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:predict

Multi-persona pre-analysis

Five expert personas independently analyze a proposed change, debate conflicts, and produce a GO, CAUTION, or STOP verdict before implementation starts.

01

Read proposal

02

Read supplied code

03

Analyze independently

04

Find agreements

Rule 01

Use before major features, significant refactors, architecture changes, competing approaches, or assumption stress tests

Rule 02

Do not use for trivial changes, already-approved work, or pure dependency upgrades with no API changes

Rule 03

Keep the five persona analyses independent before collecting agreements and conflicts

Rule 04

Resolve conflicts by worst plausible case and output actionable mitigations

Execution Map

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

01

Start

Input, scope, route

  1. 1 Read proposal Start from the feature description or change proposal provided as the command argument.
  2. 2 Read supplied code When --files is provided, inspect the relevant code paths and grep affected areas before judging the change.
02

Work

Agent / skill execution

  1. 3 Analyze independently Architect, Security, Performance, UX, and Devil’s Advocate each review the proposal without influencing one another.
  2. 4 Find agreements Collect points where all personas, or at least four of five personas, align.
03

Verify

Gate, review, validation

  1. 5 Find conflicts List meaningful disagreements across architecture, security, performance, UX, and hidden-assumption concerns.
  2. 6 Weigh tradeoffs Resolve each conflict by comparing the worst plausible case for each option, not only the expected happy path.
04

Close

Report, handoff, artifact

  1. 7 Produce verdict Return GO, CAUTION, or STOP with agreements, conflict resolutions, risk severity, early signals, mitigations, and recommendations; then run the selected chain mode if requested.

Syntax, arguments, subcommands

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

Syntax

/ak:predict <feature description or change proposal> [--files <glob>] [--chain reason|probe]

Arguments

<feature description or change proposal> Required

Proposal to review

Required feature description, refactor, competing approach, or design proposal to stress-test before implementation. Include outcome, scope, constraints, assumptions, evidence, and the decision the verdict should unblock.

/ak:predict "Add WebSocket support for real-time notifications"

Shared options

--files <glob>

Bounded code evidence

Read matching relevant code before the persona debate. Keep the glob narrow; this option grounds claims in current project evidence but does not prove runtime behavior, load limits, or exploitability.

/ak:predict "Replace REST API with GraphQL" --files src/api/**/*.ts
--chain reason

Subjective refinement

After the predict verdict, run the supported generate, critique, synthesize, and blind-judge loop when a CAUTION verdict depends on subjective design or architecture tradeoffs. It is part of the predict session, not a standalone skill.

/ak:predict "Pick auth library: Passport vs Better Auth" --chain reason
--chain probe

Constraint interrogation

After the predict verdict, harvest missing constraints and unstated assumptions when CAUTION or STOP comes from unclear requirements. Use new evidence or a viable mitigation before moving past a STOP trigger.

/ak:predict "Move from REST to GraphQL" --chain probe

Modes / Routes

--files <glob>

Attach known code paths so the prediction reads affected areas before the persona debate.

Research:
code read
Review:
persona debate
Validate:
verdict
--chain reason

After the verdict, run subjective refinement when CAUTION depends on architecture polish or design-coherence tradeoffs.

Research:
refinement loop
Review:
blind judge
Validate:
convergence
--chain probe

After the verdict, interrogate missing constraints and unstated assumptions when CAUTION or STOP comes from incomplete requirements.

Research:
requirement interrogation
Review:
assumption harvest
Validate:
saturation

Sample Prompt

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

Major feature Recommended 
/ak:predict "Add WebSocket support for real-time notifications"
Use when:
Use before implementing a major or high-risk feature.
Expected:
The skill runs independent Architect, Security, Performance, UX, and Devil’s Advocate analyses, resolves agreements and conflicts, then returns a GO, CAUTION, or STOP report with risk severities, early signals, mitigations, and recommendations.
Code-aware architecture change 
/ak:predict "Replace REST API with GraphQL" --files src/api/**/*.ts
Use when:
Use when affected files are known and should ground the persona debate.
Expected:
The prediction first reads the supplied code paths and affected areas, then each persona judges the proposal against the existing implementation before producing conflicts, resolutions, and a verdict.
Competing approaches 
/ak:predict "Pick auth library: Passport vs Better Auth" --chain reason
Use when:
Use when evaluating competing technical approaches whose tradeoffs are subjective.
Expected:
After the normal persona verdict, --chain reason runs the supported generate, critique, synthesize, blind-judge refinement loop until the subjective tradeoffs converge or the remaining disagreement is explicit.
Assumption stress test 
/ak:predict "Move from REST to GraphQL" --chain probe
Use when:
Use when stress-testing assumptions in a proposed design, especially if missing constraints may change the answer.
Expected:
After the GO, CAUTION, or STOP verdict, --chain probe harvests missing constraints and unstated assumptions so the proposal can be clarified before planning or implementation continues.

Handled Scope

  • Read proposal
  • Read supplied code
  • Analyze independently
  • Find agreements
  • Find conflicts
  • Weigh tradeoffs

Artifacts Produced

Prediction Report

Pattern

Verdict: GO | CAUTION | STOP

Outputs agreements, a conflicts-and-resolutions table, a risk summary with severity, early signal, and mitigation columns, plus numbered recommendations.

Next