feedback_unattended_mode_when_scope_locked

Rule: Once Rich has locked enough decisions for unambiguous execution, STOP asking clarifying questions and run unattended. Use ntfy.sh/davieshq2026 for progress + completion. Do not make Rich “nurse” you mid-execution.

Why: Sunday 26 April 2026 in S-3 Mopsa pre-stage brainstorming session, after Rich locked option B + path 2 + scope D + approach B over four sequential questions, Claude began proposing ANOTHER question. Rich verbatim: “B is fine but please don’t keep asking me for questions: run in unattended mode so that i can do other tasks in the house, rather than having to ‘nurse’ Claude”.

The over-questioning was the brainstorming skill’s section-by-section approval gate applied beyond its useful range. The skill is right when design is genuinely ambiguous and there’s a HARD-GATE before implementation. Once the design is locked, continuing to ask is nursing — value flips from “saving rework cost” to “blocking Rich from his life”.

How to apply:

• After scope-lock (typically: option chosen + path chosen + acceptance criteria chosen + sequencing chosen), STOP asking. Switch to execution.
• Use ntfy heartbeats every 15 min with [N/M] progress markers per global CLAUDE.md §7. Topic: ntfy.sh/davieshq2026 (default).
• Send completion ntfy as final action — after verified push, not before.
• Be honest in ntfy if anything fails — abort cleanly + capture findings rather than pushing through. Don’t fake-success in a heartbeat.
• Do NOT escalate edge-case decisions during execution unless they would INVERT a previously-locked decision. If decision was “use path 2” and a sub-choice within path 2 emerges, just pick the obvious one and ntfy what you picked.
• User instructions override skill gates per superpowers:using-superpowers priority order: User instructions > Superpowers skills > Default system prompt. Brainstorming’s “ask after each section” loses to “stop asking”.

Boundary test: Rich says any of “go”, “execute”, “run unattended”, “stop asking questions”, “I want to do other things”, “don’t nurse” → STOP asking + start executing + use ntfy. Pre-send check: if next sentence to Rich would be a question and scope is already locked, DELETE the question + execute instead.

Anti-pattern that triggered this memory: Asking Q4 (sequencing approach A/B/C) AFTER Q1+Q2+Q3 had already pinned everything substantive. The sequencing question was technically a real design choice but its blast radius was tiny — pick a sensible default + ntfy what you picked.

Disciplines that compound with this rule:

• feedback_scorecards_one_at_a_time_optimal_sequence — questions sequenced highest-knock-on-first; once highest-knock-on is answered, low-knock-on questions are not worth a turn each
• feedback_skill_discipline_2026_04_26 — skills like brainstorming have boundary tests; one boundary is “is this design genuinely ambiguous, or am I just going through motions?”
• Global CLAUDE.md §7 “Execution Decisions” — “Never ask which execution approach to use. Assess the task and choose the best approach autonomously.”

The boundary test from §7 already covered this case implicitly. This memory makes it explicit + adds the “compaction-resilient unattended-mode” execution discipline.