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
Successor to #592 at 856adb8, and a superset of draft #596's attribution/routing hoist plus #594's empty-messages hoist.
The #592 tip closed unknown-null stream_options and hoisted message-field / stream / model / sampling checks, but _validate_chat_message_passthrough_honesty still skipped non-list messages and non-object entries. Live HTTP on that head: tools + [] / omitted / null / a string billed a chat.completion (prompt_tokens: 0); tools + ["hi"] returned 500internal_error.
Call _validate_messages before proxy_completion so empty, omitted, null, non-list, and non-object messages use the same invalid_message as orchestration.
Hoist max_tokens / max_completion_tokens, attribution, and routing so unknown spend keys and negative token budgets fail closed on the tools path.
Reject routing.channel=batch and routing.latency_tolerant=true on passthrough (no batch job plane) so those hints cannot bill a silent sync completion.
Docs: docs/rest_api_design.md and README tell buyers to send a non-empty messages array of objects and to omit batch routing hints (APA: OpenAI, 2024).
Buyer next action: always send a non-empty messages array of objects; keep SDK-default null flags; omit stream (or set false) on tool-calling requests; always send a pool model; replace developer with system; omit routing.channel=batch and latency_tolerant=true on tool-calling requests. Do not send tools with empty or omitted messages.
Do not merge #587 at 9d10fa9, #589, #591, #592, #593, #594, or draft #596 in parallel. Prefer this head. 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>
Call the orchestration message, max_tokens, attribution, and routing
validators before proxy_completion so a tools or response_format body
cannot bill a completion with no prompt, crash on a non-object entry,
or silently drop unknown spend/routing keys.
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Passthrough has no batch job plane. Reject routing.channel=batch and
latency_tolerant=true before proxy_completion so a tool-calling body
cannot bill a silent sync completion.
Co-authored-by: Seongho Bae <seonghobae@users.noreply.github.com>
Unique work is sound. _validate_messages, max_tokens / max_completion_tokens, attribution, and routing now run before proxy_completion, and routing.channel=batch / latency_tolerant=true fail closed. Local HTTP on this head: empty / omitted / null / non-list / non-object messages with tools is 400 invalid_message; batch and unknown spend keys are 400.
Do not merge this head. It is still a draft, Full unit + Semgrep were queued at review time, and request knobs that already 400 on the orchestration path still bill a silent chat.completion on tools passthrough.
Live residual on dbfa0e5: tools + seed: 1 returned 200 chat.completion with prompt_tokens: 0. The same fail-open applies to stop, n>1, logprobs, logit_bias, out-of-range penalties, and unknown reasoning_effort / service_tier.
Successor that folds those knobs on top of this head: prefer that PR over #597, #596, #594, and #592. Do not merge those heads in parallel.
Buyer next action on this head: send a non-empty messages array of objects; omit routing.channel=batch and latency_tolerant=true on tool-calling requests. Do not send seed, stop, n>1, or logprobs with tools until the successor lands — those knobs are dropped while the call is still billed.
Independent non-author APPROVE + green Full unit / Semgrep are still required. This automation does not approve or merge.
python3 tests/test_chat_messages_array_tools_passthrough_http_honesty.py → ok
Do not merge this head. Request knobs that already 400 on the orchestration path still bill a silent chat.completion on tools passthrough. Live residual on dbfa0e5:
tools + seed: 1 → 200 chat.completion
tools + stop: ["END"] → 200 chat.completion
tools + n: 2 → 200 chat.completion
The same fail-open applies to logprobs, logit_bias, out-of-range penalties, and unknown reasoning_effort / service_tier. Those checks still sit after the tools early-return.
Landing vehicle: prefer #601 (18e6263) which hoists _validate_chat_passthrough_request_knobs. Cherry-pick #603e58cb2f (omit-real tool.function nulls) onto that tip. Do not merge #589, #592, #594, #596, #597, #599, #600, #603, or #605 in parallel.
Buyer next action on this head: send a non-empty messages array of objects; omit routing.channel=batch and latency_tolerant=true on tool-calling requests. Do not send seed, stop, n>1, or logprobs with tools until #601 lands — those knobs are dropped while the call is still billed.
Independent non-author APPROVE + green Full unit / Semgrep are still required. This automation does not approve or merge. Checks were still queued at review time and are not treated as a merge blocker for this residual.
The reason will be displayed to describe this comment to others. Learn more.
Batch / latency_tolerant now fail closed (live: tools + routing.channel=batch → 400). The early-return still skips the orchestration seed / stop / n / logprobs / logit_bias / penalty blocks below. Live on this head: tools + seed: 1 bills chat.completion. Hoist _validate_chat_passthrough_request_knobs before proxy_completion — that is #60118e6263. Do not merge this head.
Superseded by #607 through direct successor #601. #601 explicitly starts from this exact honesty-stack tip and extends its empty-message/spend/routing boundary with the remaining request-knob validation; #607 then carries that stack forward with the KV-backed provider-host policy. Keeping #597 open duplicates an older landing vehicle. 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
Successor to #592 at
856adb8, and a superset of draft #596's attribution/routing hoist plus #594's empty-messages hoist.The #592 tip closed unknown-null
stream_optionsand hoisted message-field / stream / model / sampling checks, but_validate_chat_message_passthrough_honestystill skipped non-listmessagesand non-object entries. Live HTTP on that head:tools+[]/ omitted /null/ a string billed achat.completion(prompt_tokens: 0);tools+["hi"]returned500internal_error._validate_messagesbeforeproxy_completionso empty, omitted, null, non-list, and non-objectmessagesuse the sameinvalid_messageas orchestration.max_tokens/max_completion_tokens,attribution, androutingso unknown spend keys and negative token budgets fail closed on the tools path.routing.channel=batchandrouting.latency_tolerant=trueon passthrough (no batch job plane) so those hints cannot bill a silent sync completion.tests/test_chat_messages_array_tools_passthrough_http_honesty.py(RED on fix(api): fail-closed unknown stream_options keys and tools-path message honesty #592: empty messages were 200; batch channel was 200).docs/rest_api_design.mdand README tell buyers to send a non-emptymessagesarray of objects and to omit batch routing hints (APA: OpenAI, 2024).Buyer next action: always send a non-empty
messagesarray of objects; keep SDK-default null flags; omitstream(or setfalse) on tool-calling requests; always send a poolmodel; replacedeveloperwithsystem; omitrouting.channel=batchandlatency_tolerant=trueon tool-calling requests. Do not sendtoolswith empty or omittedmessages.Do not merge #587 at
9d10fa9, #589, #591, #592, #593, #594, or draft #596 in parallel. Prefer this head. Merge only after Full unit + Semgrep are green and an independent non-author review lands.Test plan
python3 tests/test_chat_messages_array_tools_passthrough_http_honesty.pypython3 tests/test_message_honesty_tools_passthrough_http_honesty.pypython3 tests/test_passthrough_stream_model_http_honesty.pypython3 tests/test_stream_options_null_flags_noop_http_honesty.pypython3 tests/test_openai_passthrough.pytest_chat_attribution_routing_http_honesty.pypython3 tests/test_paper_contracts.pytest_self_check.pytest_conventions.pytest_api_contract.pyDocs
docs/rest_api_design.mdhonesty contract now includes empty/non-objectmessages, spend keys, and batch-routing reject before passthrough.messagesarray of objects and to omit batch routing hints.