Coding agents today (Claude Code, Codex, etc.) don't remember know-how - they don't get smarter in the work.

Most memory plugins only store information and facts — not workflows: the process, and the pitfalls you've already hit.

So I built Skillmaxxing. Inspired by the Hermes agent, it lets your agent learn and grow as it works — after finishing a task it automatically reflects, then creates a new skill or updates an outdated one.

And the payoff isn't just that the agent gets smarter.

Because it skips the repeated exploration, it also saves a lot of tokens.

Before building it, I wasn't sure it'd even be useful — I suspected there wouldn't be much worth learning or recording.

Then I used it myself for two days, and it had already created 4 skills — all of them reasonable.

It's open source. Try it, share feedback, or contribute!