repo-glowup

A Claude Skill that audits your repo's first impression, rewrites your README, and drafts your launch posts — grounded in what's actually in your code, nothing invented.

CI License: MIT Claude Skill

Most repos don't get ignored because the code is bad. They get ignored because a visitor can't tell what it does, whether it works, or whether it's alive, in the 10 seconds before they hit the back button. repo-glowup turns that into something fixable.

$ (inside Claude, with the skill installed)
> audit my repo and rewrite the readme if it needs it

Score: 14/25
  Clarity          2/5  — the pitch is buried after a 40-line badge wall
  Visual proof     1/5  — no screenshot, no GIF, no example output
  Instant trial    3/5  — install command works, but it's 60 lines down
  Trust signals    4/5  — LICENSE present, CI green, no CONTRIBUTING.md
  Launch readiness 4/5  — topics set, social preview missing

Top fixes (highest impact first):
  1. Move the one-liner above the badges — right now it's page 2
  2. Add a 15-second terminal GIF showing the actual payoff
  3. Set a social preview image (Settings → Social preview)

→ README.md rewritten. Diff below.

(Real output shape from an actual run — see references/ for the full rubric behind these numbers.)

What it actually does

  1. Scans your repo for the objective, checkable stuff — README structure, images and whether they have alt text, badges, LICENSE, CONTRIBUTING, CI config, recent commit activity — using a bundled script, not vibes.
  2. Scores it against five categories that map to what a first-time visitor actually does: can they tell what it is, can they see it work, can they try it in under a minute, does it look trustworthy and alive, is it set up to be found (topics, social preview, description field).
  3. Hands you a short, prioritized fix list — ranked by impact ÷ effort, not a 40-item nitpick dump.
  4. Rewrites your README, in full, using your project's real content. It won't invent a feature, a benchmark number, or a testimonial you don't have.
  5. Drafts your launch posts for Show HN, the right subreddit, and X/Twitter — three different posts, because each platform rewards a different tone and punishes a different failure mode.

It will also flag it if something looks off — like a repo that isn't actually yours, or a claim your code doesn't back up — instead of quietly going along with it. See Guardrails.

Install

Claude Code / Cowork: drop the folder into your skills directory —

git clone https://github.com/erdalacundakonuk/repo-glowup.git ~/.claude/skills/repo-glowup

Claude.ai / Cowork, from the packaged file: download the repo-glowup.skill file from this repo's Releases page and use the Save skill option where your workspace supports it.

Then just talk to Claude normally from inside (or pointed at) the repo you want audited:

"audit this repo and tell me what's hurting it" "rewrite my README so it actually gets read" "help me write a Show HN post for this"

The skill triggers on intent, not on exact wording — you don't need to name it.

Why this exists

Most advice about "getting GitHub stars" is either vague ("build in public!") or fake ("buy some stars to look credible"). Neither is useful. The actual leverage is almost always in the same handful of places — the first line, the first screen, whether there's proof it works, whether it looks alive — and those are checkable and fixable in minutes, not months. This skill just does the checking and the fixing, honestly, using what's really there.

Guardrails

This is a clarity tool, not a growth-hacking tool. It will not:

  • invent metrics, testimonials, or "used by" claims that aren't real
  • suggest buying stars, star-exchange rings, or bot traffic
  • suggest astroturfing HN/Reddit/Product Hunt comments or votes
  • write a personal launch story for a project that isn't actually yours

Full reasoning in SKILL.md.

Repo layout

repo-glowup/
├── SKILL.md                          the skill itself: triggers + workflow
├── references/
│   ├── scoring-rubric.md             the five categories, in full
│   ├── readme-patterns.md            what a good README structure looks like, and why
│   └── launch-post-patterns.md       Show HN / Reddit / X, per-platform
├── scripts/
│   └── scan_repo.py                  objective repo scan → JSON (no LLM guessing)
└── assets/
    └── readme-template.md            starting skeleton for a rewrite

Contributing

Issues and PRs welcome — especially real before/after examples from repos you've actually run this on. See CONTRIBUTING.md.

License

MIT — see LICENSE.