You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Repair for #587 at 9d10fa9, and a superset of #591's request-level hoist.
Keep stream_options.include_usage / include_obfuscation JSON null (and null+false mixes) as omit-equivalent no-ops on chat, Completions, and Responses.
Unknown stream_options keys fail closed even when their value is null.
Tools-path message honesty: weight, prefix, refusal, annotations, developer role, empty user/system content, multimodal part shape, and participant name use the same named errors before proxy_completion.
Buyer next action: send SDK-default null flags; omit stream (or set false) on tool-calling requests; always send a pool model; replace developer with system. Do not set include_usage=true / include_obfuscation=true.
Do not merge #587 at 9d10fa9. Prefer this head over #582–#587 and #591. Merge only after Full unit + Semgrep are green and an independent non-author review lands.
…closed otherwise
Chat history: message-level audio and legacy function_call are null/empty omit
no-ops; non-empty fail closed with named errors (including tools passthrough).
Tip substrate from #577 assistant refusal/annotations honesty. Local full unit:
940 passed.
…ed otherwise
OpenAI fine-tune style message weight is not applied on this gateway.
Accept null/0/1 as honest no-ops; reject other types and values with
invalid_message_weight. Tip substrate from #578. Local full unit: 943 passed.
…ion role
Reject unsupported message keys with named unknown_message_fields (not silent
strip or tools-passthrough smuggle). Reject legacy function role with
invalid_message_role migration to tool. Tip substrate from #579. Local full
unit: 947 passed.
OpenAI partial-assistant prefix flag is not applied on this gateway.
null/false are honest no-ops; true and non-booleans fail closed with
invalid_message_prefix. Tip substrate from #580. Local full unit: 950 passed.
…therwise
Named invalid_max_tool_calls on /v1/chat/completions instead of opaque
unknown_fields. Aligns with Responses max_tool_calls honesty; gateway has
no multi-step tool loop.
…losed otherwise
Legacy /v1/completions treated max_tool_calls as unknown_fields. Accept the
key for named invalid_max_tool_calls (null/empty/whitespace omit-equivalent),
matching chat/Responses honesty so SDKs get a clear migration path.
SDK clients often send include_usage/include_obfuscation as JSON null.
Drop null flag values before validation so null (and null+false mixes)
match omit / all-false no-ops on chat, Completions, and Responses.
True flags remain fail-closed with invalid_stream_options.
…age honesty
Null include_usage/include_obfuscation stay omit-equivalent, but unknown
stream_options keys no longer become no-ops just because their value is
null. Weight, prefix, refusal, annotations, developer role, empty
user/system content, and participant name now use the same named errors
on the tools passthrough path as on orchestration.
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Hoist stream, required model, stream_options, and temperature/top_p range
checks before proxy_completion so a tools or response_format body cannot
return a billed JSON completion when the SDK asked for SSE, or silently
pick a pool model when model is omitted.
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Prefer this head over #582–#587 and draft #593. Do not merge b108251 alone: that SHA still skipped stream / model / stream_options on the tools path. 856adb8 hoists those checks before proxy_completion.
What this head gets right
_normalized_stream_option_flags fail-closes unknown keys before dropping nulls. {include_continuous: null} is no longer omit-equivalent on chat, Completions, and Responses.
Tools-path message honesty (weight, prefix, refusal, annotations, developer, empty user/system content, name) runs before the passthrough return.
856adb8 then fail-closes stream=true, omitted/unknown model, stream_options, and out-of-range temperature / top_p on tools / response_format.
Do not merge yet
Independent non-author APPROVE is still required (cursor[bot] is the author; this automation cannot satisfy that gate).
Product gates (Full unit + Semgrep) were still queued when this review ran. mergeable_state: blocked.
Residual fail-open on this tip: attribution and routing still run only after the tools return. Tools + unknown attribution dimensions, or routing.channel=batch / latency_tolerant=true, still bill a sync proxy. Empty / omitted / non-list messages still skip _validate_messages — that slice is #594, do not duplicate it here.
Buyer next action
Send SDK-default stream_options as omit / null / false. Do not set include_usage=true or include_obfuscation=true.
On tool-calling requests: omit stream (or set false); always send a pool model; replace developer with system.
Do not send routing.channel=batch or latency_tolerant=true with tools until a follow-up fail-closes that path (next slice on this stack).
Reviewer seonghobae: APPROVE only after Full unit + Semgrep are green and you agree the residual attribution/routing hole can land as a follow-up.
Related heads to leave unmerged:#582–#587, #591, #593. Prefer #592 @ 856adb8 as the stream_options + tools honesty merge candidate. Empty-messages repair is #594.
The unique tip closes the two claimed defects. Do not merge this head. GitHub rejects REQUEST_CHANGES on this cursor[bot] PR; this is a COMMENT with the same merge block. Inline comments skipped: the 142-file diff is too_large.
Verified
Unknown stream_options keys fail closed before null-drop. Live HTTP 400invalid_stream_options for {include_continuous: null} on chat, Completions, Responses, and with tools.
Weight / prefix / refusal / annotations / developer / empty user content / empty name / input_audio → 400 with the named orchestration errors.
stream=true, omitted model, include_usage=true, temperature=99, top_p=2 → 400 before proxy_completion.
Unique-tip HTTP scripts are real loopback cases, not tautologies. Local runs of test_stream_options_null_flags_noop_http_honesty.py, test_message_honesty_tools_passthrough_http_honesty.py, and test_passthrough_stream_model_http_honesty.py passed.
Must-fix in _validate_chat_message_passthrough_honesty (live on this head)
tools + messages: [] / omitted / null / "hi" → 200chat.completion (prompt_tokens: 0). The helper returns when messages is not a list.
tools + messages: ["hi"] → 500internal_error (str.get in the proxy). The helper continues on non-dict entries.
tools + role: "narrator" or content: 42 → 200 (helper only type-checks list parts and empty user/system strings).
tools + unknown attribution / routing keys or max_tokens: -1 → 200 (those validators still run only after the early-return).
Overlap
Do not merge #587 at 9d10fa9 (null-drop before unknown-key check).
Do not land #591 or draft #593 in parallel (subset of this hoist, no unknown-null repair).
#589 / #594 close empty-messages on the older #586 substrate and miss this unknown-null + stream/model/sampling tip.
Repair that includes this tip plus empty-messages / non-object / spend-key honesty: #597 (169a382). Prefer #597 over merging this head.
Buyer next action
Do not merge #592. Take #597. Until that lands: always send a non-empty messages array of objects; keep SDK-default stream_options null/false; omit stream (or set false) on tool-calling requests; always send a pool model; replace developer with system. Do not send tools with empty or omitted messages — this head will still bill a completion.
Superseded by the #597 → #601 → #607 successor chain. #597 starts from this exact honesty tip and preserves its stream/model/message/stream_options safeguards while adding the later empty-message/spend/routing boundary; #601 and #607 extend that same landing line further. Keeping #592 open duplicates an older substrate. No checks, reviews, or approvals transfer.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
Summary
Repair for #587 at
9d10fa9, and a superset of #591's request-level hoist.stream_options.include_usage/include_obfuscationJSONnull(and null+false mixes) as omit-equivalent no-ops on chat, Completions, and Responses.stream_optionskeys fail closed even when their value isnull.weight,prefix,refusal,annotations,developerrole, empty user/system content, multimodal part shape, and participantnameuse the same named errors beforeproxy_completion.stream=true, omitted/unknownmodel, and out-of-rangetemperature/top_p(the fix(api): fail-closed stream, model, and sampling range on tools passthrough #591 hoist).Buyer next action: send SDK-default null flags; omit
stream(or setfalse) on tool-calling requests; always send a poolmodel; replacedeveloperwithsystem. Do not setinclude_usage=true/include_obfuscation=true.Do not merge #587 at
9d10fa9. Prefer this head over #582–#587 and #591. Merge only after Full unit + Semgrep are green and an independent non-author review lands.Test plan
python tests/test_stream_options_null_flags_noop_http_honesty.pypython tests/test_message_honesty_tools_passthrough_http_honesty.pypython tests/test_passthrough_stream_model_http_honesty.pypython tests/test_openai_passthrough.pyDocs
docs/rest_api_design.mdlists Completions / Responses / embeddings and the honesty contract (APA: OpenAI, 2024).