关于
This skill creates a new multi-agent team composition file by automating the setup of a coordinated workflow. It handles purpose definition, member selection, coordination patterns, task decomposition, and generates the required configuration and documentation. Use it when defining a multi-agent workflow or composing agents for complex, recurring collaborative tasks.
快速安装
Claude Code
推荐npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/create-team在 Claude Code 中复制并粘贴此命令以安装该技能
技能文档
Create a New Team
Multi-agent team → coord 2+ agents for multi-perspective tasks. Integrates w/ teams registry, activatable in Claude Code by name.
Use When
- Task needs multi perspectives (code review + sec audit + arch)
- Recurring collab workflow w/ consistent roles
- Existing ad-hoc composition → formalize
- Complex proc → phases / specialties
- Sprint / pipeline / parallel work
In
- Required: Team name (kebab-case,
data-pipeline-review) - Required: Purpose (1 paragraph)
- Required: Lead agent (in
agents/_registry.yml) - Optional: Coord pattern (def: hub-and-spoke). One of:
hub-and-spoke,sequential,parallel,timeboxed,adaptive - Optional: Members (def: 3-4; rec: 2-5)
- Optional: Src material
Do
Step 1: Purpose
Answer:
- What outcome delivered? (report, deploy, sprint)
- Why not 1 agent? ≥2 specialties
- When activate? Triggers
Write 1 paragraph → human/agent reads → decide activation.
Got: Clear paragraph + ≥2 specialties.
If err: Can't ID 2 specialties → no team needed. Use 1 agent w/ multi skills.
Step 2: Lead
Orchestrator. Choose from registry:
- Domain expertise → primary output
- Decompose reqs → subtasks
- Synthesize results → deliverable
# List all available agents
grep "^ - id:" agents/_registry.yml
Lead = also member (lead always member).
Got: Lead chosen + in registry.
If err: No fit → create via create-agent first. Don't create team w/ non-existent lead.
Step 3: Members
2-5 members (incl lead), non-overlap responsibilities. Each:
- id: Agent name
- role: Short title ("Quality Reviewer", "Security Auditor", "Architecture Reviewer")
- responsibilities: 1 sentence, unique
# Verify each candidate agent exists
grep "id: agent-name-here" agents/_registry.yml
Validate no overlap: same primary resp = merge / sharpen.
Got: 2-5 members, unique roles, in registry.
If err: Agent missing → create first. Overlap → rewrite / remove 1.
Step 4: Coord Pattern
| Pattern | Use When | Example |
|---|---|---|
| hub-and-spoke | Lead distrib + collect + synth. Review/audit. | r-package-review, gxp-compliance-validation, ml-data-science-review |
| sequential | Each builds on prior. Pipelines/staged. | fullstack-web-dev, tending |
| parallel | All sim, independent subtasks. No deps. | devops-platform-engineering |
| timeboxed | Fixed iters. Sprint w/ backlog. | scrum-team |
| adaptive | Self-org. Unknown/variable. | opaque-team |
Decision:
- Lead sees all results pre-output → hub-and-spoke
- B needs A's out → sequential
- All work w/o seeing others → parallel
- Iters w/ planning → timeboxed
- Unpredictable struct → adaptive
Got: Pattern + rationale.
If err: Doubt → hub-and-spoke (most common, works for review/analysis).
Step 5: Task Decomposition
Phases:
- Setup: Lead analyzes req + creates tasks
- Execution: Each member's work (par/seq/sprint)
- Synthesis: Collect + deliverable
Per member: 3-5 concrete tasks (typical req). In "Task Decomposition" prose + CONFIG tasks list.
Got: Phase-structured decomp + concrete tasks per member, matches coord pattern.
If err: Vague ("reviews things") → specific ("reviews code style vs tidyverse guide, checks coverage, evaluates err msg quality").
Step 6: Team File
cp teams/_template.md teams/<team-name>.md
Fill in order:
- YAML frontmatter:
name,description,lead,version("1.0.0"),author,created,updated,tags,coordination,members[](id, role, responsibilities) - Title:
# Team Name(readable, title case) - Intro: 1 paragraph
- Purpose: Why exist + specialties
- Team Composition: Table (Member, Agent, Role, Focus)
- Coordination Pattern: Prose + ASCII diagram
- Task Decomposition: Phased + tasks per member
- Configuration: CONFIG block (Step 7)
- Usage Scenarios: 2-3 + example prompts
- Limitations: 3-5
- See Also: Links to members + related
Got: Complete file, no placeholders.
If err: Compare existing (teams/r-package-review.md). Search placeholders ("your-team-name", "another-agent") → unfilled.
Step 7: CONFIG Block
Between <!-- CONFIG:START --> + <!-- CONFIG:END --> → YAML for tooling:
<!-- CONFIG:START -->
```yaml
team:
name: <team-name>
lead: <lead-agent-id>
coordination: <pattern>
members:
- agent: <agent-id>
role: <role-title>
subagent_type: <agent-id> # Claude Code subagent type for spawning
# ... repeat for each member
tasks:
- name: <task-name>
assignee: <agent-id>
description: <one-line description>
# ... repeat for each task
- name: synthesize-report # final task if hub-and-spoke
assignee: <lead-agent-id>
description: <synthesis description>
blocked_by: [<prior-task-names>] # for dependency ordering
```
<!-- CONFIG:END -->
subagent_type → Claude Code types. For .claude/agents/, use agent id. blocked_by = task deps (synth blocked by all reviews).
Got: Valid YAML, agents = frontmatter members, deps = valid DAG (no cycles).
If err: Validate syntax. Every assignee = agent in members. blocked_by refs earlier task names only.
Step 8: Registry
Edit teams/_registry.yml:
- id: <team-name>
path: <team-name>.md
lead: <lead-agent-id>
members: [<agent-id-1>, <agent-id-2>, ...]
coordination: <pattern>
description: <one-line description matching frontmatter>
Increment total_teams (8 → 9 after adding 1).
# Verify the entry was added
grep "id: <team-name>" teams/_registry.yml
Got: Entry in registry, count +1.
If err: Name exists → diff name / update. Verify YAML indent matches.
Step 9: README Automation
Regen:
npm run update-readmes
Updates dynamic sections in teams/README.md + files w/ <!-- AUTO:START --> markers referencing team data.
Got: Exit 0, teams/README.md has new team.
If err: npm run check-readmes → out-of-sync files. Script fail → verify package.json in root + js-yaml installed (npm install).
Step 10: Verify Activation
Test in Claude Code:
User: Use the <team-name> team to <typical task description>
Claude reads teams/<team-name>.md, extracts CONFIG, orchestrates:
TeamCreatew/ name + desc- Spawn teammates via
Agenttool (subagent_typefrom CONFIG) TaskCreatew/blocked_bydeps from CONFIG- Lead coords per pattern
Note: Teams not auto-discovered from .claude/teams/. Claude reads def directly from teams/ on request.
Got: Claude reads file, TeamCreate works, spawns correct agents, follows pattern.
If err: Verify file at teams/<team-name>.md (not subdir). All member agents exist in agents/. CONFIG YAML valid + subagent_type per member. Listed in teams/_registry.yml.
Step 11: Scaffold Translations
Required for all teams. Human + AI authors. Do not skip → backlog.
4 locales post-commit:
for locale in de zh-CN ja es; do
npm run translate:scaffold -- teams <team-name> "$locale"
done
Translate prose (code + IDs EN). Regen:
npm run translation:status
Got: 4 files at i18n/{de,zh-CN,ja,es}/teams/<team-name>.md, source_commit = HEAD. npm run validate:translations → 0 stale.
If err: Scaffold fail → team in teams/_registry.yml. Status stale → run npm run translation:status explicit.
Check
- File at
teams/<team-name>.md - YAML parses
- Required:
name,description,lead,version,author,coordination,members[] - Each member:
id,role,responsibilities - Sections: Purpose, Team Composition, Coordination Pattern, Task Decomposition, Configuration, Usage Scenarios, Limitations, See Also
- CONFIG block between markers
- CONFIG YAML valid
- All member ids in
agents/_registry.yml - Lead in members list
- No 2 members w/ same primary resp
- In
teams/_registry.ymlw/ correct path/lead/members/coord -
total_teams+1 -
npm run update-readmesno err
Traps
- Too many members: >5 → hard coord. Split team / reduce to essentials.
- Overlap resp: 2 "review code quality" → findings conflict, lead wastes time dedup. Distinct focus.
- Wrong coord: hub-and-spoke when need each other's out (should be seq) / seq when can work indep (should be par). Review Step 4.
- Missing CONFIG: Not optional decoration. Machine-readable spec for
TeamCreate/ spawn / tasks. Without → only ad-hoc prose interp (less reliable). - Lead not in members: Lead must appear as member w/ own role + resp. Coord-only lead wastes slot. Give concrete review/synth resp.
→
create-skill— parallel SKILL.md meta-patterncreate-agent— create memberscommit-changes— commit team + registry
GitHub 仓库
Frequently asked questions
What is the create-team skill?
create-team is a Claude Skill by pjt222. Skills package instructions and resources that Claude loads on demand, so Claude can perform create-team-related tasks without extra prompting.
How do I install create-team?
Use the install commands on this page: add create-team to Claude Code as a plugin, or clone its repository into your skills directory, then restart Claude so it picks up the skill.
What category does create-team belong to?
create-team is in the Meta category, tagged automation and design.
Is create-team free to use?
Yes. create-team is listed on AIMCP and free to install. It runs inside Claude, so no separate service account is required to use the skill itself.
相关推荐技能
Content Collections 是一个 TypeScript 优先的构建工具,可将本地 Markdown/MDX 文件转换为类型安全的数据集合。它专为构建博客、文档站和内容密集型 Vite+React 应用而设计,提供基于 Zod 的自动模式验证。该工具涵盖从 Vite 插件配置、MDX 编译到生产环境部署的完整工作流。
这个Claude Skill为开发者提供完整的Polymarket预测市场开发支持,涵盖API调用、交易执行和市场数据分析。关键特性包括实时WebSocket数据流,可监控实时交易、订单和市场动态。开发者可用它构建预测市场应用、实施交易策略并集成实时市场预测功能。
该Skill帮助开发者创建OpenCode插件,用于接入命令、文件、LSP等25+种事件。它提供了插件结构、事件API规范和JavaScript/TypeScript实现模式,适合需要拦截操作、扩展功能或自定义事件处理的场景。开发者可通过它快速构建响应式模块来增强OpenCode AI助手的能力。
SGLang是一个专为LLM设计的高性能推理框架,特别适用于需要结构化输出的场景。它通过RadixAttention前缀缓存技术,在处理JSON、正则表达式、工具调用等具有重复前缀的复杂工作流时,能实现极速生成。如果你正在构建智能体或多轮对话系统,并追求远超vLLM的推理性能,SGLang是理想选择。
