Support reliable cross-provider session handoff with normalized tool history

What variant of Codex are you using?

CLI

What feature would you like to see?

Feature request

Codex should support handing off a long-running session from one model provider
to another while preserving its working context.

Example:

• Continue a long GPT session with a custom Responses-compatible provider when
  the GPT quota is exhausted.
• Later hand the resulting session back to OpenAI models.
• Preserve completed conversation history, decisions, tool results, and project
  context without requiring a manual summary.

Current behavior

1. codex resume across providers is unreliable because the session retains its
   original provider-specific state.

2. codex fork -p <provider> -m <model> <session-id> copies the history, but the
   destination provider may reject valid historical tool records.

3. In my case DeepSeek returned:

   No tool output found for tool call call_mj42PDsUwFvbo5vPJM4Kunyd

4. The corresponding custom_tool_call_output was present in the rollout file,
   but a developer message occurred between the tool call and its output.
   OpenAI accepted this history; the custom provider rejected it.

5. Provider-specific pickers also make the source session difficult to locate,
   requiring its UUID.

Desired behavior

Before a cross-provider resume or fork, Codex should normalize the transcript
into a provider-neutral representation:

• preserve all user and assistant messages;
• preserve completed tool calls and outputs;
• guarantee valid call/output pairing and ordering;
• remove or transform provider-specific protocol items;
• retain the source session as an unchanged rollback point;
• clearly indicate that the destination is a new session derived from the
  original session.

Ideally:

codex handoff <session-id> --profile deepseek --model deepseek-v4-pro

Environment

• Codex CLI 0.147.0
• macOS
• Source provider: OpenAI
• Destination: custom Responses-compatible provider

Additional information

No response

Start by tracing the codex resume and codex fork entry points, then inspect the rollout file format that contains the tool call and custom_tool_call_output records. Compare the source history accepted by OpenAI with the ordering rejected by the custom provider. Done means a cross-provider handoff preserves usable messages and completed tools, leaves the source session unchanged, and clearly creates a derived session.