I'm asking DeepSeek to help with a Python service that has a locked dependency set. Some suggestions import packages we don't use; others reference methods that aren't available in our installed versions. The code can look reasonable until someone tries to run it.

Adding a package requires a separate review, so “install this library” isn't an acceptable default.

1. Would you provide the full lockfile, or a shorter list of direct dependencies and their versions?
2. For version-sensitive APIs, is a small extract from the installed package's documentation more useful than naming the version alone?
3. How would you check for new imports without rejecting valid Python standard-library modules or our own internal packages?

I'm looking for a lightweight workflow for validating AI-generated Python dependencies, not just stronger prompt wording. We can run proposed changes in a disposable environment with no network access, but I'd like to catch obvious dependency problems before running the test suite.