GitHub – blader/Claudeception: A Claude Code skill for autonomous skill extraction and continuous learning. Have Claude Code get smarter as it works.

If you’ve ever worked with an AI coding agent, you know this feeling. You finally crack a weird bug together, it takes forever, you both learn something subtle… and then the session ends. Next time? Back to square one. I’ve lived that loop more times than I’d like to admit, usually late at night with cold coffee.

That’s exactly the problem Claudeception is trying to solve.

Claudeception is a skill for Claude Code that lets the agent **remember what it learns** and reuse it later. Not in a vague, “I kind of know this” way, but as concrete, reusable skills that get automatically loaded when a similar situation comes up again.

Here’s how it works in practice. While you’re coding, Claude Code sometimes discovers something non obvious. A debugging trick. A workaround for a serverless edge case. A pattern that only shows up in your project. Claudeception runs a small retrospective at the end of sessions and asks, “Did we actually learn something worth keeping?” If the answer is yes, it writes a new skill to disk with a carefully crafted description so it can be retrieved later.

That detail matters more than it sounds. A skill called “Database fixes” is useless. A skill called “Fix for PrismaClientKnownRequestError in serverless environments” gets pulled in exactly when you need it. You feel the difference immediately.

What I really like is how picky it is. It won’t save things that came straight from documentation or one off hacks that never got verified. The bar is simple and human. Would this help you six months from now, when you’ve forgotten everything? If not, it doesn’t get saved.

The idea comes from real research on AI agents that build skill libraries over time, but here it feels grounded and practical. Less theory, more “thank you for not making me relearn this.”

If you want to explore it yourself, the project lives here:
https://github.com/blader/Claudeception

Looking ahead, this kind of persistent learning feels like the direction AI coding tools need to go. Less amnesia. More continuity. Tools that grow alongside you, instead of resetting every morning. And honestly, that sounds like a much nicer way to work.