[Bug]: Forward prompt_cache_key through the openai-chat adapter #217

Summary

A caller can send prompt_cache_key as an optional string to /v1/responses.

OpenCodex's Responses parser writes the value to options.promptCacheKey.

createOpenAIChatAdapter().buildRequest() builds the outbound /chat/completions body without that value.

The OpenAI Chat Completions Create API lists prompt_cache_key as an optional string.

OpenAI uses the key to group similar requests for prompt caching and improve cache hit rates.

OpenCodex should copy parsed.options.promptCacheKey to body.prompt_cache_key without changing the string.

OpenCodex should omit the upstream field if the caller does not send it.

Reproduction

1. Add a provider that uses the openai-chat adapter.

2. Send a Responses request with prompt_cache_key.

   curl http://127.0.0.1:10100/v1/responses \
     -H 'Content-Type: application/json' \
     -d '{
       "model": "example/example-model",
       "input": "Hello",
       "prompt_cache_key": "shared-prefix-v1",
       "stream": false
     }'

3. Capture the request that OpenCodex sends to the provider's /chat/completions endpoint.

4. Confirm that the outbound body lacks prompt_cache_key.

Logs and screenshots

src/responses/schema.ts
prompt_cache_key: z.string().optional()

src/responses/parser.ts
options.promptCacheKey = data.prompt_cache_key

src/adapters/openai-chat.ts
createOpenAIChatAdapter().buildRequest() constructs the body without prompt_cache_key

Expected outbound field:
prompt_cache_key="shared-prefix-v1"

Area

Provider adapter

Version

2.7.30

OS

macOS 26.5.2 (arm64)

Config shape

{
  "defaultProvider": "example",
  "providers": {
    "example": {
      "adapter": "openai-chat",
      "baseUrl": "https://example.invalid/v1",
      "authMode": "key",
      "apiKey": "<redacted>",
      "defaultModel": "example-model"
    }
  }
}

Checks

• I searched existing issues and docs first.
• I removed secrets, tokens, and personal data from logs and screenshots.