fix(codex): skip pass-through for unsupported response limits

[llc1123]

Summary

• add an optional outbound pass-through body policy so provider-specific veto rules do not live in the generic orchestrator pass-through middleware
• make Codex veto OpenAI Responses raw body pass-through when the client body contains token limit fields Codex rejects
• keep non-Codex channels on the existing generic pass-through rules

Root cause

Codex outbound removes token limit fields before sending upstream, but the generic pass-through middleware could later restore the original OpenAI Responses body. Requests with fields like max_output_tokens then reached Codex unchanged and failed with HTTP 400, while fallback OpenAI Responses-compatible channels such as ZenMux succeeded.

Current pass-through enablement

Body pass-through is enabled only when all generic orchestrator checks pass: current channel exists, provider request has an API format, inbound LLM API format equals outbound provider API format, original/effective stream semantics match, and channel/global pass-through setting is enabled. After that, an optional outbound policy can veto pass-through for provider-specific incompatibilities.

Tests

• go test ./internal/server/orchestrator -run 'TestApplyPassThroughBody(SkipsWhenOutboundPolicyRejects|PreservesMappedModel|PreservesAlignedStreamWithoutPatchingIt)' -count=1\n- cd llm && go test ./transformer/openai/codex -run 'TestCodexOutbound_(RejectsPassThroughBodyWithTokenLimitFields|AllowsPassThroughBodyWithoutTokenLimitFields|StreamAcceptHeader)' -count=1

Summary by CodeRabbit

• Bug Fixes
  • Improved Codex Responses request handling when unsupported token-limit fields are provided.
  • Requests containing unsupported token-limit fields are now processed through supported transformations instead of being passed through unchanged.
  • Preserved model mappings and supported token-limit settings for OpenAI Responses.
  • Requests without conflicting token-limit fields continue to use the original request body when appropriate.
  • Added safeguards to prevent unsupported request fields from reaching providers unchanged.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 108a5f88-867f-4669-8ef3-12b64f9cbf1f

📥 Commits

Reviewing files that changed from the base of the PR and between 90b377c and ac93238.

📒 Files selected for processing (1)

• internal/server/orchestrator/pass_through_test.go

🚧 Files skipped from review as they are similar to previous changes (1)

• internal/server/orchestrator/pass_through_test.go

---

📝 Walkthrough

Walkthrough

The pass-through flow now consults an optional transformer policy before reusing inbound request bodies. Codex Responses requests with token-limit fields reject pass-through. Tests cover rejection and allowed bodies.

Changes

Pass-through policy

Layer / File(s)	Summary
Policy contract and orchestration llm/transformer/interfaces.go, internal/server/orchestrator/pass_through.go	The transformer interface defines pass-through approval. The orchestrator preserves the outbound body when approval is denied and allows pass-through by default when no policy exists.
Codex token-limit filtering llm/transformer/openai/codex/outbound.go	The Codex transformer rejects pass-through for Responses bodies containing max_output_tokens, max_completion_tokens, or max_tokens. Nil, non-Responses, and bodyless requests remain allowed.
Policy validation internal/server/orchestrator/pass_through_test.go, llm/transformer/openai/codex/outbound_executor_test.go	Tests verify policy rejection, unchanged requests, absent unsupported fields, blocked token-limit fields, and allowed bodies without those fields.

Estimated code review effort: 3 (Moderate) | ~20 minutes

Mergeability Score: ⚪ Minimal · up to ac932

The PR limits Codex pass-through for unsupported token-limit request bodies while preserving existing behavior for other channels; no actionable merge-blocking risk remains after normal checks and review.

Sequence Diagram(s)

sequenceDiagram
  participant Orchestrator
  participant CodexTransformer
  participant RequestBody
  Orchestrator->>CodexTransformer: Check AllowPassThroughBody
  CodexTransformer->>RequestBody: Inspect token-limit fields
  RequestBody-->>CodexTransformer: Field presence
  CodexTransformer-->>Orchestrator: Allow or reject pass-through
  Orchestrator->>RequestBody: Reuse inbound body only when allowed

Loading

Possibly related PRs

• looplj/axonhub#2178: Both PRs modify Codex Responses pass-through handling in internal/server/orchestrator/pass_through.go; this PR adds body-policy filtering, while the related PR forwards metadata headers.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly identifies the main Codex change: skipping pass-through for unsupported response limit fields.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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.}