Approve AI Agent Skills from Telegram — Human-in-the-Loop on Your Phone

Your AI agent runs a task, reaches a manual-approval skill, and sends you a Telegram message asking for confirmation. Reply yes — the next skill runs. Reply no — it stops. No dashboard, no browser tab, just Telegram.

What You Will Build

A three-skill pipeline connected to your Telegram bot. The orchestrator runs skills in sequence — auto skills fire immediately, manual skills pause and ask you first:

The agent runs greeter and followup automatically, then pauses before review. It sends you a Telegram message asking for your approval. Reply yes — review runs. Reply no — the pipeline stops.

What You Need

• Node.js 18+ for npx
• An API key from OpenAI, Anthropic, or any supported LLM provider
• A Telegram bot token — get one free from @BotFather

Create the Project Folder

Create a folder called chain-telegram/. One root agent.yaml orchestrates all three skills — no separate chain YAML files needed:

Create agent.yaml

The root agent.yaml is the orchestrator. It lists all three skills in order — auto skills run immediately, manual skills pause for your approval.

name: Chain Telegram — Approval Demo
description: Three-skill pipeline with automatic and manual approval stages.

skills:
  # Runs first — auto (no approval needed)
  - path: ./greeter
    trigger_type: auto

  # Runs second — auto (fires immediately after greeter)
  - path: ./followup
    trigger_type: auto

  # Runs third — manual (pauses and asks for your approval in Telegram)
  - path: ./review
    trigger_type: manual

Create the SKILL.md Files

---
name: greeter
description: Greets you and shares a fun fact about today in history.
license: Apache-2.0
metadata:
  author: Open Enthrium
  version: "1.0"
---

You are a helpful assistant.

## Step 1: Greet

Say hello, mention today's date, and give one fun fact about today in history.

---
name: followup
description: Summarises the greeter's output in one sentence.
license: Apache-2.0
metadata:
  author: Open Enthrium
  version: "1.0"
---

You are a summarizer assistant.

## Step 1: Summarize

The previous skill just ran. Summarize its output in one sentence
and say "Auto skill complete ✅".

---
name: review
description: Runs only after you approve from Telegram.
license: Apache-2.0
metadata:
  author: Open Enthrium
  version: "1.0"
---

You are an assistant that confirms tasks.

## Step 1: Confirm

The user approved running you from Telegram.
Acknowledge the approval, briefly reference what the previous skills said,
and say "Manual approval complete ✅".

Create oe-config.json

This file holds your LLM key, server settings, and Telegram connector. Leave publicUrl blank for now — you will fill it in after the Cloudflare tunnel starts.

{
  "llm": {
    "provider": "openai",
    "model": "gpt-4o",
    "apiKey": "sk-..."
  },
  "server": {
    "enabled": true,
    "port": 3333,
    "publicUrl": "",
    "webhook": true
  },
  "connectors": [
    {
      "connection_name": "My Telegram Bot",
      "connection_type": "telegram",
      "baseUrl": "https://api.telegram.org/botYOUR_BOT_TOKEN",
      "auto_reply": true,
      "rate_limit": { "messages_per_minute": 20 },
      "history": { "enabled": true, "max_messages": 10 }
    }
  ]
}

Replace sk-... with your LLM API key and YOUR_BOT_TOKEN with the token from @BotFather. Anthropic users: set "provider": "anthropic" and "model": "claude-sonnet-5".

Start a Cloudflare Tunnel

Telegram needs a public HTTPS URL to deliver messages to your server. Cloudflare Tunnel gives you one for free — no account or signup needed.

npx cloudflared tunnel --url http://localhost:3333

Copy that URL. Keep this terminal open — the tunnel must stay running while you test.

Update publicUrl in oe-config.json

Paste the Cloudflare URL into publicUrl:

"publicUrl": "https://purple-mouse-1234.trycloudflare.com"

Start the Server

Open a second terminal and start OE Runtime in server mode from inside the chain-telegram/ folder:

cd chain-telegram
npx -y @openenthrium/oe-runtime@latest --serve --config oe-config.json

The server detected the Telegram connector, read the publicUrl, and called Telegram's setWebhook API automatically. Your bot is now live.

Test the Pipeline in Telegram

Open Telegram, find your bot, and send this command:

The bot ran greeter, fired followup automatically, then paused on the review skill. It is now waiting for your reply.

Approve the skill

Reply yes (or /approve) in Telegram:

Reject the skill

Reply no (or /cancel) instead:

Commands Reference

All standard OE commands work from Telegram once the bot is connected:

How It Works

When the server receives a Telegram message and the agent runs its skill pipeline:

1. Auto skills fire immediately — output from each skill is passed as context to the next one. All auto-skill results are bundled into the reply you see in Telegram.
2. Manual skills pause. The server stores the pending skill in memory, sends you a Telegram message asking for approval, and waits.
3. When you reply yes, the server looks up the pending skill for your chat, runs it with the previous output as context, and sends the result back to you in Telegram.
4. When you reply no, the skill is discarded and you get a cancellation confirmation.

Real-World Use Cases

• Email drafting — a skill drafts an email, a manual skill sends it. You approve the draft before it goes out.
• Report generation — skills gather data and format a report automatically, then pause before posting it to a shared channel.
• Database updates — skills analyse data and propose a change; you approve before anything is written.
• Daily briefings — the pipeline runs every morning, summarises your emails and calendar, and waits for you to say yes before sending the digest to your team.