  • "/work" → Execute tasks in Plans.md
  • "npm run dev" → Start dev server
  • If you need Codex CLI: run /setup codex