Four Disciplined Moves
Ask, Sort, Research, and Experiment structure every design question. Questions are sorted as world fact, user-held fact, or judgment before any work proceeds, and unsorted questions never reach the user.
Skill
Ask, sort, research, experiment — the model drives, the user rules.
/attune:explore Explore drives design discussions through four moves: Ask, Sort, Research, and Experiment. Ask raises purpose, alternative, and shortcut questions at the model first, then brings the user only well-shaped user-bound questions; Sort classifies every open question as world fact, user-held fact, or judgment; Research settles world facts through the cheapest sufficient source; Experiment creates evidence when nothing existing holds the answer. The model drives; each pass ends with labeled findings.
Ask, Sort, Research, and Experiment structure every design question. Questions are sorted as world fact, user-held fact, or judgment before any work proceeds, and unsorted questions never reach the user.
The model raises the hard questions itself and answers everything its own instruments can reach, leaving the user only their own facts and rulings — never legwork.
Research follows a source order of memory, this machine, then the web. Load-bearing recalled facts are verified against their sources before a recommendation rests on them.
Every pass presents purpose, explored alternatives, the strongest shortcut, and the recommendation, with facts, judgments, assumptions, and parked items each labeled as what they are.
Raise purpose, alternatives, and shortcut questions, then bring user-bound questions.
Raise purpose, alternatives, and shortcut questions at the model first; keep the best-guess summary and open question list visible; ask one call of up to four well-shaped user-bound questions.
Classify every open question before any work proceeds.
Mark each question as world fact, user-held fact, or judgment; mark user-bound questions; route third-party or access-gated facts as routing requests; prepare ambiguous questions as judgments.
Settle world facts through the cheapest sufficient source.
Follow memory, this machine, then the web, escalating only when a source cannot settle. Verify load-bearing recalls, run one scouting pass per settled intent, and compile a digest naming gaps.
Create evidence when nothing existing holds the answer.
Run probes against real artifacts, record commands and outputs beside briefs, and get go-ahead for trials that spend money or touch external systems. Invoke /attune:experiment for blind comparisons.
Present the pass's findings with all items labeled.
State current purpose, explored alternatives, strongest shortcut, recommendation with rationale, and unverified or parked items. Label facts, evidence, judgments, and recommendations as such.