Support GPT-5.6 provider compatibility#135
Conversation
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 6e3c3b9bee
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| } else if (json.type === 'response.failed' || json.type === 'error') { | ||
| const streamError = new Error(json.response?.error?.message || json.error?.message || json.message || 'Responses API stream failed.'); | ||
| streamError.name = 'ResponsesStreamError'; | ||
| throw streamError; |
There was a problem hiding this comment.
When an official GPT-5.6 stream ends with response.incomplete (for example from max_output_tokens or content filtering), this Responses branch does not match it, so the generator falls through and yields a normal done at EOF. In the streaming agent path that can persist a truncated answer as successful, or retry an empty response without surfacing the actual limit/filter reason; treat response.incomplete as a terminal error/status alongside response.failed.
Useful? React with 👍 / 👎.
Resolve conflicts by keeping main's Responses API path (encrypted reasoning replay, streaming, response_items) and integrating this branch's advanced model compatibility layer for OpenAI-compatible providers.
Ship GPT-5.6 Responses + advanced provider compatibility as the 24.0.0 major, with changelog and WHATS_NEW_IN_V24.md.
Keep OpenAI max reasoning effort, deep-merge nested extras without clobbering Responses fields, convert and strip response_format, treat incomplete Responses as failures, localize compatibility settings, and clear the CodeQL incomplete-regex escape finding.
|
@codex pls review |
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0dc69f30b1
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| // OpenAI documents `max` as a distinct effort above `xhigh` (GPT-5.6). | ||
| // Pass it through unchanged for the OpenAI preset and any other preset that | ||
| // does not define its own clamp above. | ||
| return effort; |
There was a problem hiding this comment.
The committed release artifacts are stale relative to this source: after unzipping dist/webbrain-{chrome,edge,firefox}-24.0.0.zip, src/providers/provider-compatibility.js still maps max to xhigh and shallow-merges extras, and src/providers/openai.js still treats response.incomplete as a normal done. Users installing the packaged extensions therefore miss the fixes in this commit and can ship truncated Responses turns or lose required include/reasoning fields, so the zips need to be rebuilt after these source changes.
Useful? React with 👍 / 👎.
Repackage Chrome, Edge, and Firefox v24.0.0 zips so packaged extensions include the Responses incomplete handling and provider-compatibility corrections from the source tree.
|
@codex pls review |
Translate the st.providers.compat.* panel strings across all non-English Chrome and Firefox locales so the new compatibility UI is not English-only.
|
Codex Review: Didn't find any major issues. Swish! Reviewed commit: ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. Codex can also answer questions or update the PR. Try commenting "@codex address that feedback". |
2 participants
No description provided.