slopshopper

The mod shop for Claude Code. slopshopper.com scrapes GitHub every day for Claude Code mods, renders a preview of what each one does to your terminal, and gives you the two commands to install it. This repository is the whole thing: the scraper, the preview harness, the static site, and six mods of our own.

Install the mods

claude plugin marketplace add ryx2/slopshopper
claude plugin install slop-detector@slopshopper

Mod	What it does
slop-detector	Refuses an edit that replaces code with // ... rest of the code; tracks TODO stubs, debug prints and other leftovers above the prompt and in /slop.
burn-rate	A fuel gauge above the prompt: context fill, tokens, dollars, dollars per hour, your plan's window, a sparkline of the last turns.
secret-shield	Redacts API keys, tokens and private keys from tool results before Claude reads them; refuses writes and commands that embed one.
tripwire	Holds rm -rf, force pushes, git reset --hard, DROP TABLE, curl | sh, sudo and .env edits behind a question; refuses the unrecoverable ones.
receipts	A ledger of files touched and commands run, counted on the spinner, printed as Markdown by /receipts.
test-pulse	Runs the test suite after every turn that changed source and keeps the result above the prompt.

Every community mod the site lists is also installable through one generated marketplace that points at each author's own repository:

claude plugin marketplace add https://slopshopper.com/community/marketplace.json
claude plugin install <name>@slopshopper-community

Mods need Claude Code 2.1.287 or later. A mod runs with your permissions; read it before you install it.

How it works

scripts/scrape.ts      GitHub code search + repo search → walk each repo's tree →
                       every hooks/hooks.json with a `modules` array is a mod →
                       data/mods.json + data/mods/<slug>.json (source files)
scripts/previews.ts    claude plugin validate --json (static facts) +
                       scripts/harness (a sandboxed replay of a scripted session) →
                       data/previews/<slug>.json
scripts/build.ts       → dist/  (index, one page per mod, /new, /about, feed.xml,
                       api/mods.json, community/marketplace.json)

The harness bundles a mod's hooks module and runs it in a fresh node:vm realm with a fake mods API: a session start, a prompt, a turn with reads, edits, a failing test run, a risky shell command and a cat .env, then turn.complete and session.measure. Every render site is then asked to draw, and a small text-grid layout engine draws the returned trees the way the terminal would. Previews show the shape of a mod, not its behaviour on your machine; the facts panel comes from claude plugin validate.

Run it

bun install
bun run scrape        # GitHub token from GITHUB_TOKEN, GH_TOKEN or `gh auth token`
bun run previews      # needs the claude CLI on PATH
bun run build
bun run dev           # http://localhost:4321

bun run scrape --repo owner/name indexes one repository; --fast skips repositories whose pushed_at is unchanged. Previews are cached by a hash of the mod's source; bun run previews --force regenerates them.

The GitHub Actions workflow in .github/workflows/site.yml runs the scrape daily, commits the refreshed data, builds the site and deploys it to GitHub Pages. Set the repository variable SITE_BASE to /slopshopper/ for a project page or / for a custom domain, and SITE_URL to match. A SCRAPE_TOKEN secret (a classic PAT with public_repo) lets code search run in CI; without it the scheduled job still refreshes known repositories.

Hosting

slopshopper.com is served by the Vercel project slopshopper (static build: vercel.json sets bun run build and dist/), which is git-connected to this repository, so every push to main, including the daily data commit, redeploys it. The project's environment sets SITE_BASE=/ and SITE_URL=https://www.slopshopper.com (the apex redirects to www, as it did before). The GitHub Actions workflow also publishes a mirror to GitHub Pages under /slopshopper/, whose pages declare slopshopper.com as canonical.

A scrape run on a laptop (bun run scrape) can take an hour; bun run scrape --discover-only followed by four --shard k/4 --out .cache/shards/k.json processes and bun run scrape --merge does the same work in parallel. scripts/publish-data.sh merges, regenerates previews and builds.

Writing a mod

Each mod under mods/ is a complete plugin: .claude-plugin/plugin.json, hooks/hooks.json naming the module, the module, a types/index.d.ts state contract, tests, and a README. Check one with:

claude plugin validate --strict mods/tripwire
claude plugin test mods/tripwire
claude --plugin-dir mods/tripwire

License

MIT for everything in this repository. Community mods are listed under their own licenses.