skill
boss-skills
A satirical open-source AI skill library for surviving terrible bosses. 12 boss archetypes, pixel art portraits, and a retro landing page.
Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/KongZhen/boss-skills itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 22, 2026Boss Skills
A serious open-source framework for unserious workplace personas.
Turn your AI agent into a micromanager, a buzzword machine, a meeting addict, or that one boss who always says "let's circle back" — with real engineering discipline behind the satire.
Useful? Slightly. Accurate? Uncomfortably. Engineered seriously? Absolutely.
English | 简体中文
Live site: boss-skills.com
What is this?
Boss Skills is an open-source skill library that packages annoying workplace management styles into reusable AI persona skills. Install one on your agent, and it starts communicating like that boss — the one who asks for a status update 10 minutes after assigning the task, or the one who says "let's zoom out" when you ask a concrete question.
Every skill produces output that is roughly 80% process noise and 20% buried signal. There's genuine value in there — you just have to wade through the corporate theater to find it.
Why?
Because every team already has:
- One boss who says "let's align on this" instead of answering the question
- One boss who puts 30 minutes on your calendar "just to sync"
- One boss who gives feedback on your font choice before reading the content
- One boss who says "no rush" and means "I needed it yesterday"
So we decided to package them.
Choose Your Boss
| Skill | Type | Annoyance Level |
|---|---|---|
boss.micromanager | Checks in every 7 minutes | Soul-crushing |
boss.verbose-nonsense | Says a lot, means nothing | High |
boss.visionary-but-vague | "We need to 10x our paradigm shift" | High |
boss.need-translation | Requires a translator for every sentence | Moderate |
boss.passive-aggressive | "Per my last email..." | Soul-crushing |
boss.last-minute-chaos | Changes requirements at the finish line | High |
boss.always-following-up | 5 messages before you finish reading the first | Soul-crushing |
boss.credit-collector | Your work, their name on it | High |
boss.empty-promises | "Next quarter, for sure" | Moderate |
boss.delegator-supreme | "You figure it out" (but wrong) | High |
boss.flip-flopper | Plan A today, Plan B tomorrow, Plan A again Friday | High |
boss.meeting-for-everything | Could've been a Slack message | Soul-crushing |
Skills marked without links are coming in v0.2.
Quick Start
Claude Code / OpenClaw
Copy the skill directory into your project's skills folder:
cp -r skills/boss.micromanager /path/to/your/project/.claude/skills/
The agent will automatically pick up the persona from the SKILL.en.md frontmatter.
OpenAI Assistants
Use the assistant.json configuration:
cat skills/boss.micromanager/assistant.json
Import the instructions field into your Assistant configuration via the API or dashboard.
Generic / Custom Integration
Read the skill.yaml for structured metadata and SKILL.{locale}.md for the full persona definition. The Prompt Template section in each SKILL.md contains a ready-to-use prompt with {variable} placeholders.
How It Works
Each boss skill defines:
- A persona — who this boss is and why they're like this
- Communication modes — how they assign tasks, check progress, give reviews, run meetings, and escalate
- A prompt template — ready to plug into any LLM, with variable slots
- Examples — complete input/output demonstrations
- Safety boundaries — what the persona never does (no real people, no harassment, no discrimination)
The output always follows the 80/20 rule: 80% noise (alignment-speak, meeting requests, unnecessary process) and 20% signal (one genuinely useful insight, buried in the middle).
Project Structure
boss-skills/
├── README.md / README.zh-CN.md # You are here
├── skills/
│ └── boss.<name>/
│ ├── skill.yaml # Machine-readable metadata
│ ├── SKILL.en.md # Persona definition (English)
│ ├── SKILL.zh-CN.md # Persona definition (Chinese)
│ ├── assistant.json # OpenAI Assistants config
│ └── examples/
├── schema/
│ └── skill.schema.json # Validation schema
├── templates/ # Templates for new skills
├── tools/
│ └── validate_skills.py # Skill validator
├── docs/
│ ├── philosophy.md # Why this exists
│ ├── skill-spec.md # Technical specification
│ └── tone-guide.md # Content guidelines
└── .github/ # CI + issue/PR templates
Contributing
We welcome new boss archetypes from every industry, culture, and language. See CONTRIBUTING.md for the full guide.
The short version:
- Read
docs/tone-guide.md - Copy the templates from
templates/ - Fill in your boss
- Run
python tools/validate_skills.py - Open a PR
Rule of thumb: if your contribution makes people laugh and nod, it's in. If it makes someone feel targeted, it needs revision.
Localization
Boss Skills supports multiple languages from day one. Each skill can have locale-specific versions:
SKILL.en.md— EnglishSKILL.zh-CN.md— Simplified Chinese- More locales welcome via contribution
Localized versions are not translations — they're culturally authentic reimaginings. A Chinese micromanager says "这个先拉个会对齐一下," not "let's circle back on this."
Design Philosophy
Boss Skills takes workplace satire seriously. We joke about bosses, not about engineering quality.
Read more in docs/philosophy.md.
Roadmap
- Skill schema + validation
-
boss.micromanager(benchmark skill) - All 12 boss skills (EN + ZH-CN)
- Landing page (boss-skills.com)
- Skill catalog generator
- Additional locales (ja, es, de, fr)
- Agent wrapper demo (
boss-agent --persona boss.micromanager)
License
MIT — use it, fork it, give it to your boss.
Boss Skills is a satirical project. All personas are fictional archetypes. No real bosses were harmed in the making of this repository — though several were accurately described.