baskduf/harness-starter-kit — reverse-engineered prompt

Build me a small open source starter kit that helps people make their code repos safer and more predictable when they use AI coding agents.

I want it to be prompt first, so the main thing is a clear adoption prompt and workflow that an agent can follow inside any existing repo. It should inspect the repo first, preserve the current tools and style, then add only useful guardrails like AGENTS.md instructions, a simple knowledge store if needed, lightweight drift checks, local verification commands, and an adoption report.

Please include prompt based commands people can paste into their agent, especially /harness doctor, /harness update, /harness refresh, /harness review, and /harness review sub-agent. These should explain what to check, when to modify files, and when to stay read only.

Add friendly docs that explain the idea of turning repeated AI mistakes into durable instructions, checks, memory, and evaluation. Include examples and templates, but make it clear agents should not blindly copy them.