BB perf-torch-cuda-graphs
Apply CUDA Graphs to PyTorch workloads — API selection (torch.compile, PyTorch make_graphed_callables, TE make_graphed_callables, MCore CudaGraphManager, FullCudaGraphWrapper, manual torch.cuda.graph), code compatibility, capture workflows, dynamic pattern handling, and troubleshooting. Triggers: CUDA graph, torch.cuda.graph, make_graphed_callables, reduce-overhead, graph capture, graph replay, kernel launch overhead, CudaGraphManager, FullCudaGraphWrapper, full-iteration graph, stream capture.
This skill covers CUDA Graphs in PyTorch — choosing between torch.compile, make_graphed_callables (multiple flavors), CudaGraphManager, and manual torch.cuda.graph. Claims to guide you through code compatibility, graph capture mechanics, dynamic pattern handling, and troubleshooting.
Seven modules and one script inside, code quality at 83, execution process at 71. No critical issues, safety checks pass. Lint found nothing. In practice, you get API navigation — which option fits your workload, how to handle dynamic shapes without crashes, where the traps hide.
Install if you're tuning LLM inference and willing to dig into CUDA details. Otherwise, overkill.
Apply CUDA Graphs to PyTorch workloads — API selection (torch.compile, PyTorch makegraphedcallables, TE makegraphedcallables, MCore CudaGraphManager…
As a process B 71/100 · Nearly there — weak spots: running it twice, progress reporting
How to improve
- The SKILL.md body is over 5,000 tokens: move reference detail into references/ and load it when needed.
- Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
- A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.
Guard findings · 0
✓ No critical or high findings
Files scanned: 7. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- warning
body-longSKILL.md body ≈ 5859 tokens (recommended < 5000); move details to references/
Process rating: all ten parameters 71/100
- 0Progress reporting. Says nothing while it works
- 30Running it twice. 5 mutating operations with no state check
- 50When it triggers. No condition that starts the skill
- 60Tools and files. Uses tools (python) that frontmatter does not declare
- 60Result and completion. Output format stated, no completion criterion
- 70Inputs and preconditions. Inputs and preconditions are listed
- 70Execution cost. Instruction body is 5859 tokens
- 100Steps. 120 steps
- 100Failures and branches. 2 branches, has a failure section
- 100Consistency. Name and required fields are in place
Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.
Quality signals
- +5Description has no quoted example phrases that should trigger the skill
- -31 of 1 scripts are never mentioned in SKILL.md
- +2Single-language instructions
- +4Description says when NOT to use the skill
- +3Description length 499: enough signal without eating the budget
- +4Structure: 24 headings
- +3Step-by-step instructions: 120 items
- +3Output format is stated explicitly
- +4Has examples (25 code blocks)
- +4Reference files are cited in the instructions (5 of 5)
- +1License stated
Quality base 70; lint remarks subtract, signals add up to 100. Result: 83.