hi-scout

name	hi:scout
description	Fast codebase scouting using parallel agents. Use for file discovery, task context gathering, searching across directories.
argument-hint	[search-target] [ext]
metadata	{"author":"baka3k","version":"2.0.0"}

Scout

Fast, token-efficient codebase scouting using parallel agents.

Arguments

• Default: Explore subagents in parallel
• ext: External Gemini/OpenCode CLI tools

Workflow

1. Analyze

Parse user prompt for targets, key directories, patterns, file types. Determine SCALE (subagent count).

2. Divide

Split codebase logically. Each agent gets distinct scope, no overlap.

3. Register Tasks

Skip if <=2 agents or Task tools unavailable. TaskList -> use existing. Else TaskCreate per agent with scope metadata. Metadata: agentType, scope, scale, agentIndex, totalAgents, toolMode, priority:P2, effort:3m. See references/task-management.md for patterns.

4. Spawn

Load internal (references/internal-scouting.md) or external (references/external-scouting.md). TaskUpdate each to in_progress before spawning. Max 3m per agent. Skip timed-out.

5. Collect

Aggregate: deduplicate paths, merge descriptions, note gaps/timeouts, list unresolved.

Report Format

# Scout Report
## Relevant Files
- path/to/file.ts - description
## Unresolved Questions
- any gaps

Configuration

Read .claude/.ck.json: gemini.model (default: gemini-3-flash-preview).

Invoke /hi:project-organization to organize outputs.