Skill Publish

Language

Match user's language: Respond in the same language the user uses.

Overview

Package a agent skill into a complete, distributable GitHub repository. Generates all the surrounding files (README, LICENSE, plugin.json, marketplace.json) and optionally creates the GitHub repo.

How It Works

1. Identify the skill to publish
2. Optionally review first with skill-review
3. Collect publishing metadata through dialogue
4. Generate the repository structure with publish.py
5. Optionally initialize git and create GitHub repo

Dialogue Flow

Progress:

• Step 1: Identify the skill
• Step 2: Pre-publish validation
• Step 3: Collect metadata
• Step 4: Generate repository
• Step 5: Post-generation

Step 1: Identify the Skill

Accept the skill location in multiple forms:

Input	Detection	Action
Directory with SKILL.md	Direct path	Use as-is
SKILL.md file path	Path ends in SKILL.md	Use parent directory
Current directory	No input, cwd has SKILL.md	Auto-detect

Step 2: Pre-Publish Validation

Check for skill-review availability:

1. First try: {SKILL_DIR}/../skill-review/scripts/validate.py
2. Fallback: ~/.agents/skills/skill-review/scripts/validate.py

If found, suggest running validation:

python3 <validate.py path> run --path <skill-path>

If there are failures, recommend fixing them before proceeding. Warnings are acceptable.

If skill-review is not available, offer:

• A) Install it: npx skills add psylch/better-skills@skill-review -g -y
• B) Continue without it (quick manual check: verify SKILL.md exists with valid frontmatter)
• C) Cancel

Step 3: Collect Metadata

Use AskUserQuestion to gather:

1. GitHub owner/org — Default: try to infer from git config user.name or gh api user -q .login. Ask user to confirm or change.

2. Repository name — Default: <skill-name>-skill (e.g., cors-audit-skill). Suggest the convention but let user override.

3. Version — Default: 1.0.0. Use semver.

4. License — Default: MIT. Options: MIT, Apache-2.0, GPL-3.0.

5. Output directory — Where to create the repo. Default: parent directory of the skill.

Step 4: Generate Repository

python3 {SKILL_DIR}/scripts/publish.py generate \
    --skill-path <path> \
    --owner <owner> \
    --repo <repo-name> \
    --version <version> \
    [--license mit|apache2|gpl3] \
    [--output <dir>] \
    [--force]

Where {SKILL_DIR} is the directory containing this SKILL.md file.

The script outputs JSON to stdout:

{
  "status": "ok",
  "output_dir": "/path/to/output",
  "repo": "owner/repo-name",
  "created": [".claude-plugin/plugin.json", "README.md", "LICENSE", ...],
  "hint": "Repository files generated. Next: git init or gh repo create."
}

Step 5: Post-Generation

Present a completion report:

[Skill Publish] Complete!

Skill: <skill-name>
Repository: <owner>/<repo>
License: <license>
Output: <directory>

Files created:
• <list from JSON "created" field>

Next Steps:
→ A) Initialize git repo
→ B) Create GitHub repo and push
→ C) Done — handle git manually

Then let the user choose:

Option A: Initialize git repo

cd <output-dir>
git init
git add .
git commit -m "Initial commit"

Option B: Create GitHub repo and push

cd <output-dir>
git init
git add .
git commit -m "Initial commit"
gh repo create <owner>/<repo> --public --source . --push

Option C: Done — User will handle git themselves.

After creation, remind the user:

• Edit README.md and README.zh.md to polish auto-generated content
• Register on skills.sh if they want public discoverability
• Test installation: npx skills add ./ -g -y

Generated Repository Structure

<repo-name>/
├── .claude-plugin/
│   ├── plugin.json
│   └── marketplace.json
├── skills/
│   └── <skill-name>/        # Copied from source
│       ├── SKILL.md
│       ├── scripts/
│       └── references/
├── README.md
├── README.zh.md
├── LICENSE
└── .gitignore

Distribution Guide Reference

For details about skills.sh registry, versioning strategy, and publishing conventions, read references/distribution_guide.md.