Hissal Agent Plugins Repo Setup

Repo Setup Skill

A guided, interactive skill for setting up a new project repository from scratch.

Overview

This skill walks the user through a series of upfront questions, then creates all files on disk and displays a text summary of the structure. It always produces the full set of standard files and wires up git (with optional remote and CI/CD).

Step 1 — Ask Questions Upfront

Before creating anything, ask the user ALL of the following. You can use the ask_user_input_v0 tool to make this interactive, grouping related questions together (max 3 per call).

Round 1 — Project basics

1. Project name — what's the repo/folder called?
2. Short description — one sentence describing what it does
3. Primary language / stack — e.g. Python, Node.js, Go, Rust, mixed, etc.

Round 2 — Structure & remote

4. Folder structure — should Claude suggest a structure based on the stack, or does the user have something specific in mind?
5. Git remote — do they have a remote URL (GitHub/GitLab/etc.) to connect to?
6. License — MIT, Apache 2.0, GPL, proprietary/none, or ask later?

Round 3 — Optional extras

7. CI/CD — do they want a basic CI/CD workflow file (e.g. GitHub Actions)?
   • If yes: what should it do? (lint, test, build, deploy, or a combo)
8. Any additional context — anything else to note in CODEBASE.md (team conventions, external services, special constraints)?

Only proceed to Step 2 once you have answers to all required fields (name, description, stack). The rest can use sensible defaults if the user skips them.

Step 2 — Show a Plan

Before creating files, print a clear summary:

📁 Project: <name>
📝 Description: <description>
🛠  Stack: <stack>

Files to be created:
  <project-name>/
  ├── README.md
  ├── .gitignore
  ├── .editorconfig
  ├── CLAUDE.md
  ├── AGENTS.md
  ├── llm-guidelines.md
  ├── CODEBASE.md
  ├── CONVENTIONS.md
  ├── <suggested folder structure>
  └── [.github/workflows/ci.yml — if CI/CD selected]

Git: init locally [+ remote: <url> if provided]

Ask: "Looks good? I'll create everything now." — wait for confirmation before proceeding.

Step 3 — Create Files on Disk

Use bash_tool to create the project. Work directory: wherever the user's session is, or /home/claude/<project-name>/ if no path specified.

3a — Folder structure

Create the root folder and any subdirectories appropriate for the stack. See references/folder-structures.md for stack-specific templates.

Stack presets — some stacks have dedicated reference files with extra setup steps:

• Unity → read references/unity-preset.md in full before creating any files. The root .gitignore goes in the Unity project root (next to Assets/). The three _ folders are created inside Assets/. Do not create a new root folder.
• C# / .NET → read references/dotnet-preset.md. Ask the user which project type they need (Console, Web API, or Class Library) before scaffolding, then use the matching template.

3b — Standard config files

README.md

# <project-name>

<description>

## Getting Started

_TODO: Add setup instructions._

## Architecture

See [CODEBASE.md](./CODEBASE.md) for a full architecture overview.

## AI / Agent Guidelines

See [CLAUDE.md](./CLAUDE.md) for AI assistant context and conventions.

.gitignore

Generate an appropriate .gitignore for the stack. For generic/unknown stacks, include:

• OS files: .DS_Store, Thumbs.db
• Editor: .vscode/, .idea/, *.swp
• Env: .env, .env.*, !.env.example
• Build: dist/, build/, out/, *.log

.editorconfig

root = true

[*]
indent_style = space
indent_size = 2
end_of_line = lf
charset = utf-8
trim_trailing_whitespace = true
insert_final_newline = true

[*.py]
indent_size = 4

[*.md]
trim_trailing_whitespace = false

[Makefile]
indent_style = tab

3c — AI Context Files

These four files form the AI/agent context layer of the project.

CLAUDE.md and AGENTS.md (mirrors — always identical)

Both files must have identical content. Keep them in sync — they serve different agents (some read CLAUDE.md, others prefer AGENTS.md).

# AI Agent Context — <project-name>

This file provides context for AI assistants and coding agents working in this repository.
It is mirrored as both `CLAUDE.md` and `AGENTS.md` — keep them in sync.

## Project Overview

<description>

**Stack:** <stack>

## Key Files

| File | Purpose |
|------|---------|
| `CODEBASE.md` | Architecture, folder map, core concepts, key types |
| `llm-guidelines.md` | Behavioral guidelines for LLMs (when to ask, how to scope changes, avoid over-engineering) |
| `CONVENTIONS.md` | Code style, naming, commits, PR process |

## Quick Reference

- Read `CODEBASE.md` before making structural changes
- Follow the conventions in `llm-guidelines.md`
- When in doubt, ask before refactoring

## Notes

_Add project-specific conventions, gotchas, or constraints here._

llm-guidelines.md

This file contains behavioral guidelines for LLMs — how to think, when to ask, how to scope changes. It is not a style guide or conventions doc (put those in CODEBASE.md).

Important: Before writing this file, check if the karpathy-guidelines skill is available:

find /mnt/skills -name "*karpathy*" 2>/dev/null

• If found: Read that skill's content and embed it as the body of llm-guidelines.md
• If not found: Use the verbatim content from references/llm-guidelines-default.md — do not summarize or rewrite it, copy it as-is

Wrap the content with this header:

# LLM Guidelines — <project-name>

Behavioral guidelines for AI assistants working in this codebase.
Derived from Andrej Karpathy's observations on common LLM coding pitfalls.

---

<content from karpathy-guidelines skill OR references/llm-guidelines-default.md>

CODEBASE.md

# Codebase Overview — <project-name>

> This file is a living document. Update it as the project evolves.

## Description

<description>

## Stack

<stack>

## Folder Map

/ ├── (fill in as project grows)

## Core Concepts

_TODO: Describe the main abstractions and mental model of this codebase._

## Key Types / Interfaces

_TODO: List the most important data structures, types, or interfaces._

## Patterns & Conventions

_TODO: Note recurring patterns, naming conventions, and architectural decisions._

## External Services & Dependencies

_TODO: List external APIs, databases, or services this project depends on._

## Known Gotchas

_TODO: Document any non-obvious behavior or tricky areas._

CONVENTIONS.md

A concise, scannable reference for how code is written in this project. Keep it short — if it can't be skimmed in 2 minutes, it won't be read. Use the template from references/conventions-template.md, filling in the stack-specific sections based on the user's answers. Leave sections as _TODO_ if not yet decided — better to be honest than prescriptive.

3d — CI/CD (optional)

If the user opted in, create .github/workflows/ci.yml. See references/ci-templates.md for templates by stack and workflow type.

3e — Git setup

cd <project-dir>
git init
git add .
git commit -m "chore: initial project setup"

If a remote URL was provided:

git remote add origin <url>
git branch -M main
git push -u origin main

Step 4 — Show Summary

After all files are created, print:

1. The full folder tree (tree command or manual listing)
2. A short text recap of what was created
3. Any next steps (e.g. "Fill in CODEBASE.md as you build", "Add your remote URL", etc.)

Also call present_files to surface the root folder to the user.

Reference Files

• references/folder-structures.md — Stack-specific folder templates (Python, Node, Go, etc.). For Unity and C#/.NET, see their preset files
• references/unity-preset.md — Full Unity game dev preset: folder structure, official .gitignore, CODEBASE.md section
• references/dotnet-preset.md — C# / .NET preset: Console, Web API, and Class Library templates with .gitignore and .editorconfig additions
• references/conventions-template.md — CONVENTIONS.md template (fill in stack-specific sections at setup time)
• references/llm-guidelines-default.md — Default LLM guidelines (used when karpathy-guidelines is not installed)
• references/ci-templates.md — GitHub Actions workflow templates by stack/purpose

Source: Hissal/agent-plugins — distributed by TomeVault.