Claude Starter Kit

Three pastes and Claude actually knows who you are, how you work, and how to handle big projects without losing the plot halfway through.

What's in it: A Personal Preferences template, a workspace CLAUDE.md template, the Research / Plan / Implement (and optional Validate) prompts, and a decision guide on when to add an Obsidian or Notion vault on top.

Why it exists: I had this scattered across three platforms: Slack saved messages, a GitHub gist, and a Google Doc. A few friends kept asking how I use Claude, and I kept copy-pasting setup instructions from different places. The canonical version was overdue.

How to use it: Paste 1 trains Claude on who you are. Paste 2 turns any folder into a workspace it can navigate. Paste 3 is the Research, Plan, Implement workflow for any project bigger than a quick task. Validate is the optional fourth step for high-stakes work.

Status check, September 2026: Barely changed since May, on purpose. The kit is stable. What changed is the advice I give alongside it.

The advice I actually give people now. Use Claude. Opus 5 on Medium or High. Then ask it to overcome every problem it runs into instead of routing around it. Can't reach the data? Connect it to the data. No integration exists? Have it build an MCP server. Blocked on access? Have it tell you exactly what to unblock and who to ask. Do every task with it for a week or two, including the ones where doing it yourself would be faster today. The gains compound. Around week two you look up and most of the work you thought was hard is automated.

What that unlocked for me: researching Jira tickets, digging through years of backlog nobody has read since it was filed, pulling customer sentiment out of recorded CSM calls, and synthesizing thousands of support tickets into a real picture of where the product actually hurts. Once you can synthesize that fast, you prioritize instead of guess.

What it isn't: A finished product. It's the version that works for me, opinionated, fork-friendly, and stripped down to what I actually use. The README walks through what each piece does and which parts to skip if they don't fit your work.

Where the prompts came from: The RPI workflow is adapted from HumanLayer's Claude Code commands, generalized for knowledge workers who aren't shipping code. The PARA folder structure is from Tiago Forte's Building a Second Brain. The Validate step pulls from the broader Claude Code community pattern.