SKILLEMALL.ai
GitHub ↗ RU Sign in

CB llm-torch-profiler-analysis

Unified LLM torch-profiler triage skill for `sglang`, `vllm`, `TensorRT-LLM`, and `TokenSpeed`. Use it to inspect an existing `trace.json(.gz)` or profile directory, or to drive live profiling against a running server when supported and return one three-table report with kernel, overlap-opportunity, and fuse-pattern tables.

sgl-project/sglang Agent Skills author: sgl-project Apache-2.0 18 files · 12 scripts body ≈ 6 272 tokens Open the sourcegithub.com↗ analyzed 5 d ago

Unified LLM torch-profiler triage skill for sglang, vllm, TensorRT-LLM, and TokenSpeed.

As a process B 74/100 · Nearly there — weak spots: progress reporting

AnalyzerAI and agentsSoftware developmenttype and topics are labelled automatically from the skill text
Runs in: Claude / Claude Code ChatGPT / Codex Gemini CLI Cursor GitHub Copilot / VS Code Qwen Code Kimi Code CLI Z.ai (GLM) DeepSeek (Deep Code) Mistral Vibe LangChain (deepagents) OpenClaw Goose OpenCode
JSON
Technical rating
C
87/100
safety, quality, tests
Safety 60%
95
Quality 40%
76
Run on models
none yet
Process rating
B
74/100
Nearly there
Progress reporting w 2
0
When it triggers w 12
50
Tools and files w 18
60
the three weakest of ten parameters · all ten

How to improve

  1. The SKILL.md body is over 5,000 tokens: move reference detail into references/ and load it when needed.
For the model run — optional
  • 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 · 5

✓ No critical or high findings

Medium and low: 5
  • low Secrets in code secret-high-entropy-token references/vllm-torch-compile-fusions.md:30
    High-entropy token-like string (may be an id, hash or a credential) (detector / deny-list definition; documentation table row)
    | `fuse_act_quant` + AITER | `Rocm…ass` | AITER SiLU-and-mul followed by FP8 group quant |
    detectortable
  • low Secrets in code secret-high-entropy-token references/vllm-torch-compile-fusions.md:43
    High-entropy token-like string (may be an id, hash or a credential) (detector / deny-list definition; documentation table row)
    | `fusion/allreduce_rms_fusion.py` | `AllReduceRMSNormPattern`, `AllReduceFusedAddRMSNormPattern`, `AllR…ern`, `AllR…ern`, `AllRe
    detectortable
  • low Secrets in code secret-high-entropy-token references/vllm-torch-compile-fusions.md:45
    High-entropy token-like string (may be an id, hash or a credential) (detector / deny-list definition; documentation table row)
    | `fusion/rocm_aiter_fusion.py` | `AiterRMSNormDynamicQuantPattern`, `AiterFusedAddRMSNormDynamicQuantPattern`, `Aite…ern`, `Aite…ern` | AITER RMSNorm/f
    detectortable
  • low Secrets in code secret-high-entropy-token references/vllm-torch-compile-fusions.md:46
    High-entropy token-like string (may be an id, hash or a credential) (documentation table row)
    | `fusion/act_quant_fusion.py` | `Silu…ern`, `Silu…ern`, `SiluMulBlockQuantPattern` | SiLU-and-mul activation output immediately quantized | fused activation-plus-q
    table
  • low Secrets in code secret-high-entropy-token references/vllm-torch-compile-fusions.md:47
    High-entropy token-like string (may be an id, hash or a credential) (documentation table row)
    | `fusion/rocm_aiter_fusion.py` | `Aite…ern` | AITER SiLU-and-mul followed by FP8 group quant | AITER `act_…ant` |
    table

Files scanned: 18. Evidence is masked. Grey chips explain why severity was lowered.

Against the Agent Skills spec

  • warning body-long SKILL.md body ≈ 6272 tokens (recommended < 5000); move details to references/
  • note edit-residue the text marks something as outdated (lines 213, 465, 506, 508): check that old rules are not kept next to new ones — the full check reads the text for contradictions

Process rating: all ten parameters 74/100

  • 0Progress reporting. Says nothing while it works
  • 50When it triggers. No condition that starts the skill
  • 60Tools and files. Uses tools (bash, 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 6272 tokens
  • 100Steps. 118 steps
  • 100Failures and branches. 1 branches, has a failure section
  • 100Consistency. Name and required fields are in place
  • 100Running it twice. Mutating operations check current state
  • low 12 top-level sections: this looks like several domains in one skill
  • low The response is described with custom markup (4 tags): a typed call is more reliable

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
  • +4Description does not say when NOT to use the skill (false activations)
  • -2localhost URLs: will not work for another user
  • -36 of 12 scripts are never mentioned in SKILL.md
  • +1No license
  • +2Single-language instructions
  • +3Description length 325: enough signal without eating the budget
  • +4Structure: 24 headings
  • +3Step-by-step instructions: 118 items
  • +3Output format is stated explicitly
  • +4Has examples (15 code blocks)
  • +4Reference files are cited in the instructions (5 of 5)

Quality base 70; lint remarks subtract, signals add up to 100. Result: 76.