feat(openai): add responses reasoning input compatibility flag

[bkidy]

Summary

• add a default-off REMOVE_RESPONSES_REASONING_INPUT flag for OpenAI Responses requests
• strip reasoning* items from input when the flag is enabled before forwarding upstream
• help partially compatible relays avoid account/session-bound errors such as Item with id 'rs_...' not found

Why

Some OpenAI-compatible relays can serve /v1/responses, but do not fully support account/session-bound reasoning state across turns. In those cases, replaying reasoning items from prior responses may fail with errors like:

• Item with id 'rs_...' not found
• invalid_encrypted_content

This PR adds a small compatibility escape hatch without changing the default behavior.

Behavior

• default remains unchanged
• when REMOVE_RESPONSES_REASONING_INPUT=true, reasoning* items are removed from the Responses input array before the request is sent upstream
• malformed input items are preserved as-is

Trade-offs

This compatibility mode may reduce some Responses-native continuity/cost benefits because account-bound reasoning items are no longer replayed.

Summary by CodeRabbit

Release Notes

• New Features
  • Added configurable option to remove reasoning data from responses via environment variable.
  • Integrated filtering mechanism with error handling for response processing.

[bkidy]

I hit this in production with OpenAI Responses routed through a partially compatible relay.
The relay can serve /v1/responses, but replaying account/session-bound reasoning items across turns may fail with errors like:

• Item with id 'rs_...' not found
• invalid_encrypted_content
  This PR keeps default behavior unchanged and adds a small compatibility escape hatch:
• disabled by default
• enabled only via REMOVE_RESPONSES_REASONING_INPUT=true
• strips reasoning* items from the Responses input array before forwarding upstream
  I kept the change intentionally small to avoid affecting the normal OpenAI Responses path.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 7acb97f1-4389-4439-8db3-99b22a806da9

📥 Commits

Reviewing files that changed from the base of the PR and between 4e1b05e and 86494be.

📒 Files selected for processing (4)

• common/init.go
• constant/env.go
• dto/openai_request.go
• relay/channel/openai/adaptor.go

---

Walkthrough

The pull request introduces an optional feature to remove reasoning-related data from OpenAI responses requests. It adds an environment variable configuration flag, a JSON filtering method, and integrates the filtering into the request processing pipeline.

Changes

Cohort / File(s)	Summary
Configuration & Feature Flag common/init.go, constant/env.go	Added environment variable REMOVE_RESPONSES_REASONING_INPUT (defaults to false) and initialized it as a package-level boolean flag.
Request Processing dto/openai_request.go, relay/channel/openai/adaptor.go	Implemented RemoveReasoningFromInput() method to filter JSON array items by type and integrated conditional call in adaptor with error handling when flag is enabled.

Sequence Diagram

sequenceDiagram
    participant Env as Environment
    participant Init as Initialization
    participant Flag as constant.RemoveResponsesReasoningInput
    participant Adaptor as OpenAI Adaptor
    participant Request as OpenAIResponsesRequest
    participant DTO as Request DTO

    Env->>Init: Read REMOVE_RESPONSES_REASONING_INPUT
    Init->>Flag: Set boolean value
    Adaptor->>Flag: Check if enabled
    alt Flag Enabled
        Adaptor->>Request: Call RemoveReasoningFromInput()
        Request->>DTO: Parse Input as JSON array
        DTO->>DTO: Filter items (remove reasoning type)
        DTO->>Request: Assign filtered array back to Input
        Request->>Adaptor: Return result
    else Flag Disabled
        Adaptor->>Adaptor: Continue without filtering
    end

Loading

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~15 minutes

Possibly related PRs

• Fix reasoning adaptor for openrouter #1577: Modifies reasoning-related field handling in the same adaptor file (relay/channel/openai/adaptor.go), involving reasoning data transformation for different API providers.

Poem

🐰 A filter whispers through the streams,
Removing reasoning from dreams,
With flags and JSON carefully dressed,
The responses flow, now cleaned and blessed! ✨

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the main change: adding an environment-driven compatibility flag (REMOVE_RESPONSES_REASONING_INPUT) for OpenAI Responses request handling to filter reasoning items.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment

📝 Coding Plan

• Generate coding plan for human review comments

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Tip

You can validate your CodeRabbit configuration file in your editor.

If your editor has YAML language server, you can enable auto-completion and validation by adding # yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json at the top of your CodeRabbit configuration file.

[RedwindA]

建议研究并使用渠道亲和性功能