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:agentkit

Skill router and workflow chain selector

Classifies ambiguous or multi-domain work, inventories installed capabilities, chooses the shortest correct skill chain, times subagent use, and applies risk-matched verification without replacing the routed skills.

01

Proportionality gate

02

Classify task

03

Inventory installs

04

Select the chain

Rule 01

This skill decides and dispatches; routed skills and agents own execution.

Rule 02

The best route is often no route: trivial work should not get workflow ceremony.

Rule 03

Installed capability inventory beats memory and wishful routing.

Rule 04

Risk determines verification and review gates; high-risk confidence is not a reason to skip review.

Execution Map

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

01

Start

Input, scope, route

  1. 1 Proportionality gate Skip ceremony for trivial, single-domain, pure conversation, or user-named skill cases; invoke the obvious owner directly.
  2. 2 Classify task Load the task taxonomy and emit class, size, risk, and domain count before shaping a workflow.
02

Work

Agent / skill execution

  1. 3 Inventory installs Use the runtime's installed skill and agent list; missing capability triggers find-skills when available or an explicit gap report.
  2. 4 Select the chain Prefer the user-named skill, then domain-specific owners, then the shortest understand → decide → execute → verify → deliver chain that actually fits.
03

Verify

Gate, review, validation

  1. 5 Time subagents Spawn subagents only when they add fresh context, specialist boundaries, or parallel wall-clock benefit; avoid splitting tiny edits.
  2. 6 Apply risk gates Low risk uses executor checks, elevated risk adds verification and self-review, high risk adds independent review and user confirmation for irreversible steps.
04

Close

Report, handoff, artifact

  1. 7 Recover links A failed link does not advance; fix, debug, or rescope it, and after two consecutive failures report the precise blocker instead of looping.
  2. 8 Report route Finish with delivered outcome, links run, agents used, verification evidence, and any remaining capability gaps.

Syntax, arguments, subcommands

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

Syntax

/ak:agentkit [task to route]

Arguments

[task to route] Required

Task to route

Natural-language work request, outcome, constraints, domains, and authority boundaries. The router uses it to choose installed skills, sequence a short workflow, and match verification to risk.

/ak:agentkit "Add subscription billing and a settings page; require tests and stop before deployment"

Don't skip the required step

Each item contrasts a tempting shortcut with the required approach.

Tempting shortcut

A big-looking task needs every workflow link.

Do this instead

Use the collapse rule; each link needs a real entry and exit reason.

Tempting shortcut

I remember a skill exists.

Do this instead

Inventory the current install; missing skills are reported or discovered, not imagined.

Tempting shortcut

The diff looks clean, so high-risk review can wait.

Do this instead

High-risk routes require independent review before delivery or irreversible action.

Sample Prompt

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

Ambiguous multi-domain feature Recommended 
/ak:agentkit "Add team billing with Stripe and a settings page; require tests and stop before deployment"
Use when:
Use at the start of multi-step or multi-domain work when the right installed skills and their order are unclear.
Expected:
Emits the route class, size, risk, and domain count, inventories installed owners, then builds a short plan such as payment plus frontend work with verification and high-risk review before irreversible steps.
Unsure owner 
/ak:agentkit "Which AgentKit workflow should handle a public API migration?"
Use when:
Use when several installed skills could apply and the decision needs routing, not just a factual answer.
Expected:
Applies the proportionality gate, classifies the migration risk and domains, confirms the installed owner or shortest chain, and names any missing capability instead of inventing one.
Workflow handoff 
/ak:agentkit "Plan and execute a docs + frontend launch page refresh"
Use when:
Use when multiple domains need sequencing, bounded handoffs, and verification, not when one obvious skill owns the whole job.
Expected:
Chooses one owner per link across understand, decide, execute, verify, and deliver stages, spawning subagents only at documented triggers such as broad investigation or completed implementation.
Collapse obvious work 
/ak:agentkit "Fix the failing CI on this branch"
Use when:
Use when a task looked like it might need routing but should collapse if one installed skill clearly owns it.
Expected:
Runs the proportionality gate first, recognizes the single-domain obvious owner, routes directly to the fix skill, and avoids an unnecessary multi-link chain or subagent split.

Handled Scope

  • Proportionality gate
  • Classify task
  • Inventory installs
  • Select the chain
  • Time subagents
  • Apply risk gates

Next