CtrlK
BlogDocsLog inGet started
Tessl Logo

flox-cuda

CUDA and GPU development with Flox. Use for NVIDIA CUDA setup, GPU computing, deep learning frameworks, cuDNN, and cross-platform GPU/CPU development.

64

Quality

77%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Low

Low-risk findings worth noting

Fix and improve this skill with Tessl

tessl review fix ./flox-plugin/skills/flox-cuda/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

65%Weight 40%Scale 1-3

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

A highly actionable, code-dense skill body with strong executable examples and clear testing/troubleshooting sections. Its main weaknesses are verbatim command/config duplication that hurts conciseness, missing explicit validation feedback loops in the setup workflows, and a monolithic structure with no bundle files to progressively disclose the longer examples.

Suggestions

De-duplicate the repeated search commands and the recurring base [install] block (cuda_nvcc/cuda_cudart/gcc-unwrapped) by defining it once and referencing it, to improve conciseness.

Add explicit validation checkpoints to the PyTorch/TensorFlow setup workflows (e.g., verify torch.cuda.is_available() after install and retry with the correct --index-url if False) to introduce a validate->fix->retry feedback loop.

Move the longer complete-environment examples (PyTorch/TensorFlow/Multi-GPU/modular) into separate reference files under references/ and link to them from a concise overview, to apply progressive disclosure and reduce the monolithic body.

DimensionReasoningScore

Conciseness

The body is dense, code-forward, and assumes Claude's competence without explaining basics, but it repeats blocks verbatim — the same 'flox search cudatoolkit --all | grep flox-cuda' / 'flox search nvcc --all | grep 12_8' / 'flox show' commands appear in both Core Commands and Package Discovery, and the same basic [install] block (cuda_nvcc priority 1, cuda_cudart priority 2, gcc-unwrapped priority 5) recurs across the Basic, PyTorch, TensorFlow, and modular-base sections. This matches the score-2 anchor 'Mostly efficient but includes some unnecessary explanation or could be tightened'. It is not a 3 due to the verbatim duplication, and not a 1 because it is lean and does not explain concepts Claude already knows.

2 / 3

Actionability

Provides fully executable TOML, bash, and Python with specific package paths, explicit priorities, systems lists, and copy-paste verification (nvcc --version, nvidia-smi, a complete hello_cuda.cu compile test, torch.cuda.is_available() check), matching the score-3 anchor 'Fully executable code/commands; specific examples; copy-paste ready'. It is not a 2 because the examples are concrete and complete rather than pseudocode.

3 / 3

Workflow Clarity

The Testing CUDA Installation and Troubleshooting sections provide clear verification sequences and checks, but the Complete CUDA Environment setup workflows (PyTorch/TensorFlow hooks) lack explicit validate->fix->retry feedback loops and the rubric notes missing validation/verification in multi-step risky setups caps workflow_clarity at 2, matching the score-2 anchor 'Steps listed but validation gaps; sequence present but checkpoints missing or implicit'. It is not a 3 because the env-setup workflows lack explicit validation checkpoints and error-recovery loops, and not a 1 because testing/troubleshooting do provide clear sequenced verification.

2 / 3

Progressive Disclosure

No bundle files exist (references/, scripts/, assets/ all missing) and the body is a single ~510-line monolithic document whose modular/debugging/examples content could live in separate reference files; it does have clear section organization and the Related Skills section, but no well-signaled one-level-deep references, matching the score-2 anchor 'Some structure but could be better organized; content that should be separate is inline'. It is not a 3 because content is not split into clearly signaled separate reference files, and not a 1 because it is organized into clear sections with no nested 2+ level references.

2 / 3

Total

9

/

12

Passed

Description

90%Weight 40%Scale 1-3

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 well-crafted description that explicitly answers both what the skill does and when to use it, with strong natural trigger-term coverage and a clear, distinct niche. The only minor gap is specificity, which lists topical areas rather than multiple distinct concrete actions.

DimensionReasoningScore

Specificity

Names the domain and several concrete sub-areas ('NVIDIA CUDA setup', 'GPU computing', 'deep learning frameworks', 'cuDNN', 'cross-platform GPU/CPU development'), but these are topical areas rather than multiple distinct concrete actions, matching the score-2 anchor 'Names domain and some actions, but not comprehensive'. It is not a 3 because it lacks a list of several specific actions like the score-3 example, and not a 1 because it does name concrete capabilities and the tool.

2 / 3

Completeness

The first sentence answers 'what' ('CUDA and GPU development with Flox') and the explicit 'Use for NVIDIA CUDA setup...' clause answers 'when', matching the score-3 anchor 'Clearly answers both what AND when with explicit triggers'. It is not a 2 because the 'when' is explicit, not merely implied.

3 / 3

Trigger Term Quality

Includes natural terms users would actually say ('NVIDIA CUDA setup', 'GPU computing', 'deep learning frameworks', 'cuDNN', 'cross-platform GPU/CPU development'), giving good coverage of likely phrasings, matching the score-3 anchor 'Good coverage of natural terms users would say'. It is not a 2 because keyword coverage is broad and natural rather than partial.

3 / 3

Distinctiveness Conflict Risk

Tightly scoped to CUDA/GPU development with Flox specifically, with distinct triggers (NVIDIA, CUDA, cuDNN, GPU computing) unlikely to fire for unrelated skills, matching the score-3 anchor 'Clear niche with distinct triggers; unlikely to conflict'. It is not a 2 because the Flox + CUDA/GPU scope is distinct rather than overlapping.

3 / 3

Total

11

/

12

Passed

Validation

93%

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

Validation15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

skill_md_line_count

SKILL.md is long (515 lines); consider splitting into references/ and linking

Warning

Total

15

/

16

Passed

Repository
flox/flox-agentic
Reviewed

Table of Contents

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.