Show HN: Tmux-agent-switcher: see which Claude/Codex agents need your attention
Tmux-agent-switcher is a tmux plugin that surfaces AI coding agent status across sessions, letting developers track which Claude or Codex agents need attention.
A developer going by ymir_e published tmux-agent-switcher, a tmux plugin designed to help users monitor multiple AI coding agents running in parallel, according to the project's GitHub repository and a Show HN post filed on August 12, 2026.
The plugin adds a sidebar, opened with Ctrl+n, that displays all tmux windows across sessions alongside icon-based status indicators showing each agent's current state. When a user selects an idle session, the plugin checkmarks it, providing a visual record of which sessions have already been reviewed. The tool also supports both Vim-style and number-based navigation, per the Show HN post.
The author built the tool around a deliberate non-wrapper philosophy: Claude and Codex are still launched normally, and the plugin infers agent state passively by inspecting tmux metadata and visible terminal output rather than taking ownership of how agents are started or managed. The author notes this approach has a known downside — UI changes or process name changes in the underlying tools can break state detection — but reports no significant issues so far.
The motivation, per the Show HN post, was remote and offline resilience. The author runs agents on a separate machine so that long-running tasks continue even when the laptop is closed or network access is unavailable, such as during a flight. Existing agent management tools were tried but tmux kept winning out for that use case. The plugin was built to improve the overview layer without replacing tmux itself as the process management backbone.
The project surfaced organically before the formal Show HN: the author mentioned it in an earlier Hacker News discussion and noted that 25 people starred it before any dedicated announcement. The repository is open on GitHub under the handle Ymirke. The author closed the post by asking what other developers use to run AI agents on remote or cloud machines, suggesting the post is partly intended to surface comparable tools or approaches in the community.
No comments yet — start the thread.