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
…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.
This PR contains 137 files, which is 37 over the limit of 100.
To get a review, reduce the PR to 100 files or fewer by splitting it into smaller PRs or changing its base branch.
Upgrade to a paid plan to raise the limit.
This review couldn't start because sufficient usage credits or metered capacity aren't available. Add credits or update usage-based reviews in the billing tab, then retry.
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro Plus
Run ID: 003be8a6-56f4-4e46-ab9d-9f01ae96df28
📥 Commits
Reviewing files that changed from the base of the PR and between 6841b71 and 3ce1a50.
The reason will be displayed to describe this comment to others. Learn more.
Stale comment
Review: validator is sound; do not merge this stack
The unique commit (3ce1a50) is the right shape: _validate_chat_assistant_tool_calls already runs before tools passthrough, so unknown tool_calls entry/function keys and invalid index fail closed on both orchestration and passthrough paths.
Do not merge this head. It is stacked on #5821a196b0, which fail-opens message weight / prefix / refusal / annotations when tools is present (those checks live only in _validate_messages after the early-return).
The entry-key checks plus a tools-path HTTP case are folded into #585 (cbd0420). Prefer #585 over #582/#583/#584.
The reason will be displayed to describe this comment to others. Learn more.
Stale comment
Do not merge #584 at 3ce1a50. Unknown tool_calls entry/function keys fail closed before passthrough, but this head inherits the #582 tools-path fail-open for weight/prefix/refusal/annotations and the #585 leftover (developer / content / name). Repair is #586 (28ef9d2).
Independent non-author APPROVE is still required on #586 before merge.
The reason will be displayed to describe this comment to others. Learn more.
Do not merge 3ce1a50. The unique tool_calls entry/function-key commit is fail-closed on both the orchestration path and tools passthrough (live: extra entry key → 400 unknown_tool_call_fields; extra function key → 400 unknown_tool_call_function_fields; index null omit / non-negative int accept / bool reject). That part is the right shape.
This stacked head still returns 200 when tools is present together with unsupported message weight, prefix, refusal, or annotations. Those named rejects live only in _validate_messages, which runs after the tools/response_format early-return. proxy_completion then forwards the original messages verbatim, so a real provider can apply prefix/weight while this gateway claims those planes fail closed. Completions max_tool_calls is still opaque unknown_fields. Omit-equivalent max_tool_calls is not popped. Shared ModelClient sampling defaults are mutated per request on ThreadingHTTPServer.
Next action: land the mergeable repair at #586 (28ef9d2) — it hoists the remaining message planes before passthrough, name-rejects Completions max_tool_calls, strips omit-equivalent max_tool_calls, and isolates sampling knobs on threading.local. Do not merge #582–#585 at their current heads; they are substrates of the same stack.
The reason will be displayed to describe this comment to others. Learn more.
Tools/response_format early-return happens before_validate_messages. Live at this head: tools + weight: 2 / prefix: true / nonempty refusal / nonempty annotations all return 200. Hoist those four checks next to _validate_chat_message_audio_function_call (already called above) so the passthrough path cannot smuggle them. #586 does this.
The reason will be displayed to describe this comment to others. Learn more.
These weight / prefix / refusal / annotations rejects are correct on the orchestration path, but they never run when tools is present. That is the commercial-honesty hole: proxy_completion forwards the original messages, so a provider can apply a plane this gateway says it rejects. Move the checks into a pre-passthrough helper and add HTTP cases with nonempty tools.
The reason will be displayed to describe this comment to others. Learn more.
Null / empty / whitespace returns without body.pop("max_tool_calls"). The key is not in _ORCHESTRATION_ONLY_KEYS, so proxy_completion forwards it. Pop on omit and assert absence on the tools path (mock echo must include the key).
The reason will be displayed to describe this comment to others. Learn more.
ALLOWED_COMPLETIONS_KEYS omits max_tool_calls, so /v1/completions still 400s as opaque unknown_fields. Add the key and call _validate_max_tool_calls(..., endpoint_path="/v1/completions") so Completions matches chat/Responses named invalid_max_tool_calls.
The reason will be displayed to describe this comment to others. Learn more.
if index is not None or index is None: is always true, so the loop never tests a missing index key. Split “omit key” vs “explicit null”, and add one nonempty-tools extra-key case so the pre-passthrough reject stays locked.
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
tool_callsentries: onlyid/type/functionplus optional streamindex; other keys fail closed (unknown_tool_call_fields).functionobject: onlyname/arguments; extra keys fail closed (unknown_tool_call_function_fields).index: null omit; non-negative int accepted; other types fail closed (invalid_tool_calls).Test plan
pytest tests/test_tool_calls_entry_keys_http_honesty.pypytest tests/test_chat_assistant_tool_calls_http_honesty.py(+ related)Product gates only: Full unit + Semgrep (Strix ignored). Independent non-author APPROVE still required.