fix(realtime): prevent ambiguous multi-agent tool dispatch

[mavrickdeveloper]

Summary

This pull request fixes ambiguous Realtime tool dispatch by rejecting duplicate model-visible names across Realtime function tools and handoffs.

Motivation

This came up while working with a multi-agent Realtime setup where tools and handoffs were composed together from different agent modules. In that kind of system, function tools and handoffs share the same callable namespace from the model's point of view, so duplicate names can become ambiguous without being obvious at setup time.

For example, a triage agent may end up with both:

function tool: transfer_to_billing
handoff tool: transfer_to_billing

or two handoffs may both be exposed as:

transfer_to_support

Before this change, Realtime could accept that configuration and later build SDK-side lookup maps keyed only by name. That means one entry could silently overwrite another, making it unclear whether a model call should run a Python function or transfer to another agent.

Changes

• Adds Realtime-specific duplicate-name validation for model-visible function tool and handoff names.
• Validates before Realtime session tools are serialized and sent to the model.
• Validates again before local tool-call dispatch, so ambiguous runtime state fails clearly.
• Raises a UserError that includes the duplicated name and where it came from, for example function tool vs handoff.
• Only checks the tools and handoffs that are actually enabled for the session, so disabled handoffs do not cause false conflicts.

Tests

• UV_FROZEN=1 bash .agents/skills/code-change-verification/scripts/run.sh

[seratch]

This looks good overall. One change before merging: please resolve and validate the initial model settings before registering the session as a model listener. Currently, validation failure during __aenter__ leaves the listener registered because __aexit__ is never called. Please also remove the listener if connect() fails, and add a regression test covering this cleanup behavior.

[mavrickdeveloper]

@codex review

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1c0280850c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: f9abf70b5c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[seratch]

Thanks for the updates. initial_model_settings["handoffs"] is validated after the settings override is applied, but those handoffs are not passed through the same is_enabled filtering used for agent handoffs. That means a disabled override handoff can now cause session startup to fail with a duplicate-name UserError, even though the PR states disabled handoffs should not participate in the model-visible namespace.

Please either filter/evaluate override handoffs before duplicate-name validation, or add a test and clarify that initial_model_settings["handoffs"] is already expected to be a final enabled list.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e95ca35a5c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[mavrickdeveloper]

@codex review

[chatgpt-codex-connector]

Note

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: e7a4af377c

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".