VividKit Crystal clear AI coding
VI
Choose Your Setup Get Started with AgentKit Continue with ClaudeKit Migrate from ClaudeKit to AgentKit Clean cutover from ClaudeKit What Is AgentKit? How AgentKit Works AgentKit Kits, Skills, Agents & Hooks AgentKit Runtime Support AgentKit Desktop App AgentKit Configuration AgentKit CLI Quick Reference AgentKit Skills Cheatsheet AgentKit Workflows AgentKit Statusline Update AgentKit CLI & Kits AgentKit Helper Clean reinstall of AgentKit AgentKit Troubleshooting What is ClaudeKit? ClaudeKit CLI Guide Command Finder Commands Cheat Sheet Workflow Recipes CLI Quick Reference Claude Mechanics Permission Modes ClaudeKit Hooks Coexistence ClaudeKit x Codex CK → Other coding tools How ClaudeKit Works Inside ClaudeKit Finding Your Unknowns Session Recovery Fix from Logs UI/UX Guide Accounts Switcher IDE Config Remote Control Codex App System Diagramming UI Review Gate Promotions Support VividKit

Quick Ref / Command

Engineer Kit

/ak:team

Agent Teams

Orchestrates coordinated multi-session Agent Teams for independent research, implementation, review, and debug workstreams through a lead, shared work state, and direct communication.

01

Discover live surface

02

Decompose work

03

Create team

04

Register work

Rule 01

Use a live team surface or abort

Rule 02

The lead owns decomposition, integration, verification, shutdown, and cleanup

Rule 03

Use the smallest team that creates genuinely independent workstreams

Rule 04

Idle is not complete, and teammate completion is not proof

Execution Map

Core lanes from input to output, aligned with the command's real execution path.

01

Start

Input, scope, route

  1. 1 Discover live surface Resolve the available lifecycle, shared-state, delegation, messaging, wait, approval, shutdown, and cleanup capabilities from the current runtime.
  2. 2 Decompose work Choose research, cook, review, or debug only when the request can be split into independent scopes with explicit outputs.
02

Work

Agent / skill execution

  1. 3 Create team Create the team through the discovered lifecycle surface instead of relying on cached command names or ordinary subagent fallbacks.
  2. 4 Register work Record work items with acceptance criteria, dependencies, owners, and durable report or plan destinations.
03

Verify

Gate, review, validation

  1. 5 Spawn teammates Delegate through the live surface, run independent scopes concurrently, and isolate concurrent writers with worktrees when supported and needed.
  2. 6 Coordinate state Use direct and team-wide messages, waits, and shared-state inspection; treat idle as waiting rather than completion.
04

Close

Report, handoff, artifact

  1. 7 Integrate and verify Integrate reports or branches only after prerequisites complete, then verify the combined result because completion messages are not proof.
  2. 8 Shutdown and clean up Collect acknowledgements or concise handoffs, shut down teammates gracefully, invoke runtime cleanup, and finish with the required journal step unless opted out.

Syntax, arguments, subcommands

How to invoke the skill: syntax, positional arguments, shared options, then each subcommand with its own syntax and outcome.

Syntax

/ak:team <template> <context> [--devs N|--researchers N|--reviewers N|--debuggers N] [--plan-approval|--no-plan-approval] [--delegate] [--worktree]

Arguments

<template> Required

Team template

Required workflow template: research, cook, review, or debug. It selects the teammate pattern; it does not prove the live team surface exists.

/ak:team review "Review checkout failure paths" --reviewers 3
<context> Required

Team context

Required outcome, scope, ownership, dependencies, evidence format, and authority boundary for the lead to decompose into independent teammate work.

/ak:team cook "Implement accepted auth plan; checkout owns src/checkout, settings owns src/settings" --devs 2

Shared options

--devs N

Developer count

Requests N implementation teammates for cook. Use only when implementation ownership can be split without overlapping files.

/ak:team cook "Implement dashboard plan" --devs 3
--researchers N

Researcher count

Requests N independent research angles before the lead synthesizes trade-offs, recommendations, and unresolved questions.

/ak:team research "Compare billing providers" --researchers 3
--reviewers N

Reviewer count

Requests N read-only review focuses with severity, evidence, impact, and recommendation for every finding.

/ak:team review "Review checkout for security, accessibility, and failures" --reviewers 3 --delegate
--debuggers N

Debugger count

Requests N competing, testable root-cause hypotheses with evidence for and against each one.

/ak:team debug "Diagnose intermittent checkout timeout" --debuggers 3
--plan-approval

Require plan approval

Keeps implementation teammates read-only until their scope, ownership, validation, risks, and rollback notes are approved through the live approval surface.

/ak:team cook "Implement accepted plan" --devs 3 --plan-approval
--no-plan-approval

Skip plan approval gate

Omits the dedicated plan gate. Normal scope, ownership, safety, and verification boundaries still apply.

/ak:team cook "Implement accepted plan" --devs 2 --no-plan-approval
--delegate

Delegate lead actions

Keeps the lead in coordination, approval, synthesis, and reporting mode without implementation edits, commands, tests, or merges.

/ak:team review "Audit checkout changes" --reviewers 4 --delegate
--worktree

Request worktree isolation

Requests isolated worktrees for implementation teammates when the live runtime supports them. It does not replace distinct file ownership.

/ak:team cook "Implement dashboard groups" --devs 3 --worktree

Modes / Routes

research

Research

Research:
Distinct angles
Review:
Alternatives and risks
Validate:
Comparison report
cook

Cook

Research:
Accepted plan
Review:
Ownership conflicts
Validate:
Combined tests and review
review

Review

Research:
Read-only scopes
Review:
Evidence and impact
Validate:
Ordered action list
debug

Debug

Research:
Competing hypotheses
Review:
Evidence against
Validate:
Root-cause report

Sample Prompt

Concrete invocations for each flag, subcommand, mode, or route available in the live workflow.

Research team Recommended 
/ak:team research compare billing providers --researchers 3
Use when:
Use for coordinated multi-session research when distinct angles can run independently.
Expected:
Discovers the live team surface, registers one owned research item per angle, waits for reports and shared state, then synthesizes recommendations and unresolved questions.
Cook team with approval 
/ak:team cook implement accepted auth plan --devs 3 --plan-approval --worktree
Use when:
Use for coordinated multi-session implementation when an accepted plan can split into non-overlapping ownership.
Expected:
Keeps developers read-only until scoped plans are approved, spawns isolated implementation work when supported, integrates completed branches, runs combined gates, and records docs impact.
Review team 
/ak:team review audit checkout changes --reviewers 4
Use when:
Use for coordinated multi-session review when security, performance, tests, and architecture can be separate focuses.
Expected:
Registers read-only reviewer scopes, requires severity with evidence and recommendations, waits for all required work, deduplicates disagreements, and produces an ordered action list.
Debug team 
/ak:team debug diagnose intermittent checkout timeout --debuggers 3
Use when:
Use for coordinated multi-session debugging when competing hypotheses predict different observable evidence.
Expected:
Assigns one debugger per hypothesis with evidence-for and evidence-against requirements, encourages direct challenges, identifies the surviving theory, and writes a durable root-cause report.

Handled Scope

  • multi-session teams
  • parallel research
  • delegated implementation
  • evidence-based review
  • competing debug hypotheses
  • worktree isolation

Related skills

Next