本文へ移動
cccskills
無料GitHub で公開

post-analiz-handoff

Use when the stakeholder asks about an analysis outcome or its implementation tasks - understand the human approval gate and review the tasks the architect created once the human approved

インストール方法を見る

含まれるファイル(1)

  • SKILL.md4.0 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Post-Analiz Handoff

Overview

The analiz has a human approval gate. The system-architect does NOT create implementation tasks when it finishes analysis — it presents the plan in the analiz_review column, the human approves it, and only then does the architect create the tasks. Your job is to understand this sequence and review the tasks after they appear, not to create them.

The sequence (who does what)

  1. Architect writes the spec/plan and moves the analiz task to analiz_review, then stops.
  2. Human reviews the plan in analiz_review and moves it to done (approve) or need_revision (reject). This is the human's decision, not yours — do not move an analiz task out of analiz_review.
  3. Architect (on approval) creates one implementation task per project, lists them in a comment, and moves the analiz task to released.

Your steps

Trigger: the stakeholder asks what an analysis found, or about the tasks it produced. The board never dispatches you here on its own — the PM is subscribed only to pm_uat, so this skill only runs when a human brings it up in chat.

  1. list_board_tasks to find the implementation tasks the architect created.
  2. Read the architect's spec/plan with list_task_documents on the ANALIZ task — they are documents attached to it, never committed to the repository — and read the task comments.
  3. Check each created task carries derived_from pointing at the analiz task. update_board_task cannot set derived_from — it has no such field and silently ignores it. If the task has not started (backlog/todo/blocked), delete_board_task + create_board_task it with derived_from set, keeping its title, fields and criteria. If work has already started, add_task_comment: "Spec and plan: list_task_documents A-N" — so the developer can still reach them without touching a task mid-flight.
  4. Check the order is declared, not just described. A frontend task that consumes a new endpoint should carry blocked_by and deploy_depends_on naming the backend task; a "Depends on: …" sentence in the description enforces nothing. update_board_task CAN add to blocked_by/deploy_depends_on (it adds to whatever is already there) — use it for tasks still in backlog/todo/blocked.
  5. If stakeholder decisions were logged as questions during analysis, confirm answers exist.
  6. Review scope and priority. Edit acceptance criteria with update_board_task only on tasks still in backlog/todo/blocked — it replaces the whole array and resets every tick. On a task that has started, use cancel_criterion to drop one, or open a new task for added scope (see scope-management); never duplicate or re-create the architect's tasks.
  7. Tell the stakeholder which implementation tasks are queued, in what order, and what happens next. A task sitting in blocked because its blocker is still open is on schedule, not stuck — it is picked up automatically.

Common Mistakes

  • Creating implementation tasks yourself — the architect owns decomposition, gated on human approval.
  • Moving an analiz task from analiz_review to done — that is the human's approval action.
  • Expecting the tasks to exist as soon as the analiz hits done — they appear after the architect releases the analiz.
  • Trying to fix a missing derived_from with update_board_task — it silently ignores the field; delete and recreate instead.
  • Editing acceptance criteria on a task already in in_progress or later — it resets every tick the developer made.

Red Flags

  • You are about to create_board_task for work the architect analyzed → stop; the architect creates it after approval.
  • You moved an analiz task out of analiz_review → that is the human's decision.
  • You called update_board_task(derived_from=...) expecting it to take effect → it won't; delete + recreate instead.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Use when writing acceptance criteria for a task - express each as an observable Given/When/Then that QA can execute, including negative cases

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月12日 更新

Use when the diff adds or changes an endpoint, resolver, RPC, job or query that takes an object id, a role check, a request binding or a tenant filter - BOLA/IDOR, function-level authorization, mass assignment and tenant scoping

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月12日 更新

Use on every UI change - semantic HTML, labels for controls, keyboard-navigable dialogs/menus, visible focus, and never color as the only signal

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月12日 更新

Use when a task changes any screen, form, dialog, menu or control - Lighthouse/axe scan of the changed screens, a keyboard walk, and the thresholds that fail a task

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月12日 更新

How to work a task returned with review, QA or UAT findings. Use when a task is in need_revision or PR review comments are in your context.

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月12日 更新

Use when deciding whether a request needs an analiz task before implementation - the conditions that require the architect's analysis versus going straight to implementation

日本語の概要は準備中です。原文の説明を表示しています。

makifbaysal/tasktrooper1122026年10月12日 更新

makifbaysal のスキルをすべて見る

このスキルの問題を報告する