planlisted
Install: claude install-skill lacerbi/dotclaude
You are planning, not executing. Explore and analyze first; probing is part of that — run scripts,
tests, or quick experiments where they resolve a real uncertainty about the task. Don't start on the
task itself until the user confirms the plan.
## Phase 1: Task Understanding
State your understanding of the task: $ARGUMENTS
If unclear, use AskUserQuestion to clarify before proceeding.
**Assess complexity** to calibrate planning depth:
- **Simple**: Single-focus task, few components → lighter exploration, concise plan
- **Moderate**: Multiple components or unknowns → targeted exploration, structured plan
- **Complex**: Cross-cutting concerns, many dependencies → thorough exploration, detailed plan
## Phase 2: Exploration
Explore the project to understand what exists and what the task involves. Scale exploration to task complexity.
**For simple tasks**: You may explore directly without subagents.
**For moderate/complex tasks**: Spawn Explore agents in parallel (Agent tool, subagent_type='Explore'), each focused on a relevant aspect:
- **Structure Explorer**: Project organization, key components, how things connect
- **Prior Work Explorer**: Existing similar work, patterns, or approaches to build on
- **Context Explorer**: Related materials, dependencies, constraints
- **Domain Explorer**: Subject matter, conventions, standards relevant to the task
- **Docs Explorer**: Documentation (READMEs, guides, API docs, comments) — what exists in or adjacent to the task area, wha