create-assistant
/create-assistant
Build the user's own personal assistant step by step - its own WhatsApp number, email, reminders, browsing, voice notes and a second brain, running on their computer or a machine they already have. Use when the user asks to create, set up, install, continue or fix their personal assistant ("create my personal assistant", "תייצר לי עוזר אישי"), or to add a part to it later (email, Telegram, a login).
- Author
- MosheOfer1 ↗
- Licence
- MIT
- Last repository change
- 08.11.2026
Assessment and security
- Files in this package
- 1
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 5
No signals detected. Repository changed on 08.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Build the user's own personal assistant step by step - its own WhatsApp number, email, reminders, browsing, voice notes and a second brain, running on their computer or a machine they already have. Use when the user asks to create, set up, install, continue or fix their personal assistant ("create my personal assistant", "תייצר לי עוזר אישי"), or to add a part to it later (email, Telegram, a login).
Triggers from the description
Original author description
Build the user's own personal assistant step by step - its own WhatsApp number, email, reminders, browsing, voice notes and a second brain, running on their computer or a machine they already have. Use when the user asks to create, set up, install, continue or fix their personal assistant ("create my personal assistant", "תייצר לי עוזר אישי"), or to add a part to it later (email, Telegram, a login).
How it works
- How to talk
- Tools
- Steps (in order; skip what is done)
- When something is wrong later
SKILL.md preview
---
name: create-assistant
description: Build the user's own personal assistant step by step - its own WhatsApp number, email, reminders, browsing, voice notes and a second brain, running on their computer or a machine they already have. Use when the user asks to create, set up, install, continue or fix their personal assistant ("create my personal assistant", "תייצר לי עוזר אישי"), or to add a part to it later (email, Telegram, a login).
---
# create-assistant
You guide the user through building their assistant. Most users are not technical.
## How to talk
- Speak the user's language (the one they wrote in) the whole way. Short, plain sentences.
- **Before every action say what you will do and why, in one or two lines.** Then do it. Then say what happened.
- One question at a time. Offer a sensible default. Every step can be skipped and done later.
- Never ask for a password, token or code in the chat. Secrets go through a local window (`setup.js secret`),
logins through a visible browser window (`chrome-step.js`). Say so when it matters: "You type it there, I never see it."
- Never spend money without explaining it first. Nothing here costs money except what the user already pays for Claude.
- If a step fails: say it plainly, show the short error, offer to retry, skip, or stop. Never loop.
## Tools
`S="node ~/.assistant/app/gateway/bin/setup.js"` (state, log, config, secrets, checks, WhatsApp pairing),
`~/.assistant/app/gateway/bin/service.js` (the service), `remote.js` (another machine), `chrome-step.js` (windows).
Every command prints `--help`. Run `$S status` at the start: it shows what is done, and resumes where they stopped.
After each step: `$S mark STEP done` (or `skipped`) and `$S log "<what happened>"`. Never log secrets.
## Steps (in order; skip what is done)
1. **language** - Confirm the language. `$S set language '"he"'` (any ISO code), `$S set ownerName '"<first name>"'`,
`$S set timezone '"<IANA zone>"'`.
2. **plan** - In 5-6 lines: what they get (an assistant they talk to on WhatsApp like a contact; it reads their
email, reminds them, browses, understands voice notes, remembers in a second brain they can read), what it needs
(a **second phone number** for the assistant - a real SIM or eSIM, not a one-time virtual number; about 30 minutes;
a computer that stays on, or a machine they already have), and that every message to other people, payment or
booking asks them first.
3. **prereqs** - First run only: if `~/.assistant/app` is missing, explain, then clone
https://github.com/MosheOfer1/create-your-own-assistant-from-claude-code there and
`npm ci --omit=dev` in `gateway/`. Then `$S check`. Node 22+ is required (offer to install it: nvm, brew, or the
nodejs.org installer); git and Claude Code are required; ffmpeg only for voice; `npx playwright-core install chromium` for the browser.
4. **host** - Ask: "this computer" or "a machine you already have (SSH)". This computer must stay on and awake:
explain how to turn off sleep for their system (macOS: Battery/Energy settings; Linux: power settings; Windows:
use WSL2 with systemd on, and turn off sleep). For SSH: `remote.js check USER@HOST` and explain any PROBLEM line.
`$S set host '"local"'` or `'"ssh:USER@HOST"'`.
5. **code** - Pin the version: `$S set version '"<latest tag of the public repo>"'` (`git -C ~/.assistant/app describe --tags --abbrev=0`).
6. **repo** - Their private GitHub repo for settings and the second brain. Needs `gh` logged in: if not, run
`gh auth login --web` in the background and tell them to finish it in the browser. Then `$S repo create my-assistant`.
If they already keep notes in Obsidian, offer to point the brain at that vault instead:
`$S set modules.brain.dir '"<vault path>"'` (a vault that is a git repo gets commits).
7. **claude** - This computer: already logged in (you run here). SSH host: print `remote.js login USER@HOST` and ask them
to type it in the prompt with a leading `!` (it is interactive, once). Or, if they prefer an API key:
`$S secret ANTHROPIC_API_KEY --url https://console.anthropic.com/settings/keys --say "<create a key and paste it>"`.
8. **whatsapp** - Ask for the assistant's number and their own number (international format).
`$S set channels.whatsapp '{"assistantNumber":"+…","ownerNumber":"+…"}'`. Explain: on the assistant's phone,
WhatsApp (or WhatsApp Business) → Linked devices → Link with phone number instead. Run `$S whatsapp` in the background;
when it prints `pairing code: XXXX-XXXX`, show the code; it prints `linked` when done. Tell them: open WhatsApp on
the assistant's phone at least every 14 days, or WhatsApp unlinks it (the assistant warns them).
9. **telegram** (optional backup) - They make a bot with @BotFather: `chrome-step.js step --url https://web.telegram.org --say "<open @BotFather, /newbot, copy the token>"`,
then `$S secret TELEGRAM_BOT_TOKEN --url https://web.telegram.org --say "<paste the token>"` and
`$S set channels.telegram '{}'`. After the service starts, its log shows `/start <code>` to send to the bot.
10. **email** (optional) - Explain app passwords (a separate password just for the assistant, revocable). Gmail needs
2-Step Verification on. `$S set modules.email '{"user":"<address>"}'`, then
`$S secret EMAIL_APP_PASSWORD --url https://myaccount.google.com/apppasswords --say "<create one named Assistant, paste it>"`.
11. **voice** (optional) - Voice notes need speech-to-text. Pick by what the host has, ask one question at most
(docs/modules/media.md has the engine table):
- A Mac with Apple silicon, or a machine with a GPU: whisper.cpp there (macOS: `brew install whisper-cpp`), a model in
`~/.assistant/data/models/` (ggml-small.bin or larger for non-English). Free, private, every language.
- No GPU: ask "simple and best quality, or nothing leaves the machine?".
Simple: a cloud service with an OpenAI-compatible API (OpenAI, Groq …):
`$S secret TRANSCRIBE_API_KEY --url <the service's key page> --say "<create a key and paste it>"`,
`$S set modules.media '{"transcriber":["http","sherpa"],"httpUrl":"<…/v1/audio/transcriptions>","httpModel":"<model>"}'`
(sherpa as the fallback only if its model is there). Say once: the audio goes to that service.
Private: sherpa (fast on a CPU, one language per model): put a streaming zipformer model for their language in
`~/.assistant/data/models/sherpa`, `$S set modules.media '{"transcriber":"sherpa","sherpaLanguages":["<code>"]}'`.
- They already run a speech server (a GPU at home over a VPN): `transcriber: ["http", …]` with its `httpUrl`; when it is
off, the next engine in the list answers.
Always: `voiceLanguage` is their chat language unless they speak several (`"auto"`).
12. **logins** (optional) - Sites the assistant should be logged in to: `chrome-step.js login --url <site>` (they log in in
the window and press Done). SSH host: then `chrome-step.js push --to USER@HOST:.assistant/data`.
Always on: `$S set modules.reminders '{}'`, `$S set modules.brain '{}'` (if not set by step 6), `$S set modules.browse '{}'`.
With WhatsApp: `$S set modules.people '{}'` (it messages others only after their yes).
Gmail, Calendar and Drive without a password: connect them at claude.ai → Settings → Connectors, and
`$S set modules.google '{}'`. It works only with its own user (step 13); say so.
Ask about a morning brief (`modules.reminders.brief {"time":"07:30"}`) and quiet hours (`quietHours`, docs/modules/core.md).
13. **service** - This computer: `service.js install`. SSH host: `$S repo push`, then `remote.js install USER@HOST --repo <owner>/my-assistant --version <tag>`.
If linger is not on, explain the one admin command it needs.
**Own user (Linux / WSL2, recommended):** explain: the assistant moves to a user of its own; it can no longer
see their files, and its code becomes read-only to it, so it may run plain commands without asking. It needs their
admin password once. Ask them to type, with a leading `!` (sudo asks for the password there, not in the chat):
`! sudo "$(command -v node)" ~/.assistant/app/gateway/bin/own-user.js install` (SSH host: `remote.js own-user USER@HOST`
prints the command). Show its `ok:` / `BAD:` lines. Then Claude must log in once for the new user (a login of its own:
a copied one breaks when either side refreshes): ask them to type the `BAD: Claude is not logged in` command it prints, in a
terminal of their own (SSH host: inside `ssh -t USER@HOST '…'`), open the link, paste the code, `/exit`. Later secrets or logins: `! sudo … own-user.js sync-from-owner`.
macOS: skip (it runs as their user, with the guards in the code).
14. **deploy** - Explain: settings and brain live in their private repo. A change made there from outside (GitHub,
Obsidian on a phone) is picked up when the assistant answers the next message. What they ask in the chat works
at once. To update the assistant: change `version` in config.json.
15. **test** - Ask them to write "hi" to the assistant's number. Watch `service.js logs` (or `remote.js logs`).
A reply means it works. Then suggest two first tasks ("remind me tomorrow at 9 to…", "what's new in my email?").
## When something is wrong later
`$S status`, `service.js status` / `logs`. WhatsApp unlinked → step 8 again (`rm` nothing; `$S whatsapp` makes a new code
after `~/.assistant/data/whatsapp-auth` is moved aside - ask first). Repeat any step any time; each is safe to run again.
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · MosheOfer1/create-your-own-assistant-from-claude-code | 90 | MIT |
| vapiai/skills | 100 | MIT |
| vapiai/skills | 100 | MIT |
Database snapshot: 13.11.2026 · package version d436a7aa56dc (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100openclaw
Use when setting up a multi-channel personal AI assistant gateway bridging chat platforms (Telegram, Discord, WhatsApp, WeChat) with LLM backends. OpenClaw: multi-channel AI assistant gateway for unified personal AI access. - 100pa-onboarding
Step-by-step onboarding guide for setting up a new AI Personal Assistant on OpenClaw. Use when: a new PA is being created, someone asks how to set up an agent, or guiding a user through the full setup process from account creation to first response. - 100openclaw-setup
OpenClaw usage expert: Helps you install, deploy, configure, and use OpenClaw personal AI assistant. Can diagnose issues, create bots, execute automated tasks, etc. Use when users need to install OpenClaw, configure Gateway, set up Channels, create Agents, troubleshoot issues,… - 100openclaw-setup
OpenClaw usage expert: Helps you install, deploy, configure, and use OpenClaw personal AI assistant. Can diagnose issues, create bots, execute automated tasks, etc. Use when users need to install OpenClaw, configure Gateway, set up Channels, create Agents, troubleshoot issues,… - 95tg-group-assistant
Use when a user wants a new Telegram group assistant with Hermes, an isolated instance, scoped read-only data access, or an optional scheduled digest and ambient routing. - 85base44-platform-agents
Base44 in-app AI agents: JSON agent configs under base44/agents/ with entity, backend-function or server-route tools, WhatsApp and Telegram channels, the agents SDK for conversation UIs, and code agents (an agent loop the app runs on the AI gateway). Use when the app needs a…