fix regresson with Legal assistant BOUNDARY VIOLATION

fix tests

2nd attempt to fix HF purpose

HF did not displayed the agent purpose. Linked to the Streamlit's value + key anti-pattern.

retrigger HF deploy

AlignmentCheck blocks on CX session, reduce FP

add more tests

fix Data Guard FP with CX session

posts only failures to slack

enable .txt

Smart trace summarization and unit test

Resolution succeeds cleanly. All packages resolve without conflicts:

face HF build errors

update with CX input sample

Caching plan

more dependencies fix

avoid slow multi-version resolution

newUI branch is also with CI

sidebar controls to open/close

\

reset message limit to 12K per Kobi request

fix test to be more resilient, less room for AlignmentCheck to get confused

fix: let uv resolve huggingface/transformers deps (not --no-deps)

fix: pin transformers==4.46.3 alongside huggingface_hub==0.26.2

fix: use huggingface_hub==0.26.2 (has both HfFolder and is_offline_mode)

fix: force-reinstall huggingface_hub==0.24.0 after all installs

pin huggingface_hub<0.25 to fix llamafirewall HfFolder import error

CI optimization: replaced pip with uv and Test Parallelization

add unit test for cumulative trace

only for consecutive assistant messages, the prior assistant excluded → no cross-contamination

Focused trace instead of cumulative trace

improve AlignmentCheck output: concise and state mismatching references to input

fix FP from DataDisclosure with JSON files

parallelism among scanners and within AlignmentCheck that can speed up to x4 on long sessions

improve UI results

parallel run of scanners

remove FP from FactsChecker, when user requests are evolving and switching tool JIRA to ServiceNow and Slack to Teams

FactsChecker is using mini gpt and produce md

protect UI, CLI and guards from oversized sessions

FactsChecker: skip oversized messages to prevent UI hang

AlignmentCheck empty-message contamination fix with scenario environment_demo_36b5bb37.json

Fix FactsChecker False Negative with session 36b5bb37 - numerical contradiction. Workaround Nemo guards and may cause regressions

fix hangs when loading environment_demo_0ea72591.json, using timeouts

fix FP of datadiscluser in 304ad747 demo scenario

fix AlignmentCheck FP with evolving user requests, see scenario environment_demo_304ad747

remove FactChecker false position with apparent contradictions in connection strings

fix test failure

AlignmentCheck warning when large message from users or data blobs

fix AlignmentCheck false positive for agent analyzing external data

fix AlignmentCheck prompt to not flag factual errors or self-correction