メニュー
Use for SHAFT PDCA, Kevin/Bob/Bruce roles, or refinement loops.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Use when analyzing SHAFT Allure results, Doctor reports, trace evidence, healer output, flaky locator/wait/assertion failures, retries, or test-fix recommendations.
Use when creating, reviewing, refactoring, repairing, or generating SHAFT web/mobile locators, smart locators, ARIA locators, XPath/CSS replacements, or codegen element identifiers.
Use when recording browser, Playwright, mobile, Appium Inspector, or user-performed flows into SHAFT tests with MCP Capture, replay, code blocks, or codegen insertion.
Use when reviewing, previewing, applying, guardrail-checking, or verifying generated SHAFT Java before or after inserting it into a repository, especially the coding-partner diff/apply/verify loop from IntelliJ.
Use when writing, reviewing, or repairing SHAFT Java tests, page objects, API tests, mobile tests, CLI/DB tests, assertions, waits, or TestNG/JUnit/Cucumber scenarios.
Forces the laziest solution that actually works, simplest, shortest, most minimal. Channels a senior dev who has seen everything: question whether the task needs to exist at all (YAGNI), reach for the standard library before custom code, native platform features before dependencies, one line before fifty. Supports intensity levels: lite, full (default), ultra. Use on ANY coding task: writing, adding, refactoring, fixing, reviewing, or designing code, and choosing libraries or dependencies. Also use whenever the user says "ponytail", "be lazy", "lazy mode", "simplest solution", "minimal solution", "yagni", "do less", or "shortest path", or complains about over-engineering, bloat, boilerplate, or unnecessary dependencies. Do NOT use for non-coding requests (general knowledge, prose, translation, summaries, recipes).
| name | agentic-pdca-loop |
| description | Use for SHAFT PDCA, Kevin/Bob/Bruce roles, or refinement loops. |
Each activity uses exactly one isolated persona: Kevin plans spec/value/acceptance/risks/Mermaid/wireframe; Bob implements the smallest cross-platform change/tests; Bruce checks E2E evidence/defects/ambiguity/confidence. Do not merge roles.
Run PDCA in Kevin->Bob->Bruce order, then repeat that order for two passes: quality/simplicity, then intuitiveness/acceptability. Rerun the smallest check each pass. Stop at >=90% confidence or a blocker.
All three personas run sequentially on the main thread of the same session -- never as spawned agents, workflows, or orchestrators (see AGENTS.md Agent Hierarchy: workflows are forbidden and .claude/workflows/ must never be recreated). Switch persona by switching phase: Kevin plans, then Bob implements the smallest change, then Bruce judges the actual diff plus real checks per AGENTS.md Validation (never Bob's self-report; hunt stubs and weakened assertions) and closes remaining gaps himself after at most three Bob rounds. Record which phase produced each commit. Decision: the marketplace ralph-loop plugin stays unadopted here -- an unbounded same-context Stop-hook loop invites token/CPU runaways given this repo's Maven fork gotchas; the capped main-thread PDCA loop is the sanctioned loop.
