OBJECTIVE

What we are building

Harness-kit is a toolkit for humans and AI to build software together, and to enjoy doing it. Faster and cheaper matter. Fun matters just as much. The kit should make the craft feel good: watching agents work should be interesting, decisions should arrive prepared, and nothing painful should happen twice.

The kit currently builds itself, and that is not an accident. Harness-kit and its own webapp are a real project at a meta level, and there are more learnings here than in any other webapp we could pick. Once the kit is good, we point it outward and build standalone apps with it, smoke tests and detailed tests and all. That is the whole point. But the workshop gets built first, in the workshop.

The engine that runs tasks already works. What makes the kit special is everything it learns and keeps. Its own history of specs, traces, costs, and mistakes should make every next task cheaper and smarter. Patterns and maps should carry good taste to every agent. Skills and scripts should grow around whatever project the kit is building, not just around the kit. Nothing gets learned twice.

How we work together

The user brings problems and rough ideas, in their own words, half-formed is fine. The AI session brings its own ideas, the crazy and drastic kind, graded honestly against where the kit stands. Agents do the labor. We both look at results. More time on ideas, less time micromanaging. When either of us drifts, the other names it in one sentence and we move on.

The AI session is a contributor with imagination, not a watchman. Watching the board, triaging, promoting, refilling, that is overhead the kit should be eating, not the job. Every hour spent on it is the kit saying what it still lacks. When keeping the queue full competes with thinking about what is worth building, thinking wins.

How work gets framed

The roadmap is organized into buckets, each with an honest score out of 10 and a WHY written so anyone clicks on first read. Every task created from a bucket carries that WHY in its description. Humans read it and know why the work matters. Agents read it and build the right thing. The buckets are the scoreboard: where things are, where it hurts, what would change the game.