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:scout

Token-efficient codebase scout

Fast, token-efficient codebase scouting with native search, scoped reads, permitted Explore agents, optional OpenCode probes, and best-of-5 verifier mode.

01

Parse target

02

Search portably

03

Divide scopes

04

Register scout work

Rule 01

Scouting finds files and context; it does not implement the change

Rule 02

Portable search and scoped reads come before agents

Rule 03

Delegation must be permitted and worth the overhead; --ultra fails closed

Rule 04

Reports should list relevant files, relationships, timed-out scopes, and honest gaps

Execution Map

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

01

Start

Input, scope, route

  1. 1 Parse target Parse the user prompt for search targets, key directories, patterns, file types, lines of code, and the right subagent scale.
  2. 2 Search portably Use search_files broadly first, then scoped read_file and small run_shell facts for local discovery.
02

Work

Agent / skill execution

  1. 3 Divide scopes Split the codebase into non-overlapping logical segments only when delegation is permitted and more than two agents are worth the overhead.
  2. 4 Register scout work Register one concise scoped work item per agent in the live task-management surface when available; otherwise update the active plan.
03

Verify

Gate, review, validation

  1. 5 Run scouts Load internal scouting for permitted Explore subagents, or external scouting for user-permitted OpenCode probes when native/local search is insufficient.
  2. 6 Collect quickly Wait up to three minutes per agent, skip non-responders, and record timed-out scopes honestly.
04

Close

Report, handoff, artifact

  1. 7 Report findings Organize outputs with project-organization, then aggregate relevant files with one-line roles, relationships, timed-out scopes, and unresolved questions.

Syntax, arguments, subcommands

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

Syntax

/ak:scout [search-target] [ext] [--ultra]

Arguments

[search-target] Required

Search target

Bounded behavior, symbol, feature, configuration, relationship, directory, or file type to map. Include scope, exclusions, expected evidence, and whether delegation or external probing is allowed when that matters.

/ak:scout "Find the editor draft persistence entry points, storage adapter, recovery UI, and focused tests. Search src/ and tests/ only."

Shared options

ext

External probe

Use user-permitted OpenCode probes only after native/local search is insufficient. It does not authorize sending secrets, unrelated private files, or broad unbounded scopes.

/ak:scout "Find payment webhook ownership across api/ and lib/; read-only" ext
--ultra

Verifier scout

Run exactly five parallel read-only candidate scout passes from the same evidence packet, then have a strongest-model verifier return only evidence-validated findings. Hard-stops if that dispatch is unavailable.

/ak:scout "Find payment webhook ownership across api/ and lib/; read-only" --ultra

Modes / Routes

default

Internal scouting

Research:
search_files, scoped read_file, then permitted Explore agents
Review:
None
Validate:
Project-organization plus concise Scout Report
ext

External OpenCode probes

Research:
User-permitted OpenCode probes after native/local search falls short
Review:
None
Validate:
Merged external scout summaries with named gaps
--ultra

Best-of-5 scout

Research:
Exactly five read-only candidate reports from one evidence packet
Review:
Strongest-model verifier scores coverage and evidence
Validate:
Evidence-validated, deduplicated union of findings

Sample Prompt

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

Find feature files Recommended 
/ak:scout payment webhook handlers
Use when:
Use at the start of a multi-directory feature or bug investigation.
Expected:
Scout searches broadly, reads scoped files, optionally uses permitted Explore agents, and returns relevant files with roles and open questions.
External probe 
/ak:scout renderer package boundaries ext
Use when:
Use when native/local search is insufficient and user-permitted OpenCode probing is allowed.
Expected:
Scout loads the external scouting path, runs scoped OpenCode probes, then merges their summaries into one report with named gaps.
Verifier scout 
/ak:scout routing layer --ultra
Use when:
Use when file discovery is high-stakes and the runtime can dispatch five parallel read-only candidates.
Expected:
Scout builds one evidence packet, runs exactly five candidate reports, and emits the verifier's validated deduplicated union of findings.

Handled Scope

  • Parse target
  • Search portably
  • Divide scopes
  • Register scout work
  • Run scouts
  • Collect quickly

Artifacts Produced

Scout Report

Pattern

Chat report or organized report path after project-organization is invoked

Relevant files with one-line roles, relationships, unresolved questions, and timed-out scopes named instead of hidden.

Next