Align the navigation with Installation, Inference and Evaluation. Present local text generation as Quick start and the pull request workflow as Agent example under Inference, and update the README checksum.
Clarify parameter-space skill use in the project introduction
Expand the introduction to explain that generated LoRA adapters let the model use a skill without explicitly including its text in the task context. Update the README checksum.
Describe the documentation entry points, explain pull requests and continuous integration, label the recorded input table and focus the evaluation section on scoring saved outputs. Keep code checks in the runtime guide and update checksums.
Unify repository layout and document setup, inference and evaluation
Move inference source to the repository root and remove the training placeholder. Reorganize the README around environment setup, concrete inference inputs and expected outputs, and evaluation of saved artifacts. Preserve the released checkpoint files.
Add the inference source for the released Qwen3.5-4B and 9B components, with the PR iteration example, regression tests, documentation and notices. Link the source from the model card and update the checksum manifest.
