Human Distill
--- name: human-distill version: 1.1.0 homepage: https://github.com/spikesubingrui-design/human-distill description: >- 蒸馏人物（博主/KOL/专家/创业者）的知识体系与观点：全网搜索（web_search/web_fetch）+ 抖音深度爬取（browser）+ 可选 LDR 深度研究。输出带 [确认]/[推断] 证据分级的人物画像。 触发词：蒸馏 / 人物画像 / 爬取博主 / human distill / KOL蒸馏 / 抖音蒸馏 metadata: openclaw: emoji: "🧠" --- # 人物蒸馏（Human Distill） 从全网公开信息 + 抖音一手文案出发，蒸馏某人在指定领域的知识体系、观点与内容风格。 ## 工作流总览 ``` - [ ] Phase 0：确认目标（人、主题、深度模式、平台线索） - [ ] Track A：全网搜索（始终执行）→ memory/[用户名]-全网素材.md - [ ] T
Human Distill — Distill Any Creator's Worldview in One Agent Run
> Turn scattered posts, Douyin captions, interviews, and course pages into a structured, evidence-graded persona dossier — with [confirmed] vs [inferred] labels so you know what to trust.
[](LICENSE)
[](https://clawhub.ai/skills/human-distill)
[](https://openclaw.ai)
[](https://cursor.com)
[](https://github.com/spikesubingrui-design/human-distill)
中文 · [什么是 Human Distill](#what-it-does) · [安装](#installation) · [用法](#usage) · [输出示例](#example-output)
---
What it does
You point an AI agent at a person (KOL, expert, founder, coach). It:
1. Track A — Web-wide search — web_search + web_fetch across Douyin, YouTube, Bilibili, WeChat articles, Zhihu, course pages
2. Track B — Douyin deep scrape (optional) — browser scroll + per-video captions + Douyin AI chapter summaries
3. Merge & distill — 7-section persona: beliefs, knowledge map, controversies, citations, style, actionable advice
4. Evidence grading — every claim tagged [确认] / [推断] / [冲突]
No API keys for Douyin. Browser + search tools only.
``mermaid`
flowchart LR
subgraph input [You provide]
P[Person + topic + mode]
end
subgraph trackA [Track A always]
WS[web_search]
WF[web_fetch]
end
subgraph trackB [Track B optional]
BR[browser Douyin]
end
subgraph out [Outputs]
M1[web-sources.md]
M2[video-transcripts.md]
M3[persona-distill.md]
end
P --> WS --> WF --> M3
P --> BR --> M2 --> M3
WF --> M1
---
Why stars matter here
| Problem | Human Distill |[冲突]
|--------|----------------|
| Watched 200 short videos, still no system | Keyword-filtered transcript archive |
| Search snippets contradict each other | section surfaces disagreements |[确认]
| Can't tell hype from evidence | vs [推断] on every bullet |
| Re-research same creator monthly | Reusable skill + markdown artifacts |
---
Depth modes
| Mode | When | Tracks |
|------|------|--------|
| quick | Fast overview, no full captions | A only |
| standard | Default — has Douyin handle | A + B |
| deep | "Comprehensive / report" | A + optional [LDR](https://github.com/eplt/local-deep-research-skill) + B |
---
Installation
ClawHub（推荐）
`bash`
clawhub login # 首次需 GitHub 登录
clawhub install human-distill
发布/更新见 [docs/LAUNCH.md](docs/LAUNCH.md)。
OpenClaw（手动）
`bash`
git clone https://github.com/spikesubingrui-design/human-distill.git \
~/.openclaw/workspace/skills/human-distill
Register triggers in your AGENTS.md (or use OpenClaw skill discovery):
`markdownskills/human-distill/SKILL.md
| "蒸馏" / "人物画像" / "human distill" / "抖音蒸馏" | |`
Cursor
`bash`
git clone https://github.com/spikesubingrui-design/human-distill.git \
~/.cursor/skills/human-distill
Then say: 「按 human-distill skill，蒸馏 [人名]，主题 [营养/投资/…]」
Optional: Local Deep Research (deep mode)
`bash
Install LDR skill separately
git clone https://github.com/eplt/local-deep-research-skill.git \
~/.openclaw/workspace/skills/local-deep-research
export LDR_BASE_URL=http://127.0.0.1:5000
---
Usage
Natural language — the agent loads SKILL.md automatically when triggers match.
`
蒸馏抖音 Mars12161209，主题营养，排除切磋和带货
`
human distill Naval Ravikant, topic wealth, quick mode
`
深度蒸馏 [人名]，主题商业，要调研报告级别
Outputs (in your workspace memory/)
| File | Content |[name]-全网素材.md
|------|---------|
| | Indexed web sources + excerpts |[name]-视频文案合集.md
| | Filtered Douyin captions (standard/deep) |[name]-[topic]观点蒸馏.md
| | Final 7-section persona dossier |
---
Example output
Distilled a 580K-follower sports-nutrition creator (nutrition topic, standard mode). Excerpt:
| Theme | Their stance | vs mainstream |
|-------|--------------|---------------|
| Post-workout protein window | Doesn't matter — total intake > timing | Must drink within 30 min |
| Carbs while cutting | No extreme low-carb required | Keto / low-carb default |
| Diet pattern | Mediterranean — whole grains, olive oil, nuts | Fad diets |
Full sample: [examples/sample-persona-excerpt.md](examples/sample-persona-excerpt.md)
---
Repository layout
`
human-distill/
├── SKILL.md # Agent instructions (main artifact)
├── README.md
├── LICENSE
├── references/
│ └── search-queries.md # Topic-specific web_search templates
└── examples/
└── sample-persona-excerpt.md
---
Requirements
| Tool | Purpose |web_search
|------|---------|
| | Discovery across platforms |web_fetch
| | Full text from articles (not Douyin video pages) |browser
| | Douyin profile scroll + per-video extraction (logged-in session) |
| LDR (optional) | Multi-cycle research in deep mode |
Douyin note: Anti-bot is aggressive. web_fetch cannot replace browser for video captions. See SKILL.md for the scroll-container selector that actually works.
---
Ethics & compliance
- Only public content. Respect platform ToS and robots rules.[推断]
- Do not scrape private or paid course material you don't have rights to.
- claims must stay labeled — never present inference as fact.
---
Related projects
- [OpenClaw](https://openclaw.ai) — agent runtime this skill targets
- [local-deep-research-skill](https://github.com/eplt/local-deep-research-skill) — optional deep mode backend
- [TrendRadar](https://github.com/jiajiaoy/TrendRadar) — trending products (this skill is trending people's ideas)
---
Contributing
PRs welcome: new references/search-queries.md domains, platform-specific scroll tricks, example dossiers (anonymized or public figures only).
1. Fork → branch → PRSKILL.md
2. Keep under ~500 lines; put long templates in references/`
---
License
[MIT](LICENSE) — use freely, attribution appreciated.
---
<p align="center">
If this saved you hours of manual note-taking,<br/>
<a href="https://github.com/spikesubingrui-design/human-distill"><strong>⭐ Star the repo</strong></a> — it helps others find it.
</p>
Natural language — the agent loads SKILL.md automatically when triggers match.
``
蒸馏抖音 Mars12161209，主题营养，排除切磋和带货
`
human distill Naval Ravikant, topic wealth, quick mode
`
深度蒸馏 [人名]，主题商业，要调研报告级别
| Tool | Purpose |
|------|---------|
| web_search | Discovery across platforms |
| web_fetch | Full text from articles (not Douyin video pages) |
| browser | Douyin profile scroll + per-video extraction (logged-in session) |
| LDR (optional) | Multi-cycle research in deep mode |
Douyin note: Anti-bot is aggressive. web_fetch cannot replace browser for video captions. See SKILL.md for the scroll-container selector that actually works.
---