메뉴
Use before any creative/design work (a feature, a package, an evolution) — explores intent and design via the spec-ladder with multi-peer adversarial review BEFORE implementation.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when bringing the harness method to a NEW or non-sentropic repo — maps the repo's conventions into a harness profile so the same scope/branch/verify kernel applies anywhere.
Use on any bug, test failure, or unexpected behaviour BEFORE proposing a fix — runs the generic evidence-first root-cause loop.
Use when you have a spec/decisions and need an executable plan — writes a lot-based BRANCH.md from the template and completes with track for realization state.
Use when completing a feature, a design step, or before merging — runs a ≥2-peer consensus review and reconciles the findings, instead of a single rubber-stamp pass.
Use when adding or evolving tests for a feature/bugfix — applies the sentropic test-pyramid, scoped-test loop, environment isolation, and AI-flaky policy.
Load FIRST in any sentropic repo — establishes the harness method (branch/scope/test/debug/review/brainstorm/plan) and supersedes superpowers for those acts.
| name | brainstorm |
| description | Use before any creative/design work (a feature, a package, an evolution) — explores intent and design via the spec-ladder with multi-peer adversarial review BEFORE implementation. |
Native sentropic ideation. Stronger than a single-pass brainstorm: it climbs the spec-ladder and gates each design step with ≥2 independent adversarial peers and batched owner decisions.
Open the act: harness brainstorm "<topic>" [--peers <n>] [--ladder study|vol|evol] (records a WorkEvent).
Pick the rung that matches the uncertainty, write the artifact under spec/:
SPEC_STUDY_<TOPIC>.md) — open question, options, trade-offs. No commitment yet.SPEC_EVOL_<TOPIC>.md) — the committed design with numbered decisions (D1…Dn) ready to plan.Don't skip rungs for a genuinely open problem; don't over-ceremony a reversible one.
Before presenting decisions, get ≥2 independent reviews from distinct angles (e.g. Opus 4.8 + Codex 5.5-xhigh, or distinct lenses: correctness / risk / user-value). Reconcile divergences explicitly — a lone reviewer rationalises; a panel surfaces failure modes. Reversible calls: decide solo. Blocking / irreversible calls: escalate.
Present owner decisions as a clean full dossier (never a delta), THEN ask in lots of ≤4 questions. Each question carries its own context inline (source + stakes + consequence) — an id like "D5" alone carries no meaning. Once answered, treat all answers final; do re-examination in prose, don't re-poll.
A ladder artifact exists, peers reconciled, decisions captured. Hand off to harness/plan to turn the
EVOL decisions into BRANCH.md lots.