CtrlK
BlogDocsLog inGet started
Tessl Logo

fla-nvidia-performance

Guidelines for NVIDIA GPU kernel / Triton / Gluon / TileLang / CUDA backend performance work in the FLA repo. Covers profiling workflow, hardware baselines, and MR-ready performance evidence requirements. Uses an installed ncu-report-skill when a task needs detailed Nsight Compute collection and diagnosis.

67

Quality

80%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Passed

No findings from the security scan

SKILL.md
Quality
Evals
Security

Quality

Content

86%Weight 40%Scale 1-5

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

The body is lean, highly actionable, and well-structured for a self-contained single-file skill, with only minor room to make MR-evidence validation checkpoints explicit.

Suggestions

Make the MR-evidence workflow's validation explicit by adding a checkpoint after the before/after benchmark (e.g. 'Verify throughput did not regress by >X% before proceeding to NCU profiling') to push workflow_clarity toward 5.

Tighten the NCU-availability preamble ('This repo intentionally does not vendor...') into a single line so the body leans closer to a 5 on conciseness.

DimensionReasoningScore

Conciseness

The body is efficient and assumes Claude's competence — no padded explanations of what Triton/CUDA are — with only minor over-explanation in the NCU-availability hedge ('This repo intentionally does not vendor ncu-report-skill...').

4 / 5

Actionability

Provides copy-paste-ready, fully executable commands (e.g. 'ncu --set full --section PmSampling... -o profile/<run_name>/reports/full_<tag> python -m benchmarks.ops.run --op chunk_kda --modes fwd') plus a concrete profile-artifact layout covering the common cases.

5 / 5

Workflow Clarity

The 'Before opening an MR' section gives a clear numbered 4-step sequence with workload-coverage and conclusion/risk steps, but validation checkpoints are implicit rather than explicit validate→fix→retry loops, so it sits clearly above 3 but below 5.

4 / 5

Progressive Disclosure

A single well-organized file under ~110 lines with clear section headers and no need for external references; per the rubric's simple-skill note, well-organized sections with no required external references score 5.

5 / 5

Total

18

/

20

Passed

Description

75%Weight 40%Scale 1-5

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

The description is specific and well-targeted to a distinct FLA/NVIDIA-performance niche, but it omits an explicit 'Use when...' trigger clause, leaving the 'when' only weakly implied and capping completeness at 3.

Suggestions

Add an explicit 'Use when...' clause naming concrete triggers, e.g. 'Use when working on Triton, Gluon, TileLANG, or CUDA kernel optimization in fla/ops/, or when preparing MR performance evidence.'

Include natural synonyms and file extensions a user might say (e.g. 'kernel tuning', 'GPU throughput', 'latency') to round out trigger-term coverage toward 5.

DimensionReasoningScore

Specificity

Lists multiple concrete actions ('profiling workflow, hardware baselines, and MR-ready performance evidence requirements', 'detailed Nsight Compute collection and diagnosis') covering the domain comprehensively, matching the comprehensive-coverage anchor.

5 / 5

Completeness

The 'what' is clear but the description field itself has no explicit 'Use when...' trigger clause — the 'when' is only weakly implied by 'Guidelines for... performance work', which caps completeness at 3 per the rubric guideline.

3 / 5

Trigger Term Quality

Strong natural terms a user would say ('NVIDIA GPU kernel', 'Triton', 'CUDA', 'performance', 'profiling', 'Nsight Compute'), but a few common synonyms/variations are missing, so it is good rather than comprehensive.

4 / 5

Distinctiveness Conflict Risk

The pairing of 'FLA repo' with 'Triton, Gluon, TileLang, CUDA' and 'ncu-report-skill' defines a clear niche with distinct triggers and minimal conflict risk with other skills.

5 / 5

Total

17

/

20

Passed

Validation

100%

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

Validation16 / 16 Passed

Validation for skill structure

No warnings or errors.

Repository
fla-org/flash-linear-attention
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.