onboard

name	onboard
description	Interactive conversational wizard that personalizes this ops template for a new owner. Fills [PLACEHOLDER] values across the repo and commits the result. Use when someone has just cloned the template and wants to set it up.
disable-model-invocation	true

Onboarding Wizard

The front door for anyone who just cloned this template. It turns the blank template into their personal ops repo — conversationally, with no terminal commands required. Works in both the Claude Code desktop app and claude.ai.

What this does

1. Choose a path — fresh start, import a local repo, or import from GitHub
2. Gather context — if importing, analyze the source so values can be pre-filled
3. Fill the gaps — ask only about what couldn't be confidently inferred
4. Commit — write every file and make one git commit

This skill only fills [PLACEHOLDER] tokens. It never deletes existing content.

The flow

Step 1 — Pick a mode

Ask the user which one fits (use AskUserQuestion):

• Fresh start — answer all placeholder questions from scratch (~5 min)
• Import a local repo — read an existing project's files and pre-fill what you can
• Import from GitHub — same, but pull the repo from a GitHub URL

Step 2 — Gather context (import modes only)

If importing, build a picture of the owner before asking anything:

• Read the source's README, package.json, pyproject.toml, CLAUDE.md, and similar
• Scan the folder structure
• Infer name, email, GitHub handle, timezone, tech stack, and focus areas

For local import, read from the filesystem with the Read/Glob tools. For GitHub import, fetch the repo's files via WebFetch. If the repo is private or unreachable, say so plainly and offer to switch to fresh mode.

Step 3 — Fill the placeholders

Fill [PLACEHOLDER] values across the template files, using inferred values where confident and asking the user only for the rest. Files that contain placeholders:

• CLAUDE.md — name, email, focus areas, timezone, tooling
• context/me.md — communication style and working preferences
• context/goals.md — career / business targets
• context/work.md — employment status / availability
• connections.md — tools in use (Notion, Slack, etc.)
• endpoints.md — GitHub username/repo, last-verified date
• MAP.md — name, focus areas, last-updated date (loaded every session via routing)
• memory/MEMORY.md, memory/overview.md, memory/glossary.md
• decisions/log.md, and any agents/*/AGENT.md
• scheduled-tasks/example-task/README.md — launchd label username

Do not fill placeholders inside instructional/example files — leave EXPANSIONS.md, GUARDRAILS.md, REVIEW.md, and the .claude/skills/ files as-is. Their brackets are documentation, not the owner's data.

The placeholder values you will need:

• Full name → [YOUR_NAME]
• Email address → [YOUR_EMAIL]
• GitHub username → [YOUR_USERNAME] (and repo name → [YOUR_REPO])
• Timezone → [TIMEZONE] (PST, EST, CET, UTC, …)
• 2–3 focus areas → [YOUR_FOCUS_AREAS] (e.g. "Job search, AI consulting")
• The three main focus areas, labeled individually (skip if same as above)
• Today's date → [DATE] (use the current date for any "last updated" fields)
• (Optional) Main services / portfolio page name

If you cannot confidently infer a value, ask for it. Never invent personal details.

Step 4 — Show and confirm

Before committing, show the user what was filled in (especially inferred values) so they can correct anything. Then write all files with the Write tool.

Step 5 — Sweep for leftovers (required)

After writing, run a final sweep so nothing personal is left blank. From the repo root:

grep -rnE '\[YOUR_[A-Z_]*\]|\[TIMEZONE\]|\[DATE\]' . \
  --include='*.md' \
  | grep -vE '\.git/|_archive/|_templates/|EXPANSIONS\.md|GUARDRAILS\.md|REVIEW\.md|\.claude/skills/'

• If it returns nothing → every owner placeholder is filled. Continue to commit.
• If it returns lines → those files still have gaps. Fill them (ask the user for any value you can't infer), then re-run the sweep until it comes back clean.

This catches files that were missed and is the safety net that makes onboarding trustworthy — never commit with owner placeholders still present.

Step 6 — Commit

Initial setup: personalized ops repo

Co-Authored-By: Claude Opus 4.8 <[e-mail]>

After the wizard

Tell the user:

1. Review CLAUDE.md and confirm everything reads correctly
2. Add their voice to longer sections like context/me.md that need real prose
3. Connect tools in connections.md if they use Notion, Slack, etc.
4. (Optional, Claude Code only) run /setup to install local tooling — graphify, plugins, hooks, and the weekly ops audit. Web users can skip this; all core features work without it.

Re-running

Safe to re-run. It only touches remaining [PLACEHOLDER] tokens, so already-filled values are left alone. To undo a run: git reset HEAD~1 (undo the commit) or git reset --hard (revert to the clean template).

Notes

• Everything happens inside Claude Code or claude.ai — no terminal needed.
• This is a manual command (disable-model-invocation): it runs only when the user types /onboard, never automatically, because it writes files and commits.