tensorrt-llm

High-throughput LLM inference on NVIDIA GPUs.

Quality

53%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Fix and improve this skill with Tessl

tessl review fix ./optional-skills/mlops/tensorrt-llm/SKILL.md

Quality

Content

61%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 a well-structured, largely executable quick-start with genuine one-level-deep reference files, but it lacks validation checkpoints in its serving/deployment workflows and carries time-sensitive version pins and a duplicate benchmark section that hurt conciseness.

Suggestions

Add validation checkpoints to serving and deployment workflows, e.g. verify GPU visibility with nvidia-smi, confirm the server health endpoint before sending client requests, and sanity-check batch outputs.

Move time-sensitive version pins (CUDA 13.2.1, Python 3.10-3.12, 'current stable GA') into a dedicated versioned section or a reference doc so the main body does not decay.

Consolidate the 'Performance benchmarks' section with the throughput figures already cited under 'When to use TensorRT-LLM' to remove duplication.

Dimension	Reasoning	Score
Conciseness	The body is mostly lean bullets and code without explaining basics Claude already knows, but it carries time-sensitive version pins ('CUDA 13.2.1', 'Python 3.10-3.12', 'current stable GA') outside any deprecated section and duplicates throughput figures in the 'Performance benchmarks' section.	3 / 5
Actionability	It provides mostly executable, copy-paste-ready guidance across install, inference, serving, quantization, multi-GPU, and batch patterns, with only minor justified placeholders such as 'x.y.z' in the Docker tag.	4 / 5
Workflow Clarity	A rough sequence exists (install → serve → query), but there are no validation checkpoints, and batch inference plus multi-GPU deployment are batch/risky operations whose missing verification steps cap workflow clarity at 3.	3 / 5
Progressive Disclosure	Structure is good with clearly signaled, one-level-deep references to real files (references/optimization.md, multi-gpu.md, serving.md), though some reference-like content (benchmarks, supported-models list, common-pattern code) remains inlined.	4 / 5
	Total	14 / 20 Passed

Description

45%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 a concise, domain-specific noun phrase that clearly identifies the technology niche but lacks concrete action verbs and any 'Use when…' trigger guidance. It is functional but under-specified for skill triggering.

Suggestions

Add a 'Use when…' clause stating when Claude should invoke this skill, e.g. 'Use when deploying or serving LLMs on NVIDIA GPUs for maximum throughput or low latency.'

Replace the single noun phrase with 2-3 concrete actions such as 'serve, optimize, and scale LLM inference on NVIDIA GPUs'.

Include common trigger terms users actually say — 'TensorRT-LLM', 'LLM serving', 'GPU deployment', and specific GPUs (A100/H100) — to sharpen distinctiveness from vLLM and llama.cpp.

Dimension	Reasoning	Score
Specificity	The phrase 'High-throughput LLM inference on NVIDIA GPUs' names the domain and platform but offers only minimal, generic action (the noun 'inference') with no concrete verbs like deploy, serve, optimize, or scale.	2 / 5
Completeness	It clearly states what the skill does but provides no 'Use when…' clause or equivalent explicit trigger guidance, so completeness is capped at 3 per the rubric guideline.	3 / 5
Trigger Term Quality	It includes some natural terms a user might say ('LLM inference', 'NVIDIA GPUs', 'high-throughput') but misses common variations and synonyms such as 'serving', 'deployment', 'TensorRT', or specific GPU names.	3 / 5
Distinctiveness Conflict Risk	The NVIDIA + high-throughput framing is somewhat specific, but LLM inference is a crowded space (vLLM, TGI, llama.cpp) with no disambiguating trigger guidance, so overlap risk remains.	3 / 5
	Total	11 / 20 Passed

Validation

81%

Warnings & errors only

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

Validation — 13 / 16 Passed

Validation for skill structure

Criteria	Description	Result
metadata_version	'metadata.version' is missing	Warning
metadata_field	'metadata' should map string keys to string values	Warning
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	13 / 16 Passed

Repository: NousResearch/hermes-agent
Path: optional-skills/mlops/tensorrt-llm/SKILL.md
Commit: 6680afb

Reviewed: about 1 hour ago

Table of Contents

Discovery Implementation Validation

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.