teammate-skill

Your teammate left. Their context didn't have to.

English | 简体中文 | Français | Deutsch | Русский | 日本語 | Italiano | Español

Your teammate quit and three years of tribal knowledge walked out the door. Your senior engineer left — no handoff doc, no runbook, just silence. Your co-founder pivoted, taking every unwritten decision with them.

Feed in their Slack messages, GitHub PRs, emails, docs, and your own description — get an AI Skill that actually works like them.

Writes code in their style. Reviews PRs with their standards. Answers questions in their voice.

How It Works · Install · Data Sources · Demo · Detailed Setup

---

How It Works

You describe your teammate (3 questions)
          ↓
Provide source materials (Slack, GitHub, email, docs — or skip)
          ↓
Dual-track AI analysis
  ├── Track A: Work Skill (systems, standards, workflows, review style)
  └── Track B: Persona (5-layer personality model)
          ↓
Generated SKILL.md — invoke anytime with /{slug}
          ↓
Evolve over time (append new data, correct mistakes, auto-version)

The generated skill has two parts that work together:

Part	What it captures
Part A — Work Skill	Systems owned · tech standards · code review focus · workflows · tribal knowledge
Part B — Persona	5-layer model: hard rules → identity → expression → decision patterns → interpersonal style

When invoked: Persona decides the attitude → Work Skill executes → output in their voice.

---

Platforms

🦞 OpenClaw

Open-source personal AI assistant by @steipete. Runs on your own hardware, answers on 25+ channels (WhatsApp, Telegram, Slack, Discord, Teams, Signal, iMessage…). Local-first, persistent memory, voice, canvas, cron jobs, and a growing skills ecosystem.

🏆 MyClaw.ai

Managed hosting for OpenClaw — skip Docker, servers, and configs. One-click deploy, always-on, automatic updates, daily backups. Your OpenClaw instance live in minutes. Perfect if you want teammate.skill running 24/7 without self-hosting.

Claude Code

Anthropic's official agentic coding CLI. Install this skill into .claude/skills/ and invoke with /create-teammate.

---

Install

🦞 OpenClaw / 🏆 MyClaw.ai

Option A — ClawHub (recommended):

openclaw skills install create-teammate

Option B — Git:

git clone https://github.com/LeoYeAI/teammate-skill ~/.openclaw/workspace/skills/create-teammate

Then start a new session (/new) and type /create-teammate.

MyClaw.ai users: SSH into your instance or use the web terminal. Same commands.

Claude Code

# Per-project
mkdir -p .claude/skills
git clone https://github.com/LeoYeAI/teammate-skill .claude/skills/create-teammate

# Global (all projects)
git clone https://github.com/LeoYeAI/teammate-skill ~/.claude/skills/create-teammate

Then type /create-teammate in Claude Code.

Other AgentSkills Agents

Clone into your agent's skill directory. Any agent that reads AgentSkills SKILL.md frontmatter will auto-detect it.

Optional Dependencies

pip3 install -r requirements.txt   # Only needed for Slack/GitHub auto-collectors

Slack auto-collection needs a Bot Token. GitHub collection uses GITHUB_TOKEN. See INSTALL.md for setup guides.

---

Supported Data Sources

Source	Type	Method
Slack	Messages, threads, reactions	✅ Auto-collect (enter username) or upload JSON export
GitHub	PRs, reviews, commits, issues	✅ Auto-collect (enter handle)
Gmail	Emails	Upload .mbox / .eml from Google Takeout
Teams / Outlook	Chat, emails	Upload export JSON
Notion	Wiki, docs	Upload Markdown/HTML export
Confluence	Wiki, docs	Upload HTML export or zip
JIRA / Linear	Issues, tickets	Upload CSV / JSON export
PDF / Images	Any document	Upload directly
Markdown / Text	Anything	Upload or paste

Mix and match any combination. Or skip entirely — even a one-sentence description generates a useful skill.

---

Demo

Creating a Teammate (one-shot)

You    ❯  /create-teammate Alex Chen, Stripe L3 backend, INTJ perfectionist,
          brutal code reviewer but usually right

Agent  ❯  👤 alex-chen | Stripe L3 Backend | INTJ, Perfectionist, Stripe-style
          Looks right? (y / change something)

You    ❯  y

Agent  ❯  Any work artifacts? Slack username, GitHub handle, files, or paste text.
          Skip is fine — I'll work with what you gave me.

You    ❯  skip

Agent  ❯  ━━━ Preview: alex-chen ━━━
          💼 Work: Payments Core · Ruby/Go/Postgres · CR: idempotency, naming
          🧠 Persona: Direct, conclusion-first · Correctness > Speed
          ━━━━━━━━━━━━━━━━━━━━━━━
          Looks right?

You    ❯  👍

Agent  ❯  ✅ alex-chen created!
          📁 teammates/alex-chen/ | 🗣️ /alex-chen
          Try it — ask Alex anything:

Using the Generated Skill

You    ❯  Alex, should we use MongoDB for this new service?

Alex   ❯  No. We need ACID for payment flows. Stick with Postgres.
          If the concern is schema flexibility, use JSONB columns.

You    ❯  Review this payment endpoint?

Alex   ❯  What problem does this solve? PR description doesn't say.
          Blocking: you're using float for amount. Use Decimal.
          Also — not idempotent. What happens on retry?

You    ❯  Skip tests, ship by Friday?

Alex   ❯  No. Tests right or we don't ship. Cut scope, not quality.

Evolving Over Time

You    ❯  /update-teammate alex-chen
          [uploads 3 months of Slack export]

Agent  ❯  Analyzed 847 messages. Found:
          • 12 new catchphrases added to Persona
          • Webhook reliability expertise added to Work Skill
          • v1 → v2 (v1 archived)

---

Platform Compatibility

This skill is built to run natively on both platforms — no adapter layer, no workarounds.

Feature	Claude Code	OpenClaw
Install	.claude/skills/	openclaw skills install or ~/.openclaw/workspace/skills/
Trigger	/create-teammate	/create-teammate
Path resolution	${CLAUDE_SKILL_DIR}	{baseDir} (auto-resolved)
File operations	Read / Write / Edit	read / write / edit
Run scripts	Bash tool	exec tool
Fetch URLs	Bash → curl	web_fetch tool
Generated output	./teammates/{slug}/	./teammates/{slug}/

The SKILL.md uses {baseDir} for all paths — both platforms resolve it automatically.

---

Evolution

Teammate skills aren't static — they improve over time:

Method	How
Append data	"I found more Slack logs" → auto-analyze delta, merge without overwriting
Correct mistakes	"They wouldn't say that, they'd say..." → instant correction
Version control	Every update auto-archives → rollback to any version

---

Quality Assurance

Every teammate goes through a 3-layer quality pipeline before you get it:

1. Quality Gate (pre-preview)

Validates generated content against 7 hard rules: Layer 0 concreteness, example count, catchphrase density, priority ordering, scope definition, no generic filler, tag→rule completeness. Failures auto-fix before you see the preview.

2. Smoke Test (post-creation)

Three automated test prompts run against every generated skill:

• Domain question — does the skill use real systems/tools, not generic advice?
• Pushback scenario — does the persona hold under pressure, or collapse into generic AI?
• Out-of-scope question — does the skill admit knowledge limits in-character?

🧪 Smoke Test: ✅ Domain ✅ Pushback ✅ Out-of-scope — 3/3 passed

3. Privacy Scan (pre-export)

Automatic detection of emails, phone numbers, API tokens, SSNs, and other PII:

python3 tools/privacy_guard.py --scan teammates/alex-chen/          # detect
python3 tools/privacy_guard.py --scan teammates/alex-chen/ --redact  # auto-fix

Raw knowledge files (knowledge/) are excluded from git and exports by default.

---

Compare Teammates

Side-by-side comparison with scenario simulation:

You    ❯  /compare alex-chen vs bob-smith

Agent  ❯  ━━━ alex-chen vs bob-smith ━━━
                         alex-chen              bob-smith
          Priority:      Correctness > Speed    Ship fast > Perfect
          CR Style:      Blocking on naming     Suggestions only
          Under Pressure: Gets quieter           Gets louder
          Says "No" by:  Direct refusal          Asking questions

You    ❯  Who should review the payments API redesign?

Agent  ❯  alex-chen: "Send me the design doc. I want to check
             idempotency and error contracts."
          bob-smith: "Let's hop on a call and walk through it."

          Recommendation: alex-chen for correctness rigor.

Also supports decision simulation — watch two teammates debate a technical decision in character.

---

Export & Share

Export teammates as portable packages:

/export-teammate alex-chen
# → alex-chen.teammate.tar.gz (skill files only, no raw data)

# Import on another machine:
tar xzf alex-chen.teammate.tar.gz -C ./teammates/

The export includes: SKILL.md, work.md, persona.md, meta.json, version history, and a manifest. Raw knowledge files are excluded by default — add --include-knowledge if needed (⚠️ contains PII).

---

Commands

Command	Description
/create-teammate	Create a new teammate skill
/list-teammates	List all generated teammates
/{slug}	Invoke teammate (full persona + work)
/{slug}-work	Work capabilities only
/{slug}-persona	Persona only
/compare {a} vs {b}	Side-by-side comparison with scenario simulation
/export-teammate {slug}	Export portable .tar.gz package for sharing
/update-teammate {slug}	Add new materials to existing teammate
/teammate-rollback {slug} {version}	Rollback to previous version
/delete-teammate {slug}	Delete a teammate skill

---

Supported Tags

Meticulous · Good-enough · Blame-deflector · Perfectionist · Procrastinator · Ship-fast · Over-engineer · Scope-creeper · Bike-shedder · Micro-manager · Hands-off · Devil's-advocate · Mentor-type · Gatekeeper · Passive-aggressive · Confrontational · Drama-free

Google-style · Meta-style · Amazon-style · Apple-style · Stripe-style · Netflix-style · Microsoft-style · Startup-mode · Agency-mode · First-principles · Open-source-native