Repository navigation
SubAgents are enclosed, prohibiting genuine task management with OpenCode. #11012
Description
Description
There is very little documentation on the SubAgent strategy, and little community discussion.
The issue: Agent → creates SubAgent to do thorough research via Task → SubAgent returns a summary to the Agent.
SubAgents operate in their own context, which makes them a de-facto task-management layer — but a SubAgent session cannot be re-entered after it returns. Example: the SubAgent produces excellent in-depth research but returns only a summary; the session closes; the Agent cannot instruct the SubAgent to write its full research to a file, improve it, or continue orchestrating. Dozens of obvious workflows are blocked by this limitation.
I understand this may be a deliberate feature-not-a-bug, but the value is unclear. It seems only a matter of time until workflows are expressed with specialty syntax anyway (like @):
Prompt: Create a %Workflow[Research, Review, Bibliography] such that Research uses @subagent/Researcher to … while Review takes the results from Research to cross-investigate …
In the meantime I am manually deleting tool responses from the exported subagent session. This is a critique of the ordering of priorities on this issue.
OpenCode v1.1.39.
Plugins
No response
OpenCode version
No response
Steps to reproduce
No response
Screenshot and/or share link
No response
Operating System
No response
Terminal
No response
Activity
github-actions commentedon Jan 28, 2026
This issue relates to several existing issues about SubAgent design limitations. Please check these related issues:
These issues all share the core theme of SubAgent enclosure and architectural limitations in task management workflows. The most directly related is #6584 which proposes a resumeSessionId parameter to continue subagent work across phases.
maurits150 commentedon Feb 20, 2026
A main agent can continue conversing with any subagent session at any moment. The session is not closed. The main agent is given a task id to reuse.
rektide commentedon Mar 10, 2026
I find it quite frustrating as well. For a month or so it was possible to communicate with subagents. You had to escape to stop them from their roll, but you could hit escape twice then talk with them. And you could keep chatting with them after.
It was really sad seeing subagents reduced to what they are now. Yes I can have the main agent talk to them again, and that's a capability I want, yes, but I also want to be able to chat with them directly.
This is a bit extreme, but to go all the way, it would be amazing to see subagents and agents all interoperate over a2a with each other, to make subagents not that much different than regular agents, and using normative protocols for exchange. #3023
Mte90 commentedon Apr 8, 2026
Also there aren't hooks to use to change the prompt and inject stuff in case.
maurits150 commentedon Apr 9, 2026
Fair, this was a useful feature, and it should come back.
github-actions commentedon Jul 6, 2026
To stay organized issues are automatically closed after 60 days of no activity. If the issue is still relevant please open a new one.