Create a practice playground

A practice playground runs the real, unmodified program in the reader's browser so every command in a Learn X article can be tried on the page. Each took one nWave feature of 9 to 34 DELIVER steps.

Discover what has shipped; don't trust a count in this file. List hugo/static/playgrounds/ (every folder except shared/) and docs/evolution/*-wasm-playground-evolution.md. A playground with an evolution doc is finalized; one without is still in DELIVER. As of 2026-10-10: jq and neovim (wasm); tmux, nushell, zsh, and fd (Linux on v86, fd finalized 2026-10-08 as a CLI tool inside zsh); zsh had no evolution doc yet. This skill turns that history into a repeatable path, so a new playground starts from the decisions already made instead of rediscovering them.

The site repo is ~/Projects/websites/jeffbaileyblog. Run every command from its root.

Ground rules that carry over from every playground

These came from real failures; the reasons matter more than the wording.

• Real program, never a simulation. Output that differs from the desktop program teaches the reader something false, which is worse than a missing feature. Any unavoidable difference goes in the bundle's deviations register (in its AGENTS.md) with a scenario that pins it.
• The program decides what a command was. The page never reads keystrokes or the screen to infer meaning; the program's own hook (zsh preexec_functions, Nushell pre_execution) writes allow-listed events that a root daemon relays.
• Expected values come from an oracle: the desktop program at the pinned version, built the same way, on x86_64. Never Homebrew, macOS, the bundle itself, or memory.
• No engine bytes before Start, no COOP/COEP or SharedArrayBuffer, no Terraform or deploy-workflow change.
• The repo is the source of truth, this skill is the map. Pins, line numbers, and step counts drift. Before copying anything, read the template bundle's AGENTS.md and its evolution doc, and trust them over this file when they disagree.
• Commit to main; never branch or open a PR. Deploy is manual; never trigger it.