[FEATURE]: background monitors

[metux]

Feature hasn't been suggested before.

• I have verified this feature I'm about to request hasn't been suggested before.

Describe the enhancement you want to request

Looking for background monitor feature similar to Claude's one.
The idea is: a separate process running in background, and new output is directly fed into the model, so it can act immediately on that.
(if the shell is idle, then it automatically picks it up, if it's busy then probably has to wait for the next turn)

For example, I'm currently working on a fleeting system, where agents (calling them "ships") can directly talk to each other via a message bus (https://github.com/mpbt-hq/starfleetctl) and so coordinate their work. So far been successful in running tranches of tasks in several projects almost automatically. Initially built it for Claude, but now like to fully port it to opencode.

Been playing around with plugins, but tricky to automatically start a new turn.
(OTOH, if there is a good way to do it via plugin - any help is welcomed :))

[dtinth]

I would love to see this feature too! It is very useful for:

• Shepherding PRs (self explanatory)

• Gangprompting in Discord - Different folks are doing gangprompting with Discord differently:

  • OpenCode team built their own Discord bot based on OpenCode 2.0, but it's not out yet (been waiting for a month)
  • Kimaki is a Discord bot that wraps the OpenCode server, injects its own plugins for deep integration.
  • Discord OpenCode Bridge (I built it) is a standalone Discord bot that connects to an existing OpenCode server without requiring custom plugins.

  However, vanilla Claude Code already supports gangprompting through the Monitor tool (the subject of this issue). Here's all I need to do:

  1. Set DISCORD_TOKEN in my env.
  2. Give Claude Code the channel ID to loop itself into.

  Claude Code then:

  1. Writes a script that polls Discord for new messages via the REST API (it just dumps the message as JSON object, an LLM knows how to parse it)
  2. Runs the script using the Monitor tool → New messages in Discord becomes a synthetic prompt
  3. Curls the Discord's REST API to respond

  Now the whole team has a direct line to the Claude Code instance. No need to use any fancy plugin.

That's why I think the Monitor tool is one missing primitive that would unlock a lot of possibilities without requiring plugins.

More links:

• OpenCode v2 has a background: true option (implemented in feat(core): implement V2 session.shell #35183) which allows running a background shell, but it only reports back when finished.
• @fl4p made PR feat(opencode): add Monitor tool (per-line background watcher) #33806 which implements this feature. However, seems like no one is looking at it right now.