Open Source Launch

Making a repo public is not the same as shipping an open-source project. Most "public" repos only run on the author's machine, leak a path or a token in history, and are invisible to the AI agents that more and more people now ask to "install this for me".

Open Source Launch gives your agent one job with three outcomes:

1. It runs anywhere — no hidden dependency on your machine, proven by a fresh clone in a temporary HOME.
2. Any agent can pick it up — the repo ships its own agent manual (AGENTS.md, CLAUDE.md, a bundled skills/<name>/SKILL.md, JSON output, safe defaults).
3. People can find it — keyword-rich description and topics, FAQ phrased as the questions people type or ask an AI, and a social preview that looks good when the link is shared.

It judges the project against those outcomes, not a fixed checklist: a small CLI gets a fraction of what a desktop app needs.

npx skills add risingjerry/open-source-launch

Or copy skills/open-source-launch/ into ~/.claude/skills/ (Claude Code) or ~/.agents/skills/ (Codex and others). Start a new session so the agent discovers it.

Just ask your agent:

Audit this repo for an open-source launch. Don't change anything.

Make this project a public repo: privacy check, make it run on any machine,
add the agent kit, write the README, then flip it public.

Refresh our already-public repo to the launch standard and draft the X post.

What happens, in order:

Posting, list submissions, package publishing and history rewrites each need your separate go-ahead — the Skill drafts them, it never sends them.

skills/open-source-launch/
├── SKILL.md                         # the workflow
├── agents/openai.yaml               # Codex display metadata
├── scripts/
│   ├── audit_open_source.py         # privacy / package / agent-kit auditor
│   ├── social_preview.py            # 1280×640 painterly social preview renderer
│   └── install_repository_pulse.py  # optional repo metrics chart
├── references/
│   ├── agent-ready.md               # how to write AGENTS.md and a bundled Skill
│   ├── portability-contract.md
│   ├── discovery-contract.md        # SEO / GEO / social preview
│   ├── adoption-contract.md         # launch channels
│   ├── repository-pulse.md
│   └── owner-defaults.example.md    # copy to owner-defaults.md for your house style
└── tests/

Your own defaults. Copy references/owner-defaults.example.md to owner-defaults.md and fill in your GitHub account, commit identity, house style and promotion rules. The Skill reads it first. Keep it out of any public copy.

Python standard library only. It reports locations, never secret values:

python3 skills/open-source-launch/scripts/audit_open_source.py . --json
python3 skills/open-source-launch/scripts/audit_open_source.py . --run-gitleaks --strict

It checks package basics, credential-shaped content, tracked env files, machine-specific references, risky artifacts, external symlinks, runtime declarations, Git state, and whether AGENTS.md / CLAUDE.md exist. A clean scan is the first gate, not "launch-ready".

python3 skills/open-source-launch/scripts/social_preview.py \
  --background bg.png --icon icon.svg --name "Product|Agent CLI" \
  --out docs/social-preview.jpg

Generate a text-free oil-painting background with any image model (prompt in owner-defaults.example.md); the script lays a white icon and wordmark inside the central square so chat apps that crop to a square still show it. Needs Google Chrome or Chromium.

• An agent that loads SKILL.md Skills (Claude Code, Codex, Cursor and others).
• Python 3.10+ for the helper scripts; git.
• Optional: gh (GitHub reads/writes), Gitleaks (history scan), Chrome (social preview), Node.js (npx skills add).
• No telemetry.

Run the privacy sweep first: the bundled auditor plus a look through git log -p, screenshots and agent instruction files. If anything private is in history, the Skill stops and asks whether to rewrite history or publish a fresh single-commit repo. It never rewrites silently.

Ship an AGENTS.md (install, test, call, never-do), a one-line CLAUDE.md with @AGENTS.md, and a bundled skills/<name>/SKILL.md. The Skill writes these for you; agent-ready.md explains what goes in them.

Use the words people search for in the repo name and description, add focused topics, phrase FAQ headings as real questions, and set a social preview. See discovery-contract.md.

Flipping visibility is covered by "make it public" unless you ask to review first. Social posts, awesome-list submissions, package releases and history rewrites are drafted only.

Yes. It doesn't pretend otherwise: platform requirements go on the first screen of the README.

npx skills remove open-source-launch

Or delete the open-source-launch folder from your skills directory.

python3 -m unittest discover -s skills/open-source-launch/tests -v
python3 -m compileall -q skills/open-source-launch
python3 skills/open-source-launch/scripts/audit_open_source.py . --strict

See CONTRIBUTING.md. Report vulnerabilities through private vulnerability reporting, not a public issue. MIT licensed.

If it helped you ship, a ⭐ helps others find it.