
claw-skeleton
πͺΌ A personal AI assistant that lives in Telegram β every topic is its own isolated Claude Code agent with layered persistent memory and proactive cron heartβ¦
Install with your AI
Paste into Claude Code, Cursor, or any agent β it reads the repo and wires the tool into your project.
Install and set up claw-skeleton (git-clone project) into my current project. Found on https://claudeers.com/claw-skeleton Repo: https://github.com/LuKresXD/claw-skeleton Homepage/docs: β Detected install method: git-clone β git clone https://github.com/LuKresXD/claw-skeleton Category: productivity. Platforms: cli, api, web. Read the repo's README for exact setup and env vars, then install it and wire it into my project. Claudeers Health Verdict: slowing; community-verified: false. Confirm the source before running anything.
git clone https://github.com/LuKresXD/claw-skeleton
// compatibility
| Platforms | cli, api, web |
|---|---|
| Operating systems | β |
| AI compatibility | claude |
| License | MIT |
| Pricing | open-source |
| Language | TypeScript |
Claw πͺΌ
A personal AI chief-of-staff that lives in Telegram.
Every chat topic is its own isolated Claude Code agent β with a persona, persistent memory, and a web of cron jobs that quietly keep an eye on your inbox, calendar, and life.
Claw is my Telegram-based personal assistant. This repo is its open skeleton β the full engine and architecture, with my personal data and integrations stripped out so you can build your own.
It's not a wrapper around a chat completion. Each Telegram forum topic spawns a long-lived, resumable Claude Code session with its own persona and memory, so "Coding," "Coach," and "Finance" are genuinely different agents that never bleed into each other. A nightly rollup distills the day's notes into durable memory, a fleet of cron jobs turns the bot from reactive to proactive, and the assistant talks back through Telegram's full native rich formatting β tables, math, live date entities, inline media β by shelling out to its own small CLI toolbelt.
Architecture
flowchart TD
You(["You Β· Telegram"]) <--> Router["grammY router<br/>auth Β· queue Β· media Β· streaming"]
Router -->|"one isolated session<br/>per forum topic"| T1["Coding<br/>Claude Code session"]
Router --> T2["Coach<br/>Claude Code session"]
Router --> T3["General<br/>Claude Code session"]
T1 --> Mem["Layered memory<br/>daily notes β nightly rollup β<br/>MEMORY.md + per-topic"]
T2 --> Mem
T3 --> Mem
Mem -.->|"reinjected at session start"| Router
T1 -.->|"shells out to"| CLI["CLI toolbelt<br/>post Β· react Β· media Β·<br/>diagrams Β· delegate Β· monitor"]
CLI -.-> You
Crons["Cron + heartbeat jobs"] -->|"inbox Β· calendar Β·<br/>proactive nudges"| Router
π Deeper dive: ARCHITECTURE.md β turn lifecycle, the memory pipeline, the rich-output path, and the CLI toolbelt, with diagrams.
What makes it interesting
- π§ Per-topic isolated agents β each forum topic is its own persistent Claude Code session with its own persona and memory file. No cross-talk, no context soup. Sub-topics in a topic-enabled supergroup get their own isolated session too, keyed by thread.
- π Memory that survives restarts β append-only daily notes get promoted by a nightly
LLM rollup into a lean long-term
MEMORY.mdplus per-topic files. The bot wakes up knowing you. - β° Proactive, not just reactive β a shared cron runner fires scheduled Claude jobs (inbox triage, calendar, digests, health) and routes the results to the right topic.
- π¬ Native rich Telegram output β replies render as real Telegram rich text: tables,
task lists, fenced code,
$LaTeX$, spoilers, collapsibles, live timezone-aware date entities, maps, and inline images/video/audio β with an automatic HTML fallback. - π§° A CLI toolbelt the bot wields itself β small standalone CLIs the agent shells out to in order to act in Telegram mid-turn: post/edit a live progress line, drop a reaction, send media, render a Mermaid diagram to an image, host a file, delegate a parallel sub-agent, or arm a watcher that wakes a fresh turn when a long background job finishes.
- π Personas & guest mode β per-topic personas, prefix routing, and a stub-then-edit
guest mode you can
@mentionfrom any chat the bot isn't even in. - β»οΈ Self-maintaining β Claw can read and edit its own source, rebuild, and restart; smart auto-clear summarizes idle sessions into memory instead of nuking them mid-thought.
- π§ Production-hardened details β model-alias indirection, periodic context re-injection, streaming Telegram message edits, per-topic queueing, and systemd units with reliability drop-ins + failure alerts.
Project structure
claw-skeleton/
βββ CLAUDE.md # the blueprint β what the assistant is and how it remembers
βββ .env.example # the environment variables you fill in
βββ src/
β βββ bot/ # grammY router, middleware, commands, guest mode
β βββ claude/ # session lifecycle: runner, auto-clear, checkpoints, prompt builder
β βββ telegram/ # sender β rich Telegram output, streaming edits, HTML fallback
β βββ tools/ # the CLI toolbelt the bot shells out to (sanitized examples)
β βββ util/ # logger, sanitizer, graceful shutdown
β βββ config.ts # topics, personas, models, constants
β βββ index.ts # entry point
βββ cron-scripts/ # shared cron runner framework + libs
βββ cron-prompts/ # example scheduled-job prompts
βββ systemd/ # unit + timer patterns (bot, cron, reliability drop-ins)
βββ personas/ # example personas (default, coding, coach)
Start with CLAUDE.md β it's the spec that defines what the assistant
is, how it remembers, and how it behaves. The TypeScript is the engine; CLAUDE.md is the soul.
Build your own
# 1. create a Telegram bot via @BotFather, make a topic-enabled supergroup
# 2. wire it up
cp .env.example .env # add BOT_TOKEN, CHAT_ID, OWNER_USER_ID
npm install
# edit src/config.ts -> your topic IDs, personas, routing
# write personas/*.md and cron-prompts/*.txt for your own life
npm run build && npm start
# 3. (optional) install the systemd units to run it 24/7 + on a schedule
This is a sanitized skeleton, not a turnkey app: the personal integrations (email, calendar, finance, healthβ¦) and data were removed by design. Point your own Claude Code at the repo and have it flesh out the parts you want β that's the whole idea.
License
MIT β see LICENSE.
// faq
What is claw-skeleton?
πͺΌ A personal AI assistant that lives in Telegram β every topic is its own isolated Claude Code agent with layered persistent memory and proactive cron heartbeats. Open skeleton of a real, running system.. It is open-source on GitHub.
Is claw-skeleton free to use?
claw-skeleton is open-source under the MIT license, so it is free to use.
What category does claw-skeleton belong to?
claw-skeleton is listed under productivity in the Claudeers registry of Claude-compatible tools.
// embed badge
[](https://claudeers.com/claw-skeleton)
// retro hit counter
[](https://claudeers.com/claw-skeleton)
// reviews
// guestbook
// related in Productivity
Agent skills for Obsidian. Teach your agent to use Obsidian CLI and open formats including Markdown, Bases, JSON Canvas.
Open source repository of plugins primarily intended for knowledge workers to use in Claude Cowork
Garry's Opinionated OpenClaw/Hermes Agent Brain
An open-source alternative to Claude Cowork (powered by opencode)