[sramkrishna]

Closes #1159.

ChairLift's applications_page.brew_bundles_group in
/usr/share/chairlift/config.yml already points bundles_paths at
/usr/share/ublue-os/homebrew and renders one installable bundle per
immediate *.Brewfile child, naming the bundle after the filename stem.
So exposing the two wallpaper apps is a matter of placing files in that
directory — no config.yml change, which is deliberate: ChairLift fails
closed on an unknown page, group, or field key and disables the whole
application, so no new group was invented.

Changes

Path	Change
system_files/shared/usr/share/ublue-os/homebrew/wallpaper-slideshow.Brewfile	new — flatpak "app.drey.Damask"
system_files/shared/usr/share/ublue-os/homebrew/video-wallpaper.Brewfile	new — flatpak "io.github.jeffshee.Hidamari"
tests/test_chairlift_config.py	3 regression tests
docs/skills/brew-lifecycle/references/package-set.md	documents the second consumer of the directory
docs/skills/brew-lifecycle/SKILL.md	red flag + references row; last_updated bumped
docs/skills/index.json, docs/skills/index.md	regenerated by scripts/generate_skill_index.py --write

Both files live in system_files/shared/ (all variants — bluefin,
bluefin-lts, dakota) and are mode 0644, matching every sibling Brewfile.
Both are direct children of the scan root with the exact .Brewfile suffix:
ChairLift does not recurse, so a file in a subdirectory would be silently
absent from the Applications page rather than an error.

Two deliberate constraints, both now documented in the skill:

• Install only. Uninstall is ChairLift's flatpak_user_group /
  flatpak_system_group. A bundle has no uninstall action, and expressing
  one by inventing a group would trip the strict schema.
• flatpak "..." is an opt-in Bluefin convention, not stock Homebrew
  Bundle. validate-brewfiles.sh and brew bundle both ignore it, so a
  flatpak-only Brewfile passes the Brewfile gate with zero package checks.
  Unit tests — not the validator — are what pin the app ids here.

Verification

Run locally in the worktree, all green:

• python3 -m pytest tests/test_chairlift_config.py tests/test_skill_docs.py -q → 42 passed
• npx --yes bats tests/test_native_recipes.bats → all ok (through ok 22)
• npx --yes bats tests/test_validate_brewfiles.bats tests/test_suite_registration.bats → all ok
• PATH="$HOME/.local/bin:$PATH" pre-commit run --all-files → all hooks passed, including "Validate docs/skills/index.json is up to date"
• python3 scripts/generate_skill_index.py --check → up to date
• just check → exit 0

Not run locally: bash scripts/validate-brewfiles.sh. The script requires
a brew binary on PATH and exits 2 with "Homebrew is required to validate
Brewfiles." in this environment. Its CI counterpart
(.github/workflows/validate-brewfiles.yaml) triggers on
system_files/shared/usr/share/ublue-os/homebrew/** and will run on this PR;
given the files contain only comment lines and flatpak declarations, it
should report zero package checks for both. Flagging this rather than
claiming it was run.

The three new tests in tests/test_chairlift_config.py assert that
bundles_paths still resolves to the directory the files live in, that each
expected bundle is a 0644 regular file directly in that directory, that each
contains exactly its intended flatpak declaration, and that the bundle
names are unique. No new test file was added — tests/test_suite_registration.bats
enforces that every suite in tests/ is registered in the Justfile, so new
assertions belong in the existing file.

— hive: backend=litellm model=deepseek/deepseek-v4.1-flash

🐝 Hive Agent: contributor | SHA: c47b3c2

[castrojo]

Rebased cleanly onto latest main.

• Conflicts resolved: docs/skills/index.json and docs/skills/index.md regenerated via scripts/generate_skill_index.py. Clean automerge of tests/test_chairlift_config.py and docs/skills/brew-lifecycle/SKILL.md.
• Tests run: pytest tests/test_chairlift_config.py (31 passed), just check.

[castrojo]

Rebased onto latest main (clearing upstream Homebrew tremor-runtime check failure from #1206) and added missing trailing newlines to video-wallpaper.Brewfile and wallpaper-slideshow.Brewfile to pass pre-commit end-of-file-fixer. Ran pytest tests/test_chairlift_config.py (31 passed).

[Danathar]

The two Brewfiles look right. brew bundle check reads both and asks for the Damask and Hidamari flatpaks, and CI is green on de7c2af. I found two problems, one in the tests and one in the docs, that I'd like fixed before I approve.

1. An existing test got deleted by accident. On main, tests/test_chairlift_config.py has def test_help_links_point_at_bluefin():. This PR replaces that line with the new tests. The old test's body is now stuck on the end of test_bundle_names_do_not_collide_with_other_discovered_brewfiles(). The help-link checks still run, but if one fails, the report will say bundle names collide. You can see it in the counts: main has 29 tests in that file and this branch has 31 (3 added, 1 gone). Putting the def test_help_links_point_at_bluefin(): line back above resources = _load_config()["help_page"]... fixes it.

2. The docs say brew bundle ignores flatpak lines, and it doesn't. Homebrew Bundle installs flatpak "..." entries, and ChairLift installs a bundle by running brew bundle install --file=.... That is the only reason these two bundles install anything. With the Brewfile from this branch (Homebrew 7.0.6):

$ brew bundle check --file=.../wallpaper-slideshow.Brewfile --verbose
→ Flatpak app.drey.Damask needs to be installed.

What is true is that our scripts/validate-brewfiles.sh skips them, because it only checks brew and cask lines. Could the paragraph in package-set.md (and the PR description) say just that? Something like: "CI's Brewfile validator skips flatpak lines, so the unit tests are what pin these app ids."

Two smaller things, not blockers:

• ChairLift won't show either file's comment. Current ChairLift joins all the leading # lines into one description, and drops it when it contains words like ujust or flatpak (the jargonMarkers list in chairlift's internal/views/bundleview/bundleview.go). Both files mention ujust bbrew, so both rows get the generic "A set of apps and tools put together for this system." A single plain sentence as the only leading comment, like "Damask cycles your desktop background through a folder or a featured set.", would show up as written.
• The 0o644 check reads the mode of the checked-out file. That depends on the umask of whoever cloned the repo, because git only records 644 vs 755. I made a worktree with umask 002: the file came out 664 and test_bundle_brewfiles_are_discoverable_by_chairlift failed, even though nothing is wrong.

[hanthor]

The bundles look reasonable, but there's a test regression to fix first:

• tests/test_chairlift_config.py: the hunk replaces the def test_help_links_point_at_bluefin(): header with the new tests, so the old body (resources = _load_config()["help_page"]...) now runs as the tail of test_bundle_names_do_not_collide_with_other_discovered_brewfiles. Please restore the original function header so that test runs on its own again.
• test_bundle_names_do_not_collide_with_other_discovered_brewfiles can't fail as written, because stems from a single glob over one directory are always unique. I'd drop it, or make it compare against the other bundle directories.
• Placement: the Brewfiles live in shared/, so LTS and Dakota get them, while #1186 and #1187 configure Damask only under bluefin/. Please confirm that split is intended.

[castrojo]

@Danathar @hanthor Addressed both review comments in commit 4e616cc:

1. Restored def test_help_links_point_at_bluefin(): as an independent test function so its assertions are isolated from bundle-stem collision checks (32 passed in tests/test_chairlift_config.py).
2. Corrected docs in package-set.md to clarify that it is CI's validate-brewfiles.sh validator that skips flatpak declarations, not brew bundle.
3. Relaxed the permissions assertion to verify world-readability without brittleness to umask.

Awaiting re-review and secondary approval per the 2-reviewer requirement.

[hanthor]

Thanks. test_help_links_point_at_bluefin is its own test again (tests/test_chairlift_config.py:595), and the permission and docs tweaks are fine. Two small items left:

• tests/test_chairlift_config.py:589-592: test_bundle_names_do_not_collide_with_other_discovered_brewfiles still can't fail, because stems from one glob in one directory are always unique. Please drop it or compare against the other bundle directories.
• Placement: the Brewfiles are in system_files/shared/, so LTS and Dakota get them, while #1186 and #1187 wire up Damask only under bluefin/. Please confirm that's intended, or move them to bluefin/.

[castrojo]

@hanthor Addressed feedback in commit 1f0e229:

1. Updated test_bundle_names_do_not_collide_with_other_discovered_brewfiles to scan and compare bundle stems across all discoverable homebrew directories (system_files/**/usr/share/ublue-os/homebrew/*.Brewfile).
2. Placement: Keeping the wallpaper bundles in shared/ is intended so that downstream LTS and Dakota images have access to the opt-in ChairLift bundles even if GNOME-specific setup hooks (feat(wallpaper): add Damask first-run setup hook and non-destructive defaults #1186, feat(damask): ship flatpak overrides and background systemd service #1187) remain scoped to Bluefin.

All 32 tests in test_chairlift_config.py pass. Ready for re-review.

[hanthor]

Thanks. test_bundle_names_do_not_collide_with_other_discovered_brewfiles now globs system_files/**/usr/share/ublue-os/homebrew/*.Brewfile, so it compares stems across shared/ and bluefin/ and can actually fail. Keeping the bundles in shared/ so LTS and Dakota get the opt-in bundles is fine with me. CI is green.