What works in each
The What works column uses the Agents page’s words:- Runs, reports and answers: sessions show under Runs, report milestones and ask you questions, and your answers reach them.
- Runs and reports; it checks for answers: sessions show under Runs and report, and the agent fetches your answers itself.
- Reports; it checks for answers: the app has no automatic hooks. Instructions tell each session to start a run, report milestones and fetch your answers.
“Not yet” means the connection was built from the app’s documentation and hasn’t been seen working in a real session yet. The Agents page and setup’s summary say the same.
For OpenCode, setup installs only for OpenCode v2. For OpenCode v1, or a version it can’t tell, it installs nothing and says so.
How your answers reach each one
An end-of-turn hook waits for an answer for five minutes by default, and only while a question or approval from the session is open. An answer that comes later reaches the session as the table says, or when the agent next fetches its answers.
The session you paste the prompt into
That session started before its hooks existed, so setup connects it directly when it can tell which session it is, and it shows under Runs straight away. Some apps load new hooks only in a new session; setup’s summary says when to start one, and the pasted session stays live through its own reports until then.An agent that isn’t listed
An agent setup doesn’t connect can still connect itself. Setup points it tospaces connect --help, which prints a short guide for the agent to follow. The agent shows you the exact change first, makes it only in its own user-level files, and writes no credential. Its sessions then show under Runs, or report and fetch your answers, under a name it chooses.
Hosted assistants, which run on their vendor’s servers, connect differently: see hosted assistants.