Using Claude as a Personal Assistant

Claude Code is a powerful AI coding agent, but it’s actually a very capable general-purpose assistant when given the right instructions, context, and tools. This guide shows how to use the Telegram bot to access Claude as your personal assistant from anywhere.

The Core Idea

Instead of building a separate “assistant app,” you:

Create a personal working folder with a CLAUDE.md that teaches Claude about you
Connect it to the bot via CLAUDE_WORKING_DIR
Add scripts, MCP tools, and Skills to extend Claude’s capabilities
Access everything from Telegram, on your phone

Claude then becomes your 24/7 assistant that knows your preferences, goals, and workflows.

Getting Started

Step 1: Create Your Personal Folder

# Create a folder for your assistant setup
mkdir ~/my-assistant
cd ~/my-assistant

# Initialize git (optional but recommended)
git init

Step 2: Write CLAUDE.md

Create CLAUDE.md in this folder. This is Claude’s “brain” — it teaches Claude about you:

# CLAUDE.md

This file provides guidance to Claude as my personal assistant.

## Quick Reference

**This folder:**
- `cli/` - Utility scripts
- `.claude/skills/` - Task workflows
- `.claude/agents/` - Specialized subagents

**Key paths:**
- Notes: `~/Documents/Notes/` (personal, health, research, etc.)
- Working dir: `~/my-assistant`

## About Me

[Your name] is a [profession] interested in [hobbies/interests].

**Personal context**: See `personal-context.md` for details about family, preferences, goals.

## How to Assist

- **Autonomy**: Handle routine tasks independently
- **Communication style**: [Your preferred style]
- **Format preference**: Bullet lists over tables
- **When to ask**: Before significant actions, big purchases, or writing to permanent files

## Important Tasks

**Keep context fresh**: When you learn something new about me, update:
- Personal goals → `personal-context.md`
- Life priorities → `pulse.md`
- Assistant behavior → this `CLAUDE.md`

## Quick Lookup

- Life goals → `personal-context.md`
- Notes system → `~/Documents/Notes/`
- Tasks → Things 3 (via MCP)

Step 3: Configure the Bot

Set your working directory in .env:

CLAUDE_WORKING_DIR=~/my-assistant

Now when the bot runs, Claude will read your CLAUDE.md and use it as context for every conversation. If you intentionally use the manual bot-folder launcher (bun run start) during development, export the same path in your shell before starting so the launcher can find .superturtle/.env:

export CLAUDE_WORKING_DIR=~/my-assistant

Building Assistant Features

Scripts and Utilities

Create cli/ scripts for tasks Claude should automate:

#!/bin/bash
# cli/health.sh
# Get health metrics from Health Auto Export

jq '.current' ~/Library/Health/Export/*.json | \
  jq -s 'add | {sleep, activity, vitals}'

Call from Telegram:

You: Check my health metrics
Claude: (runs cli/health.sh, sees sleep/activity/vitals)
Claude: Your sleep has been solid at 7.5h average.
        Resting HR at 52 (good), HRV at 71 (improving).

Skills (Auto-Triggered Workflows)

Skills are workflows that Claude can use automatically when relevant. Create them in .claude/skills/:

---
name: workout-planning
description: Create personalized workouts based on training plan and recent activity
allowed-tools: Read, Write, Bash(cli/health.sh:*), Glob
---

# Workout Planning

When asked for a workout or exercise routine:

1. **Read training plan**: `~/Documents/Notes/Health/training.md`
2. **Check recent workouts**: `~/Documents/Notes/Health/Workouts/`
3. **Get health metrics**: Run `cli/health.sh`
4. **Propose workout** based on training schedule and recovery
5. **Create** `Health/Workouts/YYYY-MM-DD-workout.md`

Format workouts as clear, numbered lists with exercise names, sets/reps, and rest periods.

Now when you ask “give me a workout”, Claude automatically:

Reads your training plan
Checks your recent activity and recovery
Creates a personalized workout file
You see it instantly in your notes app (if synced via iCloud)

MCP Servers for Tools

Connect Claude to your favorite apps via MCP. Common setups: Things 3 (macOS to-do manager):

// mcp-config.ts
things: {
  command: "npx",
  args: ["-y", "@soulmen/things-mcp"],
  env: {
    THINGS_DATABASE_PATH: `${require("os").homedir()}/Library/CloudStorage/iCloud~com~culturedcode~things/Things Database.thingsdatabase`,
  },
},

Now ask: “Add ‘Review design specs’ to my inbox” Notion (notes and databases):

notion: {
  command: "npx",
  args: ["-y", "@github/notion-mcp"],
  env: {
    NOTION_API_KEY: process.env.NOTION_API_KEY || "",
  },
},

Now ask: “What’s on my reading list?” and Claude queries your Notion database. Slack (team communication):

slack: {
  command: "npx",
  args: ["-y", "@slackhq/slack-mcp"],
  env: {
    SLACK_BOT_TOKEN: process.env.SLACK_BOT_TOKEN || "",
  },
},

Now ask: “Catch me up on what the team has been doing” and Claude reads recent Slack messages.

Example Workflows

Morning Briefing

Create a /life-pulse command that gathers data from multiple sources:

---
name: life-pulse
description: Generate executive life digest
allowed-tools: Bash, Read, Write, Task
---

# Generate Life Pulse

Gather data in parallel:
- Things: Today's tasks and upcoming
- Email: Unread and important
- Health: Sleep and recovery metrics
- Calendar: Upcoming events
- Work: Issues and blockers (if connected)

Synthesize into a brief digest covering:
- **TL;DR** - What's on your plate (3-5 bullets)
- **Now** - What needs attention right now
- **Health** - Sleep quality, activity, recovery
- **Next** - Upcoming priorities

Write to `~/Documents/Notes/life-pulse.md`

When you run this from Telegram, Claude:

Gathers all data in parallel
Synthesizes into a readable briefing
Saves to a file synced to your phone via iCloud
You read it with coffee every morning

Research and Decisions

Create a research skill for thorough decision-making:

---
name: research
description: Research topics using web search, Reddit, and Hacker News
allowed-tools: WebSearch, WebFetch, Bash, Read, Write
---

# Research Workflow

When asked to research:

1. Search the web for reviews and expert opinions
2. Search Reddit for community experiences
3. Search Hacker News for tech/startup discussions
4. Synthesize findings with pros/cons
5. Save to `~/Documents/Notes/Research/YYYY-MM-DD-topic.md`

Provide clear recommendation based on actual data, not guesses.

Message: “Research the best noise-canceling headphones under $300” Claude searches multiple sources, compares options, and saves a detailed recommendation to your Notes.

File Sync Strategy

Best practice: Sync your Notes folder to your phone via iCloud:

# On macOS
mkdir -p ~/Documents/Notes
# Enable iCloud sync: System Settings → iCloud Drive → Documents

Now files Claude creates appear on your phone instantly. Use Ulysses or iA Writer to edit them on the go. For Things 3 tasks: They sync automatically once Claude adds them. For Notion: Updates sync through the Notion app.

Security & Permissions

The bot runs Claude Code with all permission prompts disabled. This is intentional for a seamless mobile experience, but you should understand it means Claude can:

Read and write files in allowed directories
Execute shell commands without confirmation
Access external tools and APIs

See Security for defense layers in place.

Set restrictive file access: In .env:

# Only allow your assistant folder and notes
ALLOWED_PATHS=~/my-assistant,~/Documents/Notes,~/.claude

Claude can’t access other directories, protecting sensitive files.

Progressive Enhancement

Start simple and expand as you build confidence: Week 1: Basic CLAUDE.md + answer questions

You: What's my next meeting?
Claude: Checking calendar... 2pm standup

Week 2: Add one script (e.g., health.sh)

You: How did I sleep last night?
Claude: (runs health.sh, sees sleep data)

Week 3: Connect first MCP (Things 3)

You: Add project review to my inbox
Claude: ✅ Added to Things

Week 4: Create first Skill (workout planning)

You: Give me a workout for today
Claude: (reads training plan, creates workout file)

Advanced Patterns

Subagents for Complex Tasks

For complex tasks like “generate my life pulse”, use subagents that run in parallel with separate context windows:

Main Agent Task: "Give me a life pulse"
  ↓
  Subagent 1: Gather email digest
  Subagent 2: Gather task digest
  Subagent 3: Gather health digest
  Subagent 4: Gather work digest
  ↓
  Main agent synthesizes into final briefing

This keeps context windows lean and lets you handle complex workflows.

Calendar Sync via GitHub Gist

Manage calendars that sync to your phone:

# calendars/track-days.yaml (auto-generated by Claude)
events:
  - date: "2026-01-25"
    title: "Track day at Portimão"
    time: "09:00"
    # syncs to GitHub Gist
    # you subscribe in Google Calendar / Apple Calendar

Message: “Add the Belgium race to my calendar for next Saturday” Claude updates the YAML file, syncs to GitHub, and your calendar refreshes automatically.

File Organization Example

~/my-assistant/
├── CLAUDE.md                 # Your assistant's instructions
├── personal-context.md       # Your info: preferences, goals, family
├── .claude/
│   ├── skills/
│   │   ├── workout-planning/SKILL.md
│   │   ├── research/SKILL.md
│   │   └── meal-planning/SKILL.md
│   └── agents/
│       ├── health-digest/AGENT.md
│       └── email-digest/AGENT.md
├── cli/
│   ├── health.sh            # Get health metrics
│   ├── calendar.sh          # Get calendar events
│   └── slack.sh             # Get Slack messages
└── README.md                # Documentation for future you

Getting Help

Need to research something → Ask Claude with the research skill
Need a workout → Ask Claude with the workout skill
Need to update goals → Ask Claude to edit personal-context.md
Want a new workflow → Ask Claude to create a new skill

The beautiful part: when you need new capabilities, just ask Claude to build them. From Telegram, on your phone.

Next Steps

Drivers & Models — Choose between Claude Code and Codex
MCP Tools — Add more integrations
Commands — Full command reference
Security — Understanding the threat model

​The Core Idea

​Getting Started

​Step 1: Create Your Personal Folder

​Step 2: Write CLAUDE.md

​Step 3: Configure the Bot

​Building Assistant Features

​Scripts and Utilities

​Skills (Auto-Triggered Workflows)

​MCP Servers for Tools

​Example Workflows

​Morning Briefing

​Research and Decisions

​File Sync Strategy

​Security & Permissions

​Progressive Enhancement

​Advanced Patterns

​Subagents for Complex Tasks

​Calendar Sync via GitHub Gist

​File Organization Example

​Getting Help

​Next Steps