@vocmarket/voc-skill 0.3.15

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (82) hide show
  1. package/.claude-plugin/plugin.json +8 -0
  2. package/.mcp.json +11 -0
  3. package/README.md +334 -0
  4. package/bin/claude-voc.js +790 -0
  5. package/docs/boss-prompt-templates.md +682 -0
  6. package/docs/capability-map.md +195 -0
  7. package/docs/claude-code-troubleshooting-playbook.md +136 -0
  8. package/docs/context-budget-implementation-plan.md +491 -0
  9. package/docs/customer-quickstart.md +43 -0
  10. package/docs/deliverables/competitor-map-report.md +43 -0
  11. package/docs/deliverables/seven-day-content-plan.md +45 -0
  12. package/docs/deliverables/single-problem-action-plan.md +49 -0
  13. package/docs/deliverables/trend-insight-report.md +41 -0
  14. package/docs/deliverables/voc-issue-pool-report.md +42 -0
  15. package/docs/demo-runbook.md +31 -0
  16. package/docs/live-manual-acceptance-checklist.md +63 -0
  17. package/docs/npm-packaging-notes.md +93 -0
  18. package/docs/payment-package-links.md +86 -0
  19. package/install.js +208 -0
  20. package/mcp/src/core/credentials.js +113 -0
  21. package/mcp/src/core/evidence-quality.js +106 -0
  22. package/mcp/src/core/memory-store.js +31 -0
  23. package/mcp/src/core/payment-links.js +117 -0
  24. package/mcp/src/core/result-envelope.js +31 -0
  25. package/mcp/src/features/douyin-trend/live-collector.js +396 -0
  26. package/mcp/src/features/douyin-trend/preference-memory.js +130 -0
  27. package/mcp/src/features/douyin-trend/report.js +318 -0
  28. package/mcp/src/features/douyin-trend/sample-data.js +103 -0
  29. package/mcp/src/features/douyin-trend/signal-extractor.js +79 -0
  30. package/mcp/src/features/douyin-trend/topic-builder.js +1 -0
  31. package/mcp/src/features/voc-business-workflow/business-workflow.js +214 -0
  32. package/mcp/src/features/voc-competitor-map/competitor-map.js +198 -0
  33. package/mcp/src/features/voc-content-plan/content-plan.js +359 -0
  34. package/mcp/src/features/voc-issue-pool/issue-pool.js +488 -0
  35. package/mcp/src/features/voc-problem-deep-dive/deep-dive.js +662 -0
  36. package/mcp/src/features/voc-speaking-script/speaking-script.js +293 -0
  37. package/mcp/src/features/xiaohongshu-trend/assets.js +158 -0
  38. package/mcp/src/features/xiaohongshu-trend/industry-presets.js +335 -0
  39. package/mcp/src/features/xiaohongshu-trend/live-collector.js +337 -0
  40. package/mcp/src/features/xiaohongshu-trend/preference-memory.js +130 -0
  41. package/mcp/src/features/xiaohongshu-trend/report.js +319 -0
  42. package/mcp/src/features/xiaohongshu-trend/sample-data.js +79 -0
  43. package/mcp/src/features/xiaohongshu-trend/signal-extractor.js +77 -0
  44. package/mcp/src/features/xiaohongshu-trend/topic-builder.js +84 -0
  45. package/mcp/src/providers/douyin-api.js +170 -0
  46. package/mcp/src/providers/xiaohongshu-api.js +102 -0
  47. package/mcp/src/server.js +621 -0
  48. package/mcp/src/tools/douyin-preference-update.js +70 -0
  49. package/mcp/src/tools/douyin-trend-run.js +464 -0
  50. package/mcp/src/tools/voc-business-workflow-run.js +52 -0
  51. package/mcp/src/tools/voc-competitor-map-run.js +59 -0
  52. package/mcp/src/tools/voc-content-plan-run.js +59 -0
  53. package/mcp/src/tools/voc-issue-pool-run.js +59 -0
  54. package/mcp/src/tools/voc-problem-deep-dive-run.js +59 -0
  55. package/mcp/src/tools/voc-speaking-script-run.js +59 -0
  56. package/mcp/src/tools/xiaohongshu-preference-update.js +85 -0
  57. package/mcp/src/tools/xiaohongshu-trend-run.js +404 -0
  58. package/memory-templates/douyin-trend-profile.json +33 -0
  59. package/memory-templates/xiaohongshu-trend-profile.json +33 -0
  60. package/package.json +65 -0
  61. package/scripts/smoke-cross-industry.js +126 -0
  62. package/scripts/smoke-mcp.js +302 -0
  63. package/scripts/smoke-package.js +1005 -0
  64. package/skill-package-manifest.json +62 -0
  65. package/skills/douyin-trend-intelligence/SKILL.md +78 -0
  66. package/skills/douyin-trend-intelligence/references/live-mode.md +52 -0
  67. package/skills/douyin-trend-intelligence/references/output-format.md +32 -0
  68. package/skills/douyin-trend-intelligence/references/user-workflow.md +24 -0
  69. package/skills/voc-business-workflow/SKILL.md +61 -0
  70. package/skills/voc-competitor-map/SKILL.md +60 -0
  71. package/skills/voc-competitor-map/references/restaurant-competitor-framework.md +25 -0
  72. package/skills/voc-competitor-map/references/restaurant-differentiation-patterns.md +44 -0
  73. package/skills/voc-content-plan/SKILL.md +63 -0
  74. package/skills/voc-content-plan/references/restaurant-content-framework.md +41 -0
  75. package/skills/voc-content-plan/references/restaurant-speaking-script-patterns.md +34 -0
  76. package/skills/voc-issue-pool/SKILL.md +81 -0
  77. package/skills/voc-problem-deep-dive/SKILL.md +86 -0
  78. package/skills/voc-speaking-script/SKILL.md +99 -0
  79. package/skills/xiaohongshu-trend-intelligence/SKILL.md +61 -0
  80. package/skills/xiaohongshu-trend-intelligence/references/live-mode.md +65 -0
  81. package/skills/xiaohongshu-trend-intelligence/references/output-format.md +71 -0
  82. package/skills/xiaohongshu-trend-intelligence/references/user-workflow.md +60 -0
@@ -0,0 +1,195 @@
1
+ # Claude Code VOC Intelligence 能力地图
2
+
3
+ 这份文档给 Claude Code、维护者和交付人员快速定位:这套技能包能做什么、该用哪个入口、出了问题去哪里看。
4
+
5
+ ## 一句话定位
6
+
7
+ 把小红书、抖音等平台的用户真实声音,转成企业能用的趋势情报、问题优先级、经营动作、内容选题和多轮验证记忆。
8
+
9
+ ## 用户任务到入口映射
10
+
11
+ | 用户想做的事 | 推荐入口 | MCP 工具 | 输出重点 |
12
+ | --- | --- | --- | --- |
13
+ | 一次跑完整经营闭环 | `voc-business-workflow` | `voc_business_workflow_run` | 市场声音、问题池、Top 问题动作、7 天内容计划、第一条口播稿 |
14
+ | 先演示流程,不真实采集 | 小红书/抖音趋势 skill | `voc_xiaohongshu_trend_run` / `voc_douyin_trend_run`,`collectionMode=sample` | 第一轮样本观察、机会假设、待确认问题 |
15
+ | 真实采集行业趋势 | 小红书/抖音趋势 skill | `voc_xiaohongshu_trend_run` / `voc_douyin_trend_run`,`collectionMode=live` | 真实笔记/视频、评论证据、趋势假设 |
16
+ | 检查 token 是否可用 | 平台趋势 skill | `voc_xiaohongshu_token_check` / `voc_douyin_token_check` | 是否已配置、无 token 时给充值链接 |
17
+ | 用户回答校准问题 | 平台趋势 skill | `voc_xiaohongshu_preference_update` / `voc_douyin_preference_update` | 保存偏好、屏蔽方向、下一轮聚焦 |
18
+ | 把评论整理成问题池 | `voc-issue-pool` | `voc_issue_pool_run` | 问题优先级、状态、证据、影响环节、建议动作 |
19
+ | 判断哪些问题最影响生意 | `voc-issue-pool` | `voc_issue_pool_run` | 本周优先推进的问题和下一步深挖入口 |
20
+ | 深挖一个具体问题 | `voc-problem-deep-dive` | `voc_problem_deep_dive_run` | 老板视角影响、动作清单、7 天验证 |
21
+ | 用户说某动作不适合 | `voc-problem-deep-dive` | `voc_problem_deep_dive_run`,传 `feedback` | 自动解析为屏蔽动作或偏好动作 |
22
+ | 选中一条内容继续改口播 | `voc-speaking-script` | `voc_speaking_script_run` | 60 秒稿、30 秒稿、提词器版、改稿反馈、定稿记忆 |
23
+ | 下轮继续优化 | 同一 skill | 带同一 `memory`,或带 project/industry/scenario/audience 让工具自动分桶 | 应用历史记忆,避免从零开始 |
24
+
25
+ ## 技能入口
26
+
27
+ ### `voc-business-workflow`
28
+
29
+ 适合:
30
+
31
+ - 用户只想说自然语言,不想选择工具
32
+ - 一次串起市场洞察、问题优先级、单点动作、内容计划和口播稿
33
+ - 远程给客户先跑一份第一轮经营交付
34
+
35
+ 用户典型话术:
36
+
37
+ ```text
38
+ 帮我看一下南昌餐饮最近顾客在关心什么,并告诉我先改哪里、下周发什么。
39
+ ```
40
+
41
+ ### `xiaohongshu-trend-intelligence`
42
+
43
+ 适合:
44
+
45
+ - 小红书趋势情报
46
+ - 行业用户顾虑
47
+ - 内容选题
48
+ - 销售话术
49
+ - 小红书评论证据卡
50
+
51
+ 用户典型话术:
52
+
53
+ ```text
54
+ 帮我做一份家装全屋定制行业的小红书趋势情报,先用演示样例跑通流程。
55
+ ```
56
+
57
+ ### `douyin-trend-intelligence`
58
+
59
+ 适合:
60
+
61
+ - 抖音趋势情报
62
+ - 本地生活/餐饮评论洞察
63
+ - 短视频内容机会
64
+ - 门店转化线索
65
+
66
+ 用户典型话术:
67
+
68
+ ```text
69
+ 帮我看下南昌餐饮行业抖音上用户最近在吐槽什么,先小规模真实采集。
70
+ ```
71
+
72
+ ### `voc-problem-deep-dive`
73
+
74
+ 适合:
75
+
76
+ - 单点 VOC 深挖
77
+ - 差评背后问题拆解
78
+ - 老板怎么解决
79
+ - 门店动作和内容选题
80
+ - 多轮记忆优化
81
+
82
+ 用户典型话术:
83
+
84
+ ```text
85
+ 继续深挖“排队”这个问题,告诉老板今天能改什么,7 天怎么验证。
86
+ ```
87
+
88
+ ### `voc-issue-pool`
89
+
90
+ 适合:
91
+
92
+ - 评论问题池
93
+ - 用户吐槽归类
94
+ - 问题优先级排序
95
+ - 问题状态管理
96
+ - 从趋势报告进入单点深挖前的中间层
97
+
98
+ 用户典型话术:
99
+
100
+ ```text
101
+ 把这些评论整理成问题池,告诉我哪些问题最影响生意,先改哪个。
102
+ ```
103
+
104
+ ## 关键产品原则
105
+
106
+ - 首轮报告是“初步判断”,不是最终结论。
107
+ - 输出必须带证据,不只给抽象观点。
108
+ - 老板用户不需要学习提示词,Claude Code 要主动选择技能和工具。
109
+ - live 失败时给可执行下一步,不暴露原始接口错误。
110
+ - 多轮反馈要进入记忆,让建议越用越贴近客户。
111
+ - 不同行业、不同客户的记忆要隔离。
112
+
113
+ ## 记忆规则
114
+
115
+ 趋势情报记忆:
116
+
117
+ - 记录保留方向、屏蔽方向、聚焦模式、用户备注。
118
+ - 适合下一轮趋势报告排序和话术收敛。
119
+
120
+ 单点 VOC 记忆:
121
+
122
+ - 记录偏好动作、屏蔽动作、偏好内容角度、验证有效动作、验证无效动作。
123
+ - 未显式传 `memory` 时,默认按 `project/brand/store + industry + scenario + audience` 分桶。
124
+ - 显式传 `memory` 时保持兼容,用指定文件沉淀。
125
+
126
+ ## 文件索引
127
+
128
+ | 文件 | 用途 |
129
+ | --- | --- |
130
+ | `README.md` | 面向使用者和 Claude Code 的总入口说明 |
131
+ | `skills/*/SKILL.md` | Claude Code 触发规则和操作约束 |
132
+ | `mcp/src/server.js` | MCP 工具注册、schema、工具说明 |
133
+ | `mcp/src/tools/*` | CLI/MCP 可调用工具入口 |
134
+ | `mcp/src/features/*` | 趋势报告、单点深挖、记忆、信号提取 |
135
+ | `mcp/src/core/credentials.js` | token 读取口径 |
136
+ | `mcp/src/core/payment-links.js` | 充值/开通提示 |
137
+ | `mcp/src/core/evidence-quality.js` | 证据质量门槛、无样本降级建议 |
138
+ | `docs/live-manual-acceptance-checklist.md` | live 采集人工验收 |
139
+ | `docs/claude-code-troubleshooting-playbook.md` | Claude Code 故障定位、误报修复、关键词修复和发散排查 |
140
+ | `docs/payment-package-links.md` | 支付链接和套餐入口说明 |
141
+ | `docs/context-budget-implementation-plan.md` | 上下文省 token 工作流 |
142
+ | `docs/deliverables/trend-insight-report.md` | 行业用户洞察趋势清单模板 |
143
+ | `docs/deliverables/competitor-map-report.md` | 竞品图谱与错位机会报告模板 |
144
+ | `docs/deliverables/voc-issue-pool-report.md` | VOC 问题清单与优先级模板 |
145
+ | `docs/deliverables/single-problem-action-plan.md` | 单点问题 7 天行动计划模板 |
146
+ | `docs/deliverables/seven-day-content-plan.md` | 7 天内容选题与口播脚本模板 |
147
+ | `scripts/smoke-package.js` | 包级 smoke,覆盖 sample、记忆、token、MCP |
148
+ | `scripts/smoke-mcp.js` | MCP 工具发现和基本调用 smoke |
149
+
150
+ ## 工作坊输出口径
151
+
152
+ 面向餐饮老板或线下工作坊时,Claude Code 输出优先按:
153
+
154
+ 1. 你这轮能拿到什么。
155
+ 2. 我会怎么做。
156
+ 3. 第一版结果。
157
+ 4. 你需要确认什么。
158
+ 5. 下一步建议。
159
+
160
+ 每个小节都必须对应一个交付物。不要在前台讲 MCP、schema、collectionMode、提示词工程。
161
+
162
+ ## 维护检查点
163
+
164
+ 每次发布至少确认:
165
+
166
+ - `package.json`、`package-lock.json`、`skill-package-manifest.json`、`.claude-plugin/plugin.json`、`mcp/src/server.js` 版本一致。
167
+ - `npm --prefix claude-code-voc-intelligence run mcp:smoke` 通过。
168
+ - `npm --prefix claude-code-voc-intelligence run smoke:package` 通过。
169
+ - `npm run claude-voc:build` 通过。
170
+ - npm 发布后 `npm view @vocmarket/voc-skill version` 返回新版本。
171
+ - 测试工作区 `npx --yes @vocmarket/voc-skill@latest workspace --smoke` 通过。
172
+
173
+ ## 下一轮优化补充
174
+
175
+ - `voc-issue-pool` 已支持跨轮问题变化:同一记忆范围下会对比上轮/本轮证据数,并标记新增、变多、变少、持平、本轮未再出现。
176
+ - `voc-problem-deep-dive` 已支持动作权重记忆:用户反馈“有效/无效”后,会记录验证次数,后续优先推荐有效动作、降低无效动作。
177
+ - 趋势情报输出后的推荐链路应优先进入问题池:先把证据整理成 VOC 问题池,判断先改哪个问题,再进入单点 VOC 深挖。
178
+
179
+ ## 工作坊交付型入口
180
+
181
+ | 老板说法 | 推荐入口 | MCP 工具 | 交付物 |
182
+ | --- | --- | --- | --- |
183
+ | 帮我看这个品类最近顾客在关心什么 | 小红书/抖音趋势 skill | `voc_xiaohongshu_trend_run` / `voc_douyin_trend_run` | 行业用户洞察趋势清单 |
184
+ | 帮我看顾客关心什么、先改哪里、下周发什么 | `voc-business-workflow` | `voc_business_workflow_run` | VOC 经营闭环报告 |
185
+ | 我不知道竞品是谁,帮我看看别人怎么做 | `voc-competitor-map` | `voc_competitor_map_run` | 竞品图谱与错位机会报告 |
186
+ | 顾客到底在吐槽什么,先改哪个 | `voc-issue-pool` | `voc_issue_pool_run` | VOC 问题清单与优先级 |
187
+ | 这个问题老板怎么解决 | `voc-problem-deep-dive` | `voc_problem_deep_dive_run` | 单点问题 7 天行动计划 |
188
+ | 下周账号发什么,帮我写口播 | `voc-content-plan` | `voc_content_plan_run` | 7 天内容选题与口播脚本 |
189
+ | 选题 2 进入脚本共创 | `voc-speaking-script` | `voc_speaking_script_run` | 口播稿改稿与定稿记忆 |
190
+
191
+ 边界说明:
192
+
193
+ - 口播/短视频脚本属于本技能包范围,但必须来自 VOC 用户问题、竞品评论差异、问题池 Top 问题或单点深挖动作。
194
+ - 创始人故事、品牌自传、老板个人 IP 长文不属于 `claude-code-voc-intelligence` 主线。
195
+ - 现场工作坊前台不要讲 MCP、schema、提示词工程;前台只讲“看市场、看竞品、看问题、找动作、做内容”。
@@ -0,0 +1,136 @@
1
+ # Claude Code VOC Troubleshooting Playbook
2
+
3
+ This playbook helps Claude Code and maintainers quickly map live VOC collection symptoms to the right fix path. Use it when a real user conversation gets stuck, a platform collector reports the wrong state, or Claude Code does not know whether to retry, recharge, or switch tools.
4
+
5
+ ## Fast Triage
6
+
7
+ | Symptom in Claude Code | Likely Cause | User-Facing State | First Fix Path | Do Not Do |
8
+ | --- | --- | --- | --- | --- |
9
+ | Douyin live says balance is insufficient, but the account has balance | Search API or keyword request failed and was classified as billing | `needs_keyword_fix` | Check Douyin API error classification and keyword repair | Do not show recharge link unless status is really `needs_recharge` |
10
+ | Douyin returns `社交平台接口请求错误` or HTTP 5xx | Upstream search or input issue | `needs_keyword_fix`, or auto-repair then `ok` | Normalize keyword, retry with a shorter natural keyword | Do not expose raw 5xx body to user |
11
+ | User passes `茶饮18元22元` and search is unstable | Compact keyword lacks natural search spacing | Usually auto-repaired | Normalize to `茶饮 18元 22元` | Do not require user to learn CLI flags |
12
+ | User passes empty `--keywords ""` | CLI parser receives an explicit empty value | `needs_keyword_fix` | Keep explicit empty as empty; do not fall back to default keywords | Do not silently search generic defaults |
13
+ | User asks “进入真实采集 live 模式” but workflow keeps deep-diving | Skill handoff rules are too weak | Should call platform live collector | Update deep-dive/business workflow skill instructions and `nextActions` | Do not keep asking 7-day validation questions |
14
+ | MCP output validation fails because warnings contain objects | Structured output schema expects strings | Tool call error | Normalize warnings at MCP server boundary | Do not remove evidence warnings entirely |
15
+ | Live collection says no token even after user configured one | Wrong working directory or unsupported credential source | `needs_token` | Check `.env.local` search path and `~/.claude/voc-credentials.json` | Do not print token |
16
+ | First report sounds final or overconfident | Report layer ignored evidence quality | `ok`, but poor UX | Enforce “初步判断 / 机会假设 / 待校准” and evidence quality | Do not write final business conclusions from first sample |
17
+
18
+ ## Douyin Search Error Fix Pattern
19
+
20
+ When Douyin search fails, fix in this order:
21
+
22
+ 1. Classify provider errors before building user messages.
23
+ 2. Treat billing only as billing when the provider clearly says balance, quota, payment, or recharge.
24
+ 3. Treat HTTP 400/422 and generic search 5xx as search/input repair.
25
+ 4. Normalize keywords before calling the API.
26
+ 5. Retry once with a shorter natural keyword.
27
+ 6. If retry fails, return `needs_keyword_fix` with 2-3 concrete keyword examples.
28
+
29
+ Current key files:
30
+
31
+ | Concern | File |
32
+ | --- | --- |
33
+ | Provider error classification | `mcp/src/providers/douyin-api.js` |
34
+ | Keyword normalization and auto repair | `mcp/src/features/douyin-trend/live-collector.js` |
35
+ | User-facing status selection | `mcp/src/tools/douyin-trend-run.js` |
36
+ | MCP structured output normalization | `mcp/src/server.js` |
37
+ | Douyin skill behavior rules | `skills/douyin-trend-intelligence/SKILL.md` |
38
+ | Live mode reference | `skills/douyin-trend-intelligence/references/live-mode.md` |
39
+ | Manual live acceptance checks | `docs/live-manual-acceptance-checklist.md` |
40
+ | Regression smoke tests | `scripts/smoke-package.js` |
41
+
42
+ ## Status Decision Rules
43
+
44
+ Use these states consistently:
45
+
46
+ | Condition | Status | Message Strategy |
47
+ | --- | --- | --- |
48
+ | No token configured | `needs_token` | Show payment/open link and token config path |
49
+ | Token invalid | `needs_valid_token` | Ask user to reconfigure token; show payment/open link only as support path |
50
+ | Clear balance/quota/payment issue | `needs_recharge` | Show recharge/payment link |
51
+ | Search input, empty keyword, overlong keyword, generic search 5xx | `needs_keyword_fix` | Give corrected keyword examples; no payment link |
52
+ | Token valid but no usable samples | `no_live_samples` | Suggest broader/narrower keywords; no payment link |
53
+ | Some samples collected but later billing blocked | `partial_needs_recharge` | Preserve partial report and show recharge link |
54
+ | Normal collection | `ok` | Return report body and evidence quality |
55
+
56
+ ## Keyword Repair Recipes
57
+
58
+ Claude Code should prefer short, natural platform search phrases:
59
+
60
+ ```text
61
+ 品类 + 价格/问题 + 人群/场景
62
+ ```
63
+
64
+ Examples:
65
+
66
+ ```text
67
+ 茶饮 18元 22元
68
+ 茶饮 性价比 新客
69
+ 奶茶 套餐 点单 避坑
70
+ 本地茶饮 人均 点单
71
+ 茶饮 探店 值不值
72
+ ```
73
+
74
+ Repair rules:
75
+
76
+ - Insert spacing around Chinese and numbers: `茶饮18元22元` -> `茶饮 18元 22元`.
77
+ - Keep keywords under 60 characters.
78
+ - If price-specific search fails, try intent-specific search: `茶饮 性价比 新客`.
79
+ - If intent search fails, try behavior-specific search: `奶茶 套餐 点单 避坑`.
80
+ - If all fail, return `needs_keyword_fix` and tell the user what was tried.
81
+
82
+ ## Implementation Checklist For Similar Bugs
83
+
84
+ When a new misclassification appears:
85
+
86
+ 1. Reproduce with a small command or mocked provider response.
87
+ 2. Add an error `kind` at provider boundary, for example `billing`, `auth`, `request`, `upstream_search`, `permission`.
88
+ 3. Make the CLI/tool status choose from `kind`, not raw text only.
89
+ 4. Normalize user input before API calls.
90
+ 5. Add one automatic repair if it is safe and low-cost.
91
+ 6. Add smoke coverage for both the bad path and the repaired path.
92
+ 7. Update the skill instructions so Claude Code knows what to do in conversation.
93
+ 8. Update manual acceptance docs with one command a trainer can run.
94
+ 9. Run package smoke and build.
95
+
96
+ ## Divergent Debugging Ideas
97
+
98
+ If the obvious fix does not work, branch into these checks:
99
+
100
+ - Credential source: is Claude Code running from the workspace that contains `.env.local`?
101
+ - MCP schema: did the tool return object warnings where the schema expects strings?
102
+ - CLI parsing: did `--keywords ""` become boolean `true` or an empty string?
103
+ - Encoding: did Chinese error text become mojibake or `????`, breaking regex matching?
104
+ - Upstream semantics: does provider return business error codes inside HTTP 200?
105
+ - Partial success: did videos collect but comments fail, requiring partial report instead of total failure?
106
+ - Report quality: did auto-repair succeed but report still say the original keyword?
107
+ - Skill routing: did the user ask for live collection, but a deep-dive skill kept control?
108
+ - Safety: does any output accidentally echo `VOC_TOKEN`, `Bearer`, or `r:...`?
109
+
110
+ ## Verification Commands
111
+
112
+ Run these after changing collector behavior:
113
+
114
+ ```powershell
115
+ node --check claude-code-voc-intelligence\mcp\src\providers\douyin-api.js
116
+ node --check claude-code-voc-intelligence\mcp\src\features\douyin-trend\live-collector.js
117
+ node --check claude-code-voc-intelligence\mcp\src\tools\douyin-trend-run.js
118
+ node --check claude-code-voc-intelligence\scripts\smoke-package.js
119
+ npm --prefix claude-code-voc-intelligence run mcp:smoke
120
+ npm --prefix claude-code-voc-intelligence run smoke:package
121
+ npm run claude-voc:build
122
+ ```
123
+
124
+ Live manual check, only when a real token is configured:
125
+
126
+ ```powershell
127
+ node .claude\plugins\voc-intelligence\mcp\src\tools\douyin-trend-run.js --collection-mode live --industry "茶饮" --business-type "茶饮连锁" --target-audience "新客" --keywords "茶饮18元22元" --keyword-limit 1 --videos-per-keyword 1 --max-comment-pages 0 --cache-assets false --output "outputs/live-acceptance-douyin-keyword"
128
+ ```
129
+
130
+ Expected:
131
+
132
+ - Compact keyword is normalized to `茶饮 18元 22元`.
133
+ - Search 5xx is auto-repaired or returned as `needs_keyword_fix`.
134
+ - Recharge URL appears only for token, valid-token, or real balance states.
135
+ - No token value appears in output.
136
+