Skip to content
Skills Library
AI & agents · Planning

create

/create

Design and create a team-chat agent team with the person, a fixed team of subagents with roles, models, effort and a brief per member, saved to .claude/teams/ and opened in the team pane. Use when the person asks to create, set up, design or build a team.

Repository
team-chat ↗
Licence
MIT
Last repository change
11.11.2026

Assessment and security

100/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content35 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
1
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 11.11.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What it does and when to use it

Design and create a team-chat agent team with the person, a fixed team of subagents with roles, models, effort and a brief per member, saved to .claude/teams/ and opened in the team pane. Use when the person asks to create, set up, design or build a team.

Arguments

/create [what the team is for]
Original author description

Design and create a team-chat agent team with the person, a fixed team of subagents with roles, models, effort and a brief per member, saved to .claude/teams/ and opened in the team pane. Use when the person asks to create, set up, design or build a team.

How it works

  1. 1. Understand the job
  2. 2. Propose the team
  3. 3. Save it
Original SKILL.md headings · 60 lines, 724 words
SKILL.md preview
---
name: create
description: Design and create a team-chat agent team with the person, a fixed team of subagents with roles, models, effort and a brief per member, saved to .claude/teams/ and opened in the team pane. Use when the person asks to create, set up, design or build a team.
argument-hint: "[what the team is for]"
---

# Create a team-chat team

Help the person design a team of subagents, then save it with the `mcp__team-chat__save_team` tool. The person's main session is always the team's lead; never add a lead member yourself.

## 1. Understand the job

If `$ARGUMENTS` already says what the team is for, start from it. Ask only what you cannot see for yourself, in one short message:

- What the team will work on (a kind of task, a product area, a repo).
- Any roles they already have in mind, and any they do not want.
- Cost preference: lean (more haiku/sonnet, lower effort) or strong (opus/high where it matters).

Look at the project yourself before asking about it: the languages, frameworks, test setup and folder layout (README, package files, top-level folders). Use what you find to make the briefs concrete: real test commands, real folder names.

## 2. Propose the team

Propose 2–6 members. Fewer, sharper roles beat many overlapping ones. For each member give:

| field | what to write |
| --- | --- |
| `name` | handle, lowercase: `backend`, `reviewer`, `docs` |
| `role` | job title |
| `emoji` | one emoji avatar |
| `color` | one of cyan, magenta, green, yellow, blue, red, claude; different per member |
| `model` / `effort` | haiku for mechanical work (running tests, searching), sonnet for most building, opus for hard reasoning or review; effort low…max to match |
| `description` | one sentence: when the lead should hand it work |
| `purpose` | one line: what it is for |
| `responsibilities` | 2–4 lines: what it does |
| `approach` | 2–4 lines: how it works, its process, with this project's real commands and paths |
| `boundaries` | 1–3 lines: what it must never do, and who to ask instead |
| `doneWhen` | one line: its definition of done |
| `tools` | only to restrict it, e.g. a reviewer with `["Read", "Grep", "Glob", "Bash", "SendMessage"]`; otherwise leave out. An explicit list drops `Agent` (members never spawn agents); without a list the member keeps `Agent` and is only told in its brief not to use it |
| `reportsTo` | optional: the handle it answers to. Leave out for the lead (the default). Must be a member of this team, and the chain must end at the lead with no loops |
| `handsTo` | optional: handles that receive its finished work, in order. May include `lead`. Loops are allowed (builder → tester → builder) |

Give the team a short name (`dev-team`, `docs-squad`). Show the proposal as a compact table (name, role, model/effort, purpose), then each member's responsibilities and boundaries in a line or two. Make sure the boundaries fit together: every kind of work has exactly one owner.

Also propose the chain, as an indented list under the table, with `→` for each handoff:

```
- lead
  - techlead → builder, tester
    - builder → tester
    - tester → reviewer
    - reviewer → techlead
```

Check it before showing it: every `reportsTo` reaches the lead, and every `handsTo` names a member who exists (a handoff with no receiver is a mistake). Members may still message each other directly, so the chain sets the order, not a wall.

Optionally ask whether one member should be the **main** member. Untargeted prompts the person types then go to that member, and `@<member>` still goes to any member. The main member must be a member, not the lead. Leave it out to keep the main session handling untargeted prompts.

Ask the person to approve or change it. Apply their changes and show only what changed.

## 3. Save it

Once they approve, call `mcp__team-chat__save_team` with `name`, `members` and, if chosen, `main` (a member's handle). If it answers with an `Error:` (an unknown handle in a relation, a self-reference, a `reportsTo` loop, a relation on the lead, or a `main` that is not a member), fix the proposal and call again. If it answers that the team exists, ask whether to replace it, then call again with `overwrite: true`, or pick another name.

Then tell the person in two or three lines: the team is saved and open in the pane, they give it work by typing a task in the chat, and `@<member> …` talks to one member directly.

Versions

RepositoryScoreLicence
This page · ninjanazal/team-chat100MIT
verygoodopensource/vgv-wingspan100MIT
crisandrews/clawcode100MIT
digital-stoic-org/agent-skills100MIT
turalowski/agents100MIT
turalowski/agents100MIT
turalowski/agents100MIT
Borda/AI-Rig100Apache-2.0
imgly/codesign100MIT

Database snapshot: 13.11.2026 · package version 5208ff56fcee (SHA-256 digest of the entire package)

Downloads in the last 30 days: 1 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 95team-create
    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.
  • 100team-stack
    Analyze a task, propose an agent team composition with roles and responsibilities, and create the team after user confirmation. Use when the user says "team stack", "create a team", "set up agents for this", or describes a complex task that would benefit from multiple agents working together.
  • 90help
    Show the team-chat guide, how to create and start a team, read the pane, talk to members, and every command. Use when the person asks how team-chat works or for its help.
  • 86start
    Start working with a team-chat team, making it the active team and opening the team pane. Use when the person asks to start, open, switch to or work with a team.
  • 100agent-teams
    Create agent teams for parallel collaboration. Use when user mentions team, parallel, multi-agent, coordinate, collaborate, review team, multiple perspectives, or says "agent-teams".
  • 100claude-code-agent-teams
    This skill should be used when the user asks to "create an agent team", "run a multi-agent team", "coordinate multiple Claude sessions", "use agent teams", or wants to parallelize work across multiple Claude Code instances.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.