Thread by @neondatabase

Many Neon engineers use Claude Code daily. We’ve compiled a cheatsheet with the workflows and setups that work best for us 🧵

Start every session with Plan Mode (Shift + Tab + Tab).
Ask Claude to outline the approach before it writes code.
Use Opus for planning, Sonnet for implementation.

Try spinning up multiple agents with different “personalities” e.g. security reviewer, perf tuner, style checker.
For bigger setups, try Claude Swarm, Conductor, or Sculptor.

Claude is only as good as the context you give it.
Use /clear to reset, /compact to summarize, /context to inspect.
Reference large files with filename instead of pasting walls of text.

Our favorite MCP servers:
 Serena → symbol-level find/replace
 Context7 → up-to-date docs for any library
 Playwright → UI testing integration

Drop a CLAUDE .md in your repo root so Claude remembers your setup.
Keep extra memory files for routing, API, or tests.
Persistent context = fewer explanations

Use the status line:
- Add a heads-up display below the chat, e.g. with model, branch, tokens, directory
- Script it to load automatically so it’s always there when you open Claude Code

Start with your types (DB schema, Zod schema, or models), then code.
Claude gives you cleaner and more consistent output with strong contracts.

Automate recurring prompts: store Markdown files in .claude/commands
E.g. /pr-review, /create-plan, /investigate-codebase

When you need deeper reasoning, use think, think hard, think harder, or ultrathink.
Pair with larger context windows (8k–32k tokens) for complex refactors.

Useful shortcuts:
claude --resume → continue last session
/rewind or Esc Esc → undo last action
--dangerous-skip-permissions → YOLO mode (use with care)

Stay aware of compute usage:
npx ccusage → overall stats
/cost → current session spend
/usage → usage percentage

You can even add cost to your status line

These and more tips outlined in detail in this blog post: neon.com/blog/our-claud…