serving-llms-vllm

vLLM: high-throughput LLM serving, OpenAI API, quantization.

Quality

53%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./skills/mlops/inference/vllm/SKILL.md

Quality

Content

57%

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The content is a well-structured, largely executable skill with excellent progressive disclosure into real reference files. Its main weaknesses are stub comment-only steps and missing validation feedback loops in the batch workflow.

Suggestions

Replace stub comment steps with concrete executable code — provide an actual locust test_load.py and a real AWQ model-search/quantization-verification snippet.

Add an explicit validation checkpoint to the batch-inference workflow (e.g. verify output count equals input count, log failures, retry failed prompts).

Trim the copy-tracker checklist boxes and the "24x higher throughput" marketing claim to tighten token efficiency.

Dimension	Reasoning	Score
Conciseness	Mostly efficient with executable code, but the repeated copy-tracker checklists, the marketing claim "24x higher throughput", and explanatory prose add padding that could be tightened.	2 / 3
Actionability	The bulk is concrete and copy-paste ready, but several steps are stub comments rather than executable code ("Create test_load.py with sample requests", "# Search for AWQ models", and the quantization accuracy verification), leaving key details missing.	2 / 3
Workflow Clarity	Workflows use clear step sequences and checklists, and the deployment workflow has real validation thresholds, but the batch-inference workflow lacks a validate→fix→retry feedback loop, which caps this dimension at 2 per the guidelines.	2 / 3
Progressive Disclosure	The body is an overview that points to four well-signaled one-level-deep references (server-deployment.md, optimization.md, quantization.md, troubleshooting.md), all of which exist in the bundle, matching the score-3 example structure.	3 / 3
	Total	9 / 12 Passed

Description

50%

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 terse, recognizable keyword string for a specific tool, but it lacks an explicit "Use when…" trigger clause and reads as feature tags rather than concrete actions. It is adequate but not exemplary.

Suggestions

Add an explicit trigger clause, e.g. "Use when deploying production LLM APIs, serving OpenAI-compatible endpoints, or fitting large models into limited GPU memory with quantization."

Convert noun-phrase tags into concrete actions ("Serve LLMs at high throughput", "Expose OpenAI-compatible APIs", "Apply GPTQ/AWQ/FP8 quantization").

Drop the marketing claim "high-throughput" in favor of natural user terms like "inference", "deploy", and "serve".

Dimension	Reasoning	Score
Specificity	The description names the domain and several capabilities ("high-throughput LLM serving, OpenAI API, quantization") but presents them as noun-phrase keyword tags rather than concrete verbs like the score-3 example, so it is not comprehensive.	2 / 3
Completeness	It states what the skill does but contains no "Use when…" clause or equivalent explicit trigger guidance, which per the guidelines caps completeness at 2.	2 / 3
Trigger Term Quality	Terms like "vLLM", "LLM serving", "OpenAI API", and "quantization" are recognizable, but "high-throughput" is a marketing buzzword and common variations a user might say ("inference", "deploy model", "serve model") are missing.	2 / 3
Distinctiveness Conflict Risk	The named tool "vLLM" gives it a clear niche, but the generic "LLM serving" and "OpenAI API" phrasing could still overlap with other serving skills (TGI, TensorRT-LLM) and distinctive trigger language is absent.	2 / 3
	Total	8 / 12 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: skills/mlops/inference/vllm/SKILL.md
Commit: d0d116b

Reviewed: 1 day 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.