the-seed.md
A kit that makes kits. It interviews you about what you’re up against, tells you honestly when a markdown file isn’t the answer, and builds a new kit for your situation that you can keep, rerun, and hand to the next person.
fwdio.
Fine. We’ll do it ourselves.
Self-contained kits: plain markdown files that hand an ordinary person real leverage. Drop one into your AI agent and the work happens on your machine. No accounts, no backend, nothing running on ours. The file is the product.
How to run a kit
- Download a kit. It’s just a .md file.
- Drop it in a fresh folder and open your AI agent. Built for Claude Code; works anywhere.
- Say “read this and run it.”
Kits · three so far, on purpose
policy-eval-kit.md
Turns a written policy into a running evaluation. Built for the trust and safety analyst who knows exactly what a violation looks like and keeps hearing that the checking is an engineering problem. A rubric with anchors, a boundary-heavy test set, blind grading, a calibration number, and a report someone can act on.
the-council.md
Convenes five minds around your hardest decision: three cast for the problem, one who has to live with the outcome, and an empty chair for whoever isn’t in the room. Blind pitches, forced dissent, one honest quarrel, and a verdict with kill criteria and a minority report. Better than one voice agreeing with itself.
Curation over volume: five great kits beat fifty. Every kit must be able to say a markdown file isn’t the answer. Crime, HR matters, and mental-health crises route to real humans, always.