Build a small question-and-answer system for AI agents in this empty working folder. Use made-up data only; touch nothing
outside the folder, use no network, and ask me nothing while you build.
Agents write a question in a list that survives restarts, each with a proposal (what they would do themselves), and carry on
with work that does not depend on it. I answer by number, or adopt the proposal in one step. An agent can look up an answer and is
told to leave the work lying if there is none. One page I open in my browser shows only the open questions with their proposals.
Add a short paragraph I can paste into any agent's assignment: ask in the list, never in the conversation.
I check it with two made-up questions from Bel: the page shows two open, and after I answer both it shows none open.
