What:
- replace the misleading gateway test check with a source-checks build/config-options check
- remove the full upstream Vitest suite from the hard Nix promotion gate
- document that yolo validates the Nix-owned package contract, not upstream source test health
Why:
- the current full packageable upstream release builds as a package but has failing upstream Vitest cases when rebuilt from source
- nix-openclaw should block on packaging, smoke startup, config generation, module activation, and app artifacts rather than an upstream-owned unit test suite
Tests:
- git diff --check
- bash -n scripts/update-pins.sh scripts/hm-activation-macos.sh nix/modules/home-manager/openclaw-launchd-relink.sh nix/scripts/source-checks-build.sh nix/scripts/source-checks-check.sh nix/scripts/config-options-check.sh
- ruby -e 'require "yaml"; YAML.load_file(".github/workflows/ci.yml"); YAML.load_file(".github/workflows/yolo-update.yml")'
- nix eval --accept-flake-config --raw .#checks.x86_64-linux.source-checks.drvPath
- nix eval --accept-flake-config --raw .#checks.x86_64-linux.ci.drvPath
- nix flake show --accept-flake-config --all-systems --json
- scripts/check-flake-lock-owners.sh
- node scripts/select-openclaw-release.test.mjs
- scripts/hm-activation-macos.sh
- nix build --accept-flake-config -L .#checks.aarch64-darwin.ci
Blacksmith Linux hung twice during final verification while the same check passed once in between. Move Linux validation back to ubuntu-latest and keep explicit timeouts so the packaging gate favors reliability over runner speed.
Verification: git diff --check; ruby YAML parse
Add explicit CI timeouts around the Linux and macOS packaging gates so a wedged OpenClaw build fails inspectably instead of hanging indefinitely.
Verification: git diff --check; ruby YAML parse
What:
- move the Linux CI job to blacksmith-16vcpu-ubuntu-2404
- switch Linux Nix installation to cachix/install-nix-action
Why:
- the recovery line exhausted disk on ubuntu-latest during the Linux aggregator build
- the larger runner kept the normal parallel build graph intact and the cachix installer worked reliably there
Tests:
- final green CI run on main: 24217118174
What:
- restore one visible CI workflow
- add CI concurrency and Linux failure-log dumping
- disable yolo during release recovery
- delete the old split workflow_run files
Why:
- cut away the post-reset churn immediately
- keep one readable contract on main
- prevent any updater mutation during recovery
Tests:
- git diff --check
