feat(skills): add 6 due-diligence data-source skills (#111)

## Summary Adds six enterprise due-diligence data-source skills as local built-in skills (SKILL.md + SKILL.zh-CN.md + catalog-metadata sidecar), per ADR-038. | Skill | Category | Risk | Source | |---|---|---|---| | baidu-poi-search | data | low | Baidu Map Place API (REST) | | tianyancha-risk | business | low | Tianyancha risk API (REST) | | ccgp-gov-procurement | business | medium | ccgp.gov.cn (WebBridge) | | cnipa-patent-search | business | medium | CNIPA pss-system (WebBridge) | | creditchina-query | business | high | creditchina.gov.cn (WebBridge + ddddocr) | | multi-source-sentiment | research | medium | Douyin Index/Featured + WebSearch | ## Changes - 6 x skills/<id>/{SKILL.md, SKILL.zh-CN.md, catalog-metadata.v1.json} - builtin-skills.json: 34 -> 40 - manifest.json: totalSkills 63 -> 69 - README.md counts + list ## Security - Credentials replaced with env-var placeholders (BAIDU_MAP_AK, TIANYANCHA_TOKEN); verified no secrets in diff ## Validation - validate-i18n.py: 0 errors (remaining warnings pre-existing repo-wide) - zh/en heading parity verified; source_hash computed per repo algorithm Co-authored-by: yi-ge <yi-ge@desirecore.net>
2026-09-05 22:04:07 +08:00 · 2026-09-03 14:05:58 +08:00
parent 43a8cbd549
commit c83f917901
21 changed files with 2668 additions and 8 deletions
--- a/skills/multi-source-sentiment/SKILL.md
+++ b/skills/multi-source-sentiment/SKILL.md
@@ -0,0 +1,229 @@
+---
+name: multi-source-sentiment
+description: >-
+  企业尽调舆情采集——通过抖音指数、抖音精选、WebSearch 三大数据源采集企业舆情，分析负面风险（诉讼/处罚/质量/高管/劳资），输出尽调舆情报告。Use when 用户提到"查舆情"、"企业舆情"、"负面新闻"、"舆论风险"、"尽调舆情"、"企业口碑"、"舆情分析"、"网络口碑"。
+version: 1.0.0
+type: procedural
+risk_level: medium
+status: enabled
+tags:
+  - due-diligence
+  - sentiment
+  - public-opinion
+  - douyin
+metadata:
+  author: desirecore
+  updated_at: '2026-09-03'
+  i18n:
+    default_locale: en-US
+    source_locale: zh-CN
+    locales:
+      - zh-CN
+      - en-US
+    zh-CN:
+      name: 企业尽调舆情采集
+      short_desc: 抖音指数+抖音精选+WebSearch 三渠道采集，输出尽调舆情报告
+      description: >-
+        企业尽调舆情采集——通过抖音指数、抖音精选、WebSearch 三大数据源采集企业舆情，分析负面风险（诉讼/处罚/质量/高管/劳资），输出尽调舆情报告。Use when 用户提到"查舆情"、"企业舆情"、"负面新闻"、"舆论风险"、"尽调舆情"、"企业口碑"、"舆情分析"、"网络口碑"。
+      body: ./SKILL.zh-CN.md
+      source_hash: sha256:e102f3139e03fcd5
+      translated_by: human
+    en-US:
+      name: Multi-Source Sentiment
+      short_desc: Due-diligence sentiment via Douyin Index, Douyin Featured, and WebSearch
+      description: >-
+        Due-diligence sentiment collection — gather enterprise public opinion via three channels (Douyin Index, Douyin Featured, WebSearch), analyze negative risks (litigation / penalties / quality / executives / labor), and output a due-diligence sentiment report. Use when the user asks about enterprise sentiment, negative news, reputation risk, or public opinion.
+      body: ./SKILL.md
+      source_hash: sha256:e102f3139e03fcd5
+      translated_by: human
+market:
+  icon: >-
+    <svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" viewBox="0 0 24 24" fill="none"><circle cx="12" cy="12" r="2" fill="#FF2D55"/><path d="M12 7a5 5 0 0 1 5 5M12 3a9 9 0 0 1 9 9" stroke="#FF2D55" stroke-width="1.6" stroke-linecap="round"/><path d="M7 12a5 5 0 0 1 5-5M3 12a9 9 0 0 1 9-9" stroke="#FF9500" stroke-width="1.6" stroke-linecap="round"/></svg>
+  category: research
+  maintainer:
+    name: DesireCore Official
+    verified: true
+  compatible_agents: []
+  channel: latest
+  required_client_version: 10.0.115
+---
+
+# Multi-Source Sentiment (Due Diligence)
+
+## L0: One-Sentence Summary
+
+Input a company name, collect public opinion via Douyin Index (trend) + Douyin Featured (negative material) + WebSearch (web-wide news), and output a due-diligence sentiment report (negative-risk list / positive highlights / sentiment overview / overall judgment).
+
+## L1: Overview
+
+- **Purpose**: answer the core due-diligence questions — any negative publicity? litigation/penalty coverage? product-quality issues? executive misconduct? overall reputation?
+- **Channel split**:
+
+| Channel | Due-diligence use | Collection |
+|---|---|---|
+| **Douyin Index** | Quantified heat trend (is something fermenting recently?) | Kimi WebBridge visiting Oceanengine Trends |
+| **Douyin Featured** | Concrete negative material (complaint/expose videos) | Kimi WebBridge searching Douyin |
+| **WebSearch** | Web-wide negative news / court notices / penalties / complaints | DesireCore built-in WebSearch |
+
+- **Prerequisites**: Kimi WebBridge daemon running (`127.0.0.1:10086`) + browser extension connected
+
+## L2: Procedure
+
+### Source 1: Douyin Index (heat trend)
+
+#### Access Method
+
+```
+URL: https://trendinsight.oceanengine.com/arithmetic-index/analysis/keyword?keyword={company-short-name}&appName=aweme
+Method: Kimi WebBridge opens the URL and extracts page content
+Note: URL-parameterized access avoids captchas. Do not type into the search box.
+```
+
+#### Keyword Strategy
+
+| Dimension | Keyword |
+|---|---|
+| Overall heat | `{short name}` |
+| Negative heat | `{short name} 负面` |
+| Complaint heat | `{short name} 投诉` |
+
+#### Interpretation
+
+- **Sudden spike** (MoM > 200%) → a negative event may be fermenting; investigate
+- **Flat trend** → low reputation risk
+- **Persistently high** → normal for famous companies; judge by content
+
+### Source 2: Douyin Featured (negative material)
+
+#### Access Method
+
+```
+URL: https://www.douyin.com/search/{company-short-name}?type=video
+Method: Kimi WebBridge searches Douyin, extracts title / author / likes / comments
+Note: search by short name (full names rarely hit); prioritize high-engagement videos.
+```
+
+#### Collected Fields
+
+Collected fields: video title (negative signals: complaint / rights-protection / exposure / quality / scam / runaway), likes & comments (reach = impact), author (personal vent vs media report vs competitor smear), publish date (recent vs historical).
+
+#### Signal Grading
+
+| Signals | Risk |
+|---|---|
+| 投诉 / 维权 / 欺骗 / 诈骗 / 跑路 | 🔴 high |
+| 质量 / 召回 / 不合格 / 假货 | 🔴 high |
+| 欠薪 / 辞退 / 仲裁 | 🟡 medium |
+| 吐槽 / 差评 / 不满 | 🟡 medium |
+| positive / awards / innovation | 🟢 positive |
+
+### Source 3: WebSearch (web-wide)
+
+#### Search Keyword Groups
+
+```python
+# High priority (always)
+search_queries_high = [
+    "{company} 负面 OR 投诉 OR 维权 OR 欺骗 OR 跑路",
+    "{company} 起诉 OR 法院 OR 被告 OR 被执行 OR 失信",
+    "{company} 处罚 OR 违规 OR 罚款 OR 整改 OR 通报批评",
+]
+
+# Medium priority (recommended)
+search_queries_mid = [
+    "{company} 质量 OR 召回 OR 不合格 OR 安全事故",
+    "{company} 老板 OR 法人 OR 总经理 丑闻 OR 被查 OR 被抓",
+    "{company} 欠薪 OR 辞退 OR 劳动仲裁 OR 工伤",
+]
+
+# Low priority (positive control)
+search_queries_low = [
+    "{company} 获奖 OR 创新 OR 认定 OR 排名",
+]
+```
+
+#### Collected Fields
+
+Collected fields: title / source site / date / summary / URL. **Note**: distinguish the company as **plaintiff** (enforcement, not a negative signal) vs **defendant** (negative signal).
+
+### Step 1: Run the three channels
+
+1. **WebSearch** first (highest information density): run high/medium/low query groups and collect results
+2. **Douyin Index**: check the heat trend for signs of recent fermentation
+3. **Douyin Featured**: search videos and filter by negative signal words
+
+### Step 2: LLM post-processing
+
+| Step | Description |
+|---|---|
+| Sentiment classification | tag each item positive / neutral / negative |
+| Risk grading | negative → high (defendant / penalty / runaway / safety accident) / medium (complaints / quality gripes / labor) / low (generic bad reviews) |
+| Event merging | merge duplicate coverage of one event (keep the earliest) |
+| Plaintiff/defendant split | litigation where the company is plaintiff is **not** a negative risk |
+| Recency | mark "recent" (≤3 months) vs "historical" |
+
+### Step 3: Output the report
+
+```json
+{
+  "query_status": "success",
+  "source": "multi-source-sentiment",
+  "company_name": "{company}",
+  "sentiment_overview": {
+    "overall_assessment": "正面 | 中性偏正 | 中性 | 中性偏负 | 负面 | 高风险",
+    "negative_ratio": "15%",
+    "neutral_ratio": "60%",
+    "positive_ratio": "25%",
+    "trend": "稳定 | 近期发酵 | 持续负面"
+  },
+  "negative_risks": [
+    {
+      "risk_type": "诉讼报道(被告) | 行政处罚 | 产品质量 | 高管负面 | 劳资纠纷 | 消费者投诉 | 经营异常",
+      "event_title": "……",
+      "event_date": "YYYY-MM-DD",
+      "recency": "近期 | 历史",
+      "source_url": "https://……",
+      "severity": "高 | 中 | 低",
+      "summary": "1-2 sentence summary"
+    }
+  ],
+  "controversies": [],
+  "positive_highlights": [],
+  "sentiment_summary": "3-5 sentence judgment: overall reputation, main negatives and severity, fermentation status, due-diligence advice",
+  "data_sources_used": ["websearch", "douyin-index", "douyin-featured"],
+  "collection_time": "ISO8601"
+}
+```
+
+## Quality Standards
+
+- **Authenticity**: all content must come from actual collection (WebSearch results / Douyin pages); never fabricate
+- **Traceability**: every negative risk carries a source_url
+- **Plaintiff/defendant**: litigation where the company is plaintiff goes to "litigation updates", not negative risks
+- **Recency**: every negative item is marked recent (≤3 months) or historical
+- **No absolute conclusions**: sentiment is one auxiliary signal; the report must note it should be combined with business-registration, judicial, and financial data
+
+## Error Handling
+
+| Scenario | Handling |
+|---|---|
+| Douyin Index unreachable | skip; supplement with WebSearch + Featured; note "heat data missing" in the report |
+| Douyin search requires login | skip; WebSearch `{company} 抖音 负面` instead |
+| WebSearch returns nothing | mark "no negative found" — a positive signal |
+| Too many negatives | keep the top 10 by severity; count the rest |
+
+## Known Limitations
+
+- **Not real-time**: depends on search-engine indexing and Douyin pages; hour-to-day latency (7×24 monitoring requires commercial sentiment APIs)
+- **LLM-based sentiment**: no dedicated sentiment model; ~85-90% accuracy, edge cases may misjudge
+- **Douyin coverage**: WeChat/Weibo/Xiaohongshu content is only covered indirectly via WebSearch
+- **Full vs short name**: WebSearch uses the full name (precision), Douyin the short name (recall); results may differ
+
+## Troubleshooting
+
+| Problem | Cause | Fix |
+|---|---|---|
+| WebBridge daemon unreachable | daemon down | `~/.kimi-webbridge/bin/kimi-webbridge.exe start` |
+| Trends page blank | JS not rendered | wait 5 s, then evaluate innerText |
+| Douyin search empty | login required / name mismatch | skip; supplement via WebSearch |
+| WebSearch results all positive | weak negative keywords | use concrete terms (e.g. "被处罚" instead of "处罚") |
--- a/skills/multi-source-sentiment/SKILL.zh-CN.md
+++ b/skills/multi-source-sentiment/SKILL.zh-CN.md
@@ -0,0 +1,193 @@
+# 企业尽调舆情采集
+
+## L0
+
+输入企业名称，通过抖音指数（热度趋势）+ 抖音精选（负面素材）+ WebSearch（全网新闻）三渠道采集舆情，输出尽调舆情报告（负面风险清单/正面亮点/情感概览/综合判断）。
+
+## L1 概述
+
+- **目的**：回答尽调核心问题——有没有负面舆情？有没有诉讼/处罚报道？产品质量有没有问题？高管有没有负面？舆论形象如何？
+- **三渠道分工**：
+
+| 渠道 | 尽调用途 | 采集方式 |
+|---|---|---|
+| **抖音指数** | 舆论热度量化（时间序列，看是否近期发酵） | Kimi WebBridge 访问巨量算数 |
+| **抖音精选** | 具体负面内容素材（投诉视频/维权曝光/质量吐槽） | Kimi WebBridge 搜索抖音 |
+| **WebSearch** | 全网负面新闻/法院公告/监管处罚/消费者投诉 | DesireCore 内置 WebSearch |
+
+- **前置依赖**：Kimi WebBridge daemon 运行中（`127.0.0.1:10086`）+ 浏览器扩展已连接
+
+## L2 详细操作
+
+### 数据源一：抖音指数（企业名热度趋势）
+
+#### 访问方式
+
+```
+URL: https://trendinsight.oceanengine.com/arithmetic-index/analysis/keyword?keyword={企业简称}&appName=aweme
+方法: Kimi WebBridge 浏览器自动访问页面并提取内容
+注意: URL 参数化访问，不触发验证码。不要尝试在搜索框输入文字。
+```
+
+#### 尽调关键词策略
+
+| 尽调维度 | 搜索关键词 |
+|---|---|
+| 舆论整体热度 | `{企业简称}` |
+| 负面热度 | `{企业简称} 负面` |
+| 投诉热度 | `{企业简称} 投诉` |
+
+#### 尽调判断要点
+
+- **热度骤升**：近期热度陡增（环比 >200%）→ 可能存在负面事件发酵，需重点排查
+- **热度平稳**：无异常波动 → 舆论风险较低
+- **长期高热**：知名企业的常态，需结合内容素材判断正负
+
+### 数据源二：抖音精选（负面内容素材）
+
+#### 访问方式
+
+```
+URL: https://www.douyin.com/search/{企业简称}?type=video
+方法: Kimi WebBridge 搜索抖音，提取视频标题/作者/点赞数/评论数
+注意: 用企业简称搜索（全名命中率低）；优先看点赞/评论高的视频（传播广=影响大）。
+```
+
+#### 尽调采集字段
+
+- 视频标题（判断是否负面：投诉/维权/曝光/质量/欺骗/跑路）
+- 点赞数 / 评论数（传播广度 = 负面影响程度）
+- 作者昵称（个人吐槽 vs 媒体报道 vs 竞品抹黑）
+- 发布时间（近期 vs 历史）
+
+#### 负面视频筛选标准
+
+| 信号词 | 风险等级 |
+|---|---|
+| 投诉 / 维权 / 欺骗 / 诈骗 / 跑路 | 🔴 高 |
+| 质量 / 召回 / 不合格 / 假货 | 🔴 高 |
+| 欠薪 / 辞退 / 仲裁 | 🟡 中 |
+| 吐槽 / 差评 / 不满 | 🟡 中 |
+| 正面宣传 / 获奖 / 创新 | 🟢 正面 |
+
+### 数据源三：WebSearch（全网尽调舆情）
+
+#### 搜索关键词组合（按尽调维度）
+
+```python
+# 高优先级（必查）
+search_queries_high = [
+    "{企业名} 负面 OR 投诉 OR 维权 OR 欺骗 OR 跑路",
+    "{企业名} 起诉 OR 法院 OR 被告 OR 被执行 OR 失信",
+    "{企业名} 处罚 OR 违规 OR 罚款 OR 整改 OR 通报批评",
+]
+
+# 中优先级（建议查）
+search_queries_mid = [
+    "{企业名} 质量 OR 召回 OR 不合格 OR 安全事故",
+    "{企业名} 老板 OR 法人 OR 总经理 丑闻 OR 被查 OR 被抓",
+    "{企业名} 欠薪 OR 辞退 OR 劳动仲裁 OR 工伤",
+]
+
+# 低优先级（正面对照）
+search_queries_low = [
+    "{企业名} 获奖 OR 创新 OR 认定 OR 排名",
+]
+```
+
+#### 采集字段
+
+- 新闻标题 / 来源网站 / 发布时间 / 摘要 / 原文链接
+- **注意**：区分企业作为**原告**（维权，非负面信号）vs **被告**（被诉，负面信号）
+
+### Step 1：三渠道采集执行
+
+1. **WebSearch 尽调搜索**（先执行，信息量最大）：按高/中/低优先级逐组搜索，收集新闻标题+来源+日期+摘要
+2. **抖音指数热度**：访问巨量算数查企业简称热度趋势，判断是否近期发酵
+3. **抖音精选内容**：搜索抖音视频，按负面信号词筛选高互动视频
+
+### Step 2：LLM 后处理（情感分析 + 风险分级）
+
+对采集到的原始内容执行：
+
+| 分析步骤 | 说明 |
+|---|---|
+| **情感分类** | 每条标注：正面 / 中性 / 负面 |
+| **风险分级** | 负面内容分：高（诉讼被告/处罚/跑路/安全事故）/ 中（投诉/质量吐槽/劳资）/ 低（一般差评） |
+| **事件归并** | 同一事件的多条报道归并为一条（以最早报道为准） |
+| **原被告区分** | 企业作为原告的维权诉讼**不是**负面风险信号 |
+| **时效判断** | 近 3 个月的负面标记"近期"，更早的标记"历史" |
+
+### Step 3：输出尽调舆情报告
+
+```json
+{
+  "query_status": "success",
+  "source": "multi-source-sentiment",
+  "company_name": "{企业名}",
+  "sentiment_overview": {
+    "overall_assessment": "正面 | 中性偏正 | 中性 | 中性偏负 | 负面 | 高风险",
+    "negative_ratio": "15%",
+    "neutral_ratio": "60%",
+    "positive_ratio": "25%",
+    "trend": "稳定 | 近期发酵 | 持续负面"
+  },
+  "negative_risks": [
+    {
+      "risk_type": "诉讼报道(被告) | 行政处罚 | 产品质量 | 高管负面 | 劳资纠纷 | 消费者投诉 | 经营异常",
+      "event_title": "……",
+      "event_date": "YYYY-MM-DD",
+      "recency": "近期 | 历史",
+      "source_url": "https://……",
+      "severity": "高 | 中 | 低",
+      "summary": "1-2 句事件摘要"
+    }
+  ],
+  "controversies": [
+    {
+      "event_title": "……",
+      "description": "争议双方与焦点",
+      "status": "发酵中 | 已平息 | 官方已回应"
+    }
+  ],
+  "positive_highlights": [
+    "获奖/创新/贡献等正面事件（一句话）"
+  ],
+  "sentiment_summary": "3-5 句舆情综合判断：整体舆论形象、主要负面及严重程度、是否近期发酵、尽调建议",
+  "data_sources_used": ["websearch", "douyin-index", "douyin-featured"],
+  "collection_time": "ISO8601"
+}
+```
+
+## 质量标准
+
+- **数据真实性**：所有舆情内容必须来自实际采集（WebSearch 结果 / 抖音页面），绝不编造
+- **来源可溯**：每条负面风险必须附 source_url
+- **原被告区分**：企业作为原告的诉讼报道不列入负面风险（单独归类为"涉诉动态"）
+- **时效标注**：每条负面标注"近期"（3 个月内）或"历史"
+- **不做绝对结论**：舆情是辅助信号，报告末尾注明"舆情仅为尽调参考维度之一，需结合工商/司法/财务数据综合判断"
+
+## 错误处理
+
+| 场景 | 处理方式 |
+|---|---|
+| 抖音指数页面无法访问 | 跳过该源，用 WebSearch + 抖音精选补充，报告中注明"热度数据缺失" |
+| 抖音搜索需登录 | 跳过该源，WebSearch 补充搜索 `{企业名} 抖音 负面` |
+| WebSearch 无结果 | 该维度标注"未检索到相关负面"，属正面信号 |
+| 负面信息量极大 | 只取前 10 条高严重度条目，其余归入计数 |
+
+## 已知限制
+
+- **非实时**：依赖搜索引擎收录与抖音页面展示，存在小时级~天级延迟（非 7×24 实时监测——持续监测需商业舆情 API）
+- **情感分析靠 LLM**：无专用情感模型，由 LLM 后处理推断（准确率约 85-90%，边界案例可能误判）
+- **抖音覆盖**：微信公众号/微博/小红书等平台内容仅能通过 WebSearch 间接覆盖（非全平台直达）
+- **企业全名 vs 简称**：WebSearch 用全名（精确），抖音用简称（命中率）；两者结果可能有差异
+
+## 故障排查
+
+| 问题 | 原因 | 处理 |
+|---|---|---|
+| WebBridge daemon 不可达 | daemon 未运行 | `~/.kimi-webbridge/bin/kimi-webbridge.exe start` |
+| 巨量算数页面空白 | JS 渲染未完成 | 等 5 秒后 evaluate 读 innerText |
+| 抖音搜索页无结果 | 需登录 / 简称不匹配 | 跳过，WebSearch 补充 |
+| WebSearch 结果全是正面 | 负面词组合不当 | 换用具体负面词（如"被处罚"而非"处罚"） |
--- a/skills/multi-source-sentiment/catalog-metadata.v1.json
+++ b/skills/multi-source-sentiment/catalog-metadata.v1.json
@@ -0,0 +1,74 @@
+{
+  "$schema": "../../schemas/catalog-metadata.v1.schema.json",
+  "schemaVersion": 1,
+  "identity": {
+    "kind": "skill",
+    "id": "multi-source-sentiment"
+  },
+  "presentation": {
+    "defaultLocale": "en-US",
+    "i18n": {
+      "zh-CN": {
+        "name": "企业尽调舆情采集",
+        "summary": "抖音指数+抖音精选+WebSearch 三渠道采集，输出尽调舆情报告",
+        "description": "企业尽调舆情采集——通过抖音指数、抖音精选、WebSearch 三大数据源采集企业舆情，分析负面风险（诉讼/处罚/质量/高管/劳资），输出尽调舆情报告。Use when 用户提到\"查舆情\"、\"企业舆情\"、\"负面新闻\"、\"舆论风险\"、\"尽调舆情\"、\"企业口碑\"、\"舆情分析\"、\"网络口碑\"。"
+      },
+      "en-US": {
+        "name": "Multi-Source Sentiment",
+        "summary": "Due-diligence sentiment via Douyin Index, Douyin Featured, and WebSearch",
+        "description": "Due-diligence sentiment collection — gather enterprise public opinion via three channels (Douyin Index, Douyin Featured, WebSearch), analyze negative risks (litigation / penalties / quality / executives / labor), and output a due-diligence sentiment report. Use when the user asks about enterprise sentiment, negative news, reputation risk, or public opinion."
+      }
+    },
+    "category": "research",
+    "tags": [
+      "due-diligence",
+      "sentiment",
+      "public-opinion",
+      "douyin"
+    ]
+  },
+  "release": {
+    "state": "known",
+    "version": "1.0.0",
+    "versionScheme": "semver"
+  },
+  "timestamps": {
+    "catalogUpdatedAt": {
+      "state": "known",
+      "value": "2026-09-03T05:45:00Z",
+      "precision": "second"
+    },
+    "releasePublishedAt": {
+      "state": "unknown"
+    },
+    "reviewedAt": {
+      "state": "unknown"
+    },
+    "upstreamObservedAt": {
+      "state": "unknown"
+    }
+  },
+  "provenance": {},
+  "governance": {
+    "stewardship": "official",
+    "availability": "installable",
+    "license": {
+      "state": "unknown"
+    },
+    "redistribution": "verify-package-terms",
+    "listingMaintainer": {
+      "name": "DesireCore Official",
+      "verified": true
+    }
+  },
+  "compatibility": {
+    "platforms": {
+      "state": "unknown"
+    },
+    "requiredClientVersion": "10.0.115"
+  },
+  "spec": {
+    "kind": "skill",
+    "riskLevel": "medium"
+  }
+}