[idea] Integration for Deep Agents Code (dcode)

[saad-supports-langchain]

idea / problem

I've been using Deep Agents Code (dcode, LangChain's terminal coding agent) more lately, and it's the one agent in my setup that herdr doesn't recognize yet. dcode panes show up as plain shells, so I don't get the status indicators herdr provides for other agents with a few dcode sessions running side by side, I find myself clicking through panes to figure out which one is waiting on a permission prompt and which is still working through a task.

That's the something herdr solves so nicely for the agents it already supports, so I feel its absence here.

requested change

Support for dcode as an integration target, along the lines of the existing agent integrations:
• Launch/attach awareness for dcode sessions
• Pane state reporting (working / blocked / idle) driven by dcode's hooks
• Screen detection for dcode's TUI states (permission prompt, spinner) if that fits the manifest model

No strong opinions on the implementation, just flagging that the hooks system exists and looks like it was designed for exactly this kind of thing.

why you want this

Somewhat biased because I work at LangChain and would like better support for our tools, but also like being able to use a bunch of different harnesses for various tasks, dcode is especially nice because I can reference my coding agent's traces easily and am not tied to a single model provider (I know opencode provides this as well and I use them as well!)

Ultimately it would allow me to better leverage herdr with my existing tooling and not require me to maintain my own fork to support this.