CLAUDE AGENTS TEAM

AGENT TEAM GENERATOR

Claude Agents Team

Hand it a product description and it writes the whole .claude directory — orchestrator, specialists, skills, rules, hooks — then grades every file before the run exits.

team-gen · E-Shopself-eval
  • shop-orch.md
    A 100 · ship
  • api-engineer.md
    A 97 · ship
  • qa-engineer.md
    A 97 · ship
claude plugin install ./agents-team/plugins/agents-team
Read the docs
  • Claude Code plugin
  • 8 agent + 4 skill archetypes
  • 13 rule + 4 hook templates
  • Citation-backed rubric
  • MIT licensed
  • Sponsorable on GitHub

Features

/team-gen

A product spec in, a complete standards-compliant team out: orchestrator, specialists, skills, rules and hooks.

/meet

A streamed multi-agent meeting on one topic, ending in structured minutes you can act on.

/evaluate-agent

Score any agent or skill against a rubric where every line cites a published spec or prior art.

Graded before it ships

The generator lints every file it produced. One reject verdict fails the whole run.

How it grades

01

Static

  • static linter
02

Model

  • LLM-as-judge
03

Behaviour

  • behavioral harness

Three pillars

  1. 01

    Generate a team

    /team-gen ./my-product-spec.md --target .
  2. 02

    Hold a meeting

    /meet "Should we ship the v2 migration this quarter?"
  3. 03

    Lint an agent

    /evaluate-agent .claude/agents/code-reviewer.md

Why it exists

Three large agent teams — sentra-hub at ~14 agents, orch at ~42, the Orchestra team at ~40 — were hand-rolled from scratch. This plugin extracts the patterns that worked and refuses the ones below standard.

Community

GitHubmaintained