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

Marketing Kit

/ak:fix

Evidence-first bug fixing

Routes concrete bugs, errors, test failures, lint issues, UI defects, and CI failures through scout, diagnosis, fix, verification, and prevention.

01

Frame intent

02

Select mode

03

Scout context

04

Diagnose cause

Rule 01

Symptoms do not justify fixes; exact root cause does.

Rule 02

The original failing repro is the baseline for verification.

Rule 03

A fix is incomplete until blast-radius side effects are checked.

Rule 04

Three failed attempts mean the approach or architecture must be questioned.

Execution Map

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

01

Start

Input, scope, route

  1. 1 Frame intent Capture outcome, constraints, non-goals, and acceptance criteria before selecting a mode.
  2. 2 Select mode Use autonomous by default, review for high-risk work, quick for trivial lint or type errors, parallel for independent issues, or ultra for plan selection.
02

Work

Agent / skill execution

  1. 3 Scout context Map project type, symptom files, callers, tests, recent commits, and existing patterns before forming hypotheses.
  2. 4 Diagnose cause Capture the exact symptom and reproduction, then prove root cause, why now, and blast radius with file evidence.
03

Verify

Gate, review, validation

  1. 5 Assess complexity Classify simple, moderate, complex, parallel, or ultra, then pick a cause-aligned repair path.
  2. 6 Implement fix Apply minimal source changes that address the root cause while preserving non-goals and existing patterns.
04

Close

Report, handoff, artifact

  1. 7 Verify prevent Rerun the original reproduction, affected tests, blast-radius checks, and add prevention where appropriate.
  2. 8 Finalize Report confidence, root cause, changed files, evidence, prevention, side-effect sweep, docs impact, and journal handling.

Modes / Routes

--auto

Autonomous

Research:
Standard scout
Review:
Internal review
Validate:
Full affected verification
--review

Human review

Research:
Standard scout
Review:
Pause at decision points
Validate:
Approval plus affected verification
--quick

Quick

Research:
Minimal scout
Review:
Trivial issue only
Validate:
Exact repro plus direct dependents
--parallel

Parallel

Research:
Per-issue scout
Review:
Independent ownership
Validate:
Per-issue and integrated checks
--ultra

Ultra verifier

Research:
Shared diagnosis packet
Review:
Five read-only fix plans
Validate:
Winning plan only then full verify
--advice

Advisory

Research:
Selected base mode
Review:
kongming advisory supervision
Validate:
Base mode verification; advice cannot approve gates or edit code

Don't skip the required step

Each item contrasts a tempting shortcut with the required approach.

Tempting shortcut

I can see the problem, let me fix it.

Do this instead

Seeing symptoms is not root-cause evidence. Scout first.

Tempting shortcut

It is probably this line.

Do this instead

Probably means guessing. Prove the defect, trigger, and blast radius.

Tempting shortcut

Tests passed, so the fix is done.

Do this instead

The original repro, affected modules, public contracts, and prevention gate still matter.

Sample Prompt

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

Default fix Recommended 
/ak:fix "checkout submit throws TypeError" --auto
Use when:
A concrete bug can be handled autonomously.
Expected:
Frames intent, scouts the failing area, diagnoses root cause, implements the fix, verifies, and finalizes.
Trivial issue 
/ak:fix "eslint no-unused-vars in src/app.ts" --quick
Use when:
The issue is a simple lint or type failure.
Expected:
Uses the quick scout to diagnose and repair the issue without skipping focused verification afterward.
Human-in-loop 
/ak:fix "production payment regression" --review
Use when:
The bug touches critical or production-sensitive code.
Expected:
Pauses at review gates before risky decisions so the user can approve material scope changes.
Independent issues 
/ak:fix "CI failures in auth and billing" --parallel
Use when:
Two or more failures can be owned independently.
Expected:
Routes each independent issue to parallel fullstack-developer ownership with clear boundaries and merge expectations.
Ultra verifier 
/ak:fix "race in checkout retry" --ultra
Use when:
Fix selection is high-stakes and should use a verifier pass.
Expected:
Runs one diagnosis packet, five read-only candidate plans, verifier selection, then a single implementation path.
Advisory supervision 
/ak:fix "production payment regression" --advice
Use when:
You want kongming counsel after phases without letting advice edit code.
Expected:
Keeps the evidence-first fix path, adds advisory checkpoints, and still requires root-cause proof plus verification.

Handled Scope

  • Frame intent
  • Select mode
  • Scout context
  • Diagnose cause
  • Assess complexity
  • Implement fix

Related skills

Next