BC trtllm-moe-develop
Review, design, and refactor TensorRT-LLM PyTorch MoE code for architecture fit, clean code, maintainability, and testability. Always use for any modification, review, refactor, or design planning that touches MoE modules, including tensorrt_llm/_torch/moe/fused_moe, ConfigurableMoE, MoE backends, MoEScheduler/moe_scheduler.py, forward execution/chunking, communication strategies, EPLB, quantization/weight handling, routing, factories, MoE docs, or MoE tests. Also use when the user asks whether a MoE design follows the current architecture or whether a MoE refactor is reasonable.
A skill for reviewing and refactoring MoE code in TensorRT-LLM. Promises to analyze architecture, code cleanliness, and testability of expert-mixing modules—from fused_moe through scheduler and quantization.
Two files with 8285 tokens of description. Grade B: quality 79, but process only 51—the instruction is complete but lacks run examples or sandbox testing. One medium-level finding, no critical issues. The skill is narrowly scoped: activates only if your request mentions MoE modules, scheduler, or specific code paths.
Install if you're writing MoE logic in TensorRT-LLM or reviewing code against their architecture. For general refactoring or other components, it stays silent.
Review, design, and refactor TensorRT-LLM PyTorch MoE code for architecture fit, clean code, maintainability, and testability.
As a process C 51/100 · Has gaps — weak spots: result and completion, inputs and preconditions, execution cost
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 · 1
✓ No critical or high findings
Medium and low: 1
-
low Secrets in code
secret-high-entropy-tokenreferences/moe-canonical-code-examples.md:171High-entropy token-like string (may be an id, hash or a credential) (detector / deny-list definition; test fixture / example file)- `Deep…hod`: FP8 block-scale loading and scale
detectorfixture
Files scanned: 2. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
- warning
body-longSKILL.md body ≈ 8285 tokens (recommended < 5000); move details to references/ - note
edit-residuethe text marks something as outdated (lines 85, 91, 92, 210, 252, 568): check that old rules are not kept next to new ones — the full check reads the text for contradictions
Process rating: all ten parameters 51/100
- 0Result and completion. Does not say what the result is
- 0Inputs and preconditions. Does not say what the process needs to start
- 0Progress reporting. Says nothing while it works
- 40Execution cost. Instruction body is 8285 tokens: crowds the task out of the window
- 50Failures and branches. 0 branches, has a failure section
- 60Tools and files. Uses tools (web, node) that frontmatter does not declare
- 70When it triggers. States when to use, but not when not to
- 85Steps. 180 steps, 1 vague phrases
- 100Consistency. Name and required fields are in place
- 100Running it twice. Mutating operations check current state
- 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)
- +3Output format is not stated: the model decides each time
- +2Single-language instructions
- +3Description length 586: enough signal without eating the budget
- +4Structure: 20 headings
- +3Step-by-step instructions: 180 items
- +4Has examples (2 code blocks)
- +4Reference files are cited in the instructions (1 of 1)
- +1License stated
Quality base 70; lint remarks subtract, signals add up to 100. Result: 79.