| name | setup-workshop |
| description | One-shot setup for Lifestyles of the AI-Native workshop. Installs and verifies the dev tools an attendee needs — Bun (the check-in tool), the Codex CLI + plugin (Block 3 adversarial review), and git — then confirms the repo is wired up and hands off to the opening check-in. (Voice/Handy is NOT installed here — it's the first hands-on moment of Block 1.) Triggers on "set me up for the workshop", "get me set up", "workshop setup", "set up the workshop", or starting the workshop. |
Set up the workshop
The fast path. This gets an attendee from a freshly-cloned repo to ready for Block 1 in about five minutes: every tool installed, every tool verified, nothing left to discover mid-exercise. Work top to bottom; check before installing, install only what's missing, and verify each item before moving on. Report a clean checklist at the end.
The attendee should already be running Claude Code inside this repo (trusting it auto-loads the workshop skills, the ideation plugin, and the aie-coach MCP server). If they're not in the repo, stop and tell them to open it and trust it first — that's step 0. Expect first-run prompts: the trust dialog, the aie-coach MCP approval ("Use this and all future MCP servers in this project"), and per-tool-call approvals. To run hands-off, they can relaunch with claude --dangerously-skip-permissions (fine — they trust this repo).
Procedure
-
Detect the platform. Run uname -s (Darwin = macOS; otherwise treat as Windows/Linux). Pick the matching install commands below.
-
Bun — the check-in tool and skills run on it (Blocks 1 & 4).
- Check:
bun --version. If it prints a version, skip the install.
- Install (macOS/Linux):
curl -fsSL https://bun.sh/install | bash
- Install (Windows PowerShell):
powershell -c "irm bun.sh/install.ps1 | iex"
- After installing, Bun is on
~/.bun/bin; if bun --version still isn't found in this shell, tell the attendee to open a new terminal (or export PATH="$HOME/.bun/bin:$PATH").
-
Codex CLI + plugin — the adversarial-review gate in Block 3.
- Check:
codex --version. If present, skip.
- Install:
npm i -g @openai/codex (needs Node/npm — check node --version first; if missing, point them to https://nodejs.org).
- Auth is interactive — do not run it for them. Tell the attendee to run
codex login themselves once (it opens a browser / prompts for an OpenAI sign-in). Note it can wait until Block 3.
- The
codex plugin (the /codex:* commands) auto-installs from GitHub when the repo is trusted — same as ideation, no action needed (it may need the step-6 restart to finish loading). Once codex login is done, verify the whole chain with /codex:setup. Raw codex exec "…" still works as a fallback if the plugin hasn't loaded.
-
Voice (Handy) — not in setup. Don't install Handy here. Getting voice working is the first hands-on moment of Block 1 (the attendee says "Set up Handy for me" then, which runs the setup-handy skill). Mention it in the hand-off so they know it's coming; do not run it now.
-
Git — worktrees for parallel agents in Block 2.
- Check:
git --version. Almost always present. If missing: macOS xcode-select --install; Windows https://git-scm.com.
-
Confirm the repo wiring — and the coach MCP. Trusting the repo loads the .claude/skills/ skills and the ideation + codex plugins. The aie-coach MCP server (defined in .mcp.json) needs a one-time approval: on first trust Claude Code prompts to enable it — approve it, or run /mcp and enable aie-coach. It launches with bun, so if you just installed Bun in this session, restart Claude Code (or reopen the repo) so the server can start; then approve it. Without the coach MCP loaded, the AI-Native score in the check-in won't work (the plain check-in still does, via bun skills/coach-checkin/scripts/submit.ts). Report the coach as [✓] only if its tools are actually available this session; otherwise [!] with "approve via /mcp (restart if you just installed Bun)".
-
Always end with the uniform status report — print this exact block, every run, success or failure, so it's instantly clear what's ready and what isn't. One row per item, fixed-width name column, a status glyph, and a detail (version on success; the reason on failure). Then a summary line.
Status glyphs (use exactly these):
[✓] — verified working (show the version / confirmation)
[✗] — failed (show what failed and why, plus the one-line fix)
[!] — needs a manual step you can't do for them (e.g. codex login, granting macOS Accessibility) — show the exact command/action
[-] — skipped (say why, e.g. "already installed")
══════════════════════════════════════════════════════════
Lifestyles of the AI-Native · workshop setup
══════════════════════════════════════════════════════════
[✓] Bun 1.3.14
[✓] Codex CLI 0.x installed
[!] Codex login required → run: codex login (before Block 3)
[✓] Git 2.43.0
[✓] Repo trusted · skills + ideation/codex plugins + aie-coach MCP
──────────────────────────────────────────────────────────
READY: 3/4 · NEEDS YOU: Codex login · FAILED: none
══════════════════════════════════════════════════════════
Next: Block 1 sets up your voice — "Set up Handy for me".
══════════════════════════════════════════════════════════
Rules for the report:
- Include every item (Bun, Codex CLI, Git, Repo wiring) on its own row, in this order — never omit one because it failed. (Handy is not a setup item — it's a Block 1 step; only name it in the closing "Next" line.)
- On any
[✗], the detail must say why (the actual error, trimmed) and the fix. Never report [✓] for something you didn't successfully verify.
- The summary line always shows the
READY: n/total count, plus NEEDS YOU: and FAILED: lists (write none when empty).
-
Hand off to Block 1. After the report, if Bun is [✓], close with: "You're set — say "run my workshop check-in" to log your starting point (and your AI-Native score) on the live board. First thing in Block 1 we'll get your voice working — "Set up Handy for me"." If anything blocking failed, say which one to fix first and that the rest can wait.
Notes
- Install only what's missing. Always check the version first; don't reinstall a present tool.
- Don't run interactive logins (
codex login) on the attendee's behalf — surface the one command and let them do it.
- If any single tool fails, keep going with the rest and report it clearly at the end rather than aborting — a half-set-up attendee can still start Block 1 (which begins by getting voice/Handy working and otherwise needs only Bun).
- Nothing here sends anything off the machine. Tool installs are public packages; the check-in (step 8) is separately opt-in.