agentleFS
titouannwtttitouannwtt/claude-tools/skills/repo-showcase/SKILL.md

repo-showcase

Skill1 starsChanged 12 days agoMITView on GitHub

Turn a basic GitHub repository into a publish-ready showcase. Interviews the user (purpose, audience, goal, length, mascot, links), audits secrets and personal data in the files and the whole git history, removes AI co-authorship mentions, audits the file structure, then builds a short, very visual README (custom animated SVG banner, screenshots, live counters) and checks the GitHub rendering before pushing. Use when the user wants a repository to look professional, to highlight their profile, to prepare it for going public, or to check it holds no personal data.

3 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Run python3 $S doctor once per machine: it says what is possible (history rewrite, screenshots, faces, demo).

    line 13
  2. Very specific:

    Never change a repository's visibility.

    line 17
  3. Distinctive:

    Never print or store a raw secret: talk about findings with their preview and fingerprint only.

    line 19
  4. Distinctive:

    Remove every AI co-authorship mention (commits, files, pull requests, releases).

    line 23
  5. Very specific:

    No hard-coded number that changes often in a visual or the README, unless it is live (badge), synced (counts), or a dated measurement.

    line 26
  6. Very specific:

    Run python3 $S <command> -h for the options.

    line 83

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

repo-showcase

Goal: a visitor understands the project in five seconds, from the visuals alone, and nothing private leaks. Short and visual beats long: if it is long nobody reads it; if it is short and visual, people at least look at the images, and if the images are clear the project is understood at a glance.

Script: S=~/.claude/skills/repo-showcase/scripts/showcase.py (python3 $S <command>, JSON on stdout). Run python3 $S doctor once per machine: it says what is possible (history rewrite, screenshots, faces, demo).

Hard rules

  • Never change a repository's visibility. For a private repository, the last message says: going public is the user's last step (ready prints the command for them).
  • Never print or store a raw secret: talk about findings with their preview and fingerprint only.
  • Normal push and gh repo edit (description, topics, homepage) only after the user validated the preview.
  • History rewrite and force push: only after an explicit yes, asked separately each time. Never delete or move a file without agreement.
  • Remove every AI co-authorship mention (commits, files, pull requests, releases). A project that is about AI keeps its product mentions.
  • Commits: git config user.email must be a noreply address; no AI attribution in messages.
  • No hard-coded number that changes often in a visual or the README, unless it is live (badge), synced (counts), or a dated measurement.

Workflow

  1. Facts (no questions yet): inventory <repo> and structure <repo>. If the user gave only a URL, gh repo clone it into a work folder first. Read hints, project_type, run_hints, style_sources.
  2. Safety first: security <repo> (files + full history) and attribution <repo>. Report by severity and category, then ask how to handle each category with AskUserQuestion, using the options of references/security-remediation.md. Fix the working tree first. History: fix-plan (fetches the remote and refuses if it holds work the local refs lack; dry run on a temporary mirror, must report remaining: 0), show the summary, ask, then fix-apply --confirm (bundle and git directory archive first), ask again, then run the printed --force-with-lease push (existing remote refs only). A secret that was ever pushed is rotated, rewrite or not. False positive: allow <repo> <fingerprint>. Pull requests and releases: fix-prs <repo> (dry run), then --confirm.
  3. Interview: references/interview.md. Two batches of at most four questions, pre-filled from the inventory; never ask what the facts already answer. Mascot chosen: ask for the image (link or path) and its rights.
  4. Keywords: keywords <repo> --save (sources and rules in references/keywords.md; --no-remote for a repository that must stay offline). Then use WebSearch (2 to 4 queries such as "best open source <domain> tools", "<domain> github") to see the vocabulary in use and what the project relates to, merge, and show a short table (term, where to place it, why) for validation. The validated list feeds the pitch, the headings, the description and the topics (applied at Publish with gh repo edit --add-topic).
  5. Proposals: two or three README directions from references/readme-patterns.md, each with what it highlights, for whom, its length and its visuals; recommend one. The user picks.
  6. Build:
    • Visuals are custom for each repository: brief the designer agent with the template and constraints of references/visuals.md (one message per visual, palette from the product site via site-design or from the owner's style_sources).
    • Screenshots: shoot <url> --out assets/screenshots (live site, or --serve "<run hint>" for a local app); demo for a short GIF when interaction is the point. Check every capture for personal data.
    • README: follow the chosen pattern; counters through markers and counts sync; counts lint must pass. Offer counts install-action when a count changes often (it counts tracked files, so it also suits very large repositories).
    • Structure: present the structure findings as suggestions (see references/structure.md); apply only the accepted ones, with git mv.
  7. Verify: preview <repo> (static checks + renders at 880 and 390 px, light and dark). Open readme-fold-*.png and the mobile render with Read. For a flagship repository, add a visual-qa agent.
  8. Publish: show the user the fold render and the list of changes; after validation commit, push, set the description, topics and homepage with gh repo edit. Then ready <repo> must exit 0. Offer social <repo> --source assets/<banner>.svg (uploaded by hand in Settings) and pinning on the profile. Private repository: end with the final_step sentence.

Commands

CommandUse
doctordependencies and ready_for
inventory <repo> / structure <repo>facts, layout audit
keywords <repo> [--seed T] [--no-remote] [--max-search N] [--save]sourced keywords, placement, privacy
security <repo> / attribution <repo>findings; full redacted report in .git/repo-showcase/
allow <repo> <fp>...accept false positives
fix-plan / fix-apply --confirm [--push] / fix-prs [--confirm]history and remote cleanup
counts sync|lint|badge|install-action <repo>live numbers
shoot, demo, preview, socialcaptures and rendering
ready <repo>final gate, prints final_step

Run python3 $S <command> -h for the options.

Exit codes

CodeMeaningAction
0OKContinue.
4Findings or failed checksHandle them with the user, rerun.
5Dependency missingdoctor gives the install command; continue without that step if optional.
6Safety refusalRead error (dirty tree, --confirm missing, git < 2.36, refs changed).
7GitHub or networkCheck gh auth status and the URL.
3 / 2Not a repository / bad argumentsFix the call.
1Unexpected errorRepair the script (below).

Repair

Fix the module in <claude-tools>/skills/repo-showcase/scripts/, then python3 -m unittest discover -s tests -q in the skill's folder. Two attempts at most, then report the exact message to the user.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.