docs: record that system_files/nvidia/ is unconsumed and has no CDI preset — docs/skills/nvidia/{SKILL.md,references/architecture.md}, system_files/nvidia/README.md by hivecommons-hive[bot] · Pull Request #1125 · projectbluefin/common

hivecommons-hive · 2026-09-15T09:42:16Z

Docs-truth correction

Files claimed: docs/skills/nvidia/SKILL.md ("The three repos and their nvidia stacks" table + new "system_files/nvidia/ ships to nobody" subsection), docs/skills/nvidia/references/architecture.md (the projectbluefin/common bullet list under "Per-repo: where nvidia code lives"), system_files/nvidia/README.md (header status note), and the generated docs/skills/index.json / index.md (frontmatter bump only). Cluster: issue #1124 — the unconsumed system_files/nvidia/ overlay.

No code, unit, preset, script or test is touched. This PR only makes the docs describe what the build actually produces.

What was wrong

80-nvidia-container-toolkit.preset has never existed here. SKILL.md:42 and references/architecture.md:55 both named system_files/nvidia/usr/lib/systemd/system-preset/80-nvidia-container-toolkit.preset as this repo's CDI preset. git log --all -- system_files/nvidia/usr/lib/systemd/system-preset is empty. The same table recorded bluefin's CDI preset as "inherits from common"; bluefin has no such preset either.
system_files/nvidia/ reaches no image. The ctx stage publishes /system_files/nvidia, but every consumer copies shared + bluefin and stops — projectbluefin/bluefin @5729176 Containerfile:48-49, projectbluefin/bluefin-lts @2170146 Containerfile:17-18, projectbluefin/utah @636b48e Containerfile:66-67. No systemctl enable for ublue-nvidia-flatpak-runtime-sync.service exists anywhere in the org, and this repo ships no preset for it. tests/test_nvidia_flatpak_sync.bats asserts against files on disk, so it stays green either way.

The docs told contributors that edits under system_files/nvidia/ "flow into all nvidia-variant images at next build". They flow nowhere.

What this PR does not do

It does not delete the overlay and it does not wire it up. Both dispositions are recorded in #1124: wiring it requires a consumer-repo COPY plus an enablement decision (behavior change, human owner); deleting it discards an actively maintained unit (last functional change #769). This PR only stops the skill from asserting a layout that was never built, and marks the tree so the next contributor does not assume delivery.

Verified locally: scripts/check-doc-links.sh, scripts/check-skill-index.sh, pytest tests/test_skill_docs.py (10 passed). docs/skills/index.json / index.md regenerated with scripts/generate_skill_index.py --write after the frontmatter bump to version: "1.2" / last_updated: "2026-09-15" — that regeneration also refreshes generated_at, which is the script's normal behavior.

Refs #1124 — left open deliberately: this PR corrects only the documentation. The structural decision (copy /system_files/nvidia in the nvidia build path and enable the unit, or delete the tree) stays open for a human owner.

Filed by architect agent (ACMM L5 — hold-gated mode). Hold-gated: human review required.

— hive: agent=architect backend=copilot model=claude-opus-5

Danathar

Verified the factual claims: bluefin's Containerfile (testing, lines 48-49) copies only /system_files/shared and /system_files/bluefin from common; nothing in bluefin references system_files/nvidia; system_files/nvidia/usr/lib/systemd/system-preset/ does not exist in common; and the only mentions of ublue-nvidia-flatpak-runtime-sync across the org are the unit itself, its bats test, and docs — no preset or systemctl enable anywhere. Docs-only, and keeping #1124 open for the actual wire-or-delete decision is the right split.

Danathar · 2026-09-21T18:32:27Z

Hold removed: this was the Hive's ACMM level gate, which I operate, and the PR has a human approval with CI green. Ready for a maintainer to merge.

hivecommons-hive

docs-currency

(low) The new docs cite unpinned cross-repo line numbers that have already drifted. docs/skills/nvidia/SKILL.md:53-55, references/architecture.md:57-58, and system_files/nvidia/README.md:6-8 all say utah copies at Containerfile:66-67 — that was true at the SHA pinned in the PR body (@636b48e), but utah's default branch now has those COPYs at lines 76-77. Suggest either pinning the SHA in the doc text (as the PR body does) or dropping the line numbers and citing the COPY stanzas by content.

Substantive claims verified at head c74ebff: no preset exists under system_files/nvidia/ (tree has only the service + libexec helper), bluefin/bluefin-lts/utah all copy /system_files/shared + /system_files/bluefin only, and no CDI preset exists in projectbluefin/bluefin.

No findings from: correctness, security, intent-alignment, style.

— hive: agent=reviewer backend=copilot model=claude-fable-5 copilot=1.0.78

hivecommons-hive

docs-currency

New text cites a bluefin-lts path that does not exist. docs/skills/nvidia/references/architecture.md:68 (added by this PR) says CDI auto-generation is enabled per-consumer via bluefin-lts system_files_overrides/gdx/…. bluefin-lts has no gdx override directory — at both HEAD and the SHA 2170146 this PR cites, the preset is at system_files_overrides/nvidia/usr/lib/systemd/system-preset/80-nvidia-container-toolkit.preset (override dirs there are nvidia, dx, hwe, arch variants — no gdx). The unchanged SKILL.md table row for bluefin-lts (docs/skills/nvidia/SKILL.md:43) and the pre-existing gdx section (architecture.md:92-99) carry the same stale name, so a PR whose stated purpose is docs-truth is re-asserting it in fresh text. Suggest correcting line 68 to system_files_overrides/nvidia/… (and noting the stale table row in #1124 if a broader fix is out of scope here).

Every other factual claim verified at head c46a0e4: no preset dir under system_files/nvidia/ (tree contains only usr/lib/systemd/system/ and usr/libexec/), common Containerfile:90 publishes /system_files/nvidia, and none of bluefin (Containerfile:48-49), bluefin-lts (Containerfile:17-18), utah (Containerfile:76-77) copies it. Dakota's elements/bluefin-nvidia/nvidia-container-toolkit-preset.bst exists. index.json/index.md regeneration is internally consistent (version 1.2, 2026-09-15, 40 skills). #1124 is open.

No findings from: correctness, security, intent-alignment, style.

— hive: agent=reviewer backend=copilot model=claude-fable-5

hivecommons-hive

Reviewed — no findings from correctness, security, intent-alignment, style, docs-currency.

— hive: agent=reviewer backend=copilot model=claude-fable-5 copilot=1.0.78

Danathar

Re-approved at ea9631b after two follow-up commits. They correct the LTS override directory from gdx/ to nvidia/ (I checked projectbluefin/bluefin-lts: system_files_overrides has nvidia, x86_64-nvidia and aarch64-nvidia, no gdx) and replace Containerfile line-number citations with a grep the reader can run, which will not drift. Docs only, CI green.

…has no CDI preset docs/skills/nvidia/SKILL.md, docs/skills/nvidia/references/architecture.md and system_files/nvidia/README.md described a delivery path the build does not produce: - The ctx stage publishes /system_files/nvidia, but bluefin (Containerfile:48-49), bluefin-lts (Containerfile:17-18) and utah (Containerfile:66-67) copy /system_files/shared and /system_files/bluefin only. Nothing in the org enables ublue-nvidia-flatpak-runtime-sync.service and this repo ships no preset for it. - Both docs cited system_files/nvidia/usr/lib/systemd/system-preset/80-nvidia-container-toolkit.preset as this repo's CDI preset. That path has never existed (git log --all over it is empty), so bluefin's 'inherits from common' row was also false. Docs-only: records measured state and points at common#1124 for the disposition decision. Skill frontmatter bumped to 1.2 and the generated index regenerated with scripts/generate_skill_index.py --write. Refs #1124 Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: sec-check[bot] <sec-check[bot]@users.noreply.github.com>

Signed-off-by: sec-check[bot] <sec-check[bot]@users.noreply.github.com>

The cross-repo citations pinned bluefin, bluefin-lts and utah Containerfile line numbers. utah has already drifted (66-67 -> 76-77 on its default branch), so the docs were stale before merge. Replace the line-number citations with a content-based description plus a reproducible grep command. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The nvidia build overlay in projectbluefin/bluefin-lts lives in `build_scripts/overrides/nvidia/` and `system_files_overrides/nvidia/`; there is no `gdx` override directory at HEAD. Also correct the build entrypoint path to `build_scripts/build.sh` and note the arch-specific override directories. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

castrojo

LGTM. Clean rebase on main; docs accurately record that system_files/nvidia/ is unconsumed and has no CDI preset.

hivecommons-hive

docs-currency

docs/skills/nvidia/SKILL.md:4 sets last_updated: "2026-09-15" (mirrored into docs/skills/index.json:471), but the change lands 2026-09-22 (generated_at in the same regenerated index). Minor: the freshness stamp predates the edit it describes.

All factual claims in the diff were verified against the repos at head 6028704: system_files/nvidia/ contains no system-preset/ dir; common Containerfile:97 publishes /system_files/nvidia but bluefin (Containerfile:48-49), bluefin-lts (Containerfile:17-18) and utah (Containerfile:76-77) copy only shared/bluefin; bluefin-lts overrides live under nvidia/ (not gdx/) with system_files_overrides/nvidia/.../80-nvidia-container-toolkit.preset present and build_scripts/build.sh at that path; no systemctl enable of ublue-nvidia-flatpak-runtime-sync.service exists in the org; issue #1124 is open.

No findings from: correctness, security, intent-alignment, style.

Confidence: 5/5 (safe)

— hive: agent=reviewer backend=copilot model=claude-fable-5 copilot=1.0.78

hivecommons-hive Bot requested review from a team and repires as code owners September 15, 2026 09:42

hivecommons-hive Bot added hold Work is intentionally paused. architecture Structural or interface design work. agent/architect Filed or owned by the architect agent. hive/hosted-projectbluefin-knuckle-gjvq Routed by the hosted Project Bluefin Hive deployment. labels Sep 15, 2026

clubanderson mentioned this pull request Sep 16, 2026

Retitle 7 Hive PRs blocked by the Conventional Commits title gate #1132

Closed

hivecommons-hive Bot added the agent/scanner Filed or owned by the scanner agent. label Sep 17, 2026

Danathar mentioned this pull request Sep 18, 2026

kubestellar-hive files PRs with titles that fail the Conventional Commits check #939

Closed

Danathar approved these changes Sep 19, 2026

View reviewed changes

Danathar mentioned this pull request Sep 19, 2026

ci: docs-only PRs can never merge — required build checks are path-ignored on docs changes #1149

Open

Danathar removed the hold Work is intentionally paused. label Sep 21, 2026

hivecommons-hive Bot commented Sep 21, 2026

View reviewed changes

Danathar approved these changes Sep 22, 2026

View reviewed changes

sec-check[bot] and others added 4 commits September 22, 2026 21:44

chore: re-trigger validate after PR title correction

baaca7f

Signed-off-by: sec-check[bot] <sec-check[bot]@users.noreply.github.com>

castrojo force-pushed the arch/nvidia-overlay-docs-truth branch from ea9631b to 6028704 Compare September 23, 2026 01:44

castrojo approved these changes Sep 23, 2026

View reviewed changes

castrojo merged commit acf3628 into main Sep 23, 2026
9 checks passed

castrojo deleted the arch/nvidia-overlay-docs-truth branch September 23, 2026 01:57

hivecommons-hive Bot commented Sep 23, 2026

View reviewed changes

hivecommons-hive Bot mentioned this pull request Sep 23, 2026

[architect] system_files/nvidia/ is a produced-but-unconsumed overlay: no consumer copies it, no preset enables it, docs cite a preset that never existed #1124

Open

Navigation Menu

docs: record that system_files/nvidia/ is unconsumed and has no CDI preset — docs/skills/nvidia/{SKILL.md,references/architecture.md}, system_files/nvidia/README.md - #1125

docs: record that system_files/nvidia/ is unconsumed and has no CDI preset — docs/skills/nvidia/{SKILL.md,references/architecture.md}, system_files/nvidia/README.md#1125
castrojo merged 4 commits into
mainfrom
arch/nvidia-overlay-docs-truth

hivecommons-hive Bot commented Sep 15, 2026

Uh oh!

Danathar left a comment

Uh oh!

Danathar commented Sep 21, 2026

Uh oh!

hivecommons-hive Bot left a comment

Uh oh!

hivecommons-hive Bot left a comment

Uh oh!

hivecommons-hive Bot left a comment

Uh oh!

Danathar left a comment

Uh oh!

castrojo left a comment

Uh oh!

Uh oh!

hivecommons-hive Bot left a comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

hivecommons-hive Bot commented Sep 15, 2026

Docs-truth correction

What was wrong

What this PR does not do

Uh oh!

Danathar left a comment

Choose a reason for hiding this comment

Uh oh!

Danathar commented Sep 21, 2026

Uh oh!

hivecommons-hive Bot left a comment

Choose a reason for hiding this comment

Uh oh!

hivecommons-hive Bot left a comment

Choose a reason for hiding this comment

Uh oh!

hivecommons-hive Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Danathar left a comment

Choose a reason for hiding this comment

Uh oh!

castrojo left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

hivecommons-hive Bot left a comment

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants