Shows how to build and run ONNX Runtime WebGPU provider tests on Linux with no GPU, using the Mesa lavapipe software Vulkan adapter, and where that approach falls short.
$ npx skills add microsoft/onnxruntime --skill webgpu-local-testing -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install microsoft/onnxruntime webgpu-local-testing --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/microsoft/onnxruntime.git skills-src && mkdir -p .claude/skills && cp -r skills-src/.github/skills/webgpu-local-testing .claude/skills/webgpu-local-testing && rm -rf skills-srcUse ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.
Claude Code skills documentation · loads skills from .claude/skills/
Install the "webgpu-local-testing" agent skill from https://github.com/microsoft/onnxruntime/tree/main/.github/skills/webgpu-local-testing into .claude/skills/webgpu-local-testing/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "webgpu-local-testing", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/microsoft/onnxruntime/tree/main/.github/skills/webgpu-local-testingType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add microsoft/onnxruntime --skill webgpu-local-testing -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install microsoft/onnxruntime webgpu-local-testing --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/microsoft/onnxruntime.git skills-src && mkdir -p .agents/skills && cp -r skills-src/.github/skills/webgpu-local-testing .agents/skills/webgpu-local-testing && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "webgpu-local-testing" agent skill from https://github.com/microsoft/onnxruntime/tree/main/.github/skills/webgpu-local-testing into .agents/skills/webgpu-local-testing/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "webgpu-local-testing", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add microsoft/onnxruntime --skill webgpu-local-testing -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install microsoft/onnxruntime webgpu-local-testing --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/microsoft/onnxruntime.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/.github/skills/webgpu-local-testing .cursor/skills/webgpu-local-testing && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "webgpu-local-testing" agent skill from https://github.com/microsoft/onnxruntime/tree/main/.github/skills/webgpu-local-testing into .cursor/skills/webgpu-local-testing/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "webgpu-local-testing", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/microsoft/onnxruntime.git --path .github/skills/webgpu-local-testing--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add microsoft/onnxruntime --skill webgpu-local-testing -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install microsoft/onnxruntime webgpu-local-testing --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/microsoft/onnxruntime.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/.github/skills/webgpu-local-testing .gemini/skills/webgpu-local-testing && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "webgpu-local-testing" agent skill from https://github.com/microsoft/onnxruntime/tree/main/.github/skills/webgpu-local-testing into .gemini/skills/webgpu-local-testing/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "webgpu-local-testing", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install microsoft/onnxruntime webgpu-local-testingInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add microsoft/onnxruntime --skill webgpu-local-testing -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/microsoft/onnxruntime.git skills-src && mkdir -p .github/skills && cp -r skills-src/.github/skills/webgpu-local-testing .github/skills/webgpu-local-testing && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "webgpu-local-testing" agent skill from https://github.com/microsoft/onnxruntime/tree/main/.github/skills/webgpu-local-testing into .github/skills/webgpu-local-testing/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "webgpu-local-testing", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add microsoft/onnxruntime --skill webgpu-local-testing -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install microsoft/onnxruntime webgpu-local-testing --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/microsoft/onnxruntime.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/.github/skills/webgpu-local-testing .opencode/skills/webgpu-local-testing && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "webgpu-local-testing" agent skill from https://github.com/microsoft/onnxruntime/tree/main/.github/skills/webgpu-local-testing into .opencode/skills/webgpu-local-testing/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "webgpu-local-testing", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
webgpu-local-testingShows how to build and run ONNX Runtime WebGPU provider tests on Linux with no GPU, using the Mesa lavapipe software Vulkan adapter, and where that approach falls short.
On Linux the ONNX Runtime WebGPU execution provider runs on Dawn with the Vulkan backend, and Mesa lavapipe is a CPU-based Vulkan adapter that Dawn treats like any device. The skill explains why that is enough for correctness checks: many host-side validation paths, such as shape and broadcast enforcement, fire before any shader is dispatched, and element-wise and broadcasting kernels run correctly if slowly. It covers installing `mesa-vulkan-drivers` and `vulkan-loader` with `dnf` on Azure Linux, the `--use_webgpu` build flag, the `onnxruntime_provider_test` target and `VK_ICD_FILENAMES`.
The scope is strict: lavapipe validates host-side enforce and shape bugs and MatMul-free kernels only. Any graph containing a MatMul, including the expanded Attention node tests, crashes lavapipe's LLVM JIT, and for those macOS arm64 Metal is the source of truth. The skill says never to treat a green lavapipe run as validating an Attention or MatMul fix, and it is not a substitute for a real GPU on performance-sensitive paths. The excerpt is truncated.
6 steps, taken from the step headings in SKILL.md.
Read from SKILL.md and the folder at commit a571b72. It shows what the files ask for, not the result of running them.
Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.
From allowed-tools in the SKILL.md frontmatter.
Shell commands in SKILL.md call:
dnfFrom the folder's file list and the shell code blocks in SKILL.md.
No URLs in SKILL.md.
From URLs in SKILL.md, links to its own repository left out.
Names no API keys, tokens, secrets or passwords.
From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
WebGPU Provider Testing Without a GPU loads about 1.4k tokens when it runs. Until then it costs about 184 tokens; SKILL.md has 595 words of instructions outside code blocks.
Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.
The automated check found no risky patterns in SKILL.md.
Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.
The full file from microsoft/onnxruntime at commit a571b72, republished under its MIT licence (© microsoft). 595 words, ~1,435 tokens.
.claude/skills/webgpu-local-testing/SKILL.md (or your agent's skills folder).Reusable knowledge for exercising the WebGPU execution provider on a Linux box with no physical GPU.
Scope: Linux ORT WebGPU only. macOS uses the Metal backend and a real GPU; this skill is the off-Mac story for running WebGPU EP kernels in CI-less dev loops.
On Linux, ORT's WebGPU EP runs on Dawn with the Vulkan backend. Vulkan does not require a hardware GPU — Mesa lavapipe is a software (CPU) Vulkan adapter that Dawn enumerates like any other device. For EP correctness tests this is sufficient because:
ORT_ENFORCEs) fire
before any shader is dispatched. E.g. the WebGPU broadcast ORT_ENFORCE runs
host-side, so the failure is observable on a software adapter without ever touching
the GPU.You are trading speed for not needing hardware. It is not a substitute for a real GPU on perf-sensitive or driver-specific paths — but for kernel correctness it is the practical local loop.
Scope — what lavapipe can and cannot validate. Software Vulkan covers (a) host-side failures (shape/broadcast
ORT_ENFORCEs that fire before any shader dispatch) and (b) MatMul-free kernels that dispatch. It does NOT cover any graph that contains a MatMul — the MatMul family crashes lavapipe's LLVM JIT (see §5). This explicitly includes the motivating expanded-Attention node tests (test_attention_4d_softcap_neginf_mask_expanded): they decompose tosoftmax(Q·Kᵀ + bias)·V, which contains MatMuls, so they cannot run on lavapipe. For those, macOS-arm64 Metal is the source of truth. Concretely, the #28969 WebGPU broadcast-underflow fix was validated on lavapipe via a standalone Add-broadcastOpTesterproxy (a host-side enforce/shape path) — NOT via the expanded-Attention node test. Never run lavapipe green and conclude an Attention/MatMul fix is validated off-Mac.
dnf, NOT apt)dnf install -y mesa-vulkan-drivers vulkan-loader
# optional, for sanity-checking the adapter:
dnf install -y vulkan-tools && vulkaninfo | headmesa-vulkan-drivers provides lavapipe; vulkan-loader provides the ICD loader.
The lavapipe ICD manifest lands at /usr/share/vulkan/icd.d/lvp_icd.<arch>.json —
lvp_icd.x86_64.json on x86_64, lvp_icd.aarch64.json on arm64. The examples below
use the x86_64 name; substitute your arch, or glob it:
VK_ICD_FILENAMES=$(echo /usr/share/vulkan/icd.d/lvp_icd.*.json).
--use_webgpu./build.sh --config Release --parallel --use_webgpuSee the ort-build skill for general build phases and flags.
WebGPU operator/kernel tests are provider op tests — they build into the
onnxruntime_provider_test target (NOT onnxruntime_test_all; see the ort-test
skill for the executable taxonomy). Point the Vulkan loader at the lavapipe ICD and
select a subset with --gtest_filter:
cd build/Linux/Release
VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/lvp_icd.x86_64.json \
./onnxruntime_provider_test --gtest_filter="*WebGPU*"VK_ICD_FILENAMES forces Vulkan to load only lavapipe, so the run is
deterministic regardless of what else is installed.
MathOpTest.MatMulFloatType (and other MatMul-family tests) crash lavapipe with:
LLVM ERROR: Instruction Combining did not reach a fixpoint after 1 iterationsThis is a pre-existing limitation of software Vulkan (Mesa lavapipe's LLVM JIT), not an ORT bug. Exclude the MatMul family from broad lavapipe runs:
VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/lvp_icd.x86_64.json \
./onnxruntime_provider_test --gtest_filter="*WebGPU*:-*MatMul*"The Linux webgpu CI leg (.github/workflows/linux_webgpu.yml) only builds — it does not run
WebGPU kernels. A green Linux webgpu leg therefore does not mean any WebGPU test
actually executed. The macOS-arm64 webgpu leg is the only CI leg that runs WebGPU
backend node tests. So a local lavapipe run is the practical way to actually exercise
WebGPU kernels off-Mac before you push.
But mind the §1 scope: lavapipe covers host-side enforce/shape paths and MatMul-free
kernels only. Any MatMul-containing graph — including the expanded-Attention node
tests (test_attention_4d_softcap_neginf_mask_expanded) — crashes lavapipe and runs
only on the macOS-arm64 Metal leg, which is the source of truth for those. A green
lavapipe run never validates a MatMul/Attention fix off-Mac.
© microsoft, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
Just SKILL.md in .github/skills/webgpu-local-testing of microsoft/onnxruntime.
Open the folder on GitHubat commit a571b72
WebGPU Provider Testing Without a GPU next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| WebGPU Provider Testing Without a GPU this skillmicrosoft/onnxruntime | 22k | — | ~1.4k | Automated safety check: Pass | MIT | |
| Avm Tf TestingAzure/terraform-azurerm-avm-ptn-alz | 135 | — | ~1.8k | Automated safety check: Pass | MIT | |
| Testingstatic-web-server/static-web-server | 2.4k | — | ~1.5k | Automated safety check: Pass | Apache-2.0 | |
| Verify Changes389ds/389-ds-base | 294 | — | ~1.9k | Automated safety check: Pass | Custom licence | |
| Debug E2E Pipelinekubernetes-sigs/cloud-provider-azure | 294 | — | ~3.4k | Automated safety check: Pass | Apache-2.0 | |
| Debug Failing GPUfacebookexperimental/triton | 201 | — | ~709 | Automated safety check: Pass | MIT |
Azure/terraform-azurerm-avm-ptn-alz
A skill your agent uses for AVM Terraform validation, provider-mocked unit tests, real-Azure integration tests, E2E example tests, PowerShell hooks, OIDC, policy checks, and Avm.Authoring CI behavior.
static-web-server/static-web-server
Write or review tests for the Static Web Server (SWS) project — unit tests, integration tests, test fixtures, and mocking strategies
389ds/389-ds-base
Verify a 389-ds-base change by routing through the environment's build/test skill — static Python gates plus, via that skill, C compile, cmocka, RPM build and pytest suites, cargo tests, and the…
kubernetes-sigs/cloud-provider-azure
Fetch and analyze Prow e2e pipeline failures for cloud-provider-azure.
facebookexperimental/triton
Recover from GPU-busy / GPU-unavailable failures. An agent skill from facebookexperimental/triton.
getsentry/sentry-unreal
Run Sentry unit tests via Unreal automation. An agent skill from getsentry/sentry-unreal.
microsoft/onnxruntime
Finds and fixes out-of-range output writes in ONNX Runtime operator shape-inference functions where a getNumOutputs guard admits too few outputs.
microsoft/onnxruntime
Explains why editing CUTLASS fused-MHA headers in ONNX Runtime can leave stale CUDA kernels after an incremental build, and how to force and verify a real rebuild.
microsoft/onnxruntime
Builds ONNX Runtime from source with its build scripts, explaining the update, build and test phases, key flags and where the build output lands.
microsoft/onnxruntime
Triggers, re-runs and unblocks the CI checks on an ONNX Runtime pull request, after diagnosing whether a failure is transient or needs a code change.
microsoft/onnxruntime
Drafts ONNX Runtime release notes from commit history and contributor metadata using named presets for the full runtime or a scoped component.
microsoft/onnxruntime
Runs and debugs ONNX Runtime tests: Google Test executables for C++ and unittest or pytest for Python, with filters and build-directory guidance.
Works with
Categories
Shows how to build and run ONNX Runtime WebGPU provider tests on Linux with no GPU, using the Mesa lavapipe software Vulkan adapter, and where that approach falls short. On Linux the ONNX Runtime WebGPU execution provider runs on Dawn with the Vulkan backend, and Mesa lavapipe is a CPU-based Vulkan adapter that Dawn treats like any device. The skill explains why that is enough for correctness checks: many host-side validation paths, such as shape and broadcast enforcement, fire before any shader is dispatched, and element-wise and broadcasting kernels run correctly if slowly.
WebGPU Provider Testing Without a GPU fits situations like: running WebGPU kernel correctness tests on a Linux box with no GPU; checking a host-side shape or broadcast enforce fix for the WebGPU provider; debugging a lavapipe crash on a graph that contains MatMul.
Run `npx skills add microsoft/onnxruntime --skill webgpu-local-testing -a claude-code`. Or copy the skill folder (.github/skills/webgpu-local-testing in microsoft/onnxruntime) into .claude/skills/webgpu-local-testing in your project. Claude Code loads it when a task matches its description.
Run `npx skills add microsoft/onnxruntime --skill webgpu-local-testing -a codex`. Or copy the skill folder (.github/skills/webgpu-local-testing in microsoft/onnxruntime) into .agents/skills/webgpu-local-testing in your project. Codex loads it when a task matches its description.
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add microsoft/onnxruntime --skill webgpu-local-testing -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/webgpu-local-testing, .gemini/skills/webgpu-local-testing, .github/skills/webgpu-local-testing and .opencode/skills/webgpu-local-testing in your project.
Going by SKILL.md and its folder, WebGPU Provider Testing Without a GPU needs the command-line tools its instructions call (dnf). Our summary lists: A Linux host, with `dnf` on Azure Linux, to install Mesa Vulkan drivers; An ONNX Runtime source checkout built with `--use_webgpu`.
SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.
WebGPU Provider Testing Without a GPU is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.
About 1.4k tokens (SKILL.md is roughly 5.7k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.
Skills that share tags, products or a category with WebGPU Provider Testing Without a GPU: Avm Tf Testing (Azure/terraform-azurerm-avm-ptn-alz, 135 stars), Testing (static-web-server/static-web-server, 2.4k stars), Verify Changes (389ds/389-ds-base, 294 stars) and Debug E2E Pipeline (kubernetes-sigs/cloud-provider-azure, 294 stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
microsoft (a GitHub organization, an official publisher) maintains it in microsoft/onnxruntime, which has 22,035 GitHub stars. The repository holds 14 skills in this directory. The repository was last updated on October 8, 2026.
Source: microsoft/onnxruntime on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.
