Describe your idea
A request to port or wrap BMAD agents (currently configured as "Skills") into the Antigravity IDE's native "Plugins/Subagents" infrastructure to enable true parallel and asynchronous execution.
Why is this needed?
Currently, all BMAD agents are registered as Skills (.agent/skills/), which forces them into a strictly synchronous, sequential execution flow within the main IDE chat thread. Even with tools like /bmad-party-mode, the interaction simulates a sequential roundtable rather than true concurrency.
For users handling large-scale codebases, delegating heavy background tasks to different agents simultaneously (e.g., having an Amelia/Dev agent refactor a file while a Quinn/QA agent generates tests in the background) is bottlenecked because Skills lock the primary user session.
How should it work?
BMAD agents should be packaged (or provided with a wrapper configuration) that registers them under Antigravity's /plugins/agents/ directory structure instead of just /skills/. By adhering to the Plugin architecture, the Antigravity IDE would be able to launch them as true Subagents—similar to how it currently treats its native browser_subagent.
This would enable asynchronous background execution, direct filesystem manipulation without locking the main AI prompt, and genuine multi-agent software development orchestration.
PR
I am not submitting a PR at this time, but I'm opening this issue to suggest the feature and ask for community workarounds.
Additional context
I realized that solving this synchronous bottleneck requires integrating with the Antigravity Plugin framework rather than just expanding the conversational Skills logic. Any guidance on whether there is an active roadmap for this integration, or if an obscure community library already exists, would be greatly appreciated!
Describe your idea
A request to port or wrap BMAD agents (currently configured as "Skills") into the Antigravity IDE's native "Plugins/Subagents" infrastructure to enable true parallel and asynchronous execution.
Why is this needed?
Currently, all BMAD agents are registered as Skills (
.agent/skills/), which forces them into a strictly synchronous, sequential execution flow within the main IDE chat thread. Even with tools like/bmad-party-mode, the interaction simulates a sequential roundtable rather than true concurrency.For users handling large-scale codebases, delegating heavy background tasks to different agents simultaneously (e.g., having an Amelia/Dev agent refactor a file while a Quinn/QA agent generates tests in the background) is bottlenecked because Skills lock the primary user session.
How should it work?
BMAD agents should be packaged (or provided with a wrapper configuration) that registers them under Antigravity's
/plugins/agents/directory structure instead of just/skills/. By adhering to the Plugin architecture, the Antigravity IDE would be able to launch them as true Subagents—similar to how it currently treats its nativebrowser_subagent.This would enable asynchronous background execution, direct filesystem manipulation without locking the main AI prompt, and genuine multi-agent software development orchestration.
PR
I am not submitting a PR at this time, but I'm opening this issue to suggest the feature and ask for community workarounds.
Additional context
I realized that solving this synchronous bottleneck requires integrating with the Antigravity Plugin framework rather than just expanding the conversational Skills logic. Any guidance on whether there is an active roadmap for this integration, or if an obscure community library already exists, would be greatly appreciated!