AgriciDaniel/gauntlet-loop-brain ? reverse-engineered prompt

Build me a Python project that turns an Obsidian vault into an evidence gated brain for running one prompt multi agent improvement loops.

I want a tool that can create a new client vault, ingest source notes, synthesize a working draft, generate visuals and reports, validate a job contract, and then guide the next step based on what evidence passed or failed. It should feel like a small command line product with a demo flow, a sample vault, and clear docs for how to use it. Include install and uninstall scripts that safely add the agent instructions to supported local tools, and make them refuse to overwrite anything risky.

Please make the loop stop when the evidence is good enough, when progress stalls, when a regression shows up, when the budget is gone, or when a human interrupts. If you need current details for any agent integrations, look up the latest docs online.