rubpergar/krill — reverse-engineered prompt

Build me a reusable agent workspace template called Krill for OpenCode. I want it to help coding agents work in a more disciplined way before they touch product code.

The repo should start as a skeleton, with no actual app code yet. It should include clear agent rules, source of truth docs, task tracking, backlog, plans, checklists, Definition of Done, testing guidance, decision notes, debt notes, and a safe bootstrap flow for adopting it into a real project later.

Please include OpenCode commands for bootstrapping the skeleton into a project, making semantic commits, turning rough requests into better prompts, running those prompts, regenerating the README, and doing a fast path for tiny harmless changes.

The workflow should encourage spec driven development, TDD, validation, and updating durable documentation only when contracts change. Make the README polished and explain what this is, why it exists, how to install it, and that feature work is blocked until bootstrap is complete.

Look up current OpenCode docs online if you need to.