Make a Plan

Install

npx skills@latest add doughawk25/claude-skills --skill make-a-plan

Then type /make-a-plan in your coding agent.

What it does

Make a Plan is the front door for any coding task. Before writing any code, it interviews the person making the request and confirms a short, plain-language brief — goal, scope, key decisions, and what "done" means. Nothing gets built against an unconfirmed brief.

Once the brief is confirmed, it drafts a plan, gets that plan approved, and splits execution into workstreams delegated to smaller-model subagents matched by task and model tier. The top-level session stays in the planner/reviewer seat rather than doing the mechanical work itself.

The point is controlling prompt quality at the source: a short interview surfaces edge cases and audience details the original request left implicit, and over time it teaches the requester to write better prompts. Because the requester may not be a developer, the brief and plan stay jargon-free unless they've shown technical fluency first.

When to reach for it

• Any request to build, create, fix, scaffold, redesign, or change software — UI work, scripts, components, pages, integrations
• Even when the user doesn't say "plan" or "agents" — "Build X", "make me a Y", "fix these", and "clean this up" all qualify

When to skip it

Trivial one-line edits and config tweaks: skip the interview, just do them. But the skill's own caveat applies — if you find yourself three files deep in mechanical edits with no brief and no agents, stop. This skill should have fired.

How it works

Versus plan mode

Plan mode gates execution behind an approved plan — but it plans from the prompt exactly as you typed it. Make a Plan starts one step earlier: the interview and confirmed brief catch the states, edge cases, and audience the prompt left implicit, so the plan is drafted from a corrected brief instead of a raw request.

It also goes further after approval. Where plan mode hands execution back to the same session, this skill splits the work into workstreams delegated to smaller-model subagents, keeps the top-level session as planner and reviewer, and verifies each result with a different agent than the one that built it.

It's a complement, not a replacement: in agents that have a plan mode (like Claude Code), the skill uses it as the approval gate — and in agents that don't, it carries the same discipline anyway.

Compatibility

Works in Claude Code and most coding agents via the skills CLI. It delegates to subagents where the agent supports them, and falls back to sequential execution — with the same brief-and-plan discipline — where it doesn't.

Source

The skill is maintained in the open — read the raw file, browse the folder, or grab a packaged release.