How an IDE Knows Your Coding Agent Is Waiting on You: Hooks, Bells and Guesses
Book: LLM Observability Pocket Guide Also by me: Thinking in Go (2-book series) — Complete Guide to Go Programming + Hexagonal Architecture in Go My project: Hermes IDE | GitHub — an IDE for developers who ship with Claude Code and other AI coding tools Me: xgabriel.com | GitHub Two terminals, two agents, both quiet for four minutes. One is reading forty files before it writes a line. The other…
An IDE can determine whether an AI coding agent is waiting for user input through various methods. One way is by using hooks, which allow the agent to explicitly communicate its state. Another method is through terminal bells, which signal the user but provide no additional information. If an agent sends no signal, the only recourse is to observe the screen and make an educated guess.
Both Claude Code and Codex support the use of scripts to respond to certain events. For example, a script can display a desktop notification on macOS or Linux without producing any output to the terminal. This script, named 'agent-ping', can be placed in the user's bin directory and made executable. When called, it shows a notification with a customizable title and message.
There are some caveats to consider. On macOS, the osascript command used to display notifications requires Script Editor to have notification permissions. If permissions are missing, the notification command fails silently. On Linux, notify-send, the command used for notifications, requires a notification daemon, which headless servers, SSH sessions, and most containers lack.
A similar issue arises with PowerShell on Windows. The guide also provides a solution for Windows, ensuring compatibility across different operating systems.
Claude Code has 33 hook events, two of which are particularly relevant to the user's attention. The 'permission_prompt' event triggers when Claude needs user approval for tool usage, while the 'idle_prompt' event fires when Claude has finished responding and the user has remained silent for about six seconds. These hooks allow the user to decide whether to address the agent's request.
The 'Stop' hook, on the other hand, is triggered when Claude finishes responding, regardless of the time elapsed or user activity. It does not provide any additional information about the nature of Claude's reply, relying on the user to infer from the agent's final message.
By configuring these hooks, users can enhance their interaction with AI coding agents, receiving timely and relevant notifications while maintaining control over how they respond to the agents' requests.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.