| name | explore |
| description | Deep exploration of codebase, docs, and web for any topic or question |
| argument-hint | <topic or question> |
Explore #$ARGUMENTS in depth using a structured, parallel approach.
Process
-
Plan (30 seconds max):
- Break the topic into 2-3 specific sub-questions
- Identify what to search: codebase, docs, web, or all
-
Explore (parallel agents):
- Launch 1-2 agents maximum for codebase search
- Launch 1 agent for web/docs if library-specific knowledge needed
- Each agent has a focused question, not a vague "look around"
-
Synthesize:
- Merge findings from all agents
- Identify conflicts or gaps
- Produce a structured report
Report format
Exploration: [topic]
Key findings
- [Finding 1 — with file:line references where applicable]
- [Finding 2]
Architecture / Patterns
- [How the codebase handles this]
Recommendations
- [If the user asked "how should we..." — concrete recommendation]
- [If the user asked "how does..." — no recommendation needed]
Files explored
- path/to/file — [what was found]
Rules
- NEVER explore without a plan — even 30 seconds of planning saves 5 minutes of wandering
- Max 2 parallel agents during explore — more creates context fragmentation
- Always include file:line references — vague answers are worthless
- If the topic is too broad, narrow it and tell the user what you narrowed
- If stuck after 2 minutes: stop exploring, report what you found, ask the user to refine