LLM Best Practices

How-To Guides & Tutorials

5 min read

Step-by-step tutorials for the tasks developers run again every few months. Each guide names the prerequisites, walks the steps with the exact commands, ends in a verification a script could check, and links the umbrella pages it builds on.

Pages

  • launch-a-new-site: The ordered zero-to-indexed runbook: repo, deploy, DNS, discoverability files, search registration, and verification, with a copy-paste checklist.
  • deploy-quartz-site: Fork Quartz, wire the GitHub Pages workflow, point a custom domain, verify with curl.
  • set-up-claude-code: Install Claude Code, write the CLAUDE.md anchor, set up TODO.md, ship the first PR.
  • add-jsonld-to-static-site: Pick the right schema types, inject JSON-LD into <head>, validate with the Rich Results Test.
  • optimize-core-web-vitals: Measure with CrUX and web-vitals, fix LCP, CLS, and INP, re-measure at p75.
  • write-llm-friendly-content: Ship /llms.txt, /ai.txt, JSON-LD, OG cards, raw markdown links, and named heuristics agents can cite.
  • ship-llms-txt: Decide structure, write the summaries, place the file at the site root, link from robots.txt and ai.txt.
  • debug-postgres-slow-query: Enable pg_stat_statements, find the offender, run EXPLAIN ANALYZE BUFFERS, add the right index, confirm the new plan.
  • run-a-quartz-static-site-locally: Clone, install Node 22, run npx quartz build --serve, fix port, theme, and plugin-order errors.
  • audit-site-for-core-web-vitals: Lighthouse, CrUX, and web-vitals JS to measure LCP, INP, and CLS, then triage in priority order.
  • configure-typescript-strict: enable strict, fix errors starting with noImplicitAny, add noUncheckedIndexedAccess, verify a clean build.
  • deploy-to-vercel: import a project, set env vars, configure the build, attach a custom domain, enable preview deploys.
  • migrate-css-to-tailwind: install Tailwind v4, remove conflicting resets, convert components using @theme tokens, lint with the sort plugin.
  • optimize-images-for-web: resize to display dimensions, convert to WebP and AVIF, add lazy loading, set explicit width and height.
  • run-claude-code-with-mcp: configure an MCP server in settings.json, restrict permissions, verify the connection with /mcp.
  • set-up-cloudflare-tunnel: install cloudflared, create a named tunnel, route a hostname through Cloudflare DNS, run as a service.
  • set-up-postgres-locally: install Postgres, create a role and database, configure pg_hba auth, connect via psql.
  • set-up-pytest: add pytest and pytest-cov to pyproject.toml, write conftest.py, use fixtures, verify with —collect-only.
  • write-meta-descriptions: write 150-160 character descriptions with the primary keyword and a clear benefit; A/B test with GSC.
  • add-llms-txt-to-existing-site: Write a compliant /llms.txt index, place it at the site root, add the ai.txt sibling, cross-link from robots.txt, and verify with curl.
  • audit-llms-txt-with-claude: Fetch /llms.txt from a live site, send it to Claude with a validation prompt, identify format violations and missing pages, then fix and re-validate.
  • build-an-mcp-server: Choose a stack, define tools with JSON Schema, register the server, and verify it in MCP Inspector before wiring to Claude Code.
  • connect-claude-to-postgres-via-mcp: Install the MCP postgres server, configure credentials safely, connect from Claude Code, and run a test query to confirm the integration works.
  • wire-up-anthropic-sdk: Install the Anthropic Python SDK, load the API key, send a messages request, stream a response, and add a tool definition.
  • set-up-rag-pipeline: Chunk documents, embed with a hosted model, store in pgvector, retrieve top-k at query time, and ground the answer with citations.
  • configure-pgvector-on-neon: Enable the pgvector extension on a Neon database, create an embedding table, insert vectors, add an HNSW index, and run a similarity query.
  • set-up-ollama-locally: Install Ollama, pull a model, run an interactive chat, set a system prompt, and integrate with the REST API or LangChain.
  • deploy-fastapi-on-fly: Write a Dockerfile, configure fly.toml, set secrets, attach a Postgres add-on, deploy, and verify with curl.
  • write-tests-for-fastapi: Wire pytest with TestClient, override dependencies for database isolation, and write fixtures that reset state between tests.
  • migrate-postgres-database: Dump with pg_dump, restore with pg_restore, plan for downtime, verify row counts, and execute a clean cutover with minimal risk.
  • optimize-postgres-query: Find the offender with pg_stat_statements, read the plan with EXPLAIN ANALYZE, add a targeted index, and verify the new plan matches expectations.
  • migrate-react-to-server-components: Identify client/server boundaries, mark interactive subtrees with ‘use client’, move data fetching to async server components, and verify bundle size drops.
  • set-up-cloudflare-r2: Create an R2 bucket, generate API keys, attach a public domain, and access it with the AWS CLI or boto3 via the S3-compatible endpoint.
  • set-up-cloudflare-workers: Initialize a Worker with wrangler, write a fetch handler, bind KV storage, deploy to production, and verify with a curl request.
  • set-up-quartz-with-cloudflare-pages: Fork Quartz, push to GitHub, connect Cloudflare Pages, configure baseUrl, and point a custom domain without touching GitHub Pages.
  • run-pa11y-locally: Install pa11y, write a .pa11yci.json config, run a scan, fix the common issues, and plug pa11y-ci into a GitHub Actions workflow.
  • configure-pa11y-thresholds: Set per-rule threshold overrides, add ignore patterns for acceptable failures, and wire the config to CI so new violations break the build.
  • audit-dependencies: Run npm audit or pip-audit, classify findings by severity, patch or accept risk, re-test, then add a scheduled CI job to catch new CVEs.
  • write-a-cursor-rule: Write a .cursorrules file for a TypeScript project: set voice rules, scope conventions, code style, and test expectations so the AI assistant follows the project’s standards.
  • static-site-seo: Ship every SEO artifact a static site needs at build time: canonicals, sitemap, robots, JSON-LD, OG images, llms.txt, then verify with curl before deploy.
  • seo-migration-playbook: Run an SEO-safe site migration: crawl, build a redirect map, stage on noindex, flip DNS, submit sitemaps, monitor coverage for 4 weeks.
  • master-google-search-console: Verify a property, submit sitemaps, and use every major GSC report (Performance, Coverage, URL Inspection, Enhancements) for weekly health checks.

52 items under this folder.