Build a small CLI with subcommands to import, validate and export records. How do you structure it?

How to answer

The parser is five minutes of argparse. The interviewer is checking whether the customer’s operator could run your tool from a script at two in the morning and know what happened.

1. Write the command surface before any code. As usage lines: records import FILE, records validate FILE, records export --format csv|json [--out PATH]. Use add_subparsers(dest="command", required=True), so a missing command becomes a usage error naming the choices, and bind each subcommand to a function with set_defaults(func=...), so dispatch is return args.func(args).
2. Make main(argv) -> int the only place that knows about the terminal. It parses, dispatches and returns an exit code; sys.exit(main()) sits under the __main__ guard. The work lives in plain functions that return results and never print or exit.
3. Make import validate first, with the same code. One function returns the problems. Then ask: all or nothing, or load the good rows and write the bad ones to a reject file with their line numbers? Either way, write in one transaction and upsert on the record’s natural key, so a re-run never duplicates rows.
4. Say the exit codes aloud. 0 for success, 1 when the data has problems, 2 for bad usage, which parse_args raises itself as SystemExit(2). A scheduler reads the code, not the text.
5. Keep streams separate. Data goes to standard output, so export can be piped; messages go to standard error. Each error names the file, the line and the fix: orders.csv:14: amount "12,50" is not a number.
6. Test through main. Call main(["validate", str(path)]) on a temporary file; assert the return code and captured output. Usage errors raise instead, so use pytest.raises(SystemExit).

The trap is a traceback as the error message. It tells the operator nothing and tells the interviewer you never ran the tool on a bad file.