A bug report gives you two jobs to start. One helper can look for the files involved while another reviews the tests that might catch the problem. In Google Antigravity, a main agent can delegate bounded tasks to subagents and collect their findings in one session.
Define the task and the result you need
A subagent is a helper assigned part of a job by the main agent. Google documents both dynamic subagents and custom agents with their own instructions. If you regularly need a test review, a custom helper can have instructions about which folders to inspect and how to format its findings. For a one-off investigation, ask the main agent to delegate a short report with file paths and likely causes.
Use permission prompts to check the scope
Google explains that the main agent decides which tools and permissions each helper receives. When a tool action needs authorisation, the request can surface in the main interface. A helper reviewing tests may only need to read files. If it requests permission to run a command, check what the command does and where it will run. Compare that request with the scope you set at the beginning.
Combine parallel findings before editing
I’d start with one helper investigating the cause and another reviewing coverage. Giving both the same files to edit would make it harder to see whose change fixed the problem. Comparing their reports in the main session also helps avoid overlapping edits. From there, ask the main agent to propose a single change and identify the test that would verify it.
