Public Access
fix: support streaming screenshot OCR
This commit is contained in:
@@ -0,0 +1,15 @@
|
||||
## 1. Streaming screenshot OCR
|
||||
|
||||
- [x] 1.1 Add a failing screenshot-client behavior test for an SSE response with prompt and image input.
|
||||
- [x] 1.2 Route screenshot OCR through the shared Responses client and add multimodal message translation.
|
||||
|
||||
## 2. Non-streaming compatibility
|
||||
|
||||
- [x] 2.1 Add behavior coverage proving `Agent:UseStreaming=false` preserves non-streaming screenshot OCR and image input.
|
||||
- [x] 2.2 Document that screenshot OCR inherits the agent streaming setting.
|
||||
|
||||
## 3. Verification
|
||||
|
||||
- [x] 3.1 Refactor the touched screenshot and shared client paths for DRYness, SOLID boundaries, and simplicity while preserving behavior.
|
||||
- [x] 3.2 Run focused tests, the full solution tests, and strict OpenSpec validation.
|
||||
- [x] 3.3 Restart Meeting Assistant only while idle and verify screenshot OCR against the deployed LiteLLM endpoint.
|
||||
Reference in New Issue
Block a user