team-create
Skill533 starsChanged 5 months agoApache-2.0View on GitHub
Create a collaborative agent team with roles, personalities, and skills. Use when: user says "create team", "build a team", "新建团队", "创建团队", "组建团队", "搭建团队". Do NOT use for managing existing teams or assigning tasks to teams.
In this repository7 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- pr-cleaner-masterSkill
- pr-merge-runnerSkill
- release-publishSkill
- sync-gateway-docsSkill
The file
We split this skill into the rules it gives an agent
9 rules. About 0 are common, 5 are distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 132
Every team MUST have a manager agent with these properties:
- Very specific: line 198
Check ~/.agents/.skill-lock.json — if skill exists there, copy source info
- Distinctive: line 200
If not found locally, search for it (GitHub, Cloud Hub) or ask user
- Distinctive: line 245
When a team is installed, the installer should:
- Very specific: line 247
Read all agents/*/skills.json to collect every agent's skill dependencies
- Distinctive: line 251
Create OpenClaw agents via Gateway API for each agent in team.md
- Very specific: line 256
Every team MUST have a manager agent — it is the coordinator entry point
- Distinctive: line 257
All generated content in the language the user uses
- Distinctive: line 260
Do not invent skills — verify existence via lock files or web search
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Team Create
Interactive workflow for building a collaborative agent team.
A Team is an installable package — it bundles multiple OpenClaw Agents, each with its own identity (agent.md), personality (soul.md), and skill dependencies (skills.json), all self-contained and portable.
Team Protocol
Folder Structure
~/.agents/teams/{team-name}/
├── team.md # Team metadata + positioning
└── agents/
├── manager/ # MANDATORY — coordinator, never does work itself
│ ├── agent.md
│ ├── soul.md
│ └── skills.json # Skill dependencies (empty for coordinator)
├── {agent-slug}/
│ ├── agent.md # Agent identity + description
│ ├── soul.md # Personality, tone, working style
│ └── skills.json # This agent's skill dependencies with install sources
└── ...team.md Format
---
name: {team-name}
description: "{one-line team purpose}"
version: "1.0.0"
agents:
- id: manager
name: "{Manager Display Name}"
role: coordinator
- id: {agent-slug}
name: "{Display Name}"
role: worker
---
{Markdown body: team mission, positioning, collaboration model between agents}agent.md Format
Each agent maps to an OpenClaw Agent. Defines WHO this agent is — identity and responsibilities only. No skill references here.
---
id: {agent-slug}
agentId: ""
name: "{Display Role Name}"
description: "{What this agent is responsible for}"
---
{Markdown body: role positioning, core responsibilities, how this agent
collaborates with other agents in the team}id: internal role slug within the team, defined at creation time, immutableagentId: real OpenClaw Agent ID assigned by the system, backfilled at install time, left empty at creation
soul.md Format
Defines HOW this agent communicates and thinks — its personality.
# {Role Name} — Soul
## Personality
{Communication style, tone, attitude.}
## Working Style
{How this agent approaches tasks, priorities, decision-making.}
## Principles
{Core beliefs that guide decisions.}Per-Agent skills.json Format
Each agent has its OWN skills.json declaring its skill dependencies with
install sources. Format mirrors .skill-lock.json for compatibility.
{
"version": 1,
"skills": {
"{skill-id}": {
"source": "{owner/repo}",
"sourceType": "github",
"skillPath": "skills/{skill-id}/SKILL.md"
}
}
}sourceType values:
"github"— install from GitHub repo (requiressourceasowner/repo)"cloud-hub"— install from OpenClaw Cloud Hub (requiressourceas hub skill ID)"local"— already available locally, no installation needed
Manager Agent (MANDATORY)
Every team MUST have a manager agent with these properties:
- id:
manager - role:
coordinatorin team.md - Never does work itself — only dispatches, coordinates, and synthesizes
- soul.md: focuses on orchestration style, delegation principles, quality gates
- skills.json: typically empty (
"skills": {}) - The manager is the entry point when the team is invoked
Workflow
- Step 0: Parse arguments
- Step 1: Team positioning (team.md)
- Step 2: Define agents (manager auto-created + user-defined workers)
- Step 3: Define each agent's soul
- Step 4: Assign skills (per-agent skills.json)
- Step 5: Generate and save
- Step 6: Summary
Step 0: Parse Arguments
- If
$ARGUMENTScontains a team name, use it; otherwise ask - Slugify: lowercase, hyphens, no spaces (e.g., "Research Team" →
research-team) - Check
~/.agents/teams/{team-name}/does not exist - If it exists, ask: overwrite or pick a new name?
Step 1: Team Positioning (team.md)
Ask the user:
What kind of team do you want to create? Describe its purpose and mission.
- Draft team.md with frontmatter (name, description, version) and body
- Body: team mission, positioning, collaboration model
- The
manageragent is always first in the agents list withrole: coordinator - Show draft for confirmation
Step 2: Define Agents
The manager agent is auto-created. Ask the user for worker agents:
What roles does this team need? (e.g., frontend developer, tester, PM)
For each role:
- Collect: display name + brief description
- Derive slug (e.g., "Frontend Developer" →
frontend-dev) - Confirm the complete agent list (manager + workers)
Step 3: Define Each Agent's Soul (soul.md)
For each agent (including manager):
- Manager soul: orchestration style, delegation principles, how it routes work
- Worker souls: personality, working style, principles for each role
- If user says "auto", generate sensible defaults based on role
- Show drafts and let user adjust
Step 4: Assign Skills (per-agent skills.json)
For each agent, build its own skills.json:
- Recommend skills based on role heuristics
- Let user customize: add/remove per agent
- Manager typically has empty skills
- For each skill assigned to an agent, determine its install source:
- Check
~/.agents/.skill-lock.json— if skill exists there, copy source info - Check
~/.agents/skills-lock.json— same - If not found locally, search for it (GitHub, Cloud Hub) or ask user
- Check
- Write skills.json into each agent's folder
Step 5: Generate and Save
- Create
~/.agents/teams/{team-name}/ - Create
agents/{slug}/for each agent (manager + workers) - Write
team.md - For each agent write
agent.md+soul.md+skills.json - Verify with
findtree listing
Step 6: Summary
Team created successfully!
━━━━━━━━━━━━━━━━━━━━━━━━━
📁 ~/.agents/teams/{team-name}/
├── team.md — {description}
└── agents/
├── manager/
│ ├── agent.md — Coordinator
│ ├── soul.md
│ └── skills.json — 0 skills
├── {agent-1}/
│ ├── agent.md — {role}
│ ├── soul.md
│ └── skills.json — {N} skills
└── ...
Total: {N} agents (1 manager + {N-1} workers)
Per-agent skills:
• manager: (coordinator — no skills)
• {agent-1}: {skill-1}, {skill-2}, ...
• {agent-2}: {skill-1}, {skill-3}, ...
Merged skill manifest ({N} unique):
• {skill-1} — {sourceType}: {source}
• {skill-2} — {sourceType}: {source}
...Installation Flow (for consumers of this team)
When a team is installed, the installer should:
- Read all
agents/*/skills.jsonto collect every agent's skill dependencies - Merge and deduplicate skill IDs across all agents
- For each unique skill: check if already installed in
~/.agents/skills/ - Install missing skills from their declared sources
- Create OpenClaw agents via Gateway API for each agent in team.md
- Upload
agent.mdandsoul.mdto each agent's workspace viaagents.files.set
Rules
- Every team MUST have a
manageragent — it is the coordinator entry point - All generated content in the language the user uses
- Each agent folder is self-contained: agent.md + soul.md + skills.json
- skills.json MUST include source info for every skill — no orphan references
- Do not invent skills — verify existence via lock files or web search
- Agent slugs must be unique within a team
- soul.md should feel human — avoid generic corporate language
- The team folder must be self-contained and portable (installable on any machine)
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

