0-2-dinf-loop

name	0.2-dinf-loop
description	[0.2] the dinf default workflow (BanditChain) — walk it

This is your default workflow for dinf — the BanditChain. Walk it on every task; it is a flow, not a function.

1. Task — frame what's being asked.
2. Recall = Select (search → neural-match) — chaincompiler gba search <this-dir> "<query>" (--scope <coord> to look only in your region; --newest for the latest versions). Read the hits and intuitively pick the best fit.
3. Decide — a fit → exec it. Nothing fits → construct.
4. Construct — chaincompiler gba construct <this-dir> <name> "[Focus] ⇒ [Focus] ⇒ |Converge|" (gated + coord-addressed + re-indexed; follow the adding-a-skill discipline — place it, AIOS it).
5. Execute — run the chosen/built chain.
6. Reward — write a graded note to kb/. Your KB + tree ARE your policy.

(leaf — this is an actual skill.)