>
> QR refreshes every 7 days (expires 2026-05-18) โ if expired, ping me on Discord.
> ๐บ๏ธ **2026.04.13** โ **dot-skill Roadmap is live!** colleague.skill is evolving into **dot-skill** โ distill anyone, not just colleagues. ๐ **[Full Roadmap](ROADMAP.md)** ยท **[๐ฌ Discord](https://discord.gg/NVX66RxWZv)**
> ๐ **2026.04.07** โ Community gallery is live! Any skill / meta-skill can drive traffic directly to your own GitHub repo. No middleman. ๐ **[titanwings.github.io/colleague-skill-site](https://titanwings.github.io/colleague-skill-site/)**
### ๐ celebrity โ distilling Andrej Karpathy > Six-dimension research: works ยท interviews ยท decisions ยท expression DNA ยท external evaluations ยท timeline
> Case adapted from community work [alchaincyf/karpathy-skill](https://github.com/alchaincyf/karpathy-skill) ``` User โฏ Everyone's talking about Agents now โ what do you think the 2026 AI Agent looks like? celebrity.skill โฏ Everyone's doing demos, nobody's doing eval. The Agent bottleneck was never model intelligence โ 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 โ I'm betting on the best scaffolding and the strictest eval. ```
### ๐งโ๐ผ colleague โ distilling a ByteDance L2-1 backend > Input: `ByteDance L2-1 backend engineer, INTJ, blame-shifter, ByteDance-style` ``` User โฏ Can you review this API design? colleague.skill โฏ Hold on โ 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 โฏ This bug was introduced by you, right? colleague.skill โฏ Does the timeline match? That feature touched multiple places, there were other changes too. ```
### ๐ relationship โ 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 โฏ Did you think about me today? relationship.skill โฏ ...I did, a little bit. Why are you asking? ```
๐ More real-world cases in the **[community gallery](https://titanwings.github.io/colleague-skill-site/)** โ 100+ skills and counting
---
## ๐ง Features
### ๐งฑ Generated Skill Structure
dot-skill uses **Persona** as the universal base, with family-specific modules layered on top:
| Family | Persona Content | Additional Modules |
|--------|-----------------|-------------------|
| ๐งโ๐ผ **colleague** | 6-layer personality: hard rules โ identity โ expression โ decisions โ interpersonal โ Correction | โ **Work Skill**: scope, workflow, output preferences, experience knowledge base |
| ๐ **relationship** | Expression DNA ยท emotional triggers ยท conflict pattern ยท repair pattern | โ |
| ๐ **celebrity** | Mental models ยท decision heuristics ยท expression DNA ยท external-evaluation contrast | โ Six-dimension research dossier (works / interviews / decisions / timeline...) |
> **Execution**: Receive task โ Persona decides attitude & tone โ Additional modules fill in execution detail โ Output in their voice
### ๐งฌ Evolution
- ๐ฅ **Append files** โ auto-analyze delta โ merge into relevant sections, never overwrite existing conclusions
- ๐ฌ **Conversation correction** โ say "they wouldn't do that, they'd be xxx" โ writes to the Correction layer, takes effect immediately
- ๐ฐ๏ธ **Version control** โ auto-archive on every update, rollback to any previous version
- ๐ฌ **Celebrity research pipeline** โ subtitles โ transcript cleanup โ six-dimension research โ quality check
---
## ๐ Project Structure
This project follows the [AgentSkills](https://agentskills.io) open standard. The entire repo is a skill directory.
Generated colleague skills live under `./skills/colleague`:
```
dot-skill/
โโโ SKILL.md # skill entry point (official frontmatter)
โโโ prompts/ # prompt system across three families
โ โโโ intake.md # [colleague] info intake
โ โโโ work_analyzer.md # [colleague] work capability extraction
โ โโโ persona_analyzer.md # [colleague] personality extraction
โ โโโ work_builder.md # [colleague] work.md generation
โ โโโ persona_builder.md # [colleague] persona.md 6-layer structure
โ โโโ merger.md # [shared] incremental merge logic
โ โโโ correction_handler.md # [shared] conversation correction
โ โโโ relationship/ # [relationship] emotion/conflict/repair prompts
โ โโโ celebrity/ # [celebrity] six-dimension research + mental-model prompts
โโโ tools/ # Python tools
โ โโโ feishu_auto_collector.py # [colleague] Feishu auto-collector
โ โโโ dingtalk_auto_collector.py # [colleague] DingTalk auto-collector
โ โโโ slack_auto_collector.py # [colleague] Slack auto-collector
โ โโโ email_parser.py # [shared] email parser
โ โโโ research/ # [celebrity] celebrity research toolchain
โ โ โโโ download_subtitles.sh # subtitle download
โ โ โโโ transcribe_audio.py # audio โ text
โ โ โโโ srt_to_transcript.py # subtitles โ transcript
โ โ โโโ merge_research.py # six-dimension research merge
โ โ โโโ quality_check.py # quality check
โ โโโ install_*_skill.py # [shared] multi-host one-shot installers
โ โโโ skill_writer.py # [shared] skill file management
โ โโโ version_manager.py # [shared] version archive & rollback
โโโ skills/ # generated Skills (gitignored)
โ โโโ colleague/ # colleagues
โ โโโ relationship/ # close relationships
โ โโโ celebrity/ # public figures
โโโ docs/PRD.md
โโโ requirements.txt
โโโ LICENSE
```
---
## โ ๏ธ Notes
**Source material quality = Skill quality** โ and quality sources differ across families:
| Family | Source priority (high โ low) |
|--------|------------------------------|
| ๐งโ๐ผ **colleague** | Their **own long-form writing** (design docs / review comments) **โบ** **decision-making replies** **โบ** casual group chat |
| ๐ **relationship** | Complete chat history **โบ** letters / social posts / diaries **โบ** third-party descriptions |
| ๐ **celebrity** | First-person books / blogs / long interviews **โบ** decision records (launches, commits, Q&A) **โบ** third-party commentary |
- **colleague** Feishu 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 โ please file issues if you find bugs!
---
## ๐ Technical Report
> **[COLLEAGUE.SKILL: Automated AI Skill Generation via Expert Knowledge Distillation](colleague_skill.pdf)** ([arXiv](https://arxiv.org/abs/2605.31264) ยท [arXiv PDF](https://arxiv.org/pdf/2605.31264))
>
> This is the paper for **colleague.skill**, dot-skill's predecessor. It covers the Work Skill + Persona two-layer architecture, multi-source data collection, and Skill generation mechanics โ the theoretical foundation for today's `colleague` family. Separate papers on the relationship / celebrity family extensions are planned.
---
## ๐ Citation
If you use **dot-skill** 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).
---
## โญ Star History
**MIT License** ยฉ [titanwings](https://github.com/titanwings)