autocrew 0.1.0

package/skills/teardown/SKILL.md

@@ -0,0 +1,144 @@
+---
+name: teardown
+description: |
+  拆解对标内容 — 用户粘贴一段文案，输出结构化拆解报告。分析钩子、HKRR、时钟节奏、评论触发、微操技巧。结果存入 intel 系统供创作参考。
+trigger: 当用户说"拆解"、"teardown"、"分析这条"、"对标分析"时触发
+---
+
+# Competitive Teardown（对标拆解）
+
+> 免费版：分析用户粘贴的文案。Pro 版（待实现）：通过 Chrome Relay 抓取视频/账号数据。
+
+## 触发方式
+
+用户说 "拆解这条内容" / "teardown" / "帮我分析对标" 并粘贴一段文案。
+
+## 前置加载
+
+- `HAMLETDEER.md` — HKRR 框架、Clock Theory、Micro-Retention Techniques、Bang Moment Types
+- `~/.autocrew/creator-profile.json` — 用于对比自己的定位
+
+## 拆解流程
+
+### Step 1: 基本信息提取
+
+- 预估平台（根据文案风格/emoji/hashtag 特征）
+- 预估内容类型（图文 / 视频脚本 / 长文）
+- 字数统计
+
+### Step 2: 钩子分析
+
+- 开头类型：Pain point / Suspense / Ideal state / Emotional resonance / Contrast / 其他
+- 标题公式匹配：Number+Result / Contrarian / Identity+Pain / Curiosity Gap / Contrast / Resonance Question / 无匹配
+- 钩子强度评分（0-10）
+
+### Step 3: HKRR 评分
+
+对四个维度分别打分（0-10）：
+
+| 维度 | 评分 | 说明 |
+|------|------|------|
+| Happiness | X/10 | 是否有趣/有笑点？ |
+| Knowledge | X/10 | 是否有增量信息/方法论？ |
+| Resonance | X/10 | 是否触达情绪/身份认同？ |
+| Rhythm | X/10 | 节奏是否有变化？能否持续吸引？ |
+
+标注最强维度。
+
+### Step 4: Clock 映射
+
+将内容按比例映射到时钟位，分析每个位置：
+
+| Clock | 内容段落 | Bang moment 类型 | 效果评估 |
+|-------|---------|-----------------|---------|
+| 12:00 | ... | ... | 强/弱/缺失 |
+| 3:00 | ... | ... | 强/弱/缺失 |
+| 6:00 | ... | ... | 强/弱/缺失 |
+| 9:00 | ... | ... | 强/弱/缺失 |
+
+### Step 5: 评论触发点识别
+
+- 争议埋点：在哪里？什么话题？
+- 未答问题：有没有故意留白？
+- 金句钩子：哪句最值得截图？
+
+### Step 6: 微操技巧识别
+
+- 开放循环：有没有？在哪里？
+- 信息缺口：段落结尾有没有前向动力？
+- 视觉锚点：有没有高密度独立金句？
+- 断裂感：有没有突然的超短句？
+
+### Step 7: 一句话总结
+
+"这条内容有效/无效，核心原因是 ___"
+
+### Step 8: 可借鉴点
+
+列出 2-3 个可以应用到自己内容的具体技巧。
+
+## 保存拆解结果
+
+将拆解报告保存到 intel 系统：
+
+```
+~/.autocrew/data/pipeline/intel/_teardowns/{slug}-{date}.md
+```
+
+YAML frontmatter:
+```yaml
+---
+title: "拆解: {原内容标题或前20字}"
+type: teardown
+platform: "{预估平台}"
+hookType: "{钩子类型}"
+dominantHKRR: "{最强HKRR维度}"
+hookScore: 0
+overallScore: 0
+createdAt: "{ISO timestamp}"
+tags: [teardown]
+---
+```
+
+## 输出格式
+
+```
+## 🔍 对标拆解报告
+
+**内容摘要：** {前50字}...
+**预估平台：** {platform}
+**字数：** {count}
+
+### 钩子分析
+- 类型：{hook type}
+- 公式：{title formula or 无匹配}
+- 强度：{score}/10
+- 评价：{一句话评价}
+
+### HKRR 评分
+| H | K | R | R |
+|---|---|---|---|
+| {n}/10 | {n}/10 | {n}/10 | {n}/10 |
+**最强维度：** {element}
+
+### Clock 映射
+{table}
+
+### 评论触发点
+{list}
+
+### 微操技巧
+{list}
+
+### 💡 总结
+{一句话总结}
+
+### 🎯 可借鉴
+{2-3 bullet points}
+```
+
+向用户确认后保存。
+
+## 与 write-script 联动
+
+write-script 在创作时会自动搜索 `pipeline/intel/_teardowns/` 目录，查找与当前主题相关的拆解报告作为参考。无需手动操作。

package/skills/title-craft/SKILL.md

@@ -0,0 +1,234 @@
+---
+name: title-craft
+description: |
+  Domain-agnostic viral title methodology. Reference skill — not invoked standalone.
+  Other skills (topic-ideas, spawn-writer, platform-rewrite) reference this for title creation guidance.
+invokable: false
+---
+
+# Viral Title Methodology
+
+> Reference skill. Provides domain-agnostic frameworks for crafting high-performing titles across any niche. Consult this before generating or adapting titles.
+
+## Core Principle
+
+A great title speaks to ONE specific person, not a crowd. Before writing any title, answer three questions:
+
+1. **Who is reading?** — What specific person will see this in their feed?
+2. **What are they thinking right now?** — What worry, desire, or curiosity do they have at this moment?
+3. **What do they want?** — What outcome would make them stop scrolling?
+
+Build a hyper-specific reader portrait: not "people interested in AI" but "a 35-year-old traditional business owner who heard AI can cut costs, spent money on it, got burned, and now wants to know if it's all hype."
+
+---
+
+## The 8 Title Types
+
+These are psychological mechanisms, not rigid templates. Combine 2 types for differentiation. Each type has classic sub-methods listed as examples — develop new variants based on your domain and audience.
+
+### 1. Emotional Resonance — "They understand me"
+
+Make the reader feel seen and understood.
+
+**Sub-methods:**
+
+- **Honest Conversation** — Write like you're talking to a close friend. Raw, real, unfiltered.
+  - Formula: `[Frank admission] + [Real emotional expression]`
+  - Signals: first-person voice, emoji for emotion, colloquial language
+  - Core test: Does it sound like a real person talking, not a brand?
+
+- **Precision Lens** — Zoom into a specific person and situation so the target reader thinks "this is literally about me."
+  - Formula: `[Specific persona/situation] + [Concrete solution]`
+  - Signals: mention specific demographics, budgets, scenarios, life stages
+  - Core test: Could you name the exact person this is for?
+
+Develop new variants based on your domain. The underlying mechanism: reader recognition — "this is about ME."
+
+### 2. Transformation Contrast — "Change is possible"
+
+Show the gap between before and after. Spark desire for change.
+
+**Sub-methods:**
+
+- **Before/After Gap** — Highlight dramatic transformation with minimal effort implied.
+  - Formula: `[Problem state] + [Surprising result]`
+  - Signals: numbers that quantify change, words like "from X to Y"
+
+- **Small Input, Big Output** — Emphasize low cost or effort with disproportionate results.
+  - Formula: `[Minimal investment] + [Outsized outcome]`
+  - Signals: specific dollar amounts, time frames, concrete metrics
+
+Core test: Does the reader see a version of their situation that got dramatically better?
+
+### 3. Emotion Trigger — "I need to act on this feeling"
+
+Activate strong emotional response — either pain avoidance or pleasure seeking.
+
+**Sub-methods:**
+
+- **Pain Point Strike** — Hit anxiety, fear of loss, or regret directly.
+  - Formula: `[Warning word] + [Common mistake] + [Emotional punctuation]`
+  - Signals: words like "never," "avoid," "mistake," "regret"
+  - Use restraint: professional warning, not fear-mongering
+
+- **Aspiration Fulfillment** — Present the ideal state the reader craves.
+  - Formula: `[Premium/aspirational word] + [Ideal outcome] + [Simple method hint]`
+  - Signals: words suggesting quality, ease, social proof
+
+Core test: Does the reader feel an emotional pull — either "I must avoid this" or "I want that"?
+
+### 4. Perspective Elevation — "This is bigger than I thought"
+
+Connect a specific detail to a larger life improvement or insight.
+
+**Sub-methods:**
+
+- **Detail-to-Philosophy** — Elevate a tactical tip into a life principle.
+  - Formula: `[Specific detail] + [Quality of life upgrade]`
+  - Signals: connects craft to lifestyle, implies deeper wisdom
+
+- **Risk Warning** — Reveal hidden dangers the reader hasn't considered.
+  - Formula: `[Warning word] + [Hidden risk] + [Consequence hint]`
+  - Signals: protective tone, "I wish someone had told me"
+
+Core test: Does the reader walk away thinking about something bigger than the surface topic?
+
+### 5. Urgency & Scarcity — "I might miss this"
+
+Create a sense that delay equals loss.
+
+**Sub-methods:**
+
+- **Time Pressure** — Imply the content or opportunity is fleeting.
+  - Formula: `[Time-pressure word] + [Scarce resource] + [Action word]`
+  - Signals: "save now," "before it's gone," "limited"
+
+- **Insider Access** — Frame the content as exclusive or soon-to-disappear.
+  - Formula: `[Exclusivity hint] + [Valuable content] + [Urgency]`
+
+Core test: Does the reader feel they'll lose something by not engaging now? (Without being manipulative.)
+
+### 6. Pattern Breaker — "Wait, what?"
+
+Defy expectations. Create cognitive dissonance that demands resolution.
+
+**Sub-methods:**
+
+- **Surprising Reversal** — State something that contradicts common belief.
+  - Formula: `[Unexpected claim] + [Curiosity hook]`
+  - Signals: "actually," "turns out," contradiction of conventional wisdom
+
+- **Early Access** — Make the reader feel ahead of the curve.
+  - Formula: `[Ahead-of-time word] + [Future trend] + [Action word]`
+  - Signals: year references, "before everyone else," "not yet mainstream"
+
+- **Secret Sharing** — Create an insider circle feeling.
+  - Formula: `[Secret hint] + [Expert knowledge] + [Small-circle feeling]`
+  - Signals: "insiders know," "not publicly shared," "industry secret"
+
+Core test: Does the reader think "that can't be right... I need to check"?
+
+### 7. Authority & Proof — "I can trust this"
+
+Establish credibility through expertise, data, or social proof.
+
+**Sub-methods:**
+
+- **Authority Endorsement** — Borrow credibility from recognized experts or institutions.
+  - Formula: `[Authority figure/brand] + [Professional content] + [Value hint]`
+
+- **Result Promise** — Commit to specific, measurable outcomes.
+  - Formula: `[Number + method] + [Professional term] + [Clear effect]`
+  - Signals: specific numbers, actionable steps, measurable results
+
+- **Comprehensive Guide** — Promise a complete, one-stop solution.
+  - Formula: `[Completeness word] + [Core problem] + [Ultimate solution]`
+  - Signals: "complete guide," "everything you need," specific count of points
+
+Core test: Would a skeptical reader think "this person actually knows what they're talking about"?
+
+### 8. Problem-Solution — "I've been struggling with exactly this"
+
+Name the reader's problem so precisely they feel you've read their mind.
+
+**Sub-methods:**
+
+- **Deep Question** — Ask the question the reader is already asking themselves.
+  - Formula: `[Thought-provoking question] + [Value point]`
+
+- **Comparison Choice** — Frame a common decision as a clear choice.
+  - Formula: `[Option A] vs [Option B] + [Decision guidance]`
+
+- **Need Revelation** — Help the reader realize they have a problem they haven't articulated.
+  - Formula: `[Problem prompt] + [Resonance point] + [Solution direction]`
+  - Signals: "Do you ever feel...," "Why does X always...," "You probably didn't realize..."
+
+Core test: Does the reader think "YES, that's exactly my problem"?
+
+---
+
+## Combining Types
+
+Single-type titles work. Two-type combinations are stronger. Examples:
+
+- **Precision Lens + Transformation**: "Solo developer making $3K/month — how I 5x'd revenue with one change"
+- **Pattern Breaker + Authority**: "The advice every expert gives about X is wrong — here's what the data shows"
+- **Emotion Trigger + Problem-Solution**: "Stop wasting money on Y — this free alternative does it better"
+
+Pick the primary type based on the content's strongest angle. Add a secondary type if it naturally fits. Never force a combination.
+
+---
+
+## Quality Checklist (5 Questions)
+
+Every title must pass all five. If it fails any, rewrite.
+
+1. **Targeting** — Does it speak to a specific person or scenario? (Not "everyone")
+2. **Emotion** — Does it trigger concern, anticipation, curiosity, or desire?
+3. **Value** — Does it promise to solve a real problem or deliver a real upgrade?
+4. **Curiosity** — Does it create an urge to click/tap? ("I need to know more")
+5. **Credibility** — Does it feel authentic and trustworthy? (Not clickbait)
+
+Bonus check: Write 3-5 variants. The best title is rarely the first one.
+
+---
+
+## Practical Techniques
+
+- **Numbers beat vague quantifiers.** "3 methods" > "several methods." "Save $500" > "save money."
+- **First person > third person.** "I tried X" > "Someone tried X."
+- **Emoji adds visual impact** — but platform-dependent. Xiaohongshu loves it. WeChat MP less so.
+- **Front-load the hook.** On platforms that truncate (Douyin, Instagram), the first visible characters must carry the entire weight.
+- **Match platform tone.** Xiaohongshu is conversational. WeChat MP is authoritative. Bilibili is youth-friendly. Adapt.
+
+---
+
+## Platform Title Constraints
+
+Respect hard limits. Writing a beautiful 40-character title for Xiaohongshu is useless — it won't fit.
+
+| Platform | Hard Limit | Recommended | Visible Before Fold | Tone |
+|----------|-----------|-------------|--------------------:|------|
+| Xiaohongshu | 20 chars | 10-18 | 20 (no fold) | Conversational, emoji-friendly, first-person |
+| Douyin | 300 chars | 15-55 | ~55 | Hook-first, reversal words, suspense |
+| Bilibili | 80 chars | 12-24 | 20-24 | Youth-friendly, brackets for labels, detailed OK |
+| WeChat MP | 64 chars | 20-30 | 64 (no fold) | Info-dense, authoritative, colon-separated |
+| Toutiao | 30 chars (min 5) | 15-30 | 30 (no fold) | Direct, two-segment, value-forward |
+| YouTube | 100 chars | 30-70 | 60-70 | Keyword-rich, bilingual OK |
+| Twitter/X | 280 chars (whole post) | 60-140 | 280 | First line = title. Punchy. |
+| Instagram | 2200 chars (caption) | 60-125 | 125 | First line = title. Emoji-heavy. |
+
+**When adapting across platforms: re-craft, don't truncate.** A Xiaohongshu title and a Bilibili title for the same content should feel like they were written natively for each platform.
+
+---
+
+## Anti-Patterns
+
+Reject titles that fall into these traps:
+
+- **Jargon titles** — Using industry terminology the audience doesn't understand
+- **Tool review format** — "{Tool Name} Review" is boring and generic
+- **Academic tone** — Reads like a paper abstract, not a social media post
+- **Condescending tone** — Talking down to the reader, lecturing
+- **Generic superlatives** — "The Best X" / "Amazing Y" with no specificity
+- **Pure clickbait** — Promise with no substance. Destroys trust long-term.

package/skills/topic-ideas/SKILL.md

@@ -0,0 +1,105 @@
+---
+name: topic-ideas
+description: |
+  互动式选题头脑风暴。用户给出种子想法，生成5个选题方向。
+  Trigger: 帮我想选题, 想选题, 灵感, 这个方向怎么样
+---
+
+# 选题头脑风暴
+
+互动式选题头脑风暴。用户给出一个种子想法，输出 5 个经过质量门检验的选题方向。
+
+## 输入
+
+| 参数 | 来源 | 必填 | 说明 |
+|------|------|------|------|
+| seed_idea | 用户消息 | 否 | 粗略的想法、观察或方向 |
+| count | 用户消息 | 否 | 选题数量（默认 5） |
+
+## 流程
+
+### 1. 加载上下文
+
+读取 `~/.autocrew/MEMORY.md` 获取品牌背景、目标受众、人设。
+读取 `~/.autocrew/STYLE.md` 获取平台和语气偏好。
+读取 `creator-profile.json` 获取创作者画像（如有）。
+
+如果没有受众人设，先问用户：
+> 我需要先了解你的内容是给谁看的。描述一个你最想影响的人 — 他是干什么的、多大年纪、为什么会关注你？
+
+生成人设，确认后继续。
+
+### 2. 张力拆解（模糊请求时）
+
+如果用户只给了模糊请求（如 "帮我想选题"），先拆解为 3-4 个**受众侧张力**。
+
+张力 = 受众**相信的** vs **实际上是真的**，用受众的语言表述。
+
+- 差（创作者视角）："AI执行力强但判断力弱"
+- 好（受众视角）："觉得买了AI工具就能省人力，实际上要花更多时间想清楚让AI干嘛"
+
+**翻译规则**：创作者的知识必须翻译成受众听得懂的话。行业术语 → 日常语言。
+
+展示张力，问用户哪个最戳，然后从那个张力出发做头脑风暴。
+
+如果用户给了具体种子（一个故事、一个观察、一个吐槽），跳到第 3 步。
+
+### 3. 生成选题
+
+**Before generating, review the title methodology in `skills/title-craft/SKILL.md`.** Select 1-2 title types (from the 8 types) that best match the topic angle and target audience. Use the quality checklist to validate each title.
+
+生成 5 个选题。每个选题写之前，模拟人设：
+
+> [人设名], [年龄], [职业]。他刷到这条，会停下来吗？他能在3秒内理解标题在说什么吗？
+
+如果答案是否 → 重写。**绝不使用人设听不懂的术语**。
+
+每个选题必须包含：
+
+- **标题**（≤20 字）：具体、能让人停下来刷。
+- **Hook 方向**：前 3 秒怎么抓住人。
+- **角度**（内部用）：非显而易见的洞察或转折。一句话。
+- **证据**（受众可读）：支撑这个角度的具体事实/数据/案例。
+- **人设心理**：这个选题击中了受众的什么心理？解决了什么张力？
+
+### 4. 四道质量门
+
+每个选题必须全部通过：
+
+- [ ] **人设测试**：人设能在 3 秒内理解标题吗？
+- [ ] **停刷测试**：在 100 条内容中，他会为这条停下来吗？
+- [ ] **So-what 测试**：标题是否制造了悬念？读完标题会想"然后呢"？
+- [ ] **冒牌货测试**：一个没做过这件事的人也能写这个选题吗？如果能 → 不够独特，重来。
+
+没通过的选题淘汰，补新的。
+
+### 5. 反模式清单
+
+以下选题类型直接淘汰：
+
+- **术语标题**：用行业黑话当标题（受众看不懂）
+- **工具测评格式**："{工具名}测评" — 太无聊
+- **论文式标题**：读起来像学术摘要
+- **说教语气**：居高临下、教育口吻
+
+### 6. 展示与确认
+
+展示选题给用户。问哪些有感觉。
+
+### 7. 保存
+
+用户确认的选题，通过 `autocrew_topic` 工具保存：
+
+```json
+{
+  "action": "create",
+  "title": "...",
+  "description": "角度 + hook方向 + 为什么有效 + 证据",
+  "tags": [...],
+  "source": "brainstorm"
+}
+```
+
+## 翻译规则
+
+所有选题标题和描述必须用受众的语言。如果受众说中文，就写中文。不要夹杂英文术语，除非受众日常确实在用。

package/skills/video-timeline/SKILL.md

@@ -0,0 +1,117 @@
+# 视频时间轴生成
+
+> Trigger: "生成视频" / "视频素材" / "时间轴" / "video timeline" / "做视频" / "视频制作"
+
+## 概述
+
+将已完成的文案自动生成视频时间轴，包括AI配音段落规划、B-roll素材标注、知识卡片配置。
+
+## 前置条件
+
+- 内容状态为 `approved` 或 `draft_ready`
+- 文案（draft.md）已完成
+
+## 流程
+
+### 第一步：确认内容
+
+使用 `autocrew_content` 工具的 `get` 动作读取内容，确认文案已就绪。
+
+### 第二步：选择预设和比例
+
+询问用户：
+
+**预设风格：**
+- `knowledge-explainer`：知识讲解（卡片为主 + B-roll 过渡）
+- `tutorial`：教程类（步骤卡片 + 屏幕录制占位）
+
+**画面比例：**
+- `9:16` 竖屏（抖音/快手/Reels）
+- `16:9` 横屏（B站/YouTube）
+- `3:4` 小红书
+- `1:1` 正方形（朋友圈）
+- `4:3` 视频号
+
+### 第三步：AI 自动标记
+
+分析文案内容，在合适位置插入 `[card:...]` 和 `[broll:...]` 标记。
+
+**knowledge-explainer 规则：**
+- 约 60% 画面用知识卡片，40% 用 B-roll
+- 提到对比、列举、要点时用 card
+- 切换话题或需要过渡时用 broll
+- B-roll prompt 要具体（画面内容+风格+氛围）
+- card 数据从文案中提取，不编造
+
+**tutorial 规则：**
+- 每个步骤用带编号的 card 展示
+- 步骤之间用 broll 做过渡
+- 软件操作标注 `[broll:屏幕录制占位 — 操作描述]`
+
+**可用卡片模板：**
+- `comparison-table`：对比表。attrs: `title="标题" rows="名称:优点:缺点,名称:优点:缺点"`
+- `key-points`：要点列表。attrs: `items="要点1,要点2,要点3"`
+- `flow-chart`：流程图。attrs: `steps="步骤1,步骤2,步骤3"`
+- `data-chart`：数据图。attrs: `title="标题" items="标签:数值,标签:数值"`
+
+**标记语法示例：**
+```
+今天我们聊聊三个效率工具
+[card:comparison-table title="三款工具对比" rows="Notion:全能:学习曲线,Obsidian:本地:无协作"]
+
+第一个是 Notion，很多人用它来管理生活
+[broll:Notion 应用界面操作画面，暗色主题，流畅页面切换]
+
+它最强的地方在于三点
+[card:key-points items="数据库驱动,模板生态,多端同步"]
+```
+
+**跨段B-roll：** 一个 B-roll 跨越多段旁白时，用 `span=N`：
+```
+第一段话
+第二段话
+[broll:城市航拍夜景 span=2]
+```
+
+### 第四步：用户确认标记
+
+将带标记的文案展示给用户。用户可以：
+- 修改标记位置
+- 更换卡片模板
+- 调整 B-roll prompt
+- 添加或删除标记
+
+### 第五步：生成时间轴
+
+用户确认后，使用 `autocrew_timeline` 工具的 `generate` 动作：
+
+```
+action: generate
+content_id: {内容ID}
+preset: {用户选择的预设}
+aspect_ratio: {用户选择的比例}
+```
+
+将带标记文案保存为 draft.md，然后生成 timeline.json。
+
+### 第六步：引导后续操作
+
+告知用户时间轴已生成，后续可以：
+
+1. **Web UI 素材面板** — 在浏览器中打开 `/contents/{id}/assets` 查看和调整素材
+2. **自动生成素材** — 安装 `@autocrew/studio` 后，运行 `autocrew render` 自动生成 TTS 配音、B-roll 素材、知识卡片截图
+3. **导出剪映** — 运行 `autocrew render --jianying` 导出为剪映项目文件
+
+## 用户修改素材时
+
+在 Web UI 素材面板中，用户可以对每个素材：
+- 🔄 重新生成：重新调用 AI 生成
+- 📝 编辑：修改 prompt 或卡片内容
+- 📤 上传替换：用本地文件替换
+
+所有素材确认后（状态为 confirmed），即可触发最终合成。
+
+## 工具依赖
+
+- `autocrew_content` — 读取文案
+- `autocrew_timeline` — 生成和管理时间轴