botskills.sh

Skills directory for AI bot runtimes (Grok Bot, Rakazo): paste-ready BOT.md setups, ranked by verified copies. Same architecture as qaskills.sh: pnpm monorepo, shared constants package, Next.js 15 web app, file-based seed catalog, dual-registry blog engine, machine-readable by design (llms.txt, open JSON API, raw markdown per bot).

Layout

• packages/shared: constants (categories, runtimes, integrations), types, BOT.md parser and serializer
• packages/web: Next.js 15 app with leaderboard, bot pages, integration hubs, blog, API, llms.txt
  • src/lib/blog-registry.ts: one list of posts, from which the page lookup, index and sitemap derive
  • src/lib/related.ts, src/lib/integrations.ts: related links and per-tool hub pages, computed from the corpus
  • src/lib/sql.ts, src/lib/telemetry-counts.ts: the one database client and the copy/vote rollup
  • e2e/: Playwright suite (runs locally with no secrets, or against production with E2E_BASE_URL)
  • tests/: node --test unit tests for the pure libraries
• seed-bots/<slug>/BOT.md: the catalog source of truth
• scripts/: article gate, registry generator, duplicate and slop checks, fact sweep, IndexNow
• docs/seo/VERIFIED-FACTS-2026-08-25.md: what the site may assert about Grok Bot, with a dated correction section at the bottom
• docs/ROLLBACK.md and rollback.empty: how to roll production back, and the ledger of states to roll back to
• .github/workflows/ci.yml: type-check, lint, format, unit, catalogue and article gates, build, Playwright

Commands

pnpm install
pnpm build            # shared then web (Turbo order)
pnpm --filter @botskills/web dev
pnpm typecheck        # shared build + web tsc
pnpm lint             # eslint, next/core-web-vitals + typescript
pnpm format:check     # prettier (article bodies excluded)
pnpm test             # unit tests: shared, web, scripts
pnpm test:e2e         # playwright against a local dev server with no database or mail key
python3 scripts/gate.py               # every article against the publication gate
python3 scripts/register.py           # regenerate the blog registry (refuses if the gate fails)
node scripts/validate-bots.mjs        # every BOT.md against the shipping parser

Build requires zero secrets: the DB client is a lazy proxy, pages read the file catalog. Node 24.x pinned (package.json engines).

BOT.md schema

Single-line frontmatter values, inline arrays (regex-parsed, same rules as qaskills seed pipeline). Required: name, description, version, author, license, category, integrations [..], runtimes [grok-bot|rakazo], boundary. Body = the setup prompt. boundary is the one-sentence hard limit ("Never sends without approval") and is required: it renders on every surface.

Contributing

Every listing is one file: seed-bots/<slug>/BOT.md. Copy BOT-TEMPLATE.md, fill it in, and run node scripts/validate-bots.mjs. Full rules, including the three parser traps that fail a build, are in CONTRIBUTING.md.

Not a developer, or would rather not open a pull request? Use the Suggest a bot issue form and we will write the file.

Every bot must declare a boundary: the one irreversible action it will never take without a human approving that specific action. Listings without one are not merged.

Licence

MIT. See LICENSE.