| name | decomposition |
| description | This skill should be used when the user asks to "decompose", "break down task", "create detailed todos", "split into todos". Also triggers when the user says in Japanese: "タスク分解", "todo に落として", "細かい todo にして". |
Decomposition (Cursor Skill)
Break a complex task into rich, self-contained todos that can be executed independently. Run a single pass: explore, identify components, optionally ask one round of clarifying questions, then write todos via the agent's todo tool and append a summary to the active plan file.
Persistence Target
Persist the decomposition summary to the active Cursor plan file:
- Look up the currently active plan via the agent's plan tools (e.g.
read_plan / update_plan).
- If no plan is active, create
./PLAN.md and treat that as the active plan.
- Always append a
## Decomposition Summary section after writing the todos.
Process (single pass)
Step 1. Explore the codebase
Build a concrete mental model before decomposing anything:
- Read the current task or goal from the conversation context and the active plan file.
- Read
CLAUDE.md and / or AGENTS.md if present.
- Read related code, configs, or specs the task references.
- Map the directory layout for affected modules.
- Identify existing patterns for similar features (so todos can reference them).
Do not skip this step — todos generated without exploration are usually too vague to execute.
Step 2. Identify major components
Decompose the overall task into distinct components:
- What are the phases or areas of work?
- What are the dependencies between components?
- What is the natural execution order?
Keep this internal — the next steps turn it into todos.
