# Bug: Telegram integration fails with `TypeError: fetch failed` from Node.js process, but `curl` works from shell

# Bug: Telegram integration fails with `TypeError: fetch failed` from Node.js process, but `curl` works from shell

## Environment

- **Moltworker:** Deployed via "Deploy to Cloudflare" button
- **OpenClaw version:** 2026.2.17 (upgraded from 2026.2.3 — same behavior on both)
- **Container:** Cloudflare Sandbox (`standard-1`)
- **Secrets configured:** `TELEGRAM_BOT_TOKEN`, `ANTHROPIC_API_KEY`, `MOLTBOT_GATEWAY_TOKEN`, `CF_ACCESS_AUD`, `CF_ACCESS_TEAM_DOMAIN`, `CF_ACCOUNT_ID`, `R2_ACCESS_KEY_ID`, `R2_SECRET_ACCESS_KEY`, `CDP_SECRET`, `WORKER_URL`, `DEBUG_ROUTES=true`

## Description

Telegram channel shows as configured and enabled in the gateway config, and the bot token is valid. However, the grammy library's `getUpdates` polling fails continuously with `TypeError: fetch failed` from Node's built-in undici fetch.

**The key finding:** Running `curl` to `api.telegram.org` from within the container shell **works** — messages are successfully sent. But the Node.js gateway process **cannot** make outbound HTTP requests to Telegram's API.

This suggests the Cloudflare Sandbox/Firecracker environment has different network access rules for the Node.js process vs. shell commands.

## Steps to Reproduce

1. Deploy Moltworker via the "Deploy to Cloudflare" button
2. Set `TELEGRAM_BOT_TOKEN` secret with a valid bot token from @BotFather
3. Restart the gateway
4. Send a message to the bot on Telegram
5. Observe no response; check debug logs

## Expected Behavior

Telegram bot receives and responds to messages.

## Actual Behavior

Gateway starts and recognizes Telegram config:

```
[telegram] [default] starting provider (@claw_gpolzer_bot)
[telegram] autoSelectFamily=false (default-node22)
```

Then immediately begins failing with repeated fetch errors:

```
[openclaw] Non-fatal unhandled rejection (continuing): TypeError: fetch failed
at node:internal/deps/undici/undici:13502:13
at processTicksAndRejections (node:internal/process/task_queues:105:5)
```

With periodic Telegram-specific errors:

```
[telegram] network error: Network request for 'getUpdates' failed!; retrying in 2.16s.
[telegram] network error: Network request for 'getUpdates' failed!; retrying in 4.49s.
```

These errors repeat indefinitely every ~10 seconds.

## Diagnostic Evidence

**OpenClaw's own status check reports Telegram as healthy:**

```
│ Telegram │ ON │ OK │ token config (8243…MWJk · len 46) · accounts 1/1
```

**But sending a message via curl from the container shell succeeds:**

OpenClaw was able to run `curl` to `api.telegram.org/bot/sendMessage` and the message was delivered to Telegram. This confirms:
- The bot token is valid
- The container *can* reach `api.telegram.org` via shell commands
- The issue is specific to Node.js `fetch()` / undici within the gateway process

## Possible Cause

The Cloudflare Sandbox (Firecracker) environment may apply different network isolation rules to the main Node.js process vs. child processes / shell commands. The grammy library uses Node's built-in `fetch` (undici), which may be subject to stricter networking constraints in the sandbox.

## Workaround

None found. The Control UI (WebSocket-based) works correctly for all interactions — only external API calls from the Node.js process are affected. This likely also impacts Discord and Slack integrations.