Claude Plan With Me Skill

plan-with-me

A Claude skill that turns plan mode into a structured, repo-grounded design conversation instead of a quick sketch.

Before writing a single line of a plan, it researches the current (or a named target) repository thoroughly, then **always** asks several high-leverage clarifying questions at a system-design / senior-software-engineer level — each with multiple concrete, well-reasoned options and tradeoffs, never a bare yes/no. Only after you answer does it draft the implementation plan, and it always requests explicit approval before any implementation begins.

Why

Left to its own devices, plan mode tends to do one of two things: write a plan straight away and let architectural decisions get made implicitly by whichever approach occurred to it first, or ask one or two shallow clarifying questions that aren't really grounded in the codebase. Neither gives you a real say in the decisions that are expensive to reverse later — the data model, the API shape, how failures are handled, what gets tested.

This skill slows that moment down, deliberately, the way a senior engineer would if a junior engineer said "I'm going to go build X." It does the reading a good engineer does, and asks the questions a good engineer asks — as concrete, opinionated options, not open-ended prompts that push the design work back onto you.

What it does

1. Resolve scope — figures out which repo and which request it's actually planning against.
2. Research — reads the relevant parts of the codebase (via Glob/Grep/Read, or a research subagent for large repos): tech stack, existing conventions, analogous features, testing setup.
3. Ask clarifying questions — a small, adaptive set of questions (2-3 for a focused change, 6+ for a bigger design), each posed as multiple choice with real tradeoffs and a recommended default, grounded in what research turned up. See references/question-playbook.md for the categories it draws on: architecture & structure, data model &