使用方法
- 复制安装命令，在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物，并对照来源检查输出。
npx skills add https://github.com/ootto-ai/claude-content-skills --skill social-listening开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况，不等于独立质量或安全测试。
按源仓库内容呈现，保留标题、案例、代码、表格、链接以及原文引用的演示图片。
Social Listening
Turn evidence the operator already has into a usable demand map without pretending that a small sample represents the whole market.
1. Set the evidence boundary
Ask for the source material, where it came from, its date range, and the decision it should inform. Keep comments, DMs, reviews, support tickets, and interview notes labelled by source. If the user has only a claim about what people say, ask for the underlying text before analysing it.
2. Extract the signal without flattening it
For each item, capture the speaker's own wording, the situation they describe, the job they are trying to do, the obstacle, and any stated outcome. Keep a short supporting excerpt beside every finding so a reader can check it. Separate direct customer language from the operator's interpretation.
3. Cluster by decision-relevant theme
Group repeated evidence into questions, desired outcomes, objections, alternatives, proof requests, and vocabulary. Count only the supplied evidence. Mark a theme as isolated when it appears once, recurring when it appears across independent items, and unresolved when the context is too thin to tell.
4. Produce a demand map
Return a compact table with: theme, evidence count, representative wording, likely stage of the journey, confidence, and the next action. Next actions can be a research question, an FAQ update, an input for audience-personas, or a test for positioning-audit. Call out contradictions instead of averaging them away.
Hard rules
- Do not infer age, location, income, intent, or sentiment beyond what the evidence says.
- Never present supplied comments as a representative market sample without a sampling basis.
- Preserve anonymisation: do not expose names, handles, private messages, or customer details unnecessarily.
- Do not manufacture quotes or improve a speaker's wording inside quotation marks.
- Treat volume as a clue, not proof of importance; a repeated complaint from one thread is not independent corroboration.
Failure modes
| Failure | Do this instead |
|---|---|
| A loud minority becomes "the audience" | State the sample and test the theme with a wider, independently collected set. |
| Themes mix people at different stages | Split the cluster by situation or journey stage before recommending an action. |
| The result is a pile of quotes | Convert each cluster into a decision, uncertainty, and next move. |
| The user wants public-platform research | Use agent-reach to gather allowed public sources, then return here to analyse them. |
Where it sits
Use agent-reach before this skill when evidence must be collected. Use audience-personas after this skill to turn recurring patterns into grounded audience groups. Use positioning-audit after that when those groups need a message and social-proof-mining when the evidence should become approved proof.
