Claude: must-fix: The added rationale that "oneshot services have no main process" is inaccurate and contradicts this page's own description of Type=oneshot, which states the manager considers the unit up after the main process exits — i.e. oneshot units do have a tracked main process (the ExecStart= command is recorded as the service's main PID in src/core/service.c). The conclusion (ExitType=main cannot be used with Type=oneshot) is correct, but the reason is not the absence of a main process. The accurate reason is that a oneshot unit's lifetime is defined by its start command(s) running to completion, so the main-vs-cgroup distinction that ExitType= selects does not apply. Consider rephrasing so it doesn't contradict the Type= documentation above.

Claude: suggestion: Two issues with the rationale added for the cgroup case:

• "oneshot services have no main process to track" repeats the inaccurate claim from the main bullet above (oneshot's ExecStart= process is tracked as the main PID).

• "the cgroup is empty as soon as all processes of the service exit" does not actually explain an incompatibility: that is exactly the condition ExitType=cgroup is designed to detect (per the sentence just above, the service is "considered running as long as at least one process in the cgroup has not exited"), and it is true of every service type, not just oneshot. As written it could mislead readers into thinking cgroup-based exit tracking is generally non-functional.

The real mechanism is that systemd refuses Type=oneshot combined with ExitType=cgroup at unit load (service_verify() returns an error), because oneshot's run-to-completion model is incompatible with keeping the unit "running" based on cgroup occupancy. Consider stating that instead.

Claude: nit: "a file, directory, or any filesystem object referred to by the path" is slightly redundant — "any filesystem object" already subsumes files and directories. Other ConditionPath*= entries in this file phrase the checked target tersely; consider e.g. "Check for the existence of the filesystem object referred to by the path."

Claude: must-fix: The restriction stated here — that ExitType=main cannot be used with Type=oneshot — is not implemented and does not exist. service_verify() in src/core/service.c (around lines 923-947) rejects only ExitType=cgroup for oneshot units (the explicit SERVICE_EXIT_CGROUP check at ~L946: "...which isn't allowed for Type=oneshot services. Refusing."); there is no corresponding refusal for ExitType=main. Since SERVICE_EXIT_MAIN is the default, every default oneshot unit already runs with ExitType=main and loads without error, and explicitly setting Type=oneshot + ExitType=main is accepted and works (confirmed by reviewer @hpt's test on v257). My earlier review accepted this conclusion as correct — that was wrong. Both the newly added sentence and the pre-existing "Consequently, it cannot be used with Type=oneshot" line should be removed or reworded so the page no longer documents a restriction the implementation does not enforce.

Claude: suggestion: The man-page edit here is now correct: removing the "cannot be used with Type=oneshot" line for ExitType=main matches the implementation, since service_verify() refuses only ExitType=cgroup for Type=oneshot (ExitType=main is the default and loads/works). However, the commit message body still states this commit will "clarify that BOTH ExitType=main and ExitType=cgroup cannot be used with Type=oneshot," which no longer matches the final diff — the restriction text is simply removed and nothing about cgroup is added. Please update the commit message so it describes what the diff actually does (removes the inaccurate ExitType=main restriction and clarifies ConditionPathExists=).

Claude: nit: Elsewhere in this page "file system" is consistently spelled as two words; this is the only "filesystem" usage. For in-page consistency consider "any file system object referred to by the path".