本文へ移動
cccskills
無料GitHub で公開

review-cockpit

Produce and continuously maintain ONE living review document for a multi-item session — a cockpit header (Progress checklist, Working folder, Context) plus per-item review cards that you approve or request changes on directly in the doc or side panel. Use whenever a session has multiple deliverables you need to review/approve (meeting-processing + planning, multi-ticket work, briefs with several drafts, any "do X, then plan/draft Y and Z"). The doc is the interaction surface: you edit it (or reply in chat) to approve/change; the agent keeps it live as work progresses.

インストール方法を見る

含まれるファイル(2)

  • SKILL.md5.7 KB
  • references/session-review-template.md670 B

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。

review-cockpit

Purpose

A single, living review document that doubles as the control surface for a session. Instead of scattering a meeting note here, a ticket there, two drafts in chat, everything lands in one file you can open in the Obsidian side panel (or Claude side panel) and drive: see progress at a glance, review each item in place, approve or request changes inline, and watch the doc update as the agent works. Mirrors a "co-work" cockpit (Progress / Working folder / Context).

When to use

  • Any session with ≥2 things you need to review or approve (the default for "process X then plan/draft Y, Z").
  • Meeting-processing + pod planning, multi-ticket runs, briefs with several drafts, spec + plan + drafts.
  • NOT for a single trivial deliverable (one file, no approval loop) — a plain file is fine there.

Pairs with closed-loop (verification), harvest (learnings), and the V-model checkpoints. This skill governs how the deliverable is presented and driven, not the verification pipeline.

The one rule

One doc per session. It opens with a cockpit, then one review card per item. Every artifact the session produces is linked from the Working folder table (fan-out to sub-files mid-run is fine; the cockpit is the single front door). Never hand the user "see files A, B, C" — hand them the cockpit.

Structure

Copy references/session-review-template.md. Save the working copy in the relevant project folder as YYYY-MM-DD-<slug>-review.md (or -plan.md).

Cockpit (top):

  1. Status line — one line: how many items await review + last-updated timestamp.
  2. 🧭 Progress — a checklist, one row per item, each with a status glyph; a text progress bar + "X/N done · A awaiting review · B pending". Make each item a clickable anchor to its review card — [N. Title](#n-title) using GitHub-slug rules (lowercase, spaces→-, drop punctuation/emoji). Keep item headings free of + # ← → ( ) so slugs stay clean single-hyphen and the links resolve in Obsidian + the Claude side panel.
  3. 📁 Working folder — table of every artifact produced this session with its ~/vault/... path or link (meeting note, this doc, evidence dir, external issues/PRs, posted messages).
  4. 🔌 Context — sources the work is grounded in (recordings, Slack threads, issues), tools/connectors used, related tickets.

Review items (below): one card per item:

  • Status — glyph + word (see vocabulary).
  • Action — what was done or is proposed.
  • Deliverable — link/path, or an inline draft placed directly under the card so you review it in place (messages to send, doc bodies).
  • Checklist — the item's acceptance criteria as checkboxes.
  • 🗒 Your call — the approval slot you edit (approve / your requested changes).
  • Decision log — append-only record of what happened / what you decided.

Footer: "How to drive this doc" (approve/change mechanics + status vocabulary), so the doc is self-explaining.

Status vocabulary: ⏳ pending · 🔄 in progress · 📝 needs review · ✏️ changes requested · ✅ done · ⛔ blocked

Update protocol (keep it live)

  • After every meaningful step, refresh: the Progress checklist + bar, the affected item's Status, the Working folder (add new artifacts), and the item's Decision log. Bump the updated: frontmatter + status line.
  • Prefer targeted Edits over full rewrites so your inline edits/approvals are never clobbered. Before editing, re-read the file — you may have typed approvals or change requests into 🗒 Your call or ticked boxes.
  • When you approve an item, execute it, move it to ✅, and log the outcome (with the external link — issue/PR/message URL) in the Decision log.
  • When you request changes, set the item to ✏️ changes requested, apply them, re-draft in place, then return it to 📝 needs review.

Interaction contract

The doc is the shared surface; you drive it two ways, both valid:

  • In the doc / side panel: you edit 🗒 Your call or tick checkboxes. Re-read and act.
  • In chat: "3 is ok", "approve 2 and 4", "change the tone on the requirement comms". Map to item numbers and act.

Distinguish draft items (agent proposes, waits — e.g. a Slack message you own sending) from auto items (agent may execute directly — e.g. filing a GitHub issue you explicitly asked for). Draft items stay 📝 until approved; auto items go straight to ✅ with the artifact linked. When unsure whether an item is draft or auto, leave it 📝.

Post-condition

Read-only for the doc itself, but the items it tracks often mutate external state — obey each item's own post-condition (fetch back the issue, confirm the Slack message landed, etc.) and record the verified link in the Decision log. Never mark ✅ from a mutation call alone.

Optional: rendered view

The markdown doc is primary (you edit it to approve). If you want a richer at-a-glance view, additionally render an HTML Artifact of the cockpit (Progress + Working folder + Context + item statuses) — but the markdown stays the source of truth and the editable surface.

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

Deep strategic research engine — decomposes questions into parallel research threads, spawns multiple agents, and synthesizes into actionable strategic analysis

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。

huytieu/COG-second-brain1,2762026年10月2日 更新

braindump

無料

Quick capture of raw thoughts with intelligent domain classification and competitive intelligence extraction

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。

huytieu/COG-second-brain1,2762026年10月2日 更新

Run one task through the V-model verification loop: CP-2 plan → CP-3 build → CP-3v component verify → CP-4 integration verify (full lane) → CP-5 acceptance. The worker never grades its own homework; evidence rows trace back to AC-n. Opt-in: invoke with /closed-loop or by asking for the closed loop, proper verification, or an evidence trail. Ordinary work does not run this.

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。

huytieu/COG-second-brain1,2762026年10月2日 更新

Deep-dive 7-day analysis across all data sources for weekly reviews, board prep, and strategic planning

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。

huytieu/COG-second-brain1,2762026年10月2日 更新

Autonomous content pipeline - scout announcements in your field, triage by trend momentum and personal angle, produce posts/blogs/videos in your voice with ledger-based dedup, hard volume caps, and screenshot-verified publishing

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。

huytieu/COG-second-brain1,2762026年10月2日 更新

Create user stories with duplicate checking across any project tracker (Linear, GitHub Issues, Jira)

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。

huytieu/COG-second-brain1,2762026年10月2日 更新

huytieu のスキルをすべて見る

このスキルの問題を報告する