Closed
Conversation
Only emit a stable baseline of safety settings instead of iterating every HarmCategory enum member. This prevents 400 INVALID_ARGUMENT errors when the SDK adds categories that certain requests (notably multimodal/image) don't accept. Co-authored-by: jason <jason@jxnl.co>
|
Cursor Agent can help with this pull request. Just |
Deploying with
|
| Status | Name | Latest Commit | Preview URL | Updated (UTC) |
|---|---|---|---|---|
| ✅ Deployment successful! View logs |
instructor | 2002f58 | Commit Preview URL Branch Preview URL |
Jan 16 2026, 05:31 PM |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
fix(genai): stop auto-sending unsupported safety categories
Describe your changes
This PR resolves a critical production blocker where Google GenAI's
SafetySettingscaused400 INVALID_ARGUMENTerrors, particularly with multimodal (image) content.The root cause was that
instructorwas automatically sending allgoogle.genai.types.HarmCategoryenum members as safety settings, including categories unsupported for certain modalities (e.g.,HARM_CATEGORY_CIVIC_INTEGRITYor image-specific categories when no image was present).This fix refactors
update_genai_kwargsto:HARM_CATEGORY_HATE_SPEECH,HARASSMENT,DANGEROUS_CONTENTatBLOCK_ONLY_HIGH).HARM_CATEGORY_IMAGE_*categories only when the request contains image content.UNSPECIFIEDandJAILBREAKcategories.This ensures that only valid and supported safety categories are sent to the Google GenAI API, preventing the
400 INVALID_ARGUMENTerrors. Thorough regression tests have been added and updated to cover these scenarios.Issue ticket number and link
Linear: 567-261
GitHub: #1773 (#1773)
Checklist before requesting a review
Linear Issue: 567-261