[Bug] task tool returns empty response for Google/Gemini model agents

[JaeungHyun]

Bug Description

When using the task tool to invoke agents configured with Google/Gemini models, the tool returns an empty response. The same agents work correctly when called through call_omo_agent (for supported agents like explore and librarian).

Environment

• OpenCode with oh-my-opencode plugin
• Antigravity authentication configured and working
• All OAuth credentials verified (Google, Anthropic, OpenAI)

Steps to Reproduce

1. Configure an agent with a Gemini model in oh-my-opencode.json:

"ui_expert": {
  "model": "google/gemini-3-pro-high"
}

2. Call the agent using task tool:

task(subagent_type="ui_expert", prompt="Reply OK only")

3. Observe: Empty response returned

Tested Model Variants

All return empty responses via task tool:

• google/gemini-3-pro-high
• google/gemini-3-flash
• google/antigravity-gemini-3-pro-high
• google/antigravity-gemini-3-flash
• google/antigravity-gemini-3-pro-low

Working Cases

Tool	Agent	Model	Result
task	flutter_arch	anthropic/claude-opus-4-5	✅ Works
task	security_audit	openai/gpt-5.2	✅ Works
task	ui_expert	google/gemini-3-pro-high	❌ Empty
task	frontend-ui-ux-engineer	google/gemini-3-pro-high	❌ Empty
call_omo_agent	librarian	google/gemini-3-flash	✅ Works
call_omo_agent	explore	google/gemini-3-flash	✅ Works

Authentication Verification

$ opencode auth list
●  Anthropic oauth
●  OpenAI oauth  
●  Google oauth

●  Google GEMINI_API_KEY (env)

Antigravity authentication also confirmed working with valid refresh token.

Expected Behavior

task tool should return responses from Gemini model agents, similar to how it works for Anthropic and OpenAI models.

Workaround

Currently, only explore and librarian agents can use Gemini models (via call_omo_agent). Custom agents like ui_expert cannot use Gemini models.

Temporary workaround: Change custom agents to use Anthropic or OpenAI models instead.

[ea-acar]

I think this fix can be applied here as well:

#464 (comment)