finds.dev← search

// the find

coleam00/second-brain-starter

★ 799 · updated Jun 2026

Build your own AI Second Brain with Claude Code - a skill that generates a personalized PRD for a proactive, persistent AI assistant

This is a Claude Code skill, not an application — it interviews you about your tools and workflow, then has Claude generate a 9-phase PRD for a personal 'second brain' agent (memory files, hooks, integrations, a proactive heartbeat, a chat bot). It's for someone who already uses Claude Code and wants a structured starting plan for a big personal-automation project, not someone who wants a working assistant out of the box.

The architecture it recommends is sane: markdown files as memory instead of a database, a Python CLI wrapper pattern so the LLM process never sees raw API keys, and a heartbeat design that runs cheap deterministic Python to gather data before spending a Claude call to reason about it (their own numbers: ~$0.05/run vs $0.38 doing it all through MCP tool calls). The BOOTSTRAP.md self-onboarding file (interviews you once, fills in SOUL/USER/HEARTBEAT, then deletes itself) is a clever bit of UX for a markdown-only system. The proactivity-level framework (Observer/Advisor/Assistant/Partner) is a real design decision baked into the generated plan, not just a toggle.

There is essentially no working code here — the repo is a skill definition, a requirements template, and some starter markdown files. Everything described in the README (memory search, integrations, the heartbeat, the Slack bot, security hardening) is work you still have to build yourself across 9 phases; the repo only produces the plan for it. 'Security Hardening' is a template section with placeholder language, not an actual sandbox, permission system, or tested guardrail — treat anything with real credentials or irreversible actions with suspicion until you build and test that phase yourself. It's also tightly coupled to Claude Code and Anthropic's models by design, so there's no path for someone who wants to use a different agent runtime. Worth noting the README funnels toward a paid 4-hour workshop, which colors how complete you should expect the free templates to be.

View on GitHub →

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →