Bedrock: task title generation still uses unsupported gpt-5.4-mini after #26288 fix (memory path only) #28741

Summary

#26288 reported that Codex Desktop helper turns on Amazon Bedrock use the bare gpt-5.4-mini slug, which Bedrock/Mantle rejects with a 404. It was closed as completed by #27129 ("feat: use provider defaults for memory models") — but that PR only fixed the memory extraction/consolidation path. Task/thread title generation still uses gpt-5.4-mini, so the 404s continue on a clean install.

Splitting this out since #26288 is closed and the title-gen path was never addressed.

Environment

• codex-cli 0.137.0
• Provider: amazon-bedrock, model openai.gpt-5.4, region us-east-2
• Endpoint: https://bedrock-mantle.us-east-2.api.aws/openai/v1/responses

What's still broken

Mantle only serves the prefixed engines (openai.gpt-5.4, openai.gpt-5.5). The title-generation helper still submits the bare slug gpt-5.4-mini, which isn't a registered engine there → 404.

Across my local logs (~/.codex/logs_2.sqlite), the split is absolute:

Every gpt-5.4-mini call I can find today (post-#27129) is the title helper — 0/178 succeed. The prompt confirms it:

You are a helpful assistant. You will be presented with a user prompt,
and your job is to provide a short title for a task...

submitted with:

model=gpt-5.4-mini wire_api=responses transport="responses_http" http.method="POST" api.path="responses"

Request completed POST https://bedrock-mantle.us-east-2.api.aws/openai/v1/responses status=404 Not Found
Turn error: JSON-RPC error -32602: Job registration failed: Engine bad request:
Task submission failed with status 404 Not Found: Engine not found

It surfaces two ways: as a sibling title thread that 404s directly, and as a Turn error banner on the main turn even though the user-facing openai.gpt-5.4 response returned 200.

Expected

Title generation should resolve its model through the active ModelProvider, same as #27129 did for memory models — i.e. use the Bedrock-specific ID (openai.gpt-5.4) instead of the hardcoded gpt-5.4-mini. Ideally with a config override (there's currently none for title gen, unlike memories.extract_model).

Impact

Low functional impact (main agent works), but it's constant background 404 noise and the Turn error banner reads like a real failure on every fresh chat / first prompt.

Related: #26288 (closed), #27129 (the partial fix), #26808 (memory-writer model controls).