opencrow — Self - Claude Code that gets smarter the more you use it

Hey Product Hunt 👋 I built opencrow to scratch my own itch. I use Claude Code every day, and the same thing kept bugging me: every session starts from zero. It doesn't remember how I like things done, and I'd keep re-explaining the same context or re-solving the same little tasks. Claude Code actually ships the building blocks to fix this — hooks, skills, memory, MCP — but nothing wires them together into something that learns from how you use it. So I built that layer. opencrow is a thin wrapper. You run opencrow instead of claude, and the foreground is your exact same interactive session — nothing intercepted, nothing re-rendered. What changes is around it: - Each turn, it pulls relevant work from your past sessions into context (and you can full-text search your own history). - When the session ends, a background pass indexes the conversation, remembers your preferences, and turns recurring tasks into reusable skills — written into your global ~/.claude/skills/, so any claude session benefits. Two things I cared about a lot: - No extra API key. It drives the claude you're already logged into, so it stays on your subscription. - Everything is inspectable and reversible (opencrow memory / opencrow skills, archive-not-delete). Auto-generated stuff you can't undo is scary to rely on. It's MIT, and macOS (arm64) is the primary target right now. Other platforms may work but are less tested — feedback there especially welcome. I'd love to hear: what's your current setup for making Claude Code remember things across sessions? And what kinds of skills would you want it to learn for your workflow?