fix(interactive): do_probe drops a first-time probe name and can crash with UnboundLocalError

[chuenchen309]

Two bugs in GarakCommands.do_probe (garak/interactive.py):

1. The probe-name branch only covers "overwrite an existing probe" (if args.probe and self._cmd.probe) and "neither given" (elif not args.probe and not self._cmd.probe). The case of a probe name given for the first time (self._cmd.probe was falsy, args.probe is truthy) matches neither branch, so self._cmd.probe is silently never set — probe myprobe run as the very first probe command in a session does nothing.
2. If loading the generator raises ImportError or AttributeError, the except blocks only log/print and fall through — generator was never assigned, so the next line, harness.run(generator, ...), raises UnboundLocalError instead of the intended graceful message.

Fix

Add an explicit elif args.probe: self._cmd.probe = args.probe branch for the first-time-set case, and return after each except block so a failed generator load stops the command instead of falling through to code that needs a generator that was never created.

Verification

• Added tests/test_interactive.py (no prior test coverage existed for this module): calls GarakCommands.do_probe.__wrapped__ directly (bypassing the @cmd2.with_argparser decorator, so no full cmd2.Cmd registration or garak config is needed), with ThresholdEvaluator and ProbewiseHarness mocked out to isolate the control-flow bug from unrelated config-loading concerns.
• Confirmed red→green: reverting only interactive.py reproduces both symptoms exactly (probe stays None; UnboundLocalError on generator); reapplying passes.
• black --check clean on both changed files.
• xref: fix(lint): tune pylintrc and fix violations across core modules #1793 ("fix(lint): tune pylintrc and fix violations across core modules") also touches do_probe, but only for an f-string→%s logging conversion and a trailing return None, not either of these two logic bugs — not a duplicate, though a trivial rebase may be needed if it merges first.
• garak -t <target_type> -n <model_name> — not run: this is a unit-level fix and I have no live target configured; verified via the tests above rather than ticking a box I didn't exercise.
• Supporting configuration such as a generator configuration file — n/a, no config surface changed.

---

I used AI assistance (Claude) to help investigate and draft this fix, but I personally constructed the repro, reviewed the root cause, and reviewed the final diff before submitting.

Signed-off-by: Andrew Chen 48723787+chuenchen309@users.noreply.github.com

[jmartin-tech]

DCO signoff must be in the commit message

[chuenchen309]

Thanks @jmartin-tech — added the DCO sign-off; the commit now carries Signed-off-by: and the DCO check is passing. Appreciate the review.