tensorrt-llm

Optimizes LLM inference with NVIDIA TensorRT for maximum throughput and lowest latency. Use for production deployment on NVIDIA GPUs (A100/H100), when you need 10-100x faster inference than PyTorch, or for serving models with quantization (FP8/INT4), in-flight batching, and multi-GPU scaling.

Quality

76%

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 ./backend/cli/skills/ml-inference/tensorrt-llm/SKILL.md

Quality

Content

65%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.

A well-structured, action-oriented quick-start with genuine progressive disclosure into three real reference files. Its main weaknesses are inline version pins that date the content, a shell-breaking comment syntax in the serving example, and absent validation checkpoints in the workflow.

Suggestions

Move fragile version pins (tensorrt_llm==1.2.0rc3, CUDA 13.0.0, TensorRT 10.13.2) into a dedicated 'Versions / requirements' or 'Deprecated' subsection so the main flow stays evergreen.

Fix the trtllm-serve snippet: inline comments after a trailing backslash break line continuation — move comments to their own lines or drop them.

Add an explicit verification step after serving startup (e.g. a healthcheck curl or 'confirm /v1/models responds before sending traffic') to give the deploy workflow a concrete checkpoint.

Dimension	Reasoning	Score
Conciseness	The body is mostly lean with code-first sections, but inline time-sensitive version pins ('tensorrt_llm==1.2.0rc3', 'CUDA 13.0.0', 'TensorRT 10.13.2') and the redundant one-line intro ('state-of-the-art performance on NVIDIA GPUs') add tokens that will date and could be tightened.	3 / 5
Actionability	Provides concrete, mostly copy-paste-ready code for installation, inference, serving, quantization, multi-GPU, and batch cases, but the trtllm-serve block places inline comments after backslash line continuations ('--tp_size 4 \ # ...') which would break the shell command as written.	4 / 5
Workflow Clarity	A reasonable install-then-infer-then-serve progression exists, but there are no explicit validation or verification checkpoints; since none of the operations are destructive the cap-at-3 rule does not bind, yet checkpoints remain implicit.	3 / 5
Progressive Disclosure	A clear overview with well-signaled, one-level-deep references to real files (references/optimization.md, references/multi-gpu.md, references/serving.md, all present), with detail appropriately split out and easy navigation via a labeled References section.	5 / 5
	Total	15 / 20 Passed

Description

87%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.

A strong, third-person description that concretely states both what the skill does and when to use it, with hardware- and feature-specific triggers that distinguish it from general inference skills. Only minor specificity and synonym-coverage gaps keep it from being maximally tight.

Dimension	Reasoning	Score
Specificity	Names concrete capabilities ('quantization (FP8/INT4), in-flight batching, and multi-GPU scaling') plus 'maximum throughput and lowest latency', but these are framed as use-cases rather than verb-led actions, leaving minor coverage gaps relative to the comprehensive anchor.	4 / 5
Completeness	Explicitly answers 'what' ('Optimizes LLM inference with NVIDIA TensorRT') and 'when' with concrete trigger phrases ('Use for production deployment... when you need 10-100x faster inference... or for serving models with quantization...').	5 / 5
Trigger Term Quality	Includes natural user-facing terms ('NVIDIA GPUs (A100/H100)', '10-100x faster inference than PyTorch', 'quantization (FP8/INT4)', 'in-flight batching', 'multi-GPU scaling'); good coverage but missing a few common synonyms or file/library name variations.	4 / 5
Distinctiveness Conflict Risk	Carves a clear NVIDIA TensorRT-LLM niche on specific hardware (A100/H100) with distinct triggers (FP8/INT4, in-flight batching) and even contrasts against PyTorch, giving minimal conflict risk.	5 / 5
	Total	18 / 20 Passed

Validation

93%

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 — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/ml-inference/tensorrt-llm/SKILL.md
Commit: fcd2e11

Reviewed: 2 days 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.