Coding

Fixing Bugs From a Walk: He Sends a 20-Second Telegram Voice Note and His Local Coding Agent Does the Work, Asking Before Risky Commands

  • Level Advanced
  • Setup1–3 hrs (estimated)
  • Evidence Some evidence
  • CostNot stated in source
  • No-codeNo
Unnamed agentSource-linked

What did they do with Unnamed agent?

He uses Google's Antigravity coding agent every day but had to sit at his desk to use it. Typing long prompts on a phone was slow, so he built a small bridge from Telegram to the agent on his Mac. Now when he thinks of a bug fix on a walk, on the train or while making dinner, he sends a short voice note. The bridge only accepts messages from his own Telegram account, saves the audio, and hands it to the agent, which listens, plans and edits code on his machine. One status message updates with what it's doing instead of flooding the chat, and before risky commands like deleting files it pauses and sends Allow Once and Reject buttons to his phone. No router ports are opened.

How it went

dev.to post by Doni Leong (Nov 7, 2026, edited Nov 9): agy-pilot bridge from Telegram to a local Antigravity session on macOS; Telegram user ID allowlist; voice notes saved as .ogg and passed to the agent directly (no separate speech-to-text API); single status message edited at most every 1.5 s to avoid Telegram 429 errors; inline Allow Once/Reject buttons for risky tools; launchd keeps it running; long-polling so no open ports. No time-saved number given. Not affiliated with Google

Worth knowing

  • dev.to post by Doni Leong (Nov 7, 2026, edited Nov 9): agy-pilot bridge from Telegram to a local Antigravity session on macOS; Telegram user ID allowlist; voice notes saved as .ogg and passed to the agent directly (no separate speech-to-text API); single status message edited at most every 1.5 s to avoid Telegram 429 errors; inline Allow Once/Reject buttons for risky tools; launchd keeps it running; long-polling so no open ports. No time-saved number given. Not affiliated with Google.

Prompt

Cleaned version rewritten from the source — not a direct quote.

I'll send you voice notes from my phone with coding tasks for [project folder]. For each one: listen to the audio, restate the task in one line, make a short plan, then do the work and run the tests. Keep one status update going instead of many messages. Before any risky command (deleting files, force-pushing, installing packages, anything outside the project folder), stop and ask me to approve it. When you finish, send me a short summary of what changed and whether the tests passed.

Recommended

Try it in Grok Bot

  1. Copy the prompt (button below).
  2. Open Grok Bot chat and paste it. Replace any [bracketed] placeholders.
  3. Connect the listed tools (Telegram) as connectors when Grok Bot asks.
  4. If the task repeats, ask: "Make this a daily routine."

Alternatives: the same prompt works in Claude, ChatGPT Agent, OpenClaw.

Set it up

Tools: Telegram · Stack: Google Antigravity + Telegram bot

  1. Read the original source for full context.
  2. Connect the tools your agent needs.
  3. Paste the prompt and replace [bracketed] placeholders.
  4. Run once manually and check the output before scheduling it.

Generic suggested steps, not from the source.

Evidence

dev.to post by Doni Leong (Nov 7, 2026, edited Nov 9): agy-pilot bridge from Telegram to a local Antigravity session on macOS; Telegram user ID allowlist; voice notes saved as .ogg and passed to the agent directly (no separate speech-to-text API); single status message edited at most every 1.5 s to avoid Telegram 429 errors; inline Allow Once/Reject buttons for risky tools; launchd keeps it running; long-polling so no open ports. No time-saved number given. Not affiliated with Google (as of 2026-11-12)

Source: Web · Verified by us: not yet · Report broken or outdated

Related use cases

CodingX · 2026-10

Hermes Agent Nearly Doubled Speech-to-Text Accuracy on Hard Words: 99 to 183 of 240

A user turned his 15K saved Wispr Flow dictations into a test set for the local Whisper model behind his Hermes Agent. The agent turned his Wispr dictionary into a hint list, diffed raw vs corrected text to find words Whisper keeps mishearing, made find-and-fix rules, and tested every rule against the full history. With no fine-tuning, dictionary words went from 99/240 to 183/240 on Wispr clips and 19/45 to 43/45 on sentences he read aloud. It now archives the data nightly and looks for new fixes weekly.

  • Level Advanced
  • Setup1–3 hrs*
  • Evidence Strong evidence
Hermes AgentSource-linked
CodingReddit · 2026-10

1 Minute to 1 Second: Claude Code Built a Faster Plex App for a Samsung TV

On a 2022 Samsung TV the official Plex app took about a minute to open and search lagged about 10 seconds. A Redditor used Claude Code to build a free, open-source web-app client that opens in about a second, searches as you type and plays 98.7% of his library directly in hardware.

  • Level Advanced
  • Setup1–3 hrs*
  • Evidence Strong evidence
Claude CodeSource-linked
CodingReddit · 2026-10

30 MLB Clubs in a 1994 Cartridge: Claude Code Expanded Ken Griffey Jr. Baseball With a Franchise Mode

Over about two weeks, a fan and Claude Code turned the 1994 SNES game Ken Griffey Jr. Presents Major League Baseball into a free patch with all 30 of today's clubs and 2026 rosters. It adds season and franchise modes (players age and retire, 300 real prospects) that fit in the cartridge's original 8 KB save, plus browser play.

  • Level Advanced
  • Setup1–3 hrs*
  • Evidence Strong evidence
Claude CodeSource-linked