How do I prepare a GitHub repo before making it public with Claude Code?

Install launchproof and run /launchproof in the project folder. It scans the full history, writes the missing files, scores the repo and lists the publish steps for you to approve.

Does it work in Codex?

Yes. Run codex plugin marketplace add deserteaglemj/launchproof and codex plugin add launchproof@launchproof.

Make a finished project public-ready on GitHub, then plan its launch

Q: How do I check my git history for personal emails and secrets before open sourcing?

Run lp.py hygiene . --rev main from the plugin's scripts folder, or ask the agent to audit the repo. It checks every commit on that branch, including files deleted later.

launchproof is a Claude Code and Codex plugin. It scans your whole git history for personal emails, home paths, secrets and private names, writes an evidence-backed README and the files a trustworthy repo needs, scores the result and fixes gaps, then drafts posts for every launch channel. You approve every push and post.

Install View on GitHub

Install

Claude Code

claude plugin marketplace add deserteaglemj/launchproof
claude plugin install launchproof@launchproof

Codex

codex plugin marketplace add deserteaglemj/launchproof
codex plugin add launchproof@launchproof

Then run /launchproof in your project folder, or ask "make this repo ready to publish on GitHub". Needs Python 3.9 or newer; no API key.

How it works

Hygiene scan of every commit you will publish: non-noreply author emails, personal emails, home paths, 10 secret patterns, key files, your private names, and images to review.
Proof: runs your tests and evals and publishes the numbers in a results file.
Scaffold missing files: community health files, CI, CHANGELOG, release notes, llms.txt, AGENTS.md, this kind of Pages site with JSON-LD and Open Graph, topics and a social preview.
README with a demo, quickstart, sourced comparison, limitations and an FAQ written as the questions people ask AI assistants.
Launch plan with dated steps and a draft for Show HN, Reddit, X, LinkedIn, Product Hunt, Dev.to, Discord, awesome lists, plugin directories and newsletters.
Audit and fix until the 0 to 100 checklist scores 90 or more, then a report with the owner-approval commands.

Results

Each case is rebuilt from scratch and run headless with push, history rewrites and web access denied. The harness measures the score; it does not trust the agent's report.

Audit score before and after one agent run: tiny-slug 11.1 to 97.2, wordfreq 18.5 to 97.2, shotproof 77.8 to 97.2.

Eval results per case
Case	Before	After agent	Final	Result
wordfreq	18.5	97.2	100.0	pass
tiny-slug	11.1	97.2	100.0	pass
shotproof	77.8	97.2	97.2	fail*

wordfreq (Python CLI) and tiny-slug (JavaScript library) are fixtures with seeded emails, paths, keys and a client name in their history; final is after the owner-approved rewrite. shotproof is a real public repo. *It fails because a private name in its public history needs an owner rewrite.

Method, limits and the failed first round: evals/RESULTS.md.

FAQ

Does launchproof push or post anything for me?

No. Pushing, making a repo public, releases, Pages, topics, history rewrites and every launch post come back as commands for you to approve and run.

How do I check my git history for personal emails and secrets before open sourcing?

Run lp.py hygiene . --rev main from the plugin's scripts folder, or ask the agent to audit the repo. It checks every commit on that branch, including files deleted later.

Is there a README generator for Claude Code or Codex that needs no API key?

Yes. launchproof runs on the model your agent already uses, with a standard-library Python helper.

What are its limits?

The audit checks structure and evidence links, not writing quality; the secret scan has 10 patterns (it runs gitleaks too when installed); history fixes need you; the eval set is small.