VOC Growth Report

Turn exported social media comments — especially Xiaohongshu/小红书 CSV exports from 社媒助手 — into VOC insight, growth analysis, Feishu-ready operating structures...

Security Scan

The skill's requests and instructions match its stated purpose (CSV → VOC analysis → deliverable HTML/Feishu guidance); it is instruction-only, asks for no secrets, and has no install steps.

Purpose & Capability

Name/description align with the instructions: the skill focuses on ingesting social-comment CSVs (esp. 小红书/社媒助手), producing VOC/growth analysis, report prompts, Feishu/Bitable schemas, and delivery links. It does not request unrelated credentials, binaries, or install artifacts.

Instruction Scope

Runtime instructions stay within the analysis-and-delivery scope: ask for CSV path/columns, perform layered VOC analysis, produce prompts/schemas, generate/suggest HTML report and a preview link. The skill does not instruct reading unrelated system files or sending data to external endpoints. It does instruct agents to save files and start a local static preview — which is coherent with the delivery-first goal.

Install Mechanism

Instruction-only skill with no install spec and no code files. This minimizes disk/network install risk.

Credentials

No environment variables, credentials, or config paths are requested. Feishu/Bitable are mentioned only as schema and workflow guidance; the skill does not require Feishu API tokens to produce the schema or prompts.

Persistence & Privilege

Skill is not always-enabled and uses the default autonomous invocation behavior. It does not request permanent presence, system-wide changes, or modifications to other skills' configs.

Assessment

This skill is coherent and appears to do what it says: ask for a CSV, run structured VOC/growth analysis, and favor delivering a saved HTML file plus a local preview link. Before installing or running it, be aware that: 1) to produce the preview link an agent will need permission to write files and start a local static server — only run in a trusted agent environment; 2) if you later want real Feishu/Bitable imports, you'll need to provide Feishu credentials separately (the skill itself doesn't request them); 3) avoid supplying CSVs containing highly sensitive or personally identifiable data unless you trust the execution environment. If you want stricter guarantees, request an explicit no-network/local-only execution policy from the agent/harness before proceeding.

This skill converts exported social comment data into a repeatable growth-analysis workflow.

The core idea is simple:

1. ingest a CSV export,
2. analyze comments through a VOC + growth lens,
3. generate a boss-ready HTML report,
4. prefer delivering a preview link/path instead of dumping raw HTML.

Use this skill especially for 小红书 / 社媒助手 CSV exports, but it also works for similar social comment exports.

What this skill should produce

Depending on the user's ask, produce one or more of these:

• a cleaned analysis brief,
• a prompt pack for Trae / Cursor / Claude Code / Codex,
• a field schema for Feishu Bitable,
• a boss-ready HTML report prompt,
• a local preview link delivery workflow.

Default workflow

Step 1: Confirm the real deliverable

First identify which of these the user actually wants:

• analysis only: sentiment / needs / intent / opportunity
• report prompt: a prompt for another coding agent to generate the report
• report artifact: a real HTML file or preview link
• Feishu workflow: import/sync results into Feishu / Bitable
• skill/systemization: package the whole VOC workflow into a reusable system

If the user says things like:

• “不要给我代码，给我链接”
• “社媒助手抓完 csv 后怎么交给 Trae”
• “给我老板能看的报告” then optimize for delivery, not code verbosity.

Step 2: Understand the input data

Identify or ask for:

• CSV path or file
• likely columns: comment text, username, time, likes, replies, post title, link, platform
• source platform / export tool
• time range / sample size if relevant

If columns differ, infer the closest mapping instead of blocking on exact names.

This skill has already been validated against a real 社媒助手 / 小红书 comment export structure with fields like:

• 评论ID
• 评论内容
• 点赞量
• 评论时间
• IP地址
• 子评论数
• 笔记ID / 笔记链接
• 用户ID / 用户链接 / 用户名称
• 一级评论ID / 一级评论内容
• 引用的评论ID / 引用的评论内容 / 引用的用户名称

Step 3: Analyze comments in 4 layers

When doing actual VOC analysis, prefer this four-layer model:

1. Emotion

Classify into:

• 正向
• 中性
• 负向

Output:

• distribution
• positive highlights
• negative complaints

2. Intent

Classify into:

• 咨询价格
• 咨询功能
• 咨询购买
• 使用反馈
• 吐槽抱怨
• 夸赞认可
• 对比竞品
• 无效灌水
• 其他

Output:

• type distribution
• representative comments
• common questions

3. Commercial opportunity

Classify into:

• 高
• 中
• 低
• 无

Use these definitions:

• 高：明确咨询价格、购买方式、联系方式、合作、试用、下单
• 中：明确咨询功能、效果、适用人群、区别、使用方法
• 低：普通兴趣表达、轻度认可、一般互动
• 无：灌水、无关内容、纯表情

Output:

• opportunity distribution
• top high-opportunity comments
• conversion blockers

4. Need discovery

Split needs into:

• 已被满足的需求
• 未被满足的需求
• 潜在需求

Important: latent needs must be inferred from actual complaints, hesitation, comparisons, or repeated asks — never from pure imagination.

Output:

• need categories
• representative comments
• why each need is classified that way

Step 4: Upgrade analysis into growth decisions

Do not stop at “analysis”. Convert outputs into growth decisions:

• who to prioritize,
• what pain points to solve first,
• what value propositions to amplify,
• what content topics to create,
• what sales talking points to use,
• what operations team should reply to first.

When appropriate, use a Kotler-flavored framing:

• segmentation,
• need discovery,
• value proposition mapping,
• conversion opportunity,
• growth actions.

Default report structure

For boss/CEO-ready reports, prefer this structure:

1. 封面 / 数据概况
2. 用户情绪总览
3. 用户分群分析
4. 用户需求图谱
5. 商机与转化机会
6. 价值主张与增长建议
7. CEO Summary

Delivery-first rule

If the user wants a usable deliverable, do not stop at raw HTML code. Prefer to instruct the coding agent / ACP harness to:

1. generate the HTML,
2. save it to a file,
3. start a local static preview,
4. return a preview link and file path.

Use language like:

• “你的任务不是输出源码，而是完成交付”
• “最终返回访问链接、本地文件路径、报告标题、简短说明”

Output modes

Mode A: Prompt pack

When the user wants something to paste into Trae / Cursor / Claude Code / Codex, provide:

• one consolidated instruction block,
• explicit input/output contract,
• delivery requirement: link > raw code.

Mode B: Feishu workflow

When the user wants Feishu integration, provide:

• comment library field schema,
• suggested analysis fields,
• optional Bitable views,
• minimal workflow from CSV/comment sync to reporting.

Recommended 12-field base schema:

• 平台
• 帖子标题
• 帖子链接
• 评论内容
• 评论用户
• 评论时间
• 情绪倾向
• 意图类型
• 商机等级
• 是否需要回复
• 跟进状态
• 备注

Mode C: Executive summary

For direct advice in chat, use this order:

1. conclusion,
2. why,
3. next action.

Keep it concise and business-oriented.

Example trigger cases

• “帮我把社媒助手抓下来的评论 csv 做成老板能看的报告”
• “不要给我 html 代码，我要最终链接”
• “帮我做小红书 voc 分析”
• “把评论做成需求洞察 + 商机分析”
• “给 Trae 一段完整指令，从 csv 到 html 报告链接”
• “封装一个 VOC 分析 skill”

Anti-patterns

Avoid these mistakes:

• stopping at sentiment only,
• giving a word cloud as the main output,
• dumping raw HTML when the user asked for delivery,
• inventing latent needs with no textual basis,
• overcomplicating the workflow before the CSV/report path is usable.

Success standard

A strong result should make it easy for the user to go from: comment export → user insight → growth decisions → report delivery with minimal repeated prompting.

A stronger result should also be capable of producing a real executive-facing HTML demo report with sections such as:

• 封面 / 数据概况
• 用户情绪总览
• 用户分群分析
• 用户需求图谱
• 商机与转化机会
• 价值主张与增长建议
• CEO Summary