Plan For OpenClaw
Convert intent into a sequence the agent can execute without losing rigor.
Process
- Read the relevant local rules.
- Decide whether the task qualifies as a long task.
- Create or update
plan.mdwhen the workspace requires it. - Break the work into thin vertical slices.
- Order slices by dependency and risk.
- Define what evidence will mark each slice complete.
- Keep the plan updated during execution.
Planning Heuristics
- Prefer tasks that can be verified independently.
- Avoid steps that combine refactor, feature work, and cleanup in one block.
- Put the highest-uncertainty slice early enough to de-risk the task.
- Keep wording operational, not aspirational.
Red Flags
- A task list that is really just a vague summary
- Large steps with no verification story
- A plan that stops being updated after work begins