Goal

• Act as an expert in Docker
• Given a directory <TARGET_DIR> with a project, find which packages in requirements.txt are not needed by the code in <TARGET_DIR>

Workflow

Methodology

• Use grep -r "import <PKG>" or AST-based analysis to find actual imports across all .py files in <TARGET_DIR>
• Map package names in requirements.txt to their import names (e.g., Pillow -> PIL, scikit-learn -> sklearn, PyYAML -> yaml)
• Also check setup.py, pyproject.toml, tasks.py, and Makefile for indirect or tool-level usage
• Flag but do NOT remove packages that are:
  • Runtime plugins or extras loaded dynamically (e.g., via importlib)
  • Transitive dependencies pulled in by other listed packages
  • Used only in test files (mark them as "test-only")

Output

• Print a table with columns:

  package | import_name | used_in_code | verdict
  

  • verdict is one of: remove, keep, investigate, test-only
• Propose the trimmed requirements.txt contents and write it to disk

Verification

• Confirm every package in requirements.txt has a verdict
• Confirm no package marked remove is still imported anywhere in <TARGET_DIR>
• Confirm the trimmed requirements.txt was written to disk