Borrowing it
Nothing to install: this file belongs to cpuguy96/StepCOVNet. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/cpuguy96/StepCOVNet/master/.cursor/skills/wsl-gpu-stepcovnet/SKILL.mdgit clone --depth 1 https://github.com/cpuguy96/StepCOVNetWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/cpuguy96/stepcovnet/wsl-gpu-stepcovnet)<a href="https://agentmods.dev/skills/cpuguy96/stepcovnet/wsl-gpu-stepcovnet"><img src="https://agentmods.dev/badge/skills/cpuguy96/stepcovnet/wsl-gpu-stepcovnet/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/cpuguy96/stepcovnet/wsl-gpu-stepcovnet"><img src="https://agentmods.dev/badge/skills/cpuguy96/stepcovnet/wsl-gpu-stepcovnet.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00058 | $0.01417 |
| Opus 5 | $0.00029 | $0.00709 |
| Sonnet 5 | $0.00012 | $0.00283 |
| Haiku 4.5 | $0.00006 | $0.00142 |
Grade A, and why
wsl-gpu-stepcovnet scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 10d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 96 lines — stays where its author put it; the contents beside it link to each section on GitHub.
WSL GPU (StepCOVNet)
Windows development model
On Windows, CPU work stays in the clone's Windows venv; CUDA work runs in WSL. The repo clone can live on any drive or folder — dispatch code derives WSL paths from the current checkout, not from a fixed username or /mnt/c/Users/... string.
| Setting | Default | Override |
|---|---|---|
| WSL venv root | $HOME/stepcovnet-venv-wsl |
WSL_VENV |
| WSL Python | $WSL_VENV/bin/python |
STEPCOVNET_WSL_PYTHON |
| Repo root | cwd / script location | — |
Shell helpers share these via scripts/wsl_common.sh.
Rule
On native Windows, never run CUDA/GPU training in the Windows venv. From repo root (project venv activated):
python scripts/<script>.py <args>
when the script calls wsl_gpu.maybe_dispatch_for_training (or MERT --device=cuda dispatch).
Opt out: STEPCOVNET_NO_WSL=1.
GPU scheduling (one job at a time)
The WSL GPU is single-tenant for the agent:
| Rule | Detail |
|---|---|
| One training job | Do not start a second WSL GPU training run while one is already active (train, overfit, MERT extract with --device=cuda). |
| No train + infer overlap | Do not run GPU training and GPU inference (debug_*, --ar_decode, eval_dense_*, model.predict) in separate processes at the same time. |
| Before launching | assert_wsl_gpu_free_for_training() checks nvidia-smi and logs/gpu_wsl.lock (stepcovnet.wsl_gpu_lock). Windows dispatch acquires the lock before wsl; a second parallel launch fails on O_EXCL. Override: STEPCOVNET_FORCE_GPU=1. |
| Manual check | python -c "from stepcovnet import wsl_gpu; wsl_gpu.assert_wsl_gpu_free_for_training()" |
| After training | Offline decode/eval on GPU is fine sequentially once training has exited. |
CPU scripts (pytest, lint, config edits) may run while a GPU job is active — they do not use the WSL CUDA device.
Manual WSL (debugging or already inside WSL)
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 10d ago First seen · 96 lines · 58 tokens per session scan A de5f1ba8cb10
wsl-gpu-stepcovnet is a skill published in the GitHub repository cpuguy96/StepCOVNet (22 stars, last pushed 16d ago), licensed Apache-2.0. It adds 58 tokens to every session and 1,417 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
physicsnemo-discover
Official NVIDIA-authored guidance for navigating PhysicsNeMo — pick the model, datapipe, or example for a SciML/AI4Science task (surrogates, forecasting, downscaling, physics-informed, inverse, generative). Points at existing files via live repo search; never writes code. Do NOT use for installation or environment…
lepton-cli
Operate NVIDIA DGX Cloud Lepton through the globally installed lep CLI. Use when the user asks to inspect or manage Lepton workspaces, endpoints/deployments, dev pods, batch jobs, fine-tuning jobs, Ray clusters, Slurm clusters, Dynamo endpoints, storage, secrets, nodes, ingress, templates, logs, authentication, or…
depth-estimation
Real-time depth map privacy transforms using Depth Anything v2 (CoreML + PyTorch).
HomeSafe-Bench
VLM indoor safety hazard detection benchmark inspired by HomeSafeBench (arXiv 2509.23690).
ml-for-aec
Computer vision for buildings, image-to-floorplan, generative ML models, performance prediction, structural analysis ML, energy prediction, natural language to design, and point cloud ML for AEC computational design.
ml-architecture-diagram
Create accurate, publication-ready, editable machine-learning and deep-learning architecture diagrams from model code, configuration files, model summaries, graph exports, or written specifications. Use for neural-network figures, architecture schematics, model block diagrams, training/inference diagrams, paper…