Disclosure: I work on NOQT’s open-source Lumi Eggcracker. I used AI assistance to draft this walkthrough and checked it against the linked source.
A stop that reports success leaves two concrete questions: is the target cgroup hierarchy empty, and did an unrelated process survive? We published an opt-in synthetic check so those conditions can be inspected without installing on a local Linux machine.
The manually dispatched workflow runs on a disposable GitHub-hosted Ubuntu 24.04 runner. It starts a harmless two-process target tree in an owned cgroup and a separate outside canary. It invokes pidfd-stop+cgroup.kill, checks target_survivors: 0 and canary_survived: true, and emits a bounded JSON receipt with result: TERMINATED and target_processes: 2 before FORK_PROBE_RESULT=PASS. It also has bounded refusal and failure output. Preflight requires acknowledgement, public/default-branch source, a GitHub-hosted runner and the needed Linux facilities.
The try-it guide includes a manual fork/Actions route and an optional authenticated gh helper that checks fork and workflow identity. You need your own eligible GitHub account, public fork and Actions environment; there is no universal free-run promise. Do not adapt the workflow to a self-hosted runner. The guide also offers an optional way to share a pass, redacted refusal or concrete friction without private logs.
The Hugging Face Space is a static guide, not the runner. A pass tests only this synthetic kill mechanism; it is not AI-workload recognition, native v1.0.10 installation, full-product behavior, a website recording run on your machine, or production containment. No GPU or model download is needed.