[kaniini]

Two changes to the edera/mainline rebase automation. The third commit adds rebase-build.yml, which the second commit uses but left out: the kernel's .gitignore hides dot-directories from git add -A.

1. Build x86_64 on the Edera host config

The rebase acceptance build (kernel-build.sh, run by the workflow and by the rebase skill) now builds x86_64 on the Edera host config instead of bare defconfig.

• Each run fetches the tip of main in edera-dev/linux-kernel-oci and merges configs/x86_64/host.config over defconfig with merge_config.sh. This is how linux-kernel-oci applies it: the file is a fragment against x86_64_defconfig and omits SMP, NET and anything else that matches defconfig. The log records the linux-kernel-oci commit it used.
• After the merge, the Edera options are still forced on as built-ins and checked after olddefconfig. The host config sets CONFIG_OPENPAX=n and builds 9P/Xen, virtio-gpu and the Hyper-V network drivers as modules, so a vmlinux build would otherwise never compile them. OpenPaX (softmode, xattr flags, mprotect, emutramp) stays on in the test build.
• AGP is added to the x86_64 forced list because host.config has it =m, which caps DRM at m.
• arm64 is unchanged (defconfig).

2. Repair a rebase whose build fails instead of proposing it

When the independent build failed, the rebase went straight to a pull request. In practice that was a run Claude never finished. The Bash tool's two-minute default timeout cannot hold a kernel build, so the build went to the background and the session ended while it waited (#47: "Waiting for the x86 build to confirm the attach_dev breakage"; #35 the same).

• Longer timeout: the rebase step gets a 90-minute Bash timeout. The skill now says to run builds in the foreground and to keep fixing and rebuilding until both arches pass; a failing build is not a finished rebase.
• Repair round: if the independent build still fails, a new read-only fix job hands Claude its branch and the build logs. It follows a new "Repairing a failed build" section of the skill: fix the downstream commit that broke, rebuild both arches until they pass, and update the report's Build fixes and Builds sections.
• Re-check: a repaired branch is verified and built again from scratch (rebuild). The checker now runs on whichever branch is final.
• Publish: it takes the repaired branch and report when there is one, and says so in the pull request. If the repair round fixed nothing, it posts that round's report.
• Shared build job: the build job moves to a reusable rebase-build.yml, so the first build and the rebuild are the same job. Each build's log is kept as an artifact.
• Selftest: automation-selftest.yml now also runs when rebase-build.yml changes.

A repair changes downstream commits, so a repaired branch still shows drift and still goes to review rather than being pushed. What changes is that the reviewer gets a branch that builds.

Testing

• Host config: I ran kernel-build.sh x86_64 locally against edera/mainline at 017f96b23bf1, using linux-kernel-oci at db1c8e94a6d6. Every forced option survived olddefconfig, including all OPENPAX* at =y, and vmlinux linked. Two earlier attempts failed with exit 3, which led to this design: using host.config as the whole .config lost SMP/NET and everything that depends on them, and without AGP forced on, DRM stayed at m.
• Repair round: checked statically only. Every needs.* and steps.* reference resolves, the reusable workflow's inputs match its callers, and every run: block parses with bash -n. actionlint was not run. The fix/rebuild path has not run in Actions; the first nightly with a build failure will be its first real test.
• arm64: not built; its config path is unchanged.

edera/6.18-lts runs its own copy of this automation; #61 ports this PR there.

[github-actions]

Backport to edera/6.18-lts: not needed

Nothing to backport: every commit of #60 is already on edera/6.18-lts (5330eb8d79e6), changes only automation, or is no longer an Edera commit on edera/mainline.

Run log