looplisted

# Unknowns Loop — operating loop for working with strong models Core principle: **the bottleneck with strong models is not the model — it's the user's ability to keep the map (plan) matched to the territory (reality). That gap is the unknowns.** Designing the explore → question → plan → implement → log deviations → verify loop matters more than one good prompt. Origin: the 11 "Know your unknowns" examples; the loop itself is a plugin extension — see skills/loop/references/talk-source.md. Output: Artifact tool → `.unknowns/<YYYY-MM-DD>-loop-<slug>.html` → markdown. Each stage's artifact keeps its own skill's reaction control; the loop's own control is the per-stage checkpoint, which assembles into the `.unknowns/loop.json` tracker. Details: skills/loop/references/output-routing.md ## Steps (10 — scale down by size) ### 1. Define value & done criteria Fix with user before starting: purpose / value to user / done criteria / existing behavior that must never break / allowed cost & change scope. **Without success criteria, implementation volume gets mistaken for progress.** ### 2. Blind-spot investigation → `/unknowns:blindspot` (or `/blindspot` for copied installs) Investigate only, no implementation. Find unknown unknowns, conflict points, regression risks; produce improved prompt. If the **domain itself** is unfamiliar (requests vague because terms unknown), run `/unknowns:teach-me` alongside. ### 3. Interview → `/unknowns:interview` Architecture-changing questions first,