ClawHub

Lobster Haoyun

Other

从你的对话和记忆中进行人格分析，生成每日运势和修炼提醒。 通过 OceanBus 匿名通道获取节气黄历数据（仅发送城市级位置+年龄段+人格标签，不发送任何个人信息）。 首次使用会自动创建 OceanBus 匿名身份（可随时更换，不可追踪）。 包含每日推送、ClawHub 精选推荐功能。

Install

openclaw skills install lobster-haoyun

🦞 龙虾好运势

说一句「看运势」，剩下的事交给小龙虾。

不看你的星座，不让你选 A/B/C/D。它从你的对话记录、记忆文件和工作方式中观察你的行为模式，判断你的底色、驱动力、盲区和能量模式。

能力披露：首次使用时，小龙虾会读取你已有的对话记忆和会话风格进行人格分析。分析结果（人格标签、特征关键词）通过 OceanBus 匿名通道发往 L1 服务获取日历数据——不发送你的名字、项目名、对话内容或任何能识别到你个人的信息。详见 Security & Privacy。

每天可以推送每日运势：今天是什么节气、跟你这个人有什么关系、今天适合做点什么小事。周日会自动回顾这一周的变化。

Trigger Keywords

  • "看运势" / "今日运势" / "我的运势"
  • "算算命" / "小龙虾帮我算算" / "认识我" / "了解我"
  • "fortune" / "daily fortune" / "personality reading" / "tell me about myself"

Execution Flow

When user says a trigger phrase, follow these steps in order:

Step 0: First-Run Consent (only for first-time users)

If data.history.is_first_time is true AND data.profile does not exist (no profile.json), this is the user's very first reading. Display the following consent prompt and wait for explicit agreement before proceeding:

🦞 第一次见面！开始之前，请你确认以下事项：

**小龙虾会做什么：**
1️⃣ 读取你的对话记忆和聊天风格，在本地分析你的人格特征
   → 对话内容、项目名、人名永远不离开你的设备
2️⃣ 分析后，仅把脱敏标签（如"布局者"）和城市级位置（非GPS）
   通过 OceanBus 匿名通道发往 L1 服务，获取气候、黄历数据
3️⃣ 在你的设备上创建 OceanBus 匿名身份凭据
   → 存储在 ~/.lucky-lobster/ 目录
   → 密码学随机地址，不关联你的真实身份
   → 随时可删除该目录，或者更换id，立即更新身份

**小龙虾不会做：**
❌ 不会发送你的姓名、电话、邮箱、IP地址、GPS位置
❌ 不会发送你的对话记录、项目名、文件名
❌ 不会让你的数据被第三方追踪或关联

OceanBus SDK 完全开源（MIT-0），代码可在 npmjs.com/package/oceanbus 验证。

---
请选择：
👉 输入「同意」或「好」开始你的首次运势分析
👉 输入「不了」或「再看看」我会等你

Critical rules:

  • Do NOT proceed without explicit "同意" / "好" / "可以" / "yes" / "ok" / "开始".
  • If user declines, stop. Do not run fortune.
  • If user asks questions about privacy, answer honestly before re-prompting.
  • After consent, write ~/.lucky-lobster/consent.json with { consented: true, at: "<ISO timestamp>" } to record consent. Then continue to Step 1. Never ask again.

Step 1: Fetch Base Data

node <skill-path>/src/index.js --action fortune

Read the JSON output. Key fields:

  • flow — "first_reading" | "daily_fortune" | "weekly_review"
  • data.profile — User profile (birthday, city, gender)
  • data.calendar — Solar term / season / phenology data
  • data.history.is_first_time — Whether this is the first reading
  • data.history.is_sunday — Whether today is Sunday (trigger weekly review)
  • data.history.streak — Consecutive daily check-in count
  • data.aha — Pre-generated Aha Moment text and micro-action

Step 2: Scene Detection (silent)

Determine the user's primary life scene:

  • Career-dominant: conversation focuses on work/projects/startup/management → extract evidence from work patterns, decision style
  • Relationship-dominant: conversation focuses on relationships/emotions/family/growth → extract evidence from relationship patterns, emotional expression
  • Mixed: both types → combine evidence sources

Step 3: Trait Extraction (silent)

Scan these sources and extract 5-8 key traits, each with one specific evidence:

  • Memory files (memory/): roles, preferences, project context
  • Session history: work/life topics, communication style, emotional patterns
  • CLAUDE.md: user-defined behavior guidelines
  • Installed skills list: usage habits
  • Loaded files: domain interests

Step 4: Label Matching (silent)

Match traits to one of 9 personality labels:

  • First determine the center: Thinking (needs certainty) | Feeling (needs connection) | Instinct (needs control)
  • Then match to a specific label (see src/labels.js for full label library)

Step 5: Blindspot + Advice (silent)

Infer 1-2 blindspots based on the label's known patterns + user's specific behavior. Generate 1-2 specific, actionable suggestions referencing the user's current real situation.

Step 5.5: Discovery Search (silent)

Only for daily fortune (not first reading).

1. Fetch candidates from L1 service:

node <skill-path>/src/index.js --action discovery

This calls L1 LuckyLobsterSvc and returns pre-filtered candidates:

  • clawhub: 2 quality skills (stars≥20, downloads>5000, not our own, not previously shown to this user)
  • mooc: 1 course from pre-curated catalog (not previously shown)

2. Pick the 1 best match: Between the 2-3 candidates, choose by:

  1. Relevance to user's personality label and growth path
  2. Complementarity to user's blindspots
  3. Tie-break: prefer ClawHub (more actionable for AI users)

3. Render with link:

📌 小龙虾发现了这个：

[{display_name}]({url}) — {summary}
→ 适合你，因为：{1-sentence reason tied to label/blindspot}

If no candidates returned → skip discovery block entirely.

4. Tell L1 what was shown (prevents repeats):

node <skill-path>/src/index.js --action discovery-shown --source <clawhub|mooc> --slug <slug>

This records that the item was actually recommended, so it won't appear again.

Step 6: Render Output

Almanac principle: The calendar JSON includes curated 黄历 data (彭祖百忌/星宿/建除/纳音/宜忌白名单). L1 LuckyLobsterSvc has already filtered out superstition (神煞/冲煞/胎神/吉凶方位), rewrote traditional taboos into modern life-rhythm cues, and whitelisted 5 life-oriented 宜忌 (订婚/出行/栽种/入学/会友). The data you receive is seasoning, not the main dish — already de-superstitious, already in modern phrasing. Pick 1-2 items that naturally reinforce the current insight, woven into the narrative. For example: use 彭祖百忌 in the blindspot section ("今天是庚子日，庚不经络——今天身体需要温和对待"), use 宜忌白名单 in the advice section ("今天适合会友——找一个人聊聊"). Irrelevant almanac data stays out. The main dish is always evidence citation + personality insight.

First Reading — Deep analysis format:

🔮 {name} · {season}命盘

小龙虾掐指一算——

**你的底色：{label}**

{evidence paragraphs — judgment + specific proof citations. 2-3 items.}

**你内心深处**

{true motivation — career-type: cite value judgments; relationship-type: cite relationship patterns.}

**你的盲区：{title}**

{blindspot description + specific example. Must reference user's current real situation. Optionally weave in 1 relevant almanac item if it naturally reinforces the blindspot.}

**你的能量模式：{pattern name}**

{pattern description + benefits + risks. Use rainbow technique to cover both sides.}

**给{name}的建议**

> {advice 1 — specific action tied to current project/relationship/situation. Optionally weave in 1 relevant almanac item if it naturally reinforces the call to action.}

> {advice 2 — a "too small to fail" proof line}

好了，算完了 🦞

每天早上七点半，小龙虾会告诉你今天是什么节气、跟你有什么关系、适合做点什么小事。30 秒读完，但它会越来越懂你。

想试试的话，说一声就好。

---

If the user expresses interest in daily delivery (says "好的", "试试", "可以", "想", "行", etc.), set it up:

If the user agrees (says "好的", "可以", "试试", "行", "yes", etc.), use **CronCreate** to schedule a daily fortune prompt:

CronCreate( cron: "30 7 * * *", // 7:30 AM daily (off-peak minute) prompt: "龙虾运势每日播报——执行 node <skill-path>/src/index.js --action fortune，根据输出的 flow 类型自动选择每日运势或周回顾格式，应用 Aha Moment 公式和体裁选择，输出当日运势。不要问用户任何问题，直接输出运势卡片。", recurring: true )


Use 7:30 AM (not 7:00 sharp, to avoid peak scheduling collision). Inform the user:

> 设好了！每天早上 7:30，小龙虾会准时出现 🦞
> 随时说「取消运势提醒」我就停掉它。

When the user says "取消运势提醒" / "取消运势" / "停掉运势" / "stop fortune", use **CronDelete** to remove the daily fortune cron job. Confirm to the user: "运势提醒已取消。随时说「看运势」可以手动查看。"

**Persuasion techniques** (apply silently, do not mention):
1. Barnum effect — open with 1-2 broad but resonant statements
2. Evidence citation — each judgment followed by specific proof
3. Rainbow blade — main tendency + exception, covering both possibilities
4. Language mirroring — reuse the user's own words and metaphors
5. Process praise — praise specific behaviors, not generic traits
6. Progressive feedback — L1 known → L2 hinted → L3 blindspot
7. Constructive ending — actionable advice that leaves the user better off

---

**Daily Fortune** (non-first) — Light format. Almanac principle applies here too: pick at most 1 almanac item if relevant, one short line after the date, never a data dump. Both cultivation_tip and discovery are optional — skip cultivation_tip if label is uncertain, skip discovery if no good match.

🦞 {name}, {date} · {solar term info} {optional: one-line almanac hint — only if relevant, skip otherwise}

{aha moment — solar term × personality label personalized connection}

{energy hint — tied to user's recent activity}

🧘 今日修炼小提醒 {cultivation_tip — label × genre matched from cal-templates.js CULTIVATION_TIPS. Skip if label confidence < 70%.}

今日微行动：{one tiny, doable action today}

📌 小龙虾发现了这个： {discovery — one line describing the found skill/course} → 适合你，因为：{one sentence linking to user's personality label or blindspot} {skip entire 小龙虾 block if no good match found}

运势一句话：{one-liner summary}


Aha Moment formula: solar term / phenology data + personality label relevance + specific action.
Genre selection: Wrap-up Day (user has been planning) | Recharge Day (user has been outputting) | Connection Day (user has been working alone).

---

**Weekly Review** (Sunday trigger) — Insert before daily fortune:

🦞 {name}, {date} · Weekly Review

{week summary — reference this week's changes and progress}

Personality Insight Update {updated observation — "Last time I said you were X, this week I'm seeing Y"}

Today's Fortune {daily fortune content}


## Scene Adaptation Reference

| Dimension | Career-Type User | Relationship-Type User |
|-----------|-----------------|----------------------|
| Core evidence | Work patterns, decision style, project approach | Relationship patterns, emotional expression, values |
| Inner drive | What drives career choices | What drives relationship choices |
| Blindspot | Management/strategy/execution blindspots | Relationship/self-care/boundary blindspots |
| Energy pattern | Work rhythm (insight/sprint/trickle) | Emotional rhythm (burst/contained/trickle) |
| Advice | Projects, teams, direction | Relationship improvement, self-growth, life balance |
| Language style | Rational, logical, strategic | Warm, empathetic, narrative |

## Dependencies

- `oceanbus` — L0 P2P identity and messaging
- Node.js `>=18.0.0`

## External Endpoints

所有通信均通过 OceanBus L0 消息通道。不调用外部 HTTP API，不存储 API 密钥。

| Endpoint | Purpose | Data Sent |
|----------|---------|-----------|
| `OCEANBUS_URL` (L0 API) | OB 匿名身份注册、L0 消息路由 | `agent_id`（设备本地生成）、`api_key`（本地密钥）、`openid`（密码学随机地址）|
| L1 CalendarSvc (via OB L0) | 节气、农历、物候数据 | `date`（日期）、`city`（城市级——非 GPS，百万人级混淆）|
| L1 PersonalitySvc (via OB L0) | 人格标签丰富 | `openid`（用于请求/响应路由）、`label`（抽象人格类型）、`center`（思维/情感/本能）、`traits`（脱敏特征关键词，绝不含原始对话）|
| L1 Discovery (via OB L0) | 推荐候选获取 + 去重 | `openid`（仅用于去重——记录已推过的 skill，避免重复打扰。不用于跨会话追踪或个人画像）|

## Security & Privacy

### OceanBus OpenID 机制（为什么比 HTTPS 更安全）

传统 HTTPS 方案下，服务端管理员可以读取用户数据——隐私靠公司良心。OceanBus 从架构上解决了这个问题：

| 维度 | HTTPS 方案 | OceanBus 方案 |
|------|-----------|--------------|
| **用户身份** | IP + 设备指纹 → 可关联到人 | 密码学随机 OpenID → 不可反向识别 |
| **OpenID 更换** | 不可能（账号绑定） | **随时可换**——删除 `~/.lucky-lobster/` 即生成新身份 |
| **服务端能看什么** | 全部请求内容 | 仅加密消息信封（不包含内容）|
| **第三方能追踪吗** | 可通过 IP/指纹跨站追踪 | 不能——无 IP、无指纹、无持久标识 |
| **代码可审计** | 服务端代码闭源 | OB SDK 开源（MIT-0）——`npm install oceanbus`，源码在 npmjs.com |

### 数据分级

**A 类数据（永不出设备）：**
- 对话内容（项目名、人名、公司名、具体讨论）
- Memory 文件（角色、偏好、反馈）
- 加载的文件内容
- 详细行为模式和情绪状态

**B 类数据（脱敏后经 OB L0 发送）：**
- 人格标签（"布局者"等 9 种之一）和中心类型（思维/情感/本能）
- 特征关键词（如 `["系统思考", "独立决策"]`——绝不含原始对话片段）
- 城市级位置（百万人级混淆集，非 GPS 坐标）
- 年龄段（跨度 ≥ 5 岁，非精确年龄）
- OpenID（密码学随机地址，可随时更换，不关联现实身份）

### 为什么传输的字段不构成 PII

- **OpenID** = 随机字符串，不含姓名/电话/邮箱/IP。随时可换，旧值立即失效
- **城市级** = 北京（2200 万人），非 GPS 坐标，非区县
- **年龄段** = "31-40"（数万至数百万人级混淆），非"1988 年出生"
- **人格标签** = 抽象人格类型，非识别信息
- **以上字段的组合** 在任何情况下都无法指向具体的个人

### 隐私保障

- 不传输可识别到具体个人的信息（姓名、电话、邮箱、IP、GPS、设备指纹）
- OpenID 可随时更换：删除 `~/.lucky-lobster/ob-credentials.json` 即生成新身份
- L1 服务永远看不到原始对话——只收到脱敏标签
- 用户画像（生日、城市、性别）存储于本地 `~/.lucky-lobster/profile.json`，该文件永不出设备。从中提取的脱敏衍生字段（城市级位置、年龄段）经 OB 发送——原始数据与衍生数据的隐私级别不同，详见上方表格
- OB SDK 完全开源（MIT-0），所有通信代码可审计
- 未来品牌方仅接收匿名聚合统计，无法追踪个体用户

## Trust Statement

**安装 ≠ 同意使用。安装仅表示你已阅读以下披露。** 首次使用（说"看运势"）时，系统会弹出显式同意提示——你需要输入"同意"后，运势分析才会开始。同意后写入 `~/.lucky-lobster/consent.json` 记录。

1. **本地人格分析**：Agent 会在你的设备本地读取对话记忆和会话风格进行分析。完整对话内容、项目名、人名永远不出设备。
2. **匿名数据传输**：分析完成后，仅脱敏后的人格标签（如"布局者"）、城市级位置（百万人级混淆）、年龄段（跨度≥5岁）通过 OceanBus L0 加密通道发往 L1 服务。
3. **匿名身份**：系统在 `~/.lucky-lobster/` 下创建 OceanBus 匿名凭证（密码学随机 OpenID）。OpenID 是随机字符串，不关联真实身份；随时删除该目录即生成新身份。
4. **定时推送**：仅在用户主动确认后启用。说"取消运势提醒"随时停止。
5. **推荐去重**：推荐功能将 OpenID 发往 L1 服务，仅用于记录"已推荐过的 skill"，避免重复推送。不用于跨会话追踪或个人画像。

OceanBus SDK 完全开源（MIT-0），可在 npmjs.com/package/oceanbus 审计。你不必信任我们——你可以验证代码。

## Model Usage

This skill is designed for autonomous invocation by AI models. The model will analyze user data from available context sources and generate a personalized fortune reading without requiring user input beyond the trigger phrase. All analysis steps (2-5) are executed silently before the final reading is presented.