Most "AI assistant" and "second brain" templates are note-takers with extra steps, or eager yes-men that help you feel productive. This is the opposite: a Claude Code assistant that learns how you specifically self-sabotage and refuses to let you hide from the work that actually moves the needle.
Built for people who would rather hear the truth than be managed.
The model is commoditized. The harness is commoditized. The thing that isn't: an assistant that holds a real model of you — your goals, and the specific ways you get in your own way — and acts on it. "Anti-sycophancy" is having a moment as a feature. This isn't a feature. It's the operating principle, fused with accountability to your own goals.
I built it because I have a wide gap between what I can do and what I actually ship, and a documented habit of building systems to avoid the scary work. Every assistant I tried made it worse by cheerfully helping me stay busy. So I built one that calls it.
- Tells you the truth. The hard read over the comfortable one. No epistemic cowardice.
- Knows your self-sabotage. You name your failure modes on day one; it catches you in them in the moment, not as a lecture.
- Leads, then holds the line. Opens as a capable guide with a plan. The edge shows up when you start to dodge.
- Adapts to how you think. Tell it how your brain works, neurodivergence included, and how hard to push.
- Remembers you. Cross-session memory, not a goldfish.
- Clone the repo.
- Open the folder in Claude Code.
- Say
hi.
It interviews you, builds your operating manual, and gets to work. No keys to wire, no config. Your assistant takes the name you give it.
It is plain markdown and skills, about 20 files. Fork it, rip out what you do not want, add your own skills and rules. If you build something good on top, or it tells you something that stung in the right way, tell me.
Early and intentionally small. Prompts and structure, not magic. If it says something you do not want to hear, that is the point.