Publish a new placeholder package to NPM under the @liveblocks/* org. Creates a minimal package.json + index.js in a temp dir and gives the user exact commands to run. Use when publishing a brand-new @liveblocks/* package for the first time (version 0.0.0 placeholder to claim the name).
Review one or more GitHub pull requests as a progressive, understanding-first code review -- start at the highest altitude (what does it do, how, what to grasp before the details) and descend one abstraction layer at a time only when asked. Built for a senior engineer who wants to understand a PR before spending review time on it. Use when the user runs /pr-review with one or more PR links (often a frontend + backend pair), or asks to "review this PR" / "review these PRs". Treats the code diff as the source of truth and the PR description as context, reads the PR's existing discussion so it doesn't repeat points already raised, and keeps every finding at the current altitude of the review. Surfaces only interesting findings and good questions to ask the author -- never a checklist dump.
Produce a single self-contained interactive HTML document that visually explains how a *specific* system, mechanism, or design in the codebase works -- built for visual learners who find static markdown insufficient. Walks a concrete worked example through its states with toggles, steppers, diffs, and side-by-side comparisons. Invoked explicitly as /visually-explain; it is for explaining a particular thing you can point at in code or a spec, not for general-concept tutorials.
Mikado method workflow for complex refactorings. Use when the user invokes /mikado, mentions "mikado graph", or wants to break a large code change into safe, incremental steps with automatic revert-on-failure.
Open a pre-populated GitHub PR creation page in the browser, with a title and body drafted in nvie's personal writing style. Use when the user runs /pr, or asks to "open a PR", "create a PR", or "draft a PR" for the current branch. This skill never creates the PR itself -- the user reviews and submits it manually.
Solve a request strategically and architecturally rather than tactically. Use when the user invokes /proper, asks for "the Proper™ way", wants the long-term design rather than the quickest fix, or wants to refactor / rearchitect first so the requested feature fits naturally. This skill diagnoses why the request feels awkward today, proposes a design change that makes the feature easy, and -- if the strategic path is non-trivial -- hands off to /mikado to execute it in safe incremental steps.
Draft, review, pressure-test, compare, and polish technical design documents for product, API, SDK, infrastructure, storage, migration, reliability, security, privacy, performance, data model, or architecture work. Use when an engineer wants to make and defend a technical decision, turn rough notes into a review-ready TD, review an existing TD, compare alternatives, or plan rollout/testing/migration/rollback for a proposed technical change.
Scrape Claude Code session history to find repeated user corrections, preferences, and contextual guidance, then surface candidates for CLAUDE.md or memory. Use when the user wants to review past sessions for self-improvement, find patterns in their feedback, or bake recurring preferences into persistent configuration.