This is an offline analysis tool: it reads files you point it at, writes reports and charts, and talks to nothing else. It opens no network connections, runs no subprocesses, and executes no code from the files it reads.
Please open an issue at https://github.com/zero705/cell-ageing-analyzer/issues. If you would rather not discuss it in public, use GitHub's private vulnerability reporting on the same repository.
Both file formats it reads can come from somewhere else, so both are handled as untrusted input.
Dataset files (.mat)
- The header is checked before the file is parsed, and a MATLAB v7.3 (HDF5) file is rejected with an explanation instead of being guessed at.
- The parser (
scipy.io.loadmat) reads data only; the MATLAB v5 format carries no code. Any failure inside it, up to and including running out of memory, is turned into a readable error rather than a traceback. - Cell names are accepted only when they match
Cell<number>, so nothing from the file can reach the terminal as control characters.
History files (.npz)
- Loaded with
allow_pickle=False. This is the setting that matters: with pickling enabled, opening a file would be enough to run code that came with it. - The archive is checked to be a zip, and the total size it declares is checked against a 64 MB limit before anything is unpacked, so a small file cannot expand into hundreds of megabytes of memory (a "decompression bomb"). The whole Oxford campaign unpacks to about 2.3 MB.
- Every field is validated after loading: cell names must be plain names, numbers must be finite numbers, and the curve arrays must match the characterisations they belong to. Text taken from the file is stripped of control characters before it is printed, so a crafted file cannot rewrite what your terminal shows.
- A dataset file you do not trust at all. The checks above stop the common failure
modes, but the parsing itself happens inside SciPy. Treat unknown
.matfiles the way you would treat any unknown binary: open them in an environment you can throw away. - Someone who can already write to your files. The tool has no permission model; it runs with your rights.
The runtime dependencies are NumPy and SciPy, plus matplotlib for the optional charts. They are audited with pip-audit in continuous integration, and Dependabot opens pull requests when a newer version is published.