Skip to content

A sibling and a fork start on their parent's repository, checked again as a create checks one, and are refused rather than started without it

Decision 897 · accepted · 2026-10-13 · issue #147

Decision 885 gave a session one repository, named when it is created, and said that nothing else named one yet: "a fork and a sibling inherit none". That left a sibling an agent spawned, or a fork a person made to try something else, working in an empty directory on a pod while its parent worked on the code. D126 recorded it.

What is decided. session.spawn and session.fork hand the parent's repository (owner, name and ref, as the row keeps it) to session.create, the path both already take. So the new session's repository is checked exactly as a new one is: the plane's App must still have an installation for the owner, and the profile the session starts on must declare the Git host in egress.gitHosts. It is kept on the new row, and handed to the pod on every activation, as any session's is.

Refused, not dropped. When the parent's repository no longer passes (the App's installation for that owner was removed, or a fork lands on a profile that does not declare the host), the spawn or fork is refused with the same invalid_params and field: repository a create gets. The other choice, starting the new session with no repository, would be a sibling or a fork that runs without the code it was made to work on and says nothing about it.

Unchanged. A parent with no repository passes none. A private session's import has none, since a private session runs on a person's machine and is never cloned onto a pod.

Proof. github_lineage_test.exs, against the development OpenBao and the fake GitHub: a sibling spawned from a session created with a repository, and a fork of a sealed session that has one, each keep it on their row and receive it in their activation.

Governs: apps/troupe_plane/lib/troupe/plane/harness.ex