How do I install the Gpu Review skill?

Run npx skillmds@latest add allyoushawn/gpu-review in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Gpu Review safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Gpu Review?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Gpu Review free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Gpu Review?

allyoushawn (@allyoushawn) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Gpu Review by allyoushawn — Agent Skill

Name: Gpu Review
Author: allyoushawn

Gpu Review

Reviews deep learning code (.py files and Jupyter notebooks) to ensure GPU is used when available. Identifies anti-patterns like missing device detection, hardcoded CPU usage, tensors not on GPU, and performance issues. Supports PyTorch, TensorFlow/Keras, and JAX. Triggers: "review code for GPU usage", "check GPU utilization", "audit device placement", "ensure DL training runs on GPU".

allyoushawn 98c0e86 2.8 KB Updated Sep 17, 2026

File contents

GPU Code Review

Audits deep learning code for proper GPU utilization against framework-specific anti-patterns.

Inputs

Target file(s) — .py or .ipynb
Optional: severity scope (default: report Critical + Performance; runtime preflight uses Critical only)

References (load when relevant)

kb/context/colab/gpu-review-patterns.md — anti-patterns by framework (PyTorch, TensorFlow/Keras, JAX, Colab notebook-specific)

Steps

Detect framework — scan imports for torch, tensorflow / keras, or jax.
Dispatch a subagent (reading-agent for read-only audit, or general-purpose if fixes are also expected) with the file path and the framework's section of kb/context/colab/gpu-review-patterns.md as reference. The subagent checks:
- Device detection exists and uses dynamic selection (not hardcoded CPU).
- Model is moved to the detected device.
- All tensors / data are on the same device as the model before forward pass.
- DataLoader / data pipeline is GPU-optimized (pin_memory, prefetch, etc.).
- Eval / inference paths disable gradients.
- Mixed precision is considered for training code.
- Multi-GPU is considered when relevant.
- For notebooks: an early cell verifies GPU availability (!nvidia-smi).
Receive findings as a structured list with severity (Critical / Performance), the problematic snippet, and a concrete fix.

Output

## GPU Review: <filename>

**Framework:** PyTorch | TensorFlow | JAX
**Device setup found:** Yes / No
**Overall:** Good / Issues found

### Findings

#### [Critical] <title>
- **Line(s):** ...
- **Issue:** ...
- **Fix:** ...

#### [Performance] <title>
- **Line(s):** ...
- **Issue:** ...
- **Fix:** ...

### Summary
- N critical issues, M performance suggestions
- <one-line overall recommendation>

Edge cases

Code intentionally forces CPU (debugging, small models): note as informational, not Critical.
No DL framework detected: report no GPU-relevant code found.
Multi-framework files: audit each framework independently.

Notes for experiment runtime preflight

When invoked as part of an iterative experiment loop (experiment-runtime preflight), apply only Critical findings autonomously. Performance items can change training behavior (e.g. mixed precision, pin_memory) and are treated as operational downgrades per kb/context/experiments/experiment-policies.md.

---name: gpu-reviewdescription: Reviews deep learning code (.py files and Jupyter notebooks) to ensure GPU is used when available. Identifies anti-patterns like missing device detection, hardcoded CPU usage, tensors not on GPU, and performance issues. Supports PyTorch, TensorFlow/Keras, and JAX. Triggers: "review code for GPU usage", "check GPU utilization", "audit device placement", "ensure DL training runs on GPU".---# GPU Code ReviewAudits deep learning code for proper GPU utilization against framework-specific anti-patterns.## Inputs- Target file(s) — `.py` or `.ipynb`- Optional: severity scope (default: report Critical + Performance; runtime preflight uses Critical only)## References (load when relevant)- `kb/context/colab/gpu-review-patterns.md` — anti-patterns by framework (PyTorch, TensorFlow/Keras, JAX, Colab notebook-specific)## Steps1. **Detect framework** — scan imports for `torch`, `tensorflow` / `keras`, or `jax`.2. **Dispatch a subagent** (`reading-agent` for read-only audit, or `general-purpose` if fixes are also expected) with the file path and the framework's section of `kb/context/colab/gpu-review-patterns.md` as reference. The subagent checks: - Device detection exists and uses dynamic selection (not hardcoded CPU). - Model is moved to the detected device. - All tensors / data are on the same device as the model before forward pass. - DataLoader / data pipeline is GPU-optimized (`pin_memory`, prefetch, etc.). - Eval / inference paths disable gradients. - Mixed precision is considered for training code. - Multi-GPU is considered when relevant. - For notebooks: an early cell verifies GPU availability (`!nvidia-smi`).3. **Receive findings** as a structured list with severity (Critical / Performance), the problematic snippet, and a concrete fix.## Output```## GPU Review: <filename>**Framework:** PyTorch | TensorFlow | JAX**Device setup found:** Yes / No**Overall:** Good / Issues found### Findings#### [Critical] <title>- **Line(s):** ...- **Issue:** ...- **Fix:** ...#### [Performance] <title>- **Line(s):** ...- **Issue:** ...- **Fix:** ...### Summary- N critical issues, M performance suggestions- <one-line overall recommendation>```## Edge cases- Code intentionally forces CPU (debugging, small models): note as informational, not Critical.- No DL framework detected: report no GPU-relevant code found.- Multi-framework files: audit each framework independently.## Notes for experiment runtime preflightWhen invoked as part of an iterative experiment loop (`experiment-runtime` preflight), apply only **Critical** findings autonomously. Performance items can change training behavior (e.g. mixed precision, `pin_memory`) and are treated as operational downgrades per `kb/context/experiments/experiment-policies.md`.

Frequently asked questions