skilld
All skills
nvidia avatar

/tilegym-cutile-python

@129a108
by NVIDIA Corporationnvidia/skills3.5k stars
424

Expert cuTile programming assistant. Write high-performance GPU kernels using cuTile's tile-based programming model with proper validation and optimization. Supports deep agent orchestration for complex multi-kernel tasks.

Ask your Agent

Use this Skill: https://skilld.dev/gh/nvidia/skills/tilegym-cutile-python

This session only. Nothing lands on disk.

BENCHMARK.md

≈1.8k tokens on demand. Your agent reads this file only when SKILL.md points to it.

Evaluation Report

Evaluation of the tilegym-cutile-python skill before publication through NVSkills-Eval.

This benchmark summarizes 3-Tier Evaluation from NVSkills-Eval results for the skill. The goal is to document whether the skill is safe, discoverable, effective, and useful for agents before it is published for broader workflow use.

Evaluation Summary

  • Skill: tilegym-cutile-python
  • Evaluation date: 2026-06-08
  • NVSkills-Eval profile: external
  • Environment: astra-sandbox
  • Dataset: 3 evaluation tasks
  • Attempts per task: 2
  • Pass threshold: 50%
  • Overall verdict: FAIL The skill should be reviewed before NVSkills-Eval publication. Skill owners should address the applicable findings below and rerun NVSkills-Eval to refresh this benchmark.

Agents Used

  • claude-code
  • codex

Metrics Used

Reported benchmark dimensions:

  • Security: checks whether skill-assisted execution avoids unsafe behavior such as secret leakage, destructive commands, or unauthorized access.
  • Correctness: checks whether the agent follows the expected workflow and produces the correct final output.
  • Discoverability: checks whether the agent loads the skill when relevant and avoids using it when irrelevant.
  • Effectiveness: checks whether the agent performs measurably better with the skill than without it.
  • Efficiency: checks whether the agent uses fewer tokens and avoids redundant work.

Underlying evaluation signals used in this run:

  • security (Security): checks for unsafe operations, secret leakage, and unauthorized access.
  • skill_execution (Skill Execution): verifies that the agent loaded the expected skill and workflow.
  • skill_efficiency (Efficiency): checks routing quality, decoy avoidance, and redundant tool usage.
  • accuracy (Accuracy): grades final-answer correctness against the reference answer.
  • goal_accuracy (Goal Accuracy): checks whether the overall user task completed successfully.
  • behavior_check (Behavior Check): verifies expected behavior steps, including safety expectations.
  • token_efficiency (Token Efficiency): compares token usage with and without the skill.

Test Tasks

The benchmark dataset contained 3 evaluation tasks:

  • Positive tasks: 2 tasks where the skill was expected to activate.
  • Negative tasks: 1 tasks where no skill was expected.
  • Unlabeled tasks: 0 tasks where positive/negative intent could not be inferred.

Task composition is derived from the evaluation dataset when possible. Entries with expected_skill set are treated as positive skill-activation cases, while entries with expected_skill: null are treated as negative activation cases.

Results

Dimension Num claude-code codex
Security 6 100% (+0%) 100% (+0%)
Correctness 6 96% (+15%) 95% (+6%)
Discoverability 6 92% (+42%) 81% (+14%)
Effectiveness 6 83% (+1%) 86% (+12%)
Efficiency 6 78% (+34%) 70% (+12%)

Score values show skill-assisted performance. Values in parentheses show uplift versus the no-skill baseline when baseline data is available.

Tier 1: Static Validation Summary

Tier 1 validation reported findings. NVSkills-Eval ran 9 checks and found 9 total findings.

Top findings:

  • LOW QUALITY/quality_discoverability: Description very long (222 chars, recommend 50-150) (skills/tilegym-cutile-python/SKILL.md)
  • LOW QUALITY/quality_discoverability: No '## Purpose' section (skills/tilegym-cutile-python/SKILL.md)
  • LOW QUALITY/quality_reliability: No prerequisites/requirements documented (skills/tilegym-cutile-python/SKILL.md)
  • LOW QUALITY/quality_reliability: No limitations documented (skills/tilegym-cutile-python/SKILL.md)
  • LOW QUALITY/quality_efficiency: Uses complex/corporate language (skills/tilegym-cutile-python/SKILL.md)

Tier 2: Deduplication Summary

Tier 2 validation reported findings. NVSkills-Eval ran 2 checks and found 14 total findings.

Top findings:

  • HIGH DUPLICATE/duplicate: Duplicate content found across examples/convolution/conv2d_with_bias_dilation_groups.py and examples/convolution/conv3d_with_bias_dilation_groups.py and examples/convolution/conv_transpose_2d.py and examples/convolution/conv_transpose_3d.py and examples/matmul/matmul_4d_tensors.py and examples/matmul/split_k_gemm.py: "_adjust_group_size()" in examples/convolution/conv2d_with_bias_dilation_groups.py (lines 39-44) vs "_adjust_group_size()" in examples/convolution/conv3d_with_bias_dilation_groups.py (lines 42-47) vs "_adjust_group_size()" in examples/convolution/conv_transpose_2d.py (lines 48-53) vs "_adjust_group_size()" in examples/convolution/conv_transpose_3d.py (lines 49-54) vs "_adjust_group_size()" in examples/matmul/matmul_4d_tensors.py (lines 36-41) vs "_adjust_group_size()" in examples/matmul/split_k_gemm.py (lines 21-26) (examples/convolution/conv2d_with_bias_dilation_groups.py:39)
  • HIGH DUPLICATE/duplicate: Duplicate content found across examples/convolution/conv2d_with_bias_dilation_groups.py and examples/convolution/conv3d_with_bias_dilation_groups.py and examples/convolution/conv_transpose_2d.py and examples/convolution/conv_transpose_3d.py: "_select_tile_config_2d()" in examples/convolution/conv2d_with_bias_dilation_groups.py (lines 47-87) vs "_select_tile_config_3d()" in examples/convolution/conv3d_with_bias_dilation_groups.py (lines 50-88) vs "_select_tile_config_trans2d()" in examples/convolution/conv_transpose_2d.py (lines 56-94) vs "_select_tile_config_trans3d()" in examples/convolution/conv_transpose_3d.py (lines 57-95) (examples/convolution/conv2d_with_bias_dilation_groups.py:47)
  • HIGH DUPLICATE/duplicate: Duplicate content found across examples/matmul/matmul_4d_tensors.py and examples/matmul/matrix_vector_multiplication.py and examples/matmul/split_k_gemm.py: "reference_matmul()" in examples/matmul/matmul_4d_tensors.py (lines 101-103) vs "reference_matmul()" in examples/matmul/matrix_vector_multiplication.py (lines 54-56) vs "reference_gemm()" in examples/matmul/split_k_gemm.py (lines 129-131) (examples/matmul/matmul_4d_tensors.py:101)
  • HIGH DUPLICATE/duplicate: Duplicate content found across examples/convolution/conv2d_with_bias_dilation_groups.py and examples/convolution/conv3d_with_bias_dilation_groups.py and examples/convolution/conv_transpose_2d.py and examples/convolution/conv_transpose_3d.py and orchestration/composer_agent.md: "pytorch_reference()" in examples/convolution/conv2d_with_bias_dilation_groups.py (lines 305-307) vs "pytorch_reference()" in examples/convolution/conv3d_with_bias_dilation_groups.py (lines 329-331) vs "pytorch_reference()" in examples/convolution/conv_transpose_2d.py (lines 305-308) vs "pytorch_reference()" in examples/convolution/conv_transpose_3d.py (lines 336-338) vs "# ============================================================" in orchestration/composer_agent.md (lines 100-105) (examples/convolution/conv2d_with_bias_dilation_groups.py:305)
  • HIGH DUPLICATE/duplicate: Duplicate content found within orchestration/composer_agent.md: "# ============================================================" in orchestration/composer_agent.md (lines 64-71) vs "# ============================================================" in orchestration/composer_agent.md (lines 74-81) (orchestration/composer_agent.md:64)

Source: SKILL.md on GitHub

Third-party checks

1 warning3mo3 checks · Risk SAFE
  • Gen Agent Trust Hub3mo

    The skill is a cuTile programming assistant provided by NVIDIA. It guides agents through writing, optimizing, and validating GPU kernels. Security analysis found that it clones official repositories from GitHub (NVIDIA TileGym and PyTorch) and executes generated scripts as part of its validation workflow. Both behaviors are standard for its intended use case. Static analysis warnings for eval() calls were confirmed as false positives, as they refer to the PyTorch .eval() method rather than the Python eval() function.

  • Socket3mo

    No alerts

  • Snyk3mo

    Risk: MEDIUM · 2 issues

Provenance

Signed by skilld at 129a108. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 4 months ago

Capability

version
1.3.0
Other metadata
metadata
{
  "author": "TileGym Team <TileGym@nvidia.com>",
  "tags": [
    "cutile",
    "gpu-kernels",
    "cuda"
  ]
}

README badge

README badge for nvidia/skills/tilegym-cutile-python