A drop-in Claude Code skill that turns "build this" into a real plan — then guides Claude Code through a verify/repair loop. One command. It sizes the job itself, researches from the right source, locks a checkable goal, plans, and won't wave "done" through without walking every success criterion. Most planning prompts are a giant wall of text that treats a one-line copy fix the same as a new payment integration. This skill doesn't. It classifies the task into Quick / Standard / Deep first, then runs only the ceremony that tier needs — so a small change gets a one-screen plan and a real build gets the full research → refine → goal-contract → plan → execute → verify loop. This is the planner I actually run on my own projects. Free, MIT, and it's on GitHub too: github.com/JDiz00/the-super-planner-skill What it does Sizes the task before anything else. Tier 1 skips to a mini-plan and one approval. Tier 2 runs a light path with a single combined gate. Tier 3 runs the full pipeline for new projects, unfamiliar APIs, or anything touching auth / payments / third-party data. Routes research by claim type. Technical facts → official docs, never a social post. Market signal → a recent-activity search, treated as trend not truth. And it checks your dependency files first, so it verifies against your versions. Locks a Goal Contract with numbered, individually checkable success criteria — then makes every build slice trace back to one of them. Won't wave "done" through. The final review requires a passing check for each criterion. A criterion with no evidence is not done. Never invents. No API, command, path, price, or flag from memory — it verifies or says it can't. Install (30 seconds) Open START-HERE.html in the download — it walks you through it. In short: copy the jd-super-plan skill folder into ~/.claude/skills/, then in Claude Code type /jd-super-plan build a rate limiter for my API. No dependencies, nothing to build. It's opt-in only — it never fires on a casual question. What's in the download The skill (a single Markdown file — the whole product) START-HERE.html — a friendly install + use walkthrough README + MIT license Want "done" to be a check that literally exits 0 — enforced by a real Stop hook that blocks a fake "done"? That's the deeper, installable kit: the Claude Code Setup Playbook. Linked inside. This product is not affiliated with, endorsed by, or sponsored by Anthropic PBC. Claude is a trademark of Anthropic PBC.