Repo Bootstrap

Repo Bootstrap

Creates a complete, publishable skill repository from scratch. This is the foundation skill — everything else runs after this.

When to Use

Trigger phrases (exact match):

• "set up the repo"
• "create the folder pack"
• "bootstrap the repository"
• "initialize skill repo"
• "start a new prompt pack"
• "build the structure"
• "make the skeleton"

Not a trigger if:

• The user asks to edit existing files (that's editing, not bootstrapping)
• The folder structure already exists and is valid
• The user explicitly says "don't create anything, just tell me what to do"

Before Starting

Ask (or infer) these if not obvious:

1. Repo name: What is this pack called? (Used in folder name, README title, manifest)
2. Purpose: 1-sentence summary of what this pack does
3. Primary audience: Who will use this? (developers, designers, marketers, etc.)
4. Skills to include: How many skills are planned? (affects folder structure)
5. Brand file: Does a brand file already exist? If not, flag that brand-system-builder needs to run.

Default if not specified:

• Repo name: derived from purpose or "prompt-pack-{timestamp}"
• Skills count: 1 (single-skill repo)
• Brand file: create placeholder to be filled

Output Structure

Required Folder Tree

{repo-name}/
├── README.md                    # Repo overview, installation, usage
├── MANIFEST.md                  # Machine-readable index of all files
├── CLAUDE.md                    # Repo rules for Claude/AI agents
├── skills/                      # All skill definitions
│   ├── {skill-1}/
│   │   └── SKILL.md             # Primary skill file
│   ├── {skill-2}/
│   │   └── SKILL.md
│   └── {skill-N}/
│       └── SKILL.md
├── references/                  # Shared reference files (optional)
│   ├── {reference-file}.md
│   └── ...
├── scripts/                     # Helper scripts (optional)
│   └── {script}.py
└── assets/                      # Images, templates (optional)
    └── ...

Optional Folders (add only if needed)

Step-by-Step Process

Step 1: Validate the Working Directory

Check what exists:

ls -la

Three scenarios:

Scenario A — Start fresh:

mkdir -p skills/{skill-name} references scripts assets
touch README.md MANIFEST.md CLAUDE.md

Scenario B — Audit first:

ls -la skills/ 2>/dev/null || echo "No skills folder"
ls -la references/ 2>/dev/null || echo "No references folder"
cat README.md 2>/dev/null || echo "No README"

Then fill gaps based on what's missing.

Scenario C — Verify:

# Check README has required sections
# Check MANIFEST.md parses as valid list
# Check CLAUDE.md has repo rules

If valid, tell the user: "Repo structure is complete. Run 'skills health-check' to verify contents."

Step 2: Create README.md

Required sections (in order):

# {Repo Name}

{One-line description of what this pack does}

## What This Pack Contains

{Bullet list of skills/features included}

## Quick Start

{3-step installation/usage guide:
1. Clone/download
2. Copy to your prompt library
3. First usage example
}

## Skills

| Skill | Purpose | Trigger |
|-------|---------|---------|
| {skill-name} | {one-line purpose} | {trigger phrase} |
| ... | ... | ... |

## Requirements

{Any tools, software, or context needed to use this pack}

## Examples

{One concrete example of usage}

## Contributing

{How to extend or modify this pack}

---
{Optional: license, author, date}

Rules for README:

• No corporate language ("world-class", "cutting-edge")
• No filler ("In today's fast-paced world...")
• Lead with what it does, not why it matters
• Include one working example

Step 3: Create MANIFEST.md

Format: One file per line, with metadata

---
version: 1.0.0
generated: {YYYY-MM-DD}
repo: {repo-name}
---

# Files

## Core
README.md                    | text/markdown | required | overview
MANIFEST.md                  | text/markdown | required | index
CLAUDE.md                    | text/markdown | required | agent rules

## Skills
skills/{skill-name}/SKILL.md | text/markdown | required | primary skill

## References
references/{ref-name}.md     | text/markdown | optional | shared docs

## Scripts
scripts/{script-name}.py     | text/python   | optional | automation

## Assets
assets/{asset-name}.{ext}    | binary        | optional | images/media

Rules for MANIFEST:

• One entry per file
• Include type (text/binary)
• Mark required vs. optional
• Brief description in last column
• Keep sorted by category

Step 4: Create CLAUDE.md

Template:

# CLAUDE.md — Agent Rules for This Repo

## Purpose
{One paragraph explaining what this repo contains and when to use it}

## Repo Structure
{File tree overview}

## Skill Locations
- Skills: `skills/{skill-name}/SKILL.md`
- References: `references/`
- Scripts: `scripts/`

## Key Conventions
- Follow existing naming patterns
- Keep outputs modular
- Verify before committing

## When to Trigger Skills
| Situation | Trigger |
|-----------|---------|
| {situation 1} | {skill-name} |
| {situation 2} | {another-skill} |

## Current Status
{What exists vs. what's planned}

## Known Issues
{Any limitations or todos}

Step 5: Create First Skill File (Placeholder)

If no skills exist yet, create a placeholder:

---
name: {skill-name}
description: "One-line purpose statement"
---

# {Skill Name}

## When to Use

**Trigger phrases:** "..."

## What This Does

{2-3 sentences on purpose}

## Coming Soon

{Placeholder — fill in before publishing}

This ensures the folder structure is valid even if skills aren't written yet.

Step 6: Final Validation

Run this checklist:

Forbidden at root level:

• NOTES.md
• TODO.md
• CHANGELOG.md (unless intentional)
• Any file with uppercase in name

Output for User

After bootstrapping, report:

✓ Repo created: {repo-name}
✓ Folder structure: {tree output}
✓ Next steps:
  1. Add your first skill → skills/{skill-name}/SKILL.md
  2. Run brand-system-builder if no brand file exists
  3. Test with: opencode → "help" → load skill

Cross-Skill Orchestration

Calls after bootstrapping:

Depends on:

• None (this is the foundation skill)

Edge Cases

Done Checklist

Before declaring complete:

• README.md has all required sections
• MANIFEST.md is valid format and lists all files
• CLAUDE.md has repo rules and skill triggers
• skills/ folder exists with at least one skill file
• No forbidden files at root
• All folder names are lowercase, hyphenated
• User sees clear "next steps" summary