skill-forge-publish

Skill Publishing & Distribution

Process

Step 1: Pre-Publish Validation

Run the full review before publishing:

1. Execute /skill-forge review <path> and ensure score >= 80/100
2. Fix any critical or high-priority issues
3. Test with at least 5 trigger queries
4. Verify all cross-references resolve

Step 2: Create Install Script

Generate install.sh that handles:

#!/usr/bin/env bash
# Install script for [skill-name]
# Usage: bash install.sh

set -euo pipefail

SKILL_DIR="$HOME/.claude/skills"
AGENT_DIR="$HOME/.claude/agents"
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"

echo "Installing [skill-name] skill..."

# Create directories
mkdir -p "$SKILL_DIR" "$AGENT_DIR"

# Copy main skill
cp -r "$SCRIPT_DIR/skill-name" "$SKILL_DIR/"
echo "  Installed: skill-name"

# Copy sub-skills
for skill in "$SCRIPT_DIR/skills"/skill-name-*/; do
    if [ -d "$skill" ]; then
        skill_basename=$(basename "$skill")
        cp -r "$skill" "$SKILL_DIR/"
        echo "  Installed: $skill_basename"
    fi
done

# Copy agents (if any)
if [ -d "$SCRIPT_DIR/agents" ]; then
    cp "$SCRIPT_DIR/agents"/*.md "$AGENT_DIR/" 2>/dev/null || true
    echo "  Installed agents"
fi

echo ""
echo "Installation complete!"
echo "Test with: /skill-name"

Step 3: Create README.md (repo-level, NOT inside skill folder)

# [Skill Name]

[1-2 sentence description focusing on outcomes, not features]

## What it does

[Bullet list of key capabilities]

## Installation

### Claude Code

git clone https://github.com/[user]/[repo] cd [repo] bash install.sh

### Claude.ai
1. Download the latest release (.zip)
2. Go to Settings > Capabilities > Skills
3. Click "Upload skill"
4. Select the downloaded .zip file

## Commands

| Command | Description |
|---------|-------------|
| `/skill-name` | [description] |
| `/skill-name cmd` | [description] |

## Examples

### [Example 1 title]

User: "[example input]"

[Description of what happens and expected output]

## Architecture

[file tree diagram]

## License

[License type]

Step 4: Package for Distribution

For Claude.ai upload: Run python scripts/package_skill.py <path> <output-dir> to create a .skill zip file.

For GitHub:

1. Create repository with README.md at root
2. Skill folder(s) at root level
3. install.sh at root level
4. Add LICENSE file
5. Add .gitignore (exclude .tmp/, pycache/, *.pyc)

For team deployment (Claude.ai admin):

• Skills can be deployed workspace-wide by admins
• Package as .skill zip and upload through admin console

Step 5: Create .gitignore

__pycache__/
*.pyc
*.pyo
.tmp/
*.egg-info/
dist/
build/
.env
*.skill

Step 6: Release Checklist

• All files validated (score >= 80)
• install.sh tested on clean system
• README.md covers installation, usage, and examples
• LICENSE file included
• .gitignore configured
• No secrets or API keys in any file
• Test queries documented
• Version tagged (if using git)

Step 7: Post-Publish

After publishing:

1. Test installation from scratch on a clean environment
2. Run all trigger test queries
3. Collect initial user feedback
4. Plan first iteration based on feedback
5. Set up issue templates for bug reports

Distribution Channels

Channel	Best For	Format
GitHub	Open source, community	Repository + install.sh
Claude.ai upload	Personal use	.skill zip
Team admin	Organization-wide	.skill zip via admin console
Claude Plugin Marketplace	Wide distribution	.claude-plugin/ manifest