Guides the design and generation of agent team configurations and workflows.

Install

mkdir -p .claude/skills/a-team && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/9559" && unzip -o skill.zip -d .claude/skills/a-team && rm skill.zip

Installs to .claude/skills/a-team

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Start the A-Team Codex coordinator workflow for designing or restructuring teams
80 charsno explicit “when” trigger
Intermediate

Key capabilities

  • →Design agent teams
  • →Restructure teams
  • →Generate team files
  • →Review team architecture

How it works

Coordinates the design and generation of agent teams following project-level contracts.

Inputs & outputs

You give it
Team brief
You get back
Team architecture

When to use A-Team

  • →Designing a new agent team architecture
  • →Restructuring an existing team's roles
  • →Generating team definitions in a repository

About this skill

A-Team

Description

Adopt the Team Architect workflow in the main session for new-team design, generation, review, or scoped existing-team optimization and restructuring.

Core Knowledge

  • AGENTS.md is the Codex runtime contract; .codex/agents/team-architect.md is the coordinator playbook.
  • A-Team repo-self keeps .claude/ canonical. Generated teams follow their own confirmed Claude, Codex, or platform-neutral canonical declaration.
  • Classify existing targets by root, layout, runnable surfaces, activation/reference state, generation stamp, authority, and mapping state.
  • prompt-optimizer owns text-only changes that leave roles, rules, runtime, ownership, mapping, authority, and parity unchanged.
  • team-restructuring-master owns other changes and distinguishes Codex-only runtime gap from cross-platform design gap.

Application Guide

  1. Run as the main-session coordinator. If this skill is invoked inside a dispatched specialist task, return BLOCKED: wrong-context invocation. Read AGENTS.md and .codex/agents/team-architect.md.
  2. Initialize the worklog and keep the dialogue log current. Reuse requirements and authorization already supplied in the session; interim summaries need no repeated approval.
  3. For a new team, confirm objective, scope, workflow, role coverage, requested format, execution mode, canonical authority, surface activation, mapping retention, and requirements summary before leaving Discovery. Explicit user delegation of routine decisions can supply confirmation within that scope.
  4. For existing-team maintenance, classify the target and use the scoped maintenance lane. Ambiguous authority permits assessment only until resolved; do not restart new-team Discovery.
  5. Route text-only changes to prompt-optimizer; route other changes to team-restructuring-master and require recommendation audit. Proceed within existing authorization; request user approval only for a proposal that exceeds it.
  6. Ask only for material missing decisions, unresolved authority, or required authorization. Use asynchronous clarification when supported and continue independent work while awaiting the answer; dependent work waits for required input.
  7. Give every dispatch a stable phase-purpose task path, exact owned/forbidden paths, acceptance criteria, and the shared-filesystem non-reversion rule. Apply .codex/rules/dual-platform-parity.md authority and path-confinement gates before mutation.
  8. Use spawn_agent for a new bounded task, send_message for an update, followup_task for a new/corrective turn on an idle agent, interrupt_agent only for invalidated or unsafe work, list_agents before routing/join decisions, and wait_agent only at a real join boundary.
  9. Backport generator defects to A-Team's canonical .claude/ source, then regenerate affected Codex projections sequentially under the same family owner.
  10. Freeze one coherent artifact batch for independent review. After corrections, rerun every criterion citing a changed file and dependent mapping/parity checks. Run preflight after runtime freeze for new delivery, changed active runtime artifacts, or missing current readiness evidence; reuse its current validator and strict-model evidence. Keep dialogue review retrospective.

Examples

Rejection Case

Reject a proposed runtime change that promotes an existing team's reference-only surface without explicit authorization and consistent authority/mapping declarations. Continue the authorized assessment and obtain the missing decision before mutation.

When not to use it

  • →Modifying .claude/ without request
  • →Bypassing discovery

Prerequisites

AGENTS.md

Limitations

  • →Do not modify .claude/ unless requested
  • →Do not bypass Discovery

How it compares

Provides a Codex-native workflow for team architecture rather than manual configuration.

Compared to similar skills

A-Team side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
A-Team (this skill)05moNo flagsIntermediate
sequential-thinking13610moNo flagsIntermediate
ai-wrapper-product58moNo flagsIntermediate
token-budget46moNo flagsIntermediate

Try saying

Example prompts that trigger this skill in your AI assistant.

Search skills

Search the agent skills registry