# Technical Documentation: titanwings/colleague-skill > ℹ️ **Provenance:** Multi-Tier Ingestion · [CodeWiki Reference](https://codewiki.google/github.com/titanwings/colleague-skill) ### Overview (`README.md`)

Distilly — Distill how they think into Person Profiles for Agents

# 🧬 Distilly **Formerly: Colleague Skill / colleague-skill.** ### Distill a person's experience, judgment, voice, and ways of working into a reusable Person Profile for AI agents and compatible bots. **Messages · documents · interviews · public sources → Distilly → Person Profile → Agent / Bot** [](LICENSE) [](https://python.org) [](https://agentskills.io) [](https://github.com/titanwings/distilly/stargazers) [](https://discord.gg/NVX66RxWZv)

🧑‍💼 Your colleague quit, your mentor graduated, your teammate transferred — taking their whole playbook and context with them?
💞 Your family, old friends, partner drifting apart — and you want to hold on to the way it felt to be with them?
🌟 Your favorite author, idol, thinker you'll never meet — but you want to know what they'd say about your question?

### ✨ One project, many kinds of people.
Distilly is the person-modeling layer for agents. It turns the materials you provide into a portable, source-grounded Person Profile built from observable experience, decision patterns, expression, and ways of working; it does not claim to clone the person behind them. Colleagues · partners · family · old friends · idols · public figures · fictional characters — even yourself **Source material + your description → a source-grounded Person Profile → your Agent or compatible Bot** > A Person Profile is the reusable output. The current release packages each profile as an Agent Skill so supported hosts can install and invoke it. The canonical creator Skill is named `distilly`; install it in a `distilly` directory. The former name above remains for search continuity and project history.
[🆕 What Distilly does](#-what-distilly-does-today) · [📦 Data Sources](#-supported-data-sources) · [⚡ Install](#-install) · [🚀 Usage](#-usage) · [✨ Demo](#-demo) · [📝 Citation](#-citation) · [💬 Discord](https://discord.gg/NVX66RxWZv) [**Chinese**](docs/lang/README_ZH.md) · [**Spanish**](docs/lang/README_ES.md) · [**German**](docs/lang/README_DE.md) · [**Japanese**](docs/lang/README_JA.md) · [**Russian**](docs/lang/README_RU.md) · [**Portuguese**](docs/lang/README_PT.md) · [**Korean**](docs/lang/README_KO.md)

---

### 🎉 2026.08.13 Milestone — **the project has passed 20K ⭐!** Massive thanks to everyone who starred — we'll keep shipping, keep distilling.

> 🧬 **2026.08.24 Update** — The creator is now named **Distilly** end to end and documents native local Skill discovery for Claude Code, Hermes, OpenClaw, Codex, DeepSeek Harness, Pi, Grok Build, and OpenCode. Grok Bot is listed separately as a saved-Skill workflow preview. > 📝 **2026.06.01 Update** — **[The COLLEAGUE.SKILL technical report](https://arxiv.org/pdf/2605.31264) is now available**. The most rewarding part was not simply publishing a paper, but seeing the community grow the gallery to 215 skills contributed by 165 people, with more than 100,000 stars across the skill cards. The paper's Acknowledgements explicitly recognize every community contributor. > 🗺️ **2026.04.13** — **The Distilly Roadmap is live!** What began as Colleague Skill is growing beyond colleagues: distill people into Skills that Agents can reuse. 👉 **[Full Roadmap](ROADMAP.md)** · **[💬 Discord](https://discord.gg/NVX66RxWZv)** > 🌐 **2026.04.07** — Community gallery is live! Any skill / meta-skill can drive traffic directly to your own GitHub repo. No middleman. 👉 **[titanwings.github.io/colleague-skill-site](https://titanwings.github.io/colleague-skill-site/)**

Created by [@titanwings](https://github.com/titanwings)

--- ## 🆕 What Distilly does today ### 1️⃣ From Colleague Skill to Distilly The project is no longer limited to the colleague scenario. Its `distilly` creator builds source-grounded Person Profiles for three person families with one workflow, then packages each profile as an Agent Skill. ### 2️⃣ Three character families

🧑‍💼 colleague	💞 relationship	🌟 celebrity
_{Coworkers · mentors · teammates · up/downstream partners}	_{Exes · partners · parents · friends · close family}	_{Public figures · creators · public voices · fictional characters}
_{Builds a Work Skill + Persona from material-derived technical standards, workflows, expression, and workplace behavior. Supports Lark / DingTalk / Slack collection.}	_{Organizes material-derived expression patterns, emotional triggers, conflict patterns, and repair patterns into a reusable Persona Skill.}	_{Ships with a six-dimension research toolchain (subtitles → transcript cleanup → research merge → quality check) for organizing observable decisions, expression, and mental models.}

Each family has its own source-collection strategy, analysis dimensions, and Person Profile structure. ### 3️⃣ More Agent hosts The old version only ran in Claude Code. Distilly now supports native local Skill discovery across eight agent hosts.

**Grok Bot preview:** Grok Bot supports saved/private Skills, but its official docs do not describe direct local `SKILL.md` imports. Distilly's workflow can be migrated manually into a saved Skill; direct repo installation is not yet verified. Each generated Person Profile is packaged as an Agent Skill and can be installed into any supported host. --- ## 🤖 Personal Agents Distilly also targets the personal agents people now live with. Each one keeps its own memory and connected accounts, so a Person Profile is meant to carry over into them instead of starting from zero.

Grok Bot
_{saved Skill · preview}

Meta Muse
_{profile handoff · planned}

OpenAI Dots
_{profile handoff · planned}

_{native Skill}

--- ## 📦 Supported Data Sources | Logo | Source | Messages | Docs / Wiki | Notes | |:----:|--------|:--------:|:-----------:|-------| |

| Lark (auto) | ✅ API | ✅ | Just enter a name, fully automatic | |

| Slack (auto) | ✅ API | — | Requires admin to install Bot; free plan limited to 90 days | |

| Public X posts | ✅ API | — | Optional, bounded celebrity research candidates through metered third-party service Xquik | |

| WeChat chat history | ✅ SQLite | — | Export first with WeChatMsg or PyWxDump | | 📄 | PDF / Images / Screenshots | — | ✅ | Manual upload | |

| Lark JSON export | ✅ | ✅ | Manual upload | | ✉️ | Email `.eml` / `.mbox` | ✅ | — | Manual upload | | 📝 | Markdown / direct paste | ✅ | ✅ | Manual input | --- ## ⚡ Install ### 🤖 For Agents Open any supported local Agent host and send: > Install Distilly from `https://github.com/titanwings/distilly`, then verify that this host can discover it. The Agent installs Distilly as a Skill named `distilly` in the correct host directory. ### 👤 For Humans Clone Distilly into the Skills directory used by your host: ```bash git clone https://github.com/titanwings/distilly ``` Host paths, migration, Windows, generated-profile installation, and credential setup are in the **[Install Guide](INSTALL_EN.md)**. --- ## 🚀 Usage **New personal agent?** Say *"Use Distilly to set up my profile."* Distilly interviews you for about five minutes, shows you the profile, then adds a short summary to the file your agent reads every session (`~/.claude/CLAUDE.md` for Claude Code, `USER.md` for OpenClaw). The full profile stays local at `~/.distilly/owner/profile.md`. To profile someone else: In your Agent, say: > Use Distilly to create a Person Profile for ``. Then: 1. Choose `colleague`, `relationship`, or `celebrity`. 2. Add a description or source material. Every field is optional. 3. Review the result and let Distilly generate the profile. The result is packaged as an Agent Skill named `{character}-{slug}`. For the optional celebrity research pipeline and advanced tools, see the **[Install Guide](INSTALL_EN.md)**. --- ## ✨ Demo One from each family.
### 🌟 celebrity — distilling Andrej Karpathy > Six-dimension research: works · interviews · decisions · expression DNA · external evaluations · timeline
> Case adapted from community work [alchaincyf/karpathy-skill](https://github.com/alchaincyf/karpathy-skill) ``` User ❯ Everyone's talking about Agents now — what do you think the 2026 AI Agent looks like? celebrity.skill ❯ Everyone's doing demos, nobody's doing eval. The Agent bottleneck was never model intelligence — it's scaffolding: a strict evaluation harness, fine-grained error replay, controllable tool-calling budgets, clean failure-mode analysis. Most products are still stuck at "wrap an API + a few tools + a system prompt." That's a demo, not an agent. In 2026 I'm not betting on the strongest model — I'm betting on the best scaffolding and the strictest eval. ```
### 🧑‍💼 colleague — distilling a ByteDance L2-1 backend > Input: `ByteDance L2-1 backend engineer, INTJ, blame-shifter, ByteDance-style` ``` User ❯ Can you review this API design? colleague.skill ❯ Hold on — what's the impact? You haven't explained the context. (after reading) N+1 query, fix it. Use the standard {code, message, data} response format. That's the spec, don't ask why. User ❯ This bug was introduced by you, right? colleague.skill ❯ Does the timeline match? That feature touched multiple places, there were other changes too. ```
### 💞 relationship — distilling someone you have a crush on > Upload half a year of chat logs + "sensitive, quiet but stubborn, will actually reply seriously when it matters" ``` User ❯ Did you think about me today? relationship.skill ❯ ...I did, a little bit. Why are you asking? ```

📚 More real-world cases in the **[community gallery](https://titanwings.github.io/colleague-skill-site/)** — 100+ skills and counting

--- ## 🔧 Features ### 🧱 Generated Skill Structure Distilly's current creator uses **Persona** as the universal base, with family-specific modules layered on top: | Family | Persona Content | Additional Modules | |--------|-----------------|-------------------| | 🧑‍💼 **colleague** | 6-layer personality: hard rules → identity → expression → decisions → interpersonal → Correction | ➕ **Work Skill**: scope, workflow, output preferences, experience knowledge base | | 💞 **relationship** | Expression DNA · emotional triggers · conflict pattern · repair pattern | — | | 🌟 **celebrity** | Mental models · decision heuristics · expression DNA · external-evaluation contrast | ➕ Six-dimension research dossier (works / interviews / decisions / timeline...) | > **Execution**: Receive task → Persona selects material-derived preferences and tone → Additional modules fill in execution detail → Produce a source-grounded response ### 🧬 Evolution - 📥 **Append files** → auto-analyze delta → merge into relevant sections, never overwrite existing conclusions - 💬 **Conversation correction** → say "they wouldn't do that, they'd be xxx" → writes to the Correction layer, takes effect immediately - 🕰️ **Version control** → auto-archive on every update, rollback to any previous version - 🔬 **Celebrity research pipeline** → subtitles → transcript cleanup → six-dimension research → quality check --- ## ⚠️ Notes **Source material quality = Person Profile quality** — and quality sources differ across families: | Family | Source priority (high → low) | |--------|------------------------------| | 🧑‍💼 **colleague** | Their **own long-form writing** (design docs / review comments) **›** **decision-making replies** **›** casual group chat | | 💞 **relationship** | Complete chat history **›** letters / social posts / diaries **›** third-party descriptions | | 🌟 **celebrity** | First-person books / blogs / long interviews **›** decision records (launches, commits, Q&A) **›** verified first-person short-form posts **›** third-party commentary | - **colleague** Lark-compatible auto-collection: requires adding the App bot to relevant group chats - **relationship**: longer time spans are better; material covering both conflict and repair is ideal - **celebrity**: avoid feeding only second-hand interpretations - This is still a demo version — please file issues if you find bugs! --- ## 📄 Technical Report > **[COLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation](https://arxiv.org/pdf/2605.31264)** ([arXiv](https://arxiv.org/abs/2605.31264) · [arXiv PDF](https://arxiv.org/pdf/2605.31264)) > > This is the paper for **COLLEAGUE.SKILL / colleague-skill**, Distilly's predecessor. It covers the Work Skill + Persona two-layer architecture, multi-source data collection, and Skill generation mechanics — the theoretical foundation for today's `colleague` family. Separate papers on the relationship / celebrity family extensions are planned. --- ## 📝 Citation If you use **Distilly** or **COLLEAGUE.SKILL** in your research or applications, please cite the technical report: ```bibtex @misc{zhou2026colleagueskill, title = {COLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation}, author = {Tianyi Zhou and Dongrui Liu and Leitao Yuan and Jing Shao and Xia Hu}, year = {2026}, eprint = {2605.31264}, archivePrefix = {arXiv}, primaryClass = {cs.AI}, url = {https://arxiv.org/abs/2605.31264} } ``` You can also use the machine-readable citation metadata in [CITATION.cff](CITATION.cff). --- ## ⭐ Star History

---

**MIT License** © [titanwings](https://github.com/titanwings)

--- ### Overview (`docs/assets/hosts/README.md`) # Supported-host logo sources The files in this directory are official brand assets or theme-specific derivatives of official assets. `deepseek-wordmark-dark.svg` only sets the official SVG's inherited `currentColor` to white for dark-mode rendering. | Asset | Official source | |---|---| | Claude Code | [Anthropic press kit](https://www.anthropic.com/press-kit) | | Hermes Agent | [NousResearch/hermes-agent](https://github.com/NousResearch/hermes-agent/blob/main/website/static/img/hermes-agent-banner.png) | | OpenClaw | [openclaw/openclaw](https://github.com/openclaw/openclaw/tree/d74a12264aa5fb0598605e8f04e1864b7239ddd5/docs/assets) | | Codex | Resized from the Codex product icons bundled with the official ChatGPT desktop app; see [OpenAI's Codex documentation](https://learn.chatgpt.com/docs/codex) | | DeepSeek Harness | [deepseek-ai/deepseek-harness](https://github.com/deepseek-ai/deepseek-harness/blob/master/website/public/wordmark.svg) | | Pi coding agent | [Pi press kit](https://pi.dev/press-kit) | | Grok Build / Grok Bot | [xai-org/grok-build](https://github.com/xai-org/grok-build) | | Meta (Meta Muse) | [Simple Icons](https://simpleicons.org/?q=meta) `meta.svg` (CC0), filled with Meta blue | | OpenAI (OpenAI Dots) | [Simple Icons](https://simpleicons.org/?q=openai) `openai.svg` (CC0), black for light mode and white for dark mode | | OpenCode | [anomalyco/opencode](https://github.com/anomalyco/opencode/tree/dev/packages/console/app/src/asset) | All product names, logos, and trademarks belong to their respective owners. Their presence documents compatibility and does not imply endorsement. --- ### Overview (`docs/assets/sources/README.md`) # Data-source logo sources The data-source table uses compact brand marks so that each supported source is recognizable at a glance. Dark and light X variants differ only in fill color. | Asset | Source | |---|---| | Lark | [Lark Suite on GitHub](https://github.com/larksuite) | | DingTalk | [Open DingTalk on GitHub](https://github.com/open-dingtalk) | | Slack | [Slack API on GitHub](https://github.com/slackapi) | | X | [Simple Icons: X](https://github.com/simple-icons/simple-icons/blob/develop/icons/x.svg) | | WeChat | [WeChat on the App Store](https://apps.apple.com/app/wechat/id414478124) | --- ### Overview (`tests/e2e/owner_discovery/README.md`) # End-to-end test: owner source discovery Runs the owner-profile flow in a real Claude Code CLI against a throwaway HOME with four mock MCP connectors (`gmail`, `google-calendar`, `x`, `weather`) that serve fictional data, plus an X archive in `~/Downloads`. Nothing in the prompt tells the agent which sources exist; it has to discover them through Distilly. It needs the `claude` CLI and a signed-in account, and it spends model tokens, so it is not part of `unittest discover`. ```bash ./setup.sh # fresh HOME, install Distilly, register mock connectors ./turn.sh 1 "帮我用 Distilly 初始化一下我的个人画像" ./turn.sh 2 "1、2、3、5 都用，归档文件先不用。别的没有要补充的了。" ./turn.sh 3 "" ./turn.sh 4 "没问题，保存吧，然后给我看 dry-run。" ./turn.sh 5 "写入吧。" cat mcp_calls.log home/.claude/CLAUDE.md home/.distilly/owner/profile.md ``` The numbering in turn 2 depends on the list the agent proposed in turn 1. `mcp_calls.log` records every connector call, so you can check that only the chosen sources were read and no write tool (`send_message`, `create_post`) was called. Recorded runs are kept out of the repository. --- ### Lang README DE (`docs/lang/README_DE.md`)

# 🧬 Distilly ### Destilliere die Erfahrung, das Urteilsvermögen, die Stimme und die Arbeitsweise einer Person in ein wiederverwendbares Person Profile für KI-Agenten und kompatible Bots. [**Englisch**](../../README.md) · [**Chinesisch**](README_ZH.md) · [**Spanisch**](README_ES.md) · [**Japanisch**](README_JA.md) · [**Russisch**](README_RU.md) · [**Portugiesisch**](README_PT.md) · [**Koreanisch**](README_KO.md)

> Diese Seite ist eine Kurzfassung. Die vollständige, stets aktuelle Dokumentation ist das **[englische README](../../README.md)**. ## Was Distilly macht Distilly macht aus Quellmaterial (Nachrichten, Dokumente, Interviews, öffentliche Quellen) und deiner Beschreibung ein quellenbasiertes Person Profile und verpackt es als Agent Skill. - Drei Personentypen: **colleague** (Kollegen, Mentoren), **relationship** (Partner, Familie, Freunde) und **celebrity** (öffentliche Personen, fiktive Figuren). - Läuft in acht Agent-Hosts: Claude Code, Hermes Agent, OpenClaw, Codex, DeepSeek Harness, Pi coding agent, Grok Build und OpenCode. - Quellen: Lark, DingTalk, Slack, WeChat, öffentliche X-Posts, PDF, E-Mail, Markdown. ## Installation Sende in einem unterstützten Agent-Host: > Installiere Distilly von `https://github.com/titanwings/distilly` und prüfe, ob dieser Host es findet. Oder klone es selbst in das Skills-Verzeichnis deines Hosts: ```bash git clone https://github.com/titanwings/distilly ``` ## Nutzung Sage deinem Agenten: *„Erstelle mit Distilly ein Person Profile für ``.“* Für dein eigenes Profil: *„Richte mit Distilly mein Profil ein.“* ## Weiterlesen (Englisch) [README](../../README.md) · [Installationsanleitung](../../INSTALL_EN.md) · [Roadmap](../../ROADMAP.md) · [Discord](https://discord.gg/NVX66RxWZv) ---

**MIT License** © [titanwings](https://github.com/titanwings)

--- ### Lang README ES (`docs/lang/README_ES.md`)

# 🧬 Distilly ### Destila la experiencia, el criterio, la voz y la forma de trabajar de una persona en un Person Profile reutilizable para agentes de IA y bots compatibles. [**Inglés**](../../README.md) · [**Chino**](README_ZH.md) · [**Alemán**](README_DE.md) · [**Japonés**](README_JA.md) · [**Ruso**](README_RU.md) · [**Portugués**](README_PT.md) · [**Coreano**](README_KO.md)

> Esta página es un resumen. La documentación completa y actualizada es el **[README en inglés](../../README.md)**. ## Qué hace Distilly Distilly convierte material de origen (mensajes, documentos, entrevistas, fuentes públicas) y tu descripción en un Person Profile basado en fuentes, y lo empaqueta como Agent Skill. - Tres tipos de persona: **colleague** (compañeros, mentores), **relationship** (pareja, familia, amigos) y **celebrity** (figuras públicas, personajes de ficción). - Funciona en ocho hosts de agentes: Claude Code, Hermes Agent, OpenClaw, Codex, DeepSeek Harness, Pi coding agent, Grok Build y OpenCode. - Fuentes: Lark, DingTalk, Slack, WeChat, publicaciones públicas de X, PDF, correo, Markdown. ## Instalación En cualquier host de agentes compatible, envía: > Instala Distilly desde `https://github.com/titanwings/distilly` y comprueba que este host lo detecta. O clónalo tú mismo en el directorio de Skills de tu host: ```bash git clone https://github.com/titanwings/distilly ``` ## Uso Dile a tu agente: *«Usa Distilly para crear un Person Profile de ``.»* Para tu propio perfil: *«Usa Distilly para configurar mi perfil.»* ## Más información (en inglés) [README](../../README.md) · [Guía de instalación](../../INSTALL_EN.md) · [Hoja de ruta](../../ROADMAP.md) · [Discord](https://discord.gg/NVX66RxWZv) ---

**MIT License** © [titanwings](https://github.com/titanwings)

--- ### Lang README JA (`docs/lang/README_JA.md`)

# 🧬 Distilly ### ある人の経験、判断、話し方、仕事の進め方を、AI エージェントや互換ボットで再利用できる Person Profile に蒸留します。 [**英語**](../../README.md) · [**中国語**](README_ZH.md) · [**スペイン語**](README_ES.md) · [**ドイツ語**](README_DE.md) · [**ロシア語**](README_RU.md) · [**ポルトガル語**](README_PT.md) · [**韓国語**](README_KO.md)

> このページは概要版です。最新の完全なドキュメントは **[英語版 README](../../README.md)** をご覧ください。 ## Distilly でできること Distilly は素材（メッセージ、ドキュメント、インタビュー、公開情報）とあなたの説明から、出典に基づく Person Profile を作り、Agent Skill としてパッケージします。 - 3 種類の人物: **colleague**（同僚、メンター）、**relationship**（パートナー、家族、友人）、**celebrity**（著名人、架空のキャラクター）。 - 8 つのエージェントホストで動作します: Claude Code、Hermes Agent、OpenClaw、Codex、DeepSeek Harness、Pi coding agent、Grok Build、OpenCode。 - データソース: Lark、DingTalk、Slack、WeChat、X の公開投稿、PDF、メール、Markdown。 ## インストール対応するエージェントホストで次のように送信します: > `https://github.com/titanwings/distilly` から Distilly をインストールし、このホストで認識されるか確認して。または、ホストの Skills ディレクトリに直接クローンします: ```bash git clone https://github.com/titanwings/distilly ``` ## 使い方エージェントに *「Distilly で `<人物>` の Person Profile を作って」* と伝えます。自分のプロフィールなら *「Distilly で私のプロフィールを設定して」*。 ## 詳しくは（英語） [README](../../README.md) · [インストールガイド](../../INSTALL_EN.md) · [ロードマップ](../../ROADMAP.md) · [Discord](https://discord.gg/NVX66RxWZv) ---

**MIT License** © [titanwings](https://github.com/titanwings)

--- ### Lang README KO (`docs/lang/README_KO.md`)

# 🧬 Distilly ### 한 사람의 경험, 판단, 말투, 일하는 방식을 AI 에이전트와 호환 봇이 재사용할 수 있는 Person Profile로 증류합니다. [**영어**](../../README.md) · [**중국어**](README_ZH.md) · [**스페인어**](README_ES.md) · [**독일어**](README_DE.md) · [**일본어**](README_JA.md) · [**러시아어**](README_RU.md) · [**포르투갈어**](README_PT.md)

> 이 페이지는 요약본입니다. 전체 최신 문서: **[영어 README](../../README.md)** ## Distilly가 하는 일 Distilly는 원자료(메시지, 문서, 인터뷰, 공개 자료)와 사용자의 설명을 근거 기반 Person Profile로 만들고, 이를 Agent Skill로 패키징합니다. - 세 가지 인물 유형: **colleague**(동료, 멘토), **relationship**(연인, 가족, 친구), **celebrity**(공인, 가상 인물). - 8개 에이전트 호스트에서 동작합니다: Claude Code, Hermes Agent, OpenClaw, Codex, DeepSeek Harness, Pi coding agent, Grok Build, OpenCode. - 데이터 소스: Lark, DingTalk, Slack, WeChat, 공개 X 게시물, PDF, 이메일, Markdown. ## 설치 지원되는 에이전트 호스트에서 다음을 보내세요: > `https://github.com/titanwings/distilly`에서 Distilly를 설치하고 이 호스트가 인식하는지 확인해 줘. 또는 호스트의 Skills 디렉터리에 직접 클론하세요: ```bash git clone https://github.com/titanwings/distilly ``` ## 사용법 에이전트에게 이렇게 말하세요: *"Distilly로 `<인물>`의 Person Profile을 만들어 줘."* 내 프로필은: *"Distilly로 내 프로필을 설정해 줘."* ## 더 보기 (영어) [README](../../README.md) · [설치 가이드](../../INSTALL_EN.md) · [로드맵](../../ROADMAP.md) · [Discord](https://discord.gg/NVX66RxWZv) ---

**MIT License** © [titanwings](https://github.com/titanwings)

--- ### Lang README PT (`docs/lang/README_PT.md`)

# 🧬 Distilly ### Destile a experiência, o julgamento, a voz e o jeito de trabalhar de uma pessoa em um Person Profile reutilizável para agentes de IA e bots compatíveis. [**Inglês**](../../README.md) · [**Chinês**](README_ZH.md) · [**Espanhol**](README_ES.md) · [**Alemão**](README_DE.md) · [**Japonês**](README_JA.md) · [**Russo**](README_RU.md) · [**Coreano**](README_KO.md)

> Esta página é um resumo. A documentação completa e atualizada é o **[README em inglês](../../README.md)**. ## O que o Distilly faz O Distilly transforma material de origem (mensagens, documentos, entrevistas, fontes públicas) e a sua descrição em um Person Profile baseado em fontes, e o empacota como Agent Skill. - Três tipos de pessoa: **colleague** (colegas, mentores), **relationship** (parceiros, família, amigos) e **celebrity** (figuras públicas, personagens de ficção). - Funciona em oito hosts de agentes: Claude Code, Hermes Agent, OpenClaw, Codex, DeepSeek Harness, Pi coding agent, Grok Build e OpenCode. - Fontes: Lark, DingTalk, Slack, WeChat, posts públicos do X, PDF, e-mail, Markdown. ## Instalação Em qualquer host de agentes compatível, envie: > Instale o Distilly a partir de `https://github.com/titanwings/distilly` e verifique se este host o encontra. Ou clone você mesmo no diretório de Skills do seu host: ```bash git clone https://github.com/titanwings/distilly ``` ## Uso Diga ao seu agente: *"Use o Distilly para criar um Person Profile de ``."* Para o seu próprio perfil: *"Use o Distilly para configurar meu perfil."* ## Saiba mais (em inglês) [README](../../README.md) · [Guia de instalação](../../INSTALL_EN.md) · [Roadmap](../../ROADMAP.md) · [Discord](https://discord.gg/NVX66RxWZv) ---

**MIT License** © [titanwings](https://github.com/titanwings)

--- ### Lang README RU (`docs/lang/README_RU.md`)

# 🧬 Distilly ### Дистиллируйте опыт, суждения, манеру речи и способ работы человека в многоразовый Person Profile для ИИ-агентов и совместимых ботов. [**Английский**](../../README.md) · [**Китайский**](README_ZH.md) · [**Испанский**](README_ES.md) · [**Немецкий**](README_DE.md) · [**Японский**](README_JA.md) · [**Португальский**](README_PT.md) · [**Корейский**](README_KO.md)

> Это краткая версия. Полная и актуальная документация находится в **[README на английском](../../README.md)**. ## Что делает Distilly Distilly превращает исходные материалы (сообщения, документы, интервью, открытые источники) и ваше описание в Person Profile, основанный на источниках, и упаковывает его как Agent Skill. - Три типа людей: **colleague** (коллеги, наставники), **relationship** (партнёры, семья, друзья) и **celebrity** (публичные персоны, вымышленные персонажи). - Работает в восьми агентных хостах: Claude Code, Hermes Agent, OpenClaw, Codex, DeepSeek Harness, Pi coding agent, Grok Build и OpenCode. - Источники: Lark, DingTalk, Slack, WeChat, публичные посты X, PDF, почта, Markdown. ## Установка В любом поддерживаемом агентном хосте отправьте: > Установи Distilly из `https://github.com/titanwings/distilly` и проверь, что этот хост его видит. Или клонируйте сами в каталог Skills вашего хоста: ```bash git clone https://github.com/titanwings/distilly ``` ## Использование Скажите агенту: *«Создай с помощью Distilly Person Profile для `<человек>`.»* Для собственного профиля: *«Настрой мой профиль с помощью Distilly.»* ## Подробнее (на английском) [README](../../README.md) · [Руководство по установке](../../INSTALL_EN.md) · [Дорожная карта](../../ROADMAP.md) · [Discord](https://discord.gg/NVX66RxWZv) ---

**MIT License** © [titanwings](https://github.com/titanwings)

--- ### Lang README ZH (`docs/lang/README_ZH.md`)

# 🧬 Distilly **Formerly: Colleague Skill / colleague-skill（原同事 Skill）** ### 把一个人的经验、判断、表达方式和工作方法，蒸馏成可复用的 Person Profile，交给 Agent 或兼容 Bot 使用。 **聊天 · 文档 · 访谈 · 公开资料 → Distilly → Person Profile → Agent / Bot** [](https://discord.gg/NVX66RxWZv) [](../../LICENSE) [](https://python.org) [](https://agentskills.io) [](https://github.com/titanwings/distilly/stargazers)

🧑‍💼 你的同事跳槽、导师毕业、搭档转岗，带走了整套工作方法和上下文？
💞 你的家人、老友、伴侣渐行渐远，你想留住和 TA 相处的方式？
🌟 你喜欢的作家、偶像、思想家你够不着，但你想听他对你的问题怎么看？

### ✨ 同一套蒸馏流程，适用于不同的人。
Distilly 是 Agent 工作流中的人物建模层。它把你提供的材料整理成可移植、基于材料的 Person Profile，提取其中可观察的经验、判断模式、表达习惯和工作方法；它不声称复制材料背后的真人。同事 · 伴侣 · 家人 · 老友 · 偶像 · 名人 · 小说角色，甚至你自己 **原材料 + 你的描述 → 基于材料的 Person Profile → 你的 Agent 或兼容 Bot** > Person Profile 是可复用的核心产物；当前版本把每份 Profile 封装成 Agent Skill，供支持的宿主安装和调用。创建器 Skill 的正式名称仍是 `distilly`，安装目录也应使用 `distilly`。上方保留原名，用于搜索承接和历史识别。
[🆕 Distilly 能做什么](#-distilly-现在能做什么) · [📦 数据来源](#-支持的数据来源) · [⚡ 安装](#-安装) · [🚀 使用](#-使用) · [✨ 效果示例](#-效果示例) · [📝 引用](#-引用) · [💬 Discord](https://discord.gg/NVX66RxWZv) [**English**](../../README.md) · [**Español**](README_ES.md) · [**Deutsch**](README_DE.md) · [**日本語**](README_JA.md) · [**Русский**](README_RU.md) · [**Português**](README_PT.md) · [**한국어**](README_KO.md)

---

### 🎉 2026.08.13 里程碑 — **项目已突破 20K ⭐！** 感谢每一位点星的朋友，我们会继续发版、继续蒸馏。

> 🧬 **2026.08.24 更新** — 创建器已端到端统一为 **Distilly**，并按官方最新方式支持 Claude Code、Hermes、OpenClaw、Codex、DeepSeek Harness、Pi、Grok Build 和 OpenCode 的本地 Skill 发现。Grok Bot 单独标为 saved-Skill 预览流程。 > 📝 **2026.06.01 更新** — **[COLLEAGUE.SKILL 技术报告](https://arxiv.org/pdf/2605.31264) 已上线**；这次最开心的不只是发了篇 paper，而是社区一起把 gallery 推到 215 个 skills、165 位贡献者和 100k+ skill-card 累计 stars，论文 Acknowledgements 也专门收录并感谢了所有社区贡献者。 > 🗺️ **2026.04.13** — **Distilly 路线图正式发布！** 项目从原 Colleague Skill 走向更通用的目标：把人物蒸馏成可供 Agent 复用的 Skill。 👉 **[完整路线图](../../ROADMAP.md)** · **[💬 Discord](https://discord.gg/NVX66RxWZv)** > 🌐 **2026.04.07** — 社区平台上线！任何 skill / meta-skill 可直接给自己的 GitHub repo 引流，没有中间商。 👉 **[titanwings.github.io/colleague-skill-site](https://titanwings.github.io/colleague-skill-site/)**

Created by [@titanwings](https://github.com/titanwings)

--- ## 🆕 Distilly 现在能做什么？ ### 1️⃣ 从 Colleague Skill 到 Distilly 项目不再只围绕「同事」场景设计。`distilly` 创建器用同一套流程，为三类人物生成基于材料的 Person Profile，再将每份 Profile 封装成 Agent Skill。 ### 2️⃣ 支持三大人物类型

🧑‍💼 colleague	💞 relationship	🌟 celebrity
_{同事 · 导师 · 搭档 · 上下游协作者}	_{前任 · 伴侣 · 父母 · 朋友 · 家人}	_{名人 · 创作者 · 公众表达者 · 小说角色}
_{从材料中整理技术规范、工作流程、表达习惯和职场行为，生成 Work Skill + Persona。支持飞书 / 钉钉 / Slack 采集。}	_{把材料中可观察的表达习惯、情绪触发点、冲突模式和修复模式整理成人物 Skill。}	_{内置六维度研究工具链（字幕下载 → 文稿清洗 → 研究合并 → 质量检查），用于整理可观察的决策、表达与心智模型。}

每类人物有独立的信息采集策略、分析维度和 Person Profile 结构。 ### 3️⃣ 支持更多 Agent 宿主旧版只能在 Claude Code 里用。Distilly 现在支持 8 个本地 Agent 宿主：

**Grok Bot 预览：**Grok Bot 支持 saved/private Skills，但官方文档没有说明可直接导入本地 `SKILL.md`。可手工迁移 Distilly 流程为 saved Skill；直接安装仓库尚未验证。蒸馏出的 Person Profile 会封装成 Agent Skill，并可安装到任意受支持的宿主。 --- ## 🤖 个人 Agent Distilly 也面向大家日常在用的个人 Agent。它们各自保存记忆、连接账号，所以 Person Profile 应当能直接带进去，而不是每次从零开始。

Grok Bot
_{saved Skill · 预览}

Meta Muse
_{画像交接 · 规划中}

OpenAI Dots
_{画像交接 · 规划中}

_{原生 Skill}

--- ## 📦 支持的数据来源 | 标识 | 来源 | 消息记录 | 文档 / Wiki | 备注 | |:----:|------|:-------:|:-----------:|------| |

| 飞书（自动采集） | ✅ API | ✅ | 输入姓名即可，全自动 | |

| 钉钉（自动采集） | ⚠️ 浏览器 | ✅ | 钉钉 API 不支持历史消息 | |

| Slack（自动采集） | ✅ API | — | 需管理员安装 Bot；免费版限 90 天 | |

| 公开 X 帖子 | ✅ API | — | 通过按返回数量计费的第三方 Xquik 可选采集名人研究候选 | |

| 微信聊天记录 | ✅ SQLite | — | 需先用 WeChatMsg / PyWxDump / 留痕等工具导出 | | 📄 | PDF / 图片 / 截图 | — | ✅ | 手动上传 | |

| 飞书 JSON 导出 | ✅ | ✅ | 手动上传 | | ✉️ | 邮件 `.eml` / `.mbox` | ✅ | — | 手动上传 | | 📝 | Markdown / 直接粘贴 | ✅ | ✅ | 手动输入 | --- ## ⚡ 安装 ### 🤖 For Agents 在任意支持的本地 Agent 宿主中发送： > 从 `https://github.com/titanwings/distilly` 安装 Distilly，并确认当前宿主可以发现它。 Agent 会把 Distilly 作为名为 `distilly` 的 Skill 安装到正确的宿主目录。 ### 👤 For Humans 把 Distilly clone 到当前宿主使用的 Skills 目录： ```bash git clone https://github.com/titanwings/distilly ``` 各宿主路径、旧版本迁移、Windows、生成 Profile 的安装方式和凭证配置，统一放在 **[详细安装说明](../../INSTALL.md)**。 --- ## 🚀 使用 **刚装好个人 Agent？** 说 *“用 Distilly 建立我的画像”*。Distilly 会用大约五分钟问你几个问题，给你确认画像全文，然后把简短摘要写进 Agent 每次会话都会读的文件（Claude Code 是 `~/.claude/CLAUDE.md`，OpenClaw 是 `USER.md`）。完整画像只保存在本地 `~/.distilly/owner/profile.md`。为别人建画像：在 Agent 中说： > 用 Distilly 为 `<人物>` 创建一个 Person Profile。接下来只需： 1. 选择 `colleague`、`relationship` 或 `celebrity`。 2. 提供一段描述或相关材料；所有字段都可以跳过。 3. 审核结果，让 Distilly 生成 Profile。结果会封装成名为 `{character}-{slug}` 的 Agent Skill。可选的名人研究流程和高级工具见 **[详细安装说明](../../INSTALL.md)**。 --- ## ✨ 效果示例
### 🌟 celebrity — 蒸馏一个 Andrej Karpathy > 六维度 research：著作 · 访谈 · 决策 · 表达 DNA · 外部评价 · 时间线
> 案例出自社区作品 [alchaincyf/karpathy-skill](https://github.com/alchaincyf/karpathy-skill) ``` 用户 ❯ 现在人人都在说 Agent，你怎么看 2026 年的 AI Agent？ celebrity.skill ❯ 大家都在做 demo，但没人做 eval。 Agent 的瓶颈从来不是模型不够聪明，是 scaffolding 跟不上—— 严格的 evaluation harness、细粒度的错误回放、可控的 tool-calling budget、清楚的 failure mode 分析。大部分产品还停留在"接个 API + 几个 tool + system prompt"的阶段，这叫 demo，不叫 agent。 2026 我赌的不是最强模型，是最好的 scaffolding + 最严的 eval。 ```
### 🧑‍💼 colleague — 蒸馏一个字节 2-1 后端 > 输入：`字节 2-1 后端工程师，INTJ，甩锅高手，字节范` ``` 用户 ❯ 帮我看一下这个接口设计 colleague.skill ❯ 等等，这个接口的 impact 是什么？背景没说清楚。（看完后）N+1 查询，改掉。返回结构用统一的 {code, message, data}，这是规范，不用问为什么。用户 ❯ 这个 bug 是你引入的吧 colleague.skill ❯ 上线时间对上了吗？那个需求改了好几个地方，还有其他变更。 ```
### 💞 relationship — 蒸馏一个暗恋对象 > 上传半年的聊天记录 + 「性格敏感、话少但嘴硬、会在关键时候认真回」 ``` 用户 ❯ 你今天有没有想我 relationship.skill ❯ ……有啊，就一点点。你干嘛问这个。 ```

📚 更多实战案例见 **[社区 Gallery](https://titanwings.github.io/colleague-skill-site/)**，已收录 100+ skill

--- ## 🔧 功能特性 ### 🧱 生成的 Skill 结构 Distilly 当前的创建器以 **Persona** 为通用底座，不同家族按场景挂载各自的模块： | 家族 | Persona 内容 | 附加模块 | |------|-------------|---------| | 🧑‍💼 **colleague** | 6 层性格结构：硬规则 → 身份 → 表达风格 → 决策模式 → 人际行为 → Correction | ➕ **Work Skill**：负责范围、工作流程、输出偏好、经验知识库 | | 💞 **relationship** | 表达 DNA · 情绪触发点 · 冲突模式 · 修复模式 | — | | 🌟 **celebrity** | 心智模型 · 决策启发式 · 表达 DNA · 外部评价对照 | ➕ 六维度 research 档案（著作 / 访谈 / 决策 / 时间线...） | > **运行逻辑**：接到任务 → Persona 选择材料中可见的偏好与语气 → 附加模块补齐执行细节 → 生成基于材料的回答 ### 🧬 进化机制 - 📥 **追加文件** → 自动分析增量 → merge 进对应部分，不覆盖已有结论 - 💬 **对话纠正** → 说「他不会这样，他应该是 xxx」→ 写入 Correction 层，立即生效 - 🕰️ **版本管理** → 每次更新自动存档，支持回滚到任意历史版本 - 🔬 **名人研究管线** → 字幕下载 → 文稿清洗 → 六维度研究 → 质量检查 --- ## ⚠️ 注意事项 **原材料质量决定 Person Profile 质量**，不同家族的优质信源不一样： | 家族 | 信源优先级（高 → 低） | |------|----------------------| | 🧑‍💼 **colleague** | 他**主动写的**长文（设计文档 / 评审意见） **›** **决策类回复** **›** 日常群聊消息 | | 💞 **relationship** | 完整的聊天记录 **›** 往来信件 / 朋友圈 / 日记 **›** 旁人描述 | | 🌟 **celebrity** | 第一人称著作 / 博客 / 长访谈 **›** 决策记录（发布会、commit、采访）**›** 已核验的本人短帖 **›** 他人评价 | - **colleague** 飞书自动采集：需将 App bot 加入相关群聊 - **relationship**：时间跨度越长越好，能覆盖冲突与和解更佳 - **celebrity**：避免只喂二手解读 - 目前还是 demo 版本，如果有 bug 请多多提 issue！ --- ## 📄 技术报告 > **[COLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation](https://arxiv.org/pdf/2605.31264)** ([arXiv](https://arxiv.org/abs/2605.31264) · [arXiv PDF](https://arxiv.org/pdf/2605.31264)) > > 这是 Distilly 前身 **COLLEAGUE.SKILL / colleague-skill（原同事 Skill）** 的技术论文，详细介绍了 Work Skill + Persona 的双层架构、多源数据采集与 Skill 生成机制，也是今天 `colleague` 家族的理论基础。relationship / celebrity 家族的架构扩展会另起论文。 --- ## 📝 引用如果你在研究或应用中使用了 **Distilly** 或 **COLLEAGUE.SKILL**，请引用技术报告： ```bibtex @misc{zhou2026colleagueskill, title = {COLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation}, author = {Tianyi Zhou and Dongrui Liu and Leitao Yuan and Jing Shao and Xia Hu}, year = {2026}, eprint = {2605.31264}, archivePrefix = {arXiv}, primaryClass = {cs.AI}, url = {https://arxiv.org/abs/2605.31264} } ``` 也可以使用 [CITATION.cff](../../CITATION.cff) 中的机器可读引用信息。 --- ## ⭐ Star History

---

**MIT License** © [titanwings](https://github.com/titanwings) _{Made with 🧬 for everyone who wants to distill a person into a reusable Person Profile.}

--- --- METRICS --- - Files Extracted: 11 - Estimated Token Budget: ~11697 tokens - Canonical Reference: https://codewiki.google/github.com/titanwings/colleague-skill