You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Per-instance setting to allow or block a provider as a delegation target
#39747
Disabling a provider in Settings → Providers, or excluding it through the managed policy's provider allowlist, also blocks sending messages to it directly from the composer. A provider instance can't currently stay available for my own messages while other agents, such as Codex or Claude Code, are prevented from delegating work to it.
Use case
I want to use Antigravity with a personal Google OAuth sign-in directly from the composer. I don't want other agents to start it as a delegate. Google's forum says delegation from external coding agents to personal-OAuth Antigravity is not supported, even with human approval, and may lead to suspension. The forum points to AI Studio API keys or Vertex AI instead: https://discuss.ai.google.dev/t/tos-clarification-is-human-supervised-codex-to-antigravity-acp-delegation-permitted/200697
A separate instance of the same provider that uses an API key may need a different setting.
Proposal
Handle "can delegate to others" and "can receive delegated work" as separate settings for each instance. Blocking an instance as a delegate should not disable direct sends to it or delegation from it.
Minimum requirements:
An on/off setting on each provider instance for receiving delegated work
The T3 server enforces the setting before it launches or queries the provider, whether the instance is named explicitly or inherited implicitly
Blocked instances are hidden from delegation candidates or marked as unavailable
Rejections state the reason
It may also be worth applying the same rule to other automated starts that T3 manages, such as MCP thread launches, automatic sends to existing threads and scheduled tasks, so that a "direct input only" setting can't be bypassed.
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
Uh oh!
There was an error while loading. Please reload this page.
Problem
Disabling a provider in Settings → Providers, or excluding it through the managed policy's provider allowlist, also blocks sending messages to it directly from the composer. A provider instance can't currently stay available for my own messages while other agents, such as Codex or Claude Code, are prevented from delegating work to it.
Use case
I want to use Antigravity with a personal Google OAuth sign-in directly from the composer. I don't want other agents to start it as a delegate. Google's forum says delegation from external coding agents to personal-OAuth Antigravity is not supported, even with human approval, and may lead to suspension. The forum points to AI Studio API keys or Vertex AI instead:
https://discuss.ai.google.dev/t/tos-clarification-is-human-supervised-codex-to-antigravity-acp-delegation-permitted/200697
A separate instance of the same provider that uses an API key may need a different setting.
Proposal
Handle "can delegate to others" and "can receive delegated work" as separate settings for each instance. Blocking an instance as a delegate should not disable direct sends to it or delegation from it.
Minimum requirements:
It may also be worth applying the same rule to other automated starts that T3 manages, such as MCP thread launches, automatic sends to existing threads and scheduled tasks, so that a "direct input only" setting can't be bypassed.
Related: #15182 (delegation routing and defaults)
All reactions