Skip to main content

​
Guide · Claude Code

Claude 3.5 excels at reasoning-heavy engineering tasks. Here’s my setup.

​
1. Custom Instructions

- Assume TypeScript-first stack.
- Prefer step-by-step reasoning, but keep final answers concise.
- Suggest tests + observability hooks.
- Flag risks before proposing major changes.
Apply these both in Anthropic Console and Cursor.

​
2. Common Workflows

  • RFC drafting: provide context, constraints, desired outcomes; Claude returns outline + risks.
  • Code refactors: paste function + tests; ask for safer alternative.
  • Design system docs: feed component API + usage guidelines for copy polish.
  • Agent eval writing: generate adversarial prompts + scoring rubrics.

​
3. Tooling

  • Use the Claude VS Code extension or Cursor integration.
  • For long files, host gists + share URLs; Claude handles retrieval better.
  • Enable “Code Execution” in Anthropic Console for quick prototypes.

​
4. Guardrails

  • Don’t paste secrets or proprietary customer data.
  • Cross-check suggestions; treat Claude as pair-programmer, not autopilot.
  • Log key decisions in PR/Linear for audit trail.

​
5. Sample Prompt

You are pairing with me on the productivity stack. Given the following React server component...
1. Identify DX issues.
2. Propose improved structure.
3. Suggest tests + telemetry.
Code:
Happy pairing!