plan-writing — Claude Code skill: task plans with checks

What plan-writing does

The skill gives the agent a frame for planning: one task equals one clear action of 2–5 minutes, verifiable on its own. A plan opens with the goal in a single sentence, then lists tasks, and closes with done criteria. The file lands in the project root as {task-slug}.md — adding auth becomes auth-feature.md. Not inside .claude/, docs/, or temp folders.

There are no fixed templates on purpose. For a new project the plan asks about the stack and the smallest useful feature set; for a bug it asks about the root cause and the line to change; for a feature, which files are affected. Keep it to ten tasks or fewer — beyond that, split the work into several plans. Verification stays concrete: a command to run or a click to make, not a vague "check it works".

Only Read, Glob and Grep are allowed. So the skill writes the plan and leaves the code alone — edits stay with you. If you expect the agent to start executing the checklist, look elsewhere.