| name | explore-codebase |
| description | Pattern for efficiently exploring codebases using parallel subagents. Use when you need to understand code structure, find patterns, or gather context. |
| allowed-tools | Task, Read, Glob, Grep |
Explore Codebase Skill
Pattern for exploring codebases efficiently with parallel subagents.
When to Load This Skill
- You need to understand parts of the codebase
- You're gathering context before making decisions
- You need to find patterns, conventions, or existing implementations
Exploration Strategy
For Quick Searches (Do Yourself)
Glob("**/*{keyword}*")
Grep(pattern="functionName", type="ts")
Use when looking for a specific thing.
For Broad Understanding (Spawn Explorers)
Spawn parallel explorer subagents:
Task(
subagent_type: "explorer", # Custom dotagent agent (lowercase)
model: "haiku",
prompt: |
Query: {specific question}
Hints: {where to look}
Scope: {limit if needed}
Return structured YAML:
- findings (location, relevance, summary)
- patterns_observed
- related_areas
- gaps
)
Spawn multiple for different areas simultaneously.
Note: Use "explorer" (lowercase) for the custom dotagent agent defined in
@.claude/agents/explorer.md. The built-in "Explore" (capitalized) is a different
Claude Code agent with simpler behavior.
Synthesize Results
Combine explorer outputs into coherent understanding:
- Common patterns across areas
- Integration points
- Constraints discovered
- Gaps to note
Output Format
After exploration, summarize as:
codebase_context:
architecture_summary: string
existing_patterns: [string]
integration_points: [string]
constraints: [string]
gaps: [string]
Principles
- Parallel over sequential - Spawn multiple explorers at once
- Breadth over depth - Explore broadly first
- Structured output - Always produce YAML summaries
- Note gaps - What you couldn't find matters