feat: accelerator soft-slicing runtime isolation (NVIDIA HAMi-core / Ascend vcann-rt) by thxCode · Pull Request #5 · gpustack/gpustack-operator

thxCode · 2026-06-26T05:33:09Z

Summary

Implements the deferred runtime isolation Non-Goal of the shipped accelerator-resource-modes-refactor
spec, for the soft-slicing path only (hard slicing — MIG / Ascend vNPU — stays out of scope). The
DeviceManager allocator's GetContainerAllocateResponse turns a bookkeeping-only response into a real
injection: a sliced container starts with a vendor preload library activated via /etc/ld.so.preload
(NVIDIA HAMi-core libvgpu.so; Ascend vcann-rt libvruntime.so + enpu-monitor) and per-container
VRAM/compute quota derived from its .sliced.units request. The preload libraries are compiled into the
operator image per runtime version, staged onto the host by a device-manager init container, and per-pod
working directories are reclaimed by a reconciler-fed GC.

Spec: specs/2026-06-25-accelerator-soft-slicing-runtime-isolation.md.

Packaging — inline `git clone` (no submodule)

The preload sources are cloned inline at pinned commits by the Docker build stages
(ARG LIB_UBS_VIRT_ENPU_COMMIT / ARG LIB_HAMI_CORE_COMMIT; git fetch --depth 1 <SHA>) — no git
submodule, no .gitmodules. Both paths were validated locally with docker build --target:

nvbuild-12 → HAMi-core cloned from GitHub, libvgpu.so produced. ✅
cannbuild-8-910b → vcann-rt cloned from gitcode, libvruntime.so + enpu-monitor produced. ✅

Both GitHub and gitcode accept git fetch --depth 1 origin <SHA> directly.

Tasks (linear T1–T10)

T1 — copy-dir.sh host-staging helper + GPUSTACK_LIB_DIR
T2 — soft-slice quota/version helpers (SliceRatio/FloorPercent; device.RuntimeMajor)
T3 — CANN builder stages: clone + build vcann-rt per CANN/family
T4 — Ascend ld.so.preload asset
T5 — device-manager DaemonSet host staging (init container)
T6 — Ascend sliced GetContainerAllocateResponse (npu_info.config + mounts)
T7 — per-pod working-dir GC via live pod-UUID notifier
T8 — CUDA builder stages: clone + build HAMi-core per CUDA major
T9 — NVIDIA ld.so.preload asset
T10 — NVIDIA sliced GetContainerAllocateResponse (env + mounts)

Plus: CI runs on self-hosted -x4 runners.

Tests & validation

Allocator injection (env / npu_info.config / mounts) covered by table-driven unit tests
(pkg/devicemanager/allocator/{ascend,nvidia}/deviceplugin_test.go) — green.
Both inline-clone Docker build paths validated via --target (above).
⚠️ The full multi-target image build (5 CANN + 2 CUDA stages) needs CI / a disk-rich host.

Deferred to a follow-up spec

In-cluster soft-slicing injection e2e on a GPU-less node: the device-plugin Allocate path is gated on
real hardware detection (the detector probes dcmi/NVML), so a mocked Devices CR can't drive it. A
DeviceManager detector simulation mode (read a fixture instead of probing hardware) is the next-spec
seed (recorded in the spec's Follow-ups).

Status: Shipped — #5 Signed-off-by: thxCode <thxcode0824@gmail.com>

Copilot

Pull request overview

Implements soft-slicing runtime isolation in the GPUStack Operator’s device-manager allocator path by shipping vendor preload runtimes (NVIDIA HAMi-core, Ascend vcann-rt) inside the operator image, staging them onto the host via a DaemonSet init container, and injecting /etc/ld.so.preload + per-container quota/mounts during device-plugin Allocate. Adds per-pod working-dir GC and supporting helpers/tests.

Changes:

Build and package vendor preload artifacts into the operator image (multi-stage Docker builds + pinned inline clones), plus /etc/gpustack/lib/*/ld.so.preload assets and a checksum-guarded host staging helper.
Extend device-plugin allocation responders to inject preload/quota/mounts for sliced mode (Ascend + NVIDIA) and add a notifier-fed per-pod working-dir GC.
Update Helm chart DaemonSet to stage libraries onto the host and mount host /tmp; update docs/spec/test fixtures and CI runner sizing.

Reviewed changes

Copilot reviewed 37 out of 38 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
testing/sample/devices/ascend-910b.yaml	Adds Ascend 910B sample Devices fixture used for allocator testing/validation.
specs/2026-06-25-accelerator-soft-slicing-runtime-isolation.md	Full spec for soft-slicing runtime isolation packaging/injection/GC.
pkg/utils/osx/file.go	Adds `MkdirAll` helper that enforces leaf permissions (umask-independent).
pkg/utils/osx/file_test.go	Unit test for new `osx.MkdirAll`.
pkg/system/storage.go	Adjusts host path layout under `.../operator/...`; adds privileged/container detection helpers.
pkg/internalprocesses/kubernetes/embed.go	Uses `system.IsRunningAsPrivileged()`; updates air-gapped image link comment.
pkg/deviceplugin/types.go	Extends responder interface to include the allocating `*core.Container`.
pkg/deviceplugin/server.go	Threads container through Allocate/PreferredAllocation; adds sliced-mode pod-dir GC hook.
pkg/deviceplugin/server_test.go	Updates stub responder signature for new interface.
pkg/deviceplugin/helper.go	Adds soft-slicing host path helpers + `SliceRatio`/`FloorPercent`.
pkg/deviceplugin/helper_test.go	Unit tests for `SliceRatio` and `FloorPercent`.
pkg/deviceplugin/gc.go	Implements per-pod working-dir GC for sliced mode.
pkg/deviceplugin/gc_test.go	Unit tests for GC miss-streak behavior and nil/empty live sets.
pkg/deviceplugin/controller.go	Notifier payload widened to `[]string` of live pod UIDs; reconciler collects and sends UIDs.
pkg/devicemanager/detector/ascend/device.go	Updates Ascend family mapping comments/values (910B/910C/950).
pkg/devicemanager/allocator/thead/deviceplugin.go	Signature update for responder interface (container param).
pkg/devicemanager/allocator/nvidia/deviceplugin.go	Adds sliced-mode injection (HAMi-core preload + quota + mounts + dirs).
pkg/devicemanager/allocator/nvidia/deviceplugin_test.go	Table-driven tests for NVIDIA sliced injection behavior.
pkg/devicemanager/allocator/mthreads/deviceplugin.go	Signature update for responder interface (container param).
pkg/devicemanager/allocator/metax/deviceplugin.go	Signature update for responder interface (container param).
pkg/devicemanager/allocator/iluvatar/deviceplugin.go	Signature update for responder interface (container param).
pkg/devicemanager/allocator/hygon/deviceplugin.go	Signature update for responder interface (container param).
pkg/devicemanager/allocator/cambricon/deviceplugin.go	Signature update for responder interface (container param).
pkg/devicemanager/allocator/ascend/deviceplugin.go	Adds sliced-mode injection (vcann-rt preload + npu_info.config + mounts).
pkg/devicemanager/allocator/ascend/deviceplugin_test.go	Table-driven tests for Ascend sliced injection and vNPU ID assignment.
pkg/devicemanager/allocator/amd/deviceplugin.go	Signature update for responder interface (container param).
pkg/device/helper.go	Adds `device.RuntimeMajor` helper for runtime-version directory selection.
pkg/device/helper_test.go	Unit tests for `device.RuntimeMajor`.
pack/gpustack-operator/rootfs/usr/bin/copy-dir.sh	New checksum-guarded recursive staging script used by init container.
pack/gpustack-operator/rootfs/etc/gpustack/lib/nvidia/ld.so.preload	Adds NVIDIA preload asset pointing at `/usr/local/vgpu/libvgpu.so`.
pack/gpustack-operator/rootfs/etc/gpustack/lib/ascend/ld.so.preload	Adds Ascend preload asset pointing at `/opt/enpu/vcann-rt/lib/libvruntime.so`.
pack/gpustack-operator/external/nvidia/build-libvgpu.sh	Build script for HAMi-core `libvgpu.so` (build-stage only).
pack/gpustack-operator/external/ascend/build-libvnpu.sh	Build script for vcann-rt artifacts with dcmi/HAL build-time stubbing.
pack/gpustack-operator/Dockerfile	Adds GPUSTACK_LIB_DIR, pinned inline clones, nvbuild/cannbuild stages, and installs assets/scripts.
docs/architecture.md	Documents sliced-mode runtime isolation injection + packaging/staging/GC behavior.
deploy/gpustack-operator/chart/templates/device-manager/daemonset.yaml	Adds init container staging `/etc/gpustack/lib` to host; mounts host `/tmp`.
.github/workflows/ci.yml	Moves CI builds to `*-x4` runners for larger multi-stage image builds.
.claude/skills/gpustack-operator-chart-e2e/SKILL.md	Clarifies chart-e2e skill scope (chart contract vs feature e2e).

Copilot

Pull request overview

Copilot reviewed 37 out of 38 changed files in this pull request and generated 4 comments.

Signed-off-by: thxCode <thxcode0824@gmail.com>

Add a recursive, sha256-guarded copy-dir.sh into the operator image rootfs (installed to /usr/bin, 0755) for the device-manager init container to stage the soft-slicing library tree onto the host idempotently. Declare the new GPUSTACK_LIB_DIR build arg (= GPUSTACK_CONF_DIR/lib). Task 1 of accelerator-soft-slicing-runtime-isolation. Signed-off-by: thxCode <thxcode0824@gmail.com>

Status: Shipped — #5 Signed-off-by: thxCode <thxcode0824@gmail.com>

Copilot

Pull request overview

Copilot reviewed 37 out of 38 changed files in this pull request and generated 2 comments.

Copilot

Pull request overview

Copilot reviewed 38 out of 39 changed files in this pull request and generated 2 comments.

Copilot

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

Copilot

Pull request overview

Copilot reviewed 54 out of 55 changed files in this pull request and generated 3 comments.

Ship /etc/gpustack/lib/ascend/ld.so.preload (0644) pointing at the in-container vcann-rt mount /opt/enpu/vcann-rt/lib/libvruntime.so, installed in the final stage alongside the cann-* libs so the device-manager init container stages the whole /etc/gpustack/lib tree onto the host. Task 4 of accelerator-soft-slicing-runtime-isolation. Signed-off-by: thxCode <thxcode0824@gmail.com>

…t container Add a stage-libs init container that runs copy-dir.sh to idempotently stage the in-image /etc/gpustack/lib tree onto the host at /var/lib/gpustack/operator/lib, and mount host /tmp into the device-manager so the allocator can create the shared /tmp/vgpulock the NVIDIA soft-slicing runtime expects. Task 5 of accelerator-soft-slicing-runtime-isolation. Signed-off-by: thxCode <thxcode0824@gmail.com>

…ntainers Register the Ascend Sliced device-plugin server and rewrite its GetContainerAllocateResponse to apply real soft-slicing isolation: - render a per-container npu_info.config (aicore/memory quota from the .sliced.units request, shm-id from the accelerator ID, scheduling-policy=2); - assign the lowest-free virtual-npu-id per physical NPU by scanning on-disk configs (level-based, idempotent on re-allocation); - mount the staged libvruntime.so + enpu-monitor + ld.so.preload + the config + /dev/shm into the container. vcann-rt's npu_info.config is single-NPU, so a sliced Ascend container maps to one card. Shared OperatorLibDir/OperatorPodsDir/PodWorkDir helpers land in pkg/deviceplugin for the NVIDIA path to reuse. Task 6 of accelerator-soft-slicing-runtime-isolation. Signed-off-by: thxCode <thxcode0824@gmail.com>

Copilot

Pull request overview

Copilot reviewed 54 out of 55 changed files in this pull request and generated 2 comments.

…ifier Widen the DevicesReconciler notifier from chan struct{} to chan []string so it carries the node's live pod-UUID set on each reconcile, and add a level-based podDirGC that the Sliced server runs in ListAndWatch: it seeds from on-disk pod directories and removes one only after its UUID has been absent from the live set for 3 consecutive reconciles. Exclusive/Shared servers keep emitting the ListAndWatch response and ignore the payload. Task 7 of accelerator-soft-slicing-runtime-isolation. Signed-off-by: thxCode <thxcode0824@gmail.com>

Add the nvbuild-12 / nvbuild-13 stages that compile HAMi-core (libvgpu.so) against the nvidia/cuda 12.9.2 / 13.0.3 cudnn-devel-ubi8 images (install cmake, run HAMi-core build.sh via build-libvgpu.sh) and copy the products into ${GPUSTACK_LIB_DIR}/nvidia/cuda-{12,13} in the final image. Both stages verified-built locally (libvgpu.so ~700KB each); HAMi-core compiles cleanly against cuda-13. Task 8 of accelerator-soft-slicing-runtime-isolation. Signed-off-by: thxCode <thxcode0824@gmail.com>

Ship /etc/gpustack/lib/nvidia/ld.so.preload (0644) pointing at the in-container HAMi-core mount /usr/local/vgpu/libvgpu.so, installed in the final stage alongside the cuda-* libs so the device-manager init container stages it onto the host. Task 9 of accelerator-soft-slicing-runtime-isolation. Signed-off-by: thxCode <thxcode0824@gmail.com>

…ontainers Rewrite the NVIDIA Sliced GetContainerAllocateResponse to apply real soft-slicing isolation: - set CUDA_DEVICE_SM_LIMIT (floor(R*100)) and a per-card CUDA_DEVICE_MEMORY_LIMIT_<i> (MiB->KiB, scaled by R) from the .sliced.units request, plus the shared cache env; - create the per-container /tmp/vgpulock, pods/<X>, and pods/<X>/tmp/vgpu dirs (0777); - mount the staged libvgpu.so + ld.so.preload + the vgpu cache + /dev/shm, keeping the GPU visible via NVIDIA_VISIBLE_DEVICES so HAMi-core enforces the limits. Completes the soft-slicing runtime-isolation spec (Status: Built). Task 10 of accelerator-soft-slicing-runtime-isolation. Signed-off-by: thxCode <thxcode0824@gmail.com>

- chart-e2e: clarify it validates the chart contract (install/startup/image/ version), not feature/behavioral e2e - architecture.md: note the device-plugin allocator soft-slicing runtime isolation in Stage 2 - spec: align task descriptions with the inline-clone packaging (no submodule) and record the GPU-less injection-e2e blocker as a next-spec follow-up The sliced soft-slicing injection e2e (CASE 6) is intentionally NOT added here: GPU-less simulation can't drive the device-plugin Allocate (detector is hardware-gated), so it's deferred to a spec that adds a detector simulation mode. Signed-off-by: thxCode <thxcode0824@gmail.com>

Copilot

Pull request overview

Copilot reviewed 54 out of 55 changed files in this pull request and generated 3 comments.

+    command -v git >/dev/null 2>&1 || { apt-get update -y && apt-get install -y --no-install-recommends git ca-certificates && rm -rf /var/lib/apt/lists/*; }
+    rm -rf /tmp/ubs-virt && git init -q /tmp/ubs-virt && cd /tmp/ubs-virt
+    git remote add origin https://gitcode.com/openeuler/ubs-virt.git
+    git fetch --depth 1 origin "${LIB_UBS_VIRT_ENPU_COMMIT}" || git fetch origin


+    set -exo pipefail
+    rm -rf /tmp/hami-core-src && git init -q /tmp/hami-core-src && cd /tmp/hami-core-src
+    git remote add origin https://github.com/Project-HAMi/HAMi-core.git
+    git fetch --depth 1 origin "${LIB_HAMI_CORE_COMMIT}" || git fetch origin


+	// Sliced containers leave per-pod working directories on the host; reclaim them
+	// as their pods disappear from the reconciler's live pod-UUID set.
+	var gc *podDirGC
+	if s.AllocationMode == workercore.DeviceAllocationModeSliced {
+		gc = newPodDirGC(OperatorPodsDir)
+	}
+
 	// Watch for updates and send ListAndWatch response whenever there's a change.
+	s.Logger.Info("watching for device updates")
 	for {
 		select {
 		case <-ctx.Done():
 			return ctx.Err()
-		case <-notifier:
+		case livePodUIDs := <-notifier:
 			resp, err := s.getListAndWatchResponse(ctx)
 			if err != nil {
-				s.Logger.Error(err, "get list and watch response")
+				s.Logger.Error(err, "get list and watch response on update")
 				return err
 			}
 			if err = srv.Send(resp); err != nil {
 				s.Logger.Error(err, "send list and watch response")
 				return err
 			}
+			s.Logger.Info("sent list and watch response")
+			if gc != nil {
+				gc.reconcile(livePodUIDs)
+			}


Copilot

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

Signed-off-by: thxCode <thxcode0824@gmail.com>

Copilot

Pull request overview

Copilot reviewed 54 out of 55 changed files in this pull request and generated 4 comments.

+        - name: gpustack-nvidia-vgpulock-dir
+          hostPath:
+            path: /tmp/vgpulock
+            type: Directory
+        {{- end }}


+// FloorPercent converts a per-card fraction R into the integer compute percent the
+// soft-slicing runtimes expect (HAMi-core CUDA_DEVICE_SM_LIMIT, vcann-rt aicore-quota),
+// rounding down: floor(R*100).
+func FloorPercent(r float64) int {
+	return int(r * 100)


+		"CUDA_DEVICE_SM_LIMIT":            strconv.Itoa(deviceplugin.FloorPercent(ratio)),
+		"CUDA_DEVICE_MEMORY_SHARED_CACHE": ctrVgpuSharedCache,
+	}
+	for i := range accels {
+		limit := int64(float64(accels[i].group.Memory) * ratio)


+	_, _ = fmt.Fprintf(&b, "physical-npu-id=%d\n", npuId)
+	_, _ = fmt.Fprintf(&b, "virtual-npu-id=%d\n", vnpuId)
+	_, _ = fmt.Fprintf(&b, "aicore-quota=%d\n", deviceplugin.FloorPercent(ratio))
+	_, _ = fmt.Fprintf(&b, "memory-quota=%d\n", int64(float64(memoryMiB)*ratio))
+	_, _ = fmt.Fprintf(&b, "shm-id=%s\n", shmID)


Copilot AI review requested due to automatic review settings June 26, 2026 05:33

Copilot started reviewing on behalf of thxCode June 26, 2026 05:33 View session

thxCode added a commit that referenced this pull request Jun 26, 2026

docs(spec): mark soft-slicing runtime isolation Shipped

97a433c

Status: Shipped — #5 Signed-off-by: thxCode <thxcode0824@gmail.com>

Copilot AI reviewed Jun 26, 2026

View reviewed changes

Comment thread pkg/devicemanager/allocator/ascend/deviceplugin.go

Comment thread docs/architecture.md Outdated

Comment thread pkg/system/storage.go

Comment thread pkg/internalprocesses/kubernetes/embed.go

Copilot AI review requested due to automatic review settings June 26, 2026 05:38

Copilot started reviewing on behalf of thxCode June 26, 2026 05:38 View session

Copilot AI reviewed Jun 26, 2026

View reviewed changes

Comment thread pkg/internalprocesses/kubernetes/embed.go

Comment thread docs/architecture.md Outdated

Comment thread pkg/devicemanager/allocator/ascend/deviceplugin.go

Comment thread pkg/devicemanager/allocator/nvidia/deviceplugin.go Outdated

thxCode added 6 commits June 26, 2026 13:53

refactor: system stroage

1566b9a

Signed-off-by: thxCode <thxcode0824@gmail.com>

refactor(devicemanager): pass container during allocation

622abc4

Signed-off-by: thxCode <thxcode0824@gmail.com>

refactor(devicemanager): adjust cann variant

857608d

Signed-off-by: thxCode <thxcode0824@gmail.com>

refactor(devicemanager): adjust controller predication

8c1294b

Signed-off-by: thxCode <thxcode0824@gmail.com>

chore(testing): sample devices

b8f1571

Signed-off-by: thxCode <thxcode0824@gmail.com>

thxCode added a commit that referenced this pull request Jun 26, 2026

docs(spec): mark soft-slicing runtime isolation Shipped

ed6c627

Status: Shipped — #5 Signed-off-by: thxCode <thxcode0824@gmail.com>

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from 97a433c to ed6c627 Compare June 26, 2026 06:08

thxCode requested a review from Copilot June 26, 2026 06:11

Copilot started reviewing on behalf of thxCode June 26, 2026 06:12 View session

Copilot AI reviewed Jun 26, 2026

View reviewed changes

Comment thread pkg/deviceplugin/controller.go Outdated

Comment thread pkg/deviceplugin/server.go

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from ed6c627 to c17541b Compare June 26, 2026 06:29

Copilot AI review requested due to automatic review settings June 26, 2026 07:37

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from c17541b to d017f64 Compare June 26, 2026 07:37

Copilot started reviewing on behalf of thxCode June 26, 2026 07:38 View session

Copilot AI reviewed Jun 26, 2026

View reviewed changes

Comment thread .github/workflows/ci-chart.yml Outdated

Comment thread pkg/utils/osx/file.go Outdated

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from d017f64 to 76b62e9 Compare June 26, 2026 11:58

thxCode requested a review from Copilot June 26, 2026 12:00

Copilot started reviewing on behalf of thxCode June 26, 2026 12:00 View session

Copilot AI reviewed Jun 26, 2026

thxCode requested a review from Copilot June 26, 2026 12:05

Copilot started reviewing on behalf of thxCode June 26, 2026 12:05 View session

Copilot AI reviewed Jun 26, 2026

View reviewed changes

Comment thread .claude/skills/gpustack-operator-xbuild-and-verify/scripts/preflight.sh Outdated

Comment thread specs/2026-06-25-accelerator-soft-slicing-runtime-isolation.md Outdated

Comment thread specs/2026-06-25-accelerator-soft-slicing-runtime-isolation.md Outdated

thxCode added 3 commits June 26, 2026 20:41

Copilot AI review requested due to automatic review settings June 26, 2026 12:43

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from 0f4536a to bbb65fb Compare June 26, 2026 12:43

Copilot started reviewing on behalf of thxCode June 26, 2026 12:43 View session

Copilot AI reviewed Jun 26, 2026

View reviewed changes

Comment thread pkg/deviceplugin/controller.go Outdated

Comment thread pkg/deviceplugin/gc.go

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from bbb65fb to 4bb7501 Compare June 26, 2026 13:21

thxCode added 5 commits June 26, 2026 21:33

Copilot AI review requested due to automatic review settings June 26, 2026 13:33

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from 4bb7501 to babb2cd Compare June 26, 2026 13:33

Copilot started reviewing on behalf of thxCode June 26, 2026 13:34 View session

Copilot AI reviewed Jun 26, 2026

View reviewed changes

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from babb2cd to 9640796 Compare June 26, 2026 16:09

Copilot AI review requested due to automatic review settings June 26, 2026 16:12

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from 9640796 to 9dc17b9 Compare June 26, 2026 16:12

Copilot started reviewing on behalf of thxCode June 26, 2026 16:12 View session

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from 9dc17b9 to 4a8136e Compare June 26, 2026 16:14

Copilot AI reviewed Jun 26, 2026

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from 4a8136e to e0e7386 Compare June 26, 2026 16:28

refactor: ci and skills

250d2d2

Signed-off-by: thxCode <thxcode0824@gmail.com>

Copilot AI review requested due to automatic review settings June 26, 2026 16:37

thxCode force-pushed the feat/accelerator-soft-slicing-runtime-isolation-v2 branch from e0e7386 to 250d2d2 Compare June 26, 2026 16:37

Copilot started reviewing on behalf of thxCode June 26, 2026 16:37 View session

thxCode merged commit 5829fd5 into main Jun 26, 2026
7 checks passed

Copilot AI reviewed Jun 26, 2026

View reviewed changes

Uh oh!

Conversation

thxCode commented Jun 26, 2026

Summary

Packaging — inline git clone (no submodule)

Tasks (linear T1–T10)

Tests & validation

Deferred to a follow-up spec

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Packaging — inline `git clone` (no submodule)