optimize-for-gpu

GPU-accelerate Python code using CuPy, Numba CUDA, Warp, cuDF, cuML, cuGraph, KvikIO, cuCIM, cuxfilter, cuVS, cuSpatial, and RAFT. Use whenever the user mentions GPU/CUDA/NVIDIA acceleration, or wants to speed up NumPy, pandas, scikit-learn, scikit-image, NetworkX, GeoPandas, or Faiss workloads. Covers physics simulation, differentiable rendering, mesh ray casting, particle systems (DEM/SPH/fluids), vector/similarity search, GPUDirect Storage file IO, interactive dashboards, geospatial analysis, medical imaging, and sparse eigensolvers. Also use when you see CPU-bound Python code (loops, large arrays, ML pipelines, graph analytics, image processing) that would benefit from GPU acceleration, even if not explicitly requested.

Quality

92%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Quality

Content

85%

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A well-structured, highly actionable skill body with strong progressive disclosure and a clear optimization workflow. Its main weakness is verbosity from redundant trigger listings and the exhaustive library-combination catalog.

Suggestions

Collapse the 'When This Skill Applies' bullet list — it duplicates the frontmatter description and the per-library 'Use X when...' sections; a short pointer is enough.

Trim the 25-entry 'Combining Libraries' list to the handful of combinations that are non-obvious, since most interoperate trivially via the CUDA Array Interface already stated once.

Move the long install/availability-check block to a reference file (e.g. references/installation.md) and keep only the `uv add` essentials plus the CUDA-12/13 note inline.

Dimension	Reasoning	Score
Conciseness	Mostly actionable and efficient, but the ~700-line body carries notable redundancy — the 27-item 'When This Skill Applies' list largely restates the description, and the 25-pair 'Combining Libraries' list could be trimmed — so not every token earns its place.	2 / 3
Actionability	Provides copy-paste-ready before/after code transformations for every library, exact `uv add` install commands, and concrete profiling snippets — fully executable guidance rather than pseudocode.	3 / 3
Workflow Clarity	The 'Optimization Workflow' is a clear five-step sequence with explicit validation checkpoints (profile first, profile the GPU version) and an error-recovery feedback loop ('Test numerical correctness against CPU results').	3 / 3
Progressive Disclosure	Overview points to 12 one-level-deep reference files, each signaled inline with 'Read: references/x.md' and summarized in a final 'Reference Files' table; all referenced files exist in ./references/.	3 / 3
	Total	11 / 12 Passed

Description

100%

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A dense, highly specific description that clearly states capabilities, names the exact libraries and workload domains, and gives explicit trigger guidance in third person. It is long but every clause carries concrete information rather than fluff.

Dimension	Reasoning	Score
Specificity	Lists many concrete actions and tools — 'GPU-accelerate Python code using CuPy, Numba CUDA, Warp, cuDF...' and enumerates specific workload types (physics simulation, mesh ray casting, vector search) rather than vague language.	3 / 3
Completeness	Explicitly answers both what (GPU-accelerate Python with the listed libraries across the listed workloads) and when ('Use whenever the user mentions GPU/CUDA/NVIDIA acceleration...' plus the CPU-bound catch-all trigger).	3 / 3
Trigger Term Quality	Covers natural terms users actually say — 'GPU/CUDA/NVIDIA acceleration', 'speed up NumPy, pandas, scikit-learn, scikit-image, NetworkX, GeoPandas, or Faiss workloads' — with good breadth of library and workload keywords.	3 / 3
Distinctiveness Conflict Risk	The GPU/CUDA/NVIDIA niche with these specific RAPIDS libraries is a clear, distinct scope unlikely to trigger for unrelated skills.	3 / 3
	Total	12 / 12 Passed

Validation

87%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
skill_md_line_count	SKILL.md is long (708 lines); consider splitting into references/ and linking	Warning
relative_links	Relative link issues: 1 missing	Warning

	Total	14 / 16 Passed

Repository: K-Dense-AI/scientific-agent-skills
Path: skills/optimize-for-gpu/SKILL.md
Commit: 757b63b

Reviewed: about 11 hours ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.