playbooks

home / skills / chachamaru127 / claude-code-harness / harness-init

harness-init skill

/skills/harness-init

This skill initializes a project for harness setup using natural language, delivering ready-to-run plans and defaults for immediate development.

npx playbooks add skill chachamaru127/claude-code-harness --skill harness-init

Review the files below or copy the command above to add this skill to your agents.

Files (4)
SKILL.md
3.6 KB
---
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."
description-en: "Initialize project to Harness spec. Development warm-up here. 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."
description-ja: "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."
allowed-tools: ["Read", "Write", "Edit", "Grep", "Glob", "Bash", "Task"]
argument-hint: "[project-name] [--mode=solo|2agent] [--stack=next-supabase] [--name=app-name]"
disable-model-invocation: true
---

# 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 diagnosis** → **SSOT initialization** → **Final validation**
- → **Ready to run Plan→Work→Review immediately**

## Usage

```bash
/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](references/execution-flow.md) |
| **Smart Defaults** | See [references/smart-defaults.md](references/smart-defaults.md) |
| **Generated Files** | See [references/generated-files.md](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`

Overview

This skill initializes a project to the Harness specification so development can begin with natural language only. It prepares a runnable workspace, generates initial files (Plans.md, AGENTS.md, CLAUDE.md), and performs an environment diagnosis and SSOT initialization. The setup completes with a minimum of one interactive question and yields a ready Plan→Work→Review cycle.

How this skill works

The skill asks a single unified question to capture project intent, audience, and whether to use smart defaults or explicit configuration. It can generate a new project (real scaffolding like create-next-app), introduce the harness into an existing repo by analyzing source code, or apply a specified tech stack such as next + supabase. After generation it runs environment checks, initializes single source of truth files, and returns a completion report with next steps.

When to use it

  • When you type /harness-init to start a new project quickly
  • To introduce the harness into an existing codebase and auto-configure agents
  • When you want immediate defaults and minimal interaction to launch development
  • If you need a reproducible Plan→Work→Review workflow ready to run
  • When specifying stack or mode (e.g., --stack=next-supabase or --mode=solo)

Best practices

  • Prefer the single-question flow for fastest setup; use 'leave it to you' to skip confirmation
  • Specify --stack and --name when you need deterministic scaffolding and repo names
  • Run the skill from the project root so existing files can be analyzed and preserved
  • Review the generated Plans.md and AGENTS.md before running automated work commands
  • Use the optional Codex CLI setup (/setup codex) only if you need Codex compatibility

Example use cases

  • Launch a new Next.js + Supabase app with harness scaffolding using --stack=next-supabase
  • Introduce harness to an existing JavaScript project to enable autonomous Plan→Work→Review cycles
  • Quickly scaffold a solo-mode project with smart defaults by saying 'leave it to you'
  • Initialize SSOT and agent configuration to hand off to automated agents or human developers
  • Prepare a repository for immediate /plan-with-agent and /work commands after setup

FAQ

How many questions will the skill ask?

Typically one unified question; a second confirmation appears only if you opted out of defaults.

Can it analyze an existing project?

Yes. When run from a repo, it inspects code to integrate the harness without overwriting key files.

What stacks are supported by default?

Smart defaults use next-supabase; you can override with --stack or a detailed response.