Design: {title}

Generated by /office-hours on {date} Branch: {branch} Repo: {owner/repo} Status: DRAFT Mode: Startup Supersedes: {prior filename — omit this line if first design on this branch}

Problem Statement

{from Phase 2A}

Demand Evidence

{from Q1 — specific quotes, numbers, behaviors demonstrating real demand}

Status Quo

{from Q2 — concrete current workflow users live with today}

Target User & Narrowest Wedge

{from Q3 + Q4 — the specific human and the smallest version worth paying for}

Constraints

{from Phase 2A}

Premises

{from Phase 3}

Cross-Model Perspective

{If second opinion ran in Phase 3.5 (Codex or Claude subagent): independent cold read — steelman, key insight, challenged premise, prototype suggestion. Verbatim or close paraphrase. If second opinion did NOT run (skipped or unavailable): omit this section entirely — do not include it.}

Approaches Considered

Approach A: {name}

{from Phase 4}

Approach B: {name}

{from Phase 4}

Recommended Approach

{chosen approach with rationale}

Open Questions

{any unresolved questions from the office hours}

Success Criteria

{measurable criteria from Phase 2A}

Distribution Plan

{how users get the deliverable — binary download, package manager, container image, web service, etc.} {CI/CD pipeline for building and publishing — GitHub Actions, manual release, auto-deploy on merge?} {omit this section if the deliverable is a web service with existing deployment pipeline}

Dependencies

{blockers, prerequisites, related work}

The Assignment

{one concrete real-world action the founder should take next — not "go build it"}

What I noticed about how you think

{observational, mentor-like reflections referencing specific things the user said during the session. Quote their words back to them — don't characterize their behavior. 2-4 bullets.}

模板各章节与技能主体阶段一一对应：Problem Statement/Constraints/Success Criteria 来自 Phase 2A 诊断；Demand Evidence 对应 Q1、Status Quo 对应 Q2、Target User & Narrowest Wedge 对应 Q3+Q4；Premises 来自 Phase 3 前提挑战；Approaches Considered 来自 Phase 4 强制备选一；Cross-Model Perspective 只在 Phase 3.5 第二意见**实际运行**时才包含，否则整节省略——这是硬性约束，防止模板空壳。

### Builder 模式设计文档模板

Builder 模式服务个人项目、黑客松、学习与开源场景，模板重心从"需求证据"转向"爽点与构建路径"：

```markdown
# Design: {title}

Generated by /office-hours on {date}
Branch: {branch}
Repo: {owner/repo}
Status: DRAFT
Mode: Builder
Supersedes: {prior filename — omit this line if first design on this branch}

## Problem Statement
{from Phase 2B}

## What Makes This Cool
{the core delight, novelty, or "whoa" factor}

## Constraints
{from Phase 2B}

## Premises
{from Phase 3}

## Cross-Model Perspective
{If second opinion ran in Phase 3.5 (Codex or Claude subagent): independent cold read — coolest version, key insight, existing tools, prototype suggestion. Verbatim or close paraphrase. If second opinion did NOT run (skipped or unavailable): omit this section entirely — do not include it.}

## Approaches Considered
### Approach A: {name}
{from Phase 4}
### Approach B: {name}
{from Phase 4}

## Recommended Approach
{chosen approach with rationale}

## Open Questions
{any unresolved questions from the office hours}

## Success Criteria
{what "done" looks like}

## Distribution Plan
{how users get the deliverable — binary download, package manager, container image, web service, etc.}
{CI/CD pipeline for building and publishing — or "existing deployment pipeline covers this"}

## Next Steps
{concrete build tasks — what to implement first, second, third}

## What I noticed about how you think
{observational, mentor-like reflections referencing specific things the user said during the session. Quote their words back to them — don't characterize their behavior. 2-4 bullets.}

两模式的关键差异：Startup 模板有 Demand Evidence / Status Quo / Target User & Narrowest Wedge / The Assignment（给创始人布置一个具体现实行动，而不是"go build it"），Builder 模板则用 What Makes This Cool 与 Next Steps（第一、第二、三步具体构建任务）替代。两种模式下 "What I noticed about how you think" 都要求引用用户原话而非刻画其行为，这是与后文交接话术共用的 anti-slop 纪律。

写后告知

写入完成后必须对用户说：

"Design doc saved to: {repo path if written, else ~/.gstack path}{when both: ' (cross-session copy in ~/.gstack)'}. Other skills (/plan-ceo-review, /plan-eng-review) will find it automatically."

即明确告知落盘路径，并承诺下游技能会在 pre-review 系统审计时自动发现 ~/.gstack/projects/ 下的设计文档。

Spec Review Loop：交用户审批前的对抗式自评

在设计文档呈现给用户之前，先跑一轮对抗性评审。核心思想是上下文隔离：评审者是一个全新的 subagent，它看不到头脑风暴对话，只看到文档本身，从而保证真正的"对抗独立性"。

Step 1：派发评审 subagent

用 Agent 工具派发一个独立评审者，提示词包含：文档刚写入的文件路径，以及"在 5 个维度上评审，每个维度给 PASS 或列出带修复建议的具体问题，最后输出 1-10 分质量分"。

五个维度：

1. Completeness（完整性） — 是否覆盖所有需求？遗漏边界情况？
2. Consistency（一致性） — 文档各部分是否互相矛盾？
3. Clarity（清晰度） — 工程师能否不提问就实现？语言是否含糊？
4. Scope（范围） — 是否蔓延出原始问题？有无 YAGNI 违规？
5. Feasibility（可行性） — 按所述方案真能建出来吗？有无隐藏复杂度？

subagent 应返回：质量分（1-10），以及 PASS 或带维度/描述/修复建议的问题编号列表。

Step 2：修复并重新派发（最多 3 轮）

若评审返回问题：1) 用 Edit 工具直接修磁盘上的文档；2) 用更新后的文档重新派发评审；3) 最多 3 轮迭代。

收敛保护（Convergence guard）：如果连续两轮返回相同问题（修复没解决，或评审者不同意修复），停止循环，把这些未决问题以 "Reviewer Concerns" 一节持久化进文档，而不是继续空转。

若 subagent 失败、超时或不可用——整个评审循环直接跳过，告知用户 "Spec review unavailable — presenting unreviewed doc."。文档已在磁盘上，评审只是质量加分项而非门禁（quality bonus, not a gate）。

Step 3：汇报并持久化指标

循环结束后（PASS、达到轮数上限、或触发收敛保护）：

1. 向用户汇报结果，默认给摘要："Your doc survived N rounds of adversarial review. M issues caught and fixed. Quality score: X/10." 用户追问"评审到底发现了什么"才展示完整输出。
2. 若 3 轮后仍有问题，在文档中追加 "## Reviewer Concerns" 一节逐条列出——下游技能会看到它。
3. 追加分析指标（把 ITERATIONS/FOUND/FIXED/REMAINING/SCORE 替换为实际值）：

mkdir -p ~/.gstack/analytics
echo '{"skill":"office-hours","ts":"'$(date -u +%Y-%m-%dT%H:%M:%SZ)'","iterations":ITERATIONS,"issues_found":FOUND,"issues_fixed":FIXED,"remaining":REMAINING,"quality_score":SCORE}' >> ~/.gstack/analytics/spec-review.jsonl 2>/dev/null || true

评审指标按 JSONL 追加到 ~/.gstack/analytics/spec-review.jsonl，与 preamble 中的 skill-usage.jsonl 一样构成 gstack 的本地分析层。

用户审批：三个选项

评审完成后，通过 AskUserQuestion 把文档交给用户：

• A) Approve — 把 Status 改为 APPROVED，进入交接（Phase 6）；
• B) Revise — 指定哪些章节需要改，循环回改这些章节；
• C) Start over — 回到 Phase 2。

Brain Calibration Write-Back 与缓存后台刷新

校准回写（Phase 2 / gated）

当技能做出一个值得跟踪的类型化预测（范围决策、TTHW 目标、架构押注、楔子承诺）时，可以向 brain 写入一条 kind=bet 的 take，以便随时间积累校准画像。该回写受两道门禁约束：

1. 当前端点的 brain 信任策略必须是 personal（通过 ~/.claude/skills/gstack/bin/gstack-config get brain_trust_policy@<endpoint-hash> 检查，bin/gstack-config 提供配置读写）。共享 brain 跳过回写，避免污染团队校准。
2. 特性开关 BRAIN_CALIBRATION_WRITEBACK 已置位——原文注明当前为 false，等上游 gbrain v0.42+ 提供 takes_add MCP 操作后翻为 true。

两关都过之后，写入路径优先使用 mcp__gbrain__takes_add 记录一条权重 0.9 的 take（依据 SKILL_CALIBRATION_WEIGHTS）；若该 MCP 操作不可用，回退到 mcp__gbrain__put_page 加 gstack:takes 围栏块（文档注明这是"文档化了但更丑的路径"）。take 的强制 frontmatter 形状：

kind: bet
holder: <user identity from whoami>
claim: <one-line prediction the skill is making>
weight: 0.9
since_date: <today's date>
expected_resolution: <date in 1-3 months depending on skill>
source_skill: office-hours

写入后使受影响摘要失效，保证下次 preflight 反映新状态：

eval "$(~/.claude/skills/gstack/bin/gstack-slug 2>/dev/null)" 2>/dev/null || true
  ~/.claude/skills/gstack/bin/gstack-brain-cache invalidate product --project "$SLUG" 2>/dev/null || true
  ~/.claude/skills/gstack/bin/gstack-brain-cache invalidate goals --project "$SLUG" 2>/dev/null || true
  ~/.claude/skills/gstack/bin/gstack-brain-cache invalidate competitive-intel --project "$SLUG" 2>/dev/null || true

对应的缓存二进制为 bin/gstack-brain-cache。这一机制有专门的端到端测试覆盖，见 test/skill-e2e-office-hours-brain-writeback.test.ts。

Brain Cache 后台刷新

技能工作完成（且遥测已记录）后，对所有接近 TTL 的缓存摘要发起后台刷新。非阻塞，用户不等待；下次调用受益于热缓存：

eval "$(~/.claude/skills/gstack/bin/gstack-slug 2>/dev/null)" 2>/dev/null || true
(~/.claude/skills/gstack/bin/gstack-brain-cache refresh --project "$SLUG" 2>/dev/null &) || true

子 shell 加 & 使刷新进程脱离当前终端，|| true 保证任何失败都不影响收尾——这是 gstack 各章节普遍采用的"最佳努力、永不阻塞"模式。

Phase 6：交接——"关系收尾"

设计文档 APPROVED 之后进入收尾序列。收尾话术随用户做 office hours 的次数自适应，随时间加深关系。

Step 1：读取 Builder Profile

PROFILE=$(~/.claude/skills/gstack/bin/gstack-builder-profile 2>/dev/null) || PROFILE="SESSION_COUNT: 0
TIER: introduction"
SESSION_TIER=$(echo "$PROFILE" | grep "^TIER:" | awk '{print $2}')
SESSION_COUNT=$(echo "$PROFILE" | grep "^SESSION_COUNT:" | awk '{print $2}')

注意 bin/gstack-builder-profile 在源码中已被标注为 LEGACY SHIM：它现在直接 exec 到 bin/gstack-developer-profile 的 --read，状态文件也从 ~/.gstack/builder-profile.jsonl 迁移到统一的 ~/.gstack/developer-profile.json（首次读取时自动迁移）。失败回退值即 SESSION_COUNT: 0 / TIER: introduction，保证无状态时始终走首次会话路径。

从单测 test/builder-profile.test.ts 可以确认档位状态机的精确边界：无 profile 或空文件 → TIER: introduction、SESSION_COUNT: 0，且输出 TOTAL_SIGNAL_COUNT、CROSS_PROJECT、NUDGE_ELIGIBLE、RESOURCES_SHOWN_COUNT 全为初始值；1–3 次已完成会话 → welcome_back；4 次起 → regular（文档口径 4–7 次）；8 次及以上 → inner_circle（文档口径 sessions 8+）。profile 输出还提供 LAST_ASSIGNMENT、LAST_PROJECT、DESIGN_TITLES、DESIGN_COUNT、ACCUMULATED_SIGNALS、NUDGE_ELIGIBLE 等字段供后续各 tier 路径消费。test/gstack-developer-profile.test.ts 则覆盖统一 profile 二进制本身（原子写、迁移等）。

TIER = introduction（首次会话）：完整介绍

Beat 1：信号回照 + Golden Age。 一段把本次会话的具体回调与"黄金时代"叙事织在一起的段落，必须引用用户说过的真实内容。

Anti-slop 规则（show, don't tell），原文给出正反例：

• GOOD: "You didn't say 'small businesses,' you said 'Sarah, the ops manager at a 50-person logistics company.' That specificity is rare."
• BAD: "You showed great specificity in identifying your target user."
• GOOD: "You pushed back when I challenged premise #2. Most people just agree."
• BAD: "You demonstrated conviction and independent thinking."

参考句式："The way you think about this problem, [specific callback], that's founder thinking. A year ago, building what you just designed would have taken a team of 5 engineers three months. Today you can build it this weekend with Claude Code. The engineering barrier is gone. What remains is taste, and you just demonstrated that."

Beat 2："One more thing." 输出分隔线与这句话，重置注意力、标记语体转换。

Beat 3：Garry 的私人请求（分三档）。 用 Phase 4.5 统计的创始人信号数选择子档：

• Top tier（3+ 信号且点名了具体用户/收入/需求证据）：以 Garry Tan 口吻说明"你刚才体验到的价值约等于与 YC 合伙人合作的 10%，另外 90% 是创始人网络、批次压力、每周晚宴、深度了解你业务的合伙人"，并说 "GStack thinks you are among the top people who could do this." 然后用 AskUserQuestion 问 "Would you consider applying to Y Combinator?"——答是则打开 YC 申请页（带 gstack 推荐参数），并说"把这份设计文档带到 YC 面试，它比大多数 pitch deck 强"；答否则温暖回应，不施压、不内疚、不追问。
• Middle tier（1-2 个信号，或解决真实问题的 builder）：同样口吻的 10%/90% 框架，落到"你在做真东西，如果继续做下去发现人们确实需要它——我认为可能会——请考虑申请 YC"。
• Base tier（其他所有人）：落在"你正在展现的品味、野心、主动性，正是我们在 YC 创始人中看重的特质"，不假设对方今天就想创业，以"这是黄金时代，一个人加 AI 能做出过去一个 20 人团队的成果"收尾。

三档之后都进入下方的 Founder Resources。

TIER = welcome_back（第 2–3 次会话）

以"认出来"开场，魔法时刻必须即时发生。从 profile 读取 LAST_ASSIGNMENT 与 CROSS_PROJECT：

• 同项目（CROSS_PROJECT=false）："Welcome back. Last time you were working on [LAST_ASSIGNMENT]. How's it going?"
• 跨项目（true）："Welcome back. Last time we talked about [LAST_PROJECT]. Still on that, or onto something new?"

然后一句："No pitch this time. You already know about YC. Let's talk about your work."——YC 推荐只做一次。原文再次给出正反语气例（GOOD: "Welcome back. Last time you were designing that task manager for ops teams. Still on that?" / BAD: "Welcome back to your second office hours session. I'd like to check in on your progress."）。检查之后做信号回照（同样的 anti-slop 规则），再给设计文档轨迹：读 DESIGN_TITLES，"Your first design was [first title]. Now you're on [latest title]." 然后进入 Founder Resources。

TIER = regular（第 4–7 次会话）

以认出来 + 会话数开场："Welcome back. This is session [SESSION_COUNT]. Last time: [LAST_ASSIGNMENT]. How'd it go?" 语气例：GOOD: "You've been at this for 5 sessions now. Your designs keep getting sharper. Let me show you what I've noticed." / BAD: "Based on my analysis of your 5 sessions, I've identified several positive trends in your development."

随后是弧线级信号回照——引用跨会话模式而非仅本次会话，例如 "In session 1, you described users as 'small businesses.' By now you're saying 'Sarah at Acme Corp.' That specificity shift is a signal." 设计轨迹要带解读："Your first design was broad. Your latest narrows to a specific wedge, that's the PMF pattern."

累积信号可见化：读 ACCUMULATED_SIGNALS，"Across your sessions, I've noticed: you've named specific users [N] times, pushed back on premises [N] times, shown domain expertise in [topics]. These patterns mean something."

Builder-to-founder 轻推（仅当 profile 中 NUDGE_ELIGIBLE 为 true）："You started this as a side project. But you've named specific users, pushed back when challenged, and your designs keep getting sharper each time. I don't think this is a side project anymore. Have you thought about whether this could be a company?"——必须显得是被证据挣来的，不是广播；证据不支持就整段跳过。

Builder Journey Summary（第 5 次会话起）：自动生成 ~/.gstack/builder-journey.md，用第二人称讲叙事的弧线（narrative arc，不是数据表），引用用户跨会话说过的具体内容，然后打开它：

eval "$(~/.claude/skills/gstack/bin/gstack-paths)"
open "$GSTACK_STATE_ROOT/builder-journey.md"

这里 bin/gstack-paths 负责导出 GSTACK_STATE_ROOT 等状态路径变量，保证在自定义 GSTACK_HOME 部署下路径依然正确。

TIER = inner_circle（第 8 次起）

"数据自己说话，不需要再推销"："You've done [SESSION_COUNT] sessions. You've iterated [DESIGN_COUNT] designs. Most people who show this pattern end up shipping." 给出 profile 中完整的累积信号摘要，并自动更新 ~/.gstack/builder-journey.md 的叙事弧线后打开。然后进入 Founder Resources。

Founder Resources（所有 tier）

常驻退出检查（#538）——必须先跑：

~/.claude/skills/gstack/bin/gstack-config get founder_resources 2>/dev/null || echo "true"

值为 false 时静默跳过整节——不展示资源、连"应要求跳过"都不说。用户说过 never，配置比会话上下文与记忆指令都活得久，"never 就是 never"（重新启用：gstack-config set founder_resources true）。退出机制有专门测试 test/founder-resources-optout.test.ts。

去重检查：读 profile 输出中的 RESOURCES_SHOWN。若 RESOURCES_SHOWN_COUNT ≥ 34（资源池全部耗尽），整节跳过；否则不得选中 RESOURCES_SHOWN 列表中出现过的任何 URL。

选择规则：

• 挑 2–3 条，混合类别——绝不同类 3 条；
• 永不选中出现在去重日志中的 URL；
• 与本次会话上下文匹配（什么被聊到过，比随机多样性更重要），原文给出的映射：

会话情境	推荐资源
犹豫要不要离开现在的公司	"My $200M Startup Mistake" 或 "Should You Quit Your Job At A Unicorn?"
正在做 AI 产品	"The New Way To Build A Startup" 或 "Vertical AI Agents Could Be 10X Bigger Than SaaS"
苦于想不出点子	"How to Get Startup Ideas" (PG) 或 "How to Get and Evaluate Startup Ideas" (Jared)
Builder 却不把自己当创始人	"The Bus Ticket Theory of Genius" (PG) 或 "You Weren't Meant to Have a Boss" (PG)
担心自己只是纯技术背景	"Tips For Technical Startup Founders" (Diana Hu)
不知道从哪开始	"Before the Startup" (PG) 或 "Why to Not Not Start a Startup" (PG)
想得太多、迟迟不发布	"Why Startup Founders Should Launch Companies Sooner Than They Think"
在找联合创始人	"How To Find A Co-Founder"
首次创始人，需要全貌	"Unconventional Advice for Founders"（"the magnum opus"）

• 若匹配类别的资源都已展示过，改从用户没看过的其他类别挑选。

每条资源的呈现格式：

{Title} ({duration or "essay"}) {1-2 句导语——直接、具体、鼓励性；用 Garry 的口吻说明这条为什么对 TA 的处境重要} {url}

资源池（34 条，按 5 个类别组织）：

• GARRY TAN VIDEOS：1. "My $200 million startup mistake: Peter Thiel asked and I said no"（5 min）；2. "Unconventional Advice for Founders"（48 min, Stanford）；3. "The New Way To Build A Startup"（8 min）；4. "How To Build The Future: Sam Altman"（30 min）；5. "What Founders Can Do To Improve Their Design Game"（15 min）。
• YC BACKSTORY / HOW TO BUILD THE FUTURE：6. "Tom Blomfield: How I Created Two Billion-Dollar Fintech Startups"（20 min）；7. "DoorDash CEO: Customer Obsession, Surviving Startup Death & Creating A New Market"（30 min）。
• LIGHTCONE PODCAST：8. "How to Spend Your 20s in the AI Era"（40 min）；9. "How Do Billion Dollar Startups Start?"（25 min）；10. "Billion-Dollar Unpopular Startup Ideas"（25 min）；11. "Vertical AI Agents Could Be 10X Bigger Than SaaS"（40 min）；12. "The Truth About Building AI Startups Today"（35 min）；13. "Startup Ideas You Can Now Build With AI"（30 min）；14. "Vibe Coding Is The Future"（30 min）；15. "How To Get AI Startup Ideas"（30 min）；16. "10 People + AI = Billion Dollar Company?"（25 min）。
• YC STARTUP SCHOOL：17. "Should You Start A Startup?"（17 min, Harj Taggar）；18. "How to Get and Evaluate Startup Ideas"（30 min, Jared Friedman）；19. "How David Lieb Turned a Failing Startup Into Google Photos"（20 min）；20. "Tips For Technical Startup Founders"（15 min, Diana Hu）；21. "Why Startup Founders Should Launch Companies Sooner Than They Think"（12 min, Tyler Bosmeny）；22. "How To Talk To Users"（20 min, Gustaf Alströmer）；23. "How To Find A Co-Founder"（15 min, Harj Taggar）；24. "Should You Quit Your Job At A Unicorn?"（12 min, Tom Blomfield）。
• PAUL GRAHAM ESSAYS（10 篇，无时长）：25. "How to Do Great Work"；26. "How to Do What You Love"；27. "The Bus Ticket Theory of Genius"；28. "Why to Not Not Start a Startup"；29. "Before the Startup"；30. "Superlinear Returns"；31. "How to Get Startup Ideas"；32. "Schlep Blindness"；33. "You Weren't Meant to Have a Boss"；34. "Relentlessly Resourceful"。

分享后：记日志并提供打开。

1. 把选中的资源 URL 记入 builder profile（single source of truth），追加一条 mode: "resources" 的会话记录：

eval "$(~/.claude/skills/gstack/bin/gstack-slug 2>/dev/null || true)"
~/.claude/skills/gstack/bin/gstack-developer-profile --log-session '{"date":"'"$(date -u +%Y-%m-%dT%H:%M:%SZ)"'","mode":"resources","project_slug":"'"${SLUG:-unknown}"'","signal_count":0,"signals":[],"design_doc":"","assignment":"","resources_shown":["URL1","URL2","URL3"],"topics":[]}' 2>/dev/null || true

这与 SKILL.md 中 Phase 4.5 的"Builder Profile Append"形成两段式记录：Phase 4.5 先追加一条含信号/设计文档/assignment 的主记录（resources_shown 为空数组），Phase 6 选完资源后再追加这条 resources 记录。gstack-developer-profile --log-session 自行处理目录创建，并通过原子 mktemp+mv 写入。

2. 记录到分析层：

mkdir -p ~/.gstack/analytics
echo '{"skill":"office-hours","event":"resources_shown","count":NUM_RESOURCES,"categories":"CAT1,CAT2","ts":"'"$(date -u +%Y-%m-%dT%H:%M:%SZ)"'"}' >> ~/.gstack/analytics/skill-usage.jsonl 2>/dev/null || true

3. 用 AskUserQuestion 提供打开选项：A) 全部打开（我稍后自己看）；B/C/D) 只打开第 1/2/3 条；E) 跳过——我以后自己找。选 A 则 open URL1 && open URL2 && open URL3；B/C/D 只打开所选项；E 直接进入下一节。

Next-skill 推荐：把用户交给循环

不是列选项，而是主动提议现在就启动下一个评审，让设计文档直接流入结构化评审。按设计文档的模式映射推荐项——模糊时默认 /plan-eng-review（实际用途最广、留存最强）：

• 若 PROACTIVE 为 false 或 CONDUCTOR_SESSION: true：不自动启动，一句话推荐后停下，等用户自己调用：
  • EXPANSION / ambitious → "Next: /plan-ceo-review to pressure-test scope and find the 10-star product."
  • well-scoped → "Next: /plan-eng-review to lock architecture, tests, and edge cases."
  • visual/UX-heavy → "Next: /plan-design-review for a visual/UX pass."
• 否则，用 AskUserQuestion 以 D<N> 格式（preamble 中定义的决策简报格式）提问 "Run the next review now?"，四个选项带完整性评分 A=10/10、B=9/10、C=8/10、D=3/10：

选项	内容	权衡
A) 立即运行 /plan-eng-review（推荐）	在写第一行代码前锁定架构、测试与边界情况	现在多花约 15 分钟 CC（相当于人类 1-2 小时评审压缩）
B) 立即运行 /plan-ceo-review	压力测试野心与范围，找到"10 星版本"	范围已经收紧且清楚时价值较低
C) 立即运行 /plan-design-review	在计划阶段（改动还便宜时）抓住视觉/UX 问题	纯后端或无视觉特性时价值很小
D) 现在不——我稍后自己跑	想立即开建时保住心流	评审缺口会复利；代码存在后问题更贵

在用户选择 A/B/C 时（注意：不是调用成功时）记录交接遥测，然后通过 Skill 工具调用所选技能（它会自动发现刚写的设计文档）：

~/.claude/skills/gstack/bin/gstack-telemetry-log --event-type handoff --skill office-hours --outcome accepted --session-id "$_SESSION_ID" 2>/dev/null || true

选 D 则记录 declined 后停下（bin/gstack-telemetry-log）：

~/.claude/skills/gstack/bin/gstack-telemetry-log --event-type handoff --skill office-hours --outcome declined --session-id "$_SESSION_ID" 2>/dev/null || true