jupyter-notebook-writing

Write Milvus application-level Jupyter notebook examples through a Markdown-first developer-relations workflow. The core idea is that raw `.ipynb` files are awkward JSON documents full of metadata, execution counts, and cached outputs, so the skill keeps a clean `.md` file as the source of truth and converts it to a runnable notebook with the `jupyter-switch` tool invoked via `uvx`. Python fenced code blocks become code cells while all other Markdown becomes markdown cells; conversion runs in both directions and never deletes the original input, writing a `.bak` backup if an output file already exists.

Reach for it when creating or editing a notebook that demonstrates Milvus in an application context such as RAG, semantic search, or hybrid search. Prerequisites are Python 3.10 or newer and uv. A typical notebook follows a fixed template: title, prerequisites, setup with a MilvusClient connection, data preparation, collection creation and insertion, query or search, and cleanup.

The skill also details how to execute notebooks reliably. Because the system default environment may lack the required packages, it walks through detecting conda or mamba environments and registered Jupyter kernels, asking the user which environment to use, and registering a kernel with ipykernel when needed before running `jupyter execute --kernel_name=<name>`. Before a test run it comments out setup-only cells such as `pip install` cells, API-key placeholder cells, and mock or demo cells so they stay in the notebook structure but do nothing, with credentials supplied externally as environment variables instead. Two bundled references, bootcamp-format.md and milvus-code-style.md, cover tutorial layout and pymilvus coding conventions and should be read when the task touches those topics.