[diegosouzapw]

Summary

• Adds open-sse/utils/diagnostics.ts with four exports: reportMalformed200(), synthOpenAIErrorChunk(), synthResponsesFailure(), and detectMalformedNonStream() — all TypeScript, all messages sanitized through sanitizeErrorMessage() per Hard Rule fix(ui): fix Select dropdown dark theme inconsistency #12.
• Wires detectMalformedNonStream() in chatCore.ts after response translation + sanitization + tool-call execution, so providers that return a structurally valid raw body that translates into choices:[] or output:[] now surface as HTTP 502 with a [MALFORMED-200] structured log entry instead of a silent 200 with no content.
• Covers both Chat Completions shape (choices) and Responses API shape (object:"response" + output), including reasoning-only and tool-call-only completions (which are intentionally allowed as valid output).
• Adds tests/unit/diagnostics.test.ts — 18 tests, all passing.

Thanks to @haipham22 for the upstream contribution.

Deferred (not ported)

The following upstream slices were intentionally NOT ported because they require invasive cross-file signature changes that ripple to many call sites:

1. parseSSEToOpenAIResponse return-type change — upstream changes null → {parsed, empty, recvBytes} which requires updating all callers in nonStreamingHandler.js and sseToJsonHandler.js. OmniRoute's equivalent lives across sseParser.ts + nonStreamingSse.ts with multiple callers — a separate, larger PR.
2. getAbortReason() plumbing through streamHandler — upstream adds abort-reason forwarding through createStreamController and down into createSSEStream. OmniRoute's streamHandler.ts has different structure; safe to add but requires its own PR.
3. convertResponsesStreamToJson sawTerminal/textDeltas accumulation — upstream adds terminal-event tracking and text-delta backfill to streamToJsonConverter.js. OmniRoute doesn't have a direct equivalent; deferred for a focused Responses-API improvement pass.
4. stream.js empty-passthrough detection — upstream adds producedOutput() tracking with synthOpenAIErrorChunk/synthResponsesFailure emission in the SSE transform flush. Invasive; deferred.

All four are self-contained follow-ups that can be opened as separate PRs once this core lands.

Test plan

• node --import tsx/esm --test tests/unit/diagnostics.test.ts — 18 tests, all pass
• npm run typecheck:core — clean
• npx eslint open-sse/utils/diagnostics.ts tests/unit/diagnostics.test.ts — no issues
• npm run check:cycles — no cycles
• Pre-commit + pre-push hooks pass (lint-staged, docs-sync, any-budget, tracked-artifacts)

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[gemini-code-assist]

Code Review

This pull request introduces a diagnostics utility to detect and report malformed HTTP-200 upstream responses, integrating it into the chat core handler and adding comprehensive unit tests. The review feedback highlights two key improvements: first, ensuring that valid termination reasons such as 'content_filter' and 'length' are not incorrectly flagged as malformed responses; second, simplifying the synthesized OpenAI error chunk to exclude successful completion fields like 'choices' to prevent client SDKs from ignoring the error.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

If the upstream provider returns a response with empty content but a valid finish_reason such as "content_filter" (indicating the response was filtered for safety) or "length" (indicating it hit the token limit immediately), this function will classify it as "empty_choices". This causes chatCore.ts to return a 502 Bad Gateway error to the client, which is incorrect and prevents the client from properly handling content filtering or length limits.\n\nWe should explicitly allow these valid termination reasons as non-malformed.

  const anyHasOutput = choices.some((choice) => {\n    const c = choice as Record<string, unknown>;\n    const msg = c?.message as Record<string, unknown> | undefined;\n    if (typeof msg?.content === "string" && (msg.content as string).length > 0) return true;\n    if (Array.isArray(msg?.tool_calls) && (msg.tool_calls as unknown[]).length > 0) return true;\n    if (typeof msg?.reasoning_content === "string" && (msg.reasoning_content as string).length > 0)\n      return true;\n    if (c?.finish_reason === "content_filter" || c?.finish_reason === "length") return true;\n    return false;\n  });

[gemini-code-assist]

Synthesizing an error chunk that contains both a top-level error object and a valid choices array with finish_reason: "stop" is contradictory and can cause client SDKs (such as the official OpenAI Python/Node SDKs) to behave unexpectedly. Some client implementations or user loops may stop processing immediately upon seeing finish_reason: "stop" and successfully terminate the stream, completely missing the error field.\n\nTo ensure that stream errors are reliably detected and thrown by client SDKs, the synthesized error chunk should be a pure error object without choices or other successful completion fields.\n\nNote: If you apply this change, remember to update the corresponding test in tests/unit/diagnostics.test.ts (lines 33-42) to match the new error shape.

  const body = {\n    error: {\n      message: safeMessage,\n      type: "upstream_empty_response",\n    },\n  };