𝗖𝗼𝗱𝗶𝗻𝗴 𝗔𝗴𝗲𝗻𝘁𝘀 𝗼𝘃𝗲𝗿 𝗧𝗲𝗹𝗲𝗴𝗿𝗮𝗺: 𝗧𝗵𝗲 𝗢𝗽𝗲𝗿𝗮𝘁𝗶𝗻𝗴 𝗖𝗼𝗻𝘁𝗿𝗮𝗰𝘁
You have your coding agent running in a terminal pane. Now you need to know how to talk to it.
Think of the relay as a courier. Most of your messages go straight to the agent. A few specific words tell the relay to do something else.
These are your local control commands:
• status: The relay summarizes what the agent is doing right now. Use this to check progress. Note that status shows what the agent claims is happening. You must verify the results yourself.
• send
• compact / new session: The relay sends commands to clean up or start fresh.
• interrupt: The relay stops the current task. Use this if the agent makes a mistake.
• restart: The relay restarts the agent in the same session. Use this if the terminal freezes.
Everything else is just a message for the agent. If you type "Add a retry to the upload client," the relay simply types that into the terminal for the agent to read.
One rule often confuses people. When the agent asks you a question like "Yes or No?", your reply goes straight to the agent. If you type "A," the relay passes "A" to the terminal. It does not try to run "A" as a command.
Safety is your responsibility. Follow these rules to avoid disasters:
Production and staging are read-only. Only inspect and debug. Never edit or delete anything in these environments.
Never push or merge to shared branches like main or dev. Use feature branches only.
Do not change shared infrastructure without human approval.
Never paste secrets or credentials into the chat.
The relay helps, but you are the first line of defense.
You can now run a full coding loop from your phone. Send a task, check the status, answer the agent's questions, and restart if it gets stuck. No SSH or terminal required.
Source: https://dev.to/jerilk/coding-agents-over-telegram-part-3-the-day-to-day-operating-contract-i1a
Optional learning community: https://t.me/GyaanSetuAi