sglang

Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, or when you need 5× faster inference than vLLM with prefix sharing. Powers 300,000+ GPUs at xAI, AMD, NVIDIA, and LinkedIn.

Quality

80%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Critical

Do not install without reviewing

Quality

Content

77%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 content is highly actionable with executable examples and clean progressive disclosure to real reference files. Its main weakness is conciseness, where explanatory prose and capability lists pad the body beyond what Claude needs.

Suggestions

Trim or move the RadixAttention 'How it works' explanation and the 'Supported models'/'Hardware support' lists into references, leaving only the decision-relevant signal in SKILL.md.

Condense the benchmark prose into the throughput table alone, dropping the redundant narrative restatements of the same numbers.

Add a quick validation step (e.g., curl the /v1/models endpoint or check server startup logs) before running client examples so first-run failures are caught early.

Dimension	Reasoning	Score
Conciseness	The body is largely code-driven and efficient, but the RadixAttention explanation, performance benchmark prose, and 'Supported models'/'Hardware support' lists restate concepts Claude already knows and could be trimmed or moved to references.	3 / 5
Actionability	Copy-paste ready installation commands, server launch flags, and complete executable Python snippets covering JSON, regex, grammar, agents, multi-turn, multi-modal, and batching across the common cases.	5 / 5
Workflow Clarity	Quick start flows logically from install → launch → inference → structured output, and benchmark tables clarify decision-making; however there are no explicit validation/checkpoint steps for destructive or batch operations (e.g., verifying server health before running examples).	4 / 5
Progressive Disclosure	Clear overview structure with well-signaled one-level-deep references to real files (structured-generation.md, radix-attention.md, deployment.md) that exist in the bundle; the bulk detail is appropriately split out of SKILL.md.	5 / 5
	Total	17 / 20 Passed

Description

83%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 strong: it answers both 'what' and 'when' with concrete triggers and positions the skill against vLLM. It is specific and distinct, with only minor room to add more natural trigger synonyms.

Dimension	Reasoning	Score
Specificity	Names the domain (structured generation and serving) and several concrete capabilities (JSON/regex outputs, constrained decoding, agentic workflows with tool calls, prefix sharing); coverage is good but not exhaustive of all serving features.	4 / 5
Completeness	Clearly states what it does ('Fast structured generation and serving for LLMs with RadixAttention prefix caching') and explicit 'when' guidance ('Use for JSON/regex outputs, constrained decoding, agentic workflows with tool calls, or when you need 5× faster inference than vLLM with prefix sharing').	5 / 5
Trigger Term Quality	Includes natural user-facing terms like 'JSON/regex outputs', 'constrained decoding', 'agentic workflows with tool calls', and 'prefix sharing'; missing some synonyms and version-specific trigger phrasings.	4 / 5
Distinctiveness Conflict Risk	The SGLang niche and explicit vLLM comparison make it distinct from general serving skills, though the broad 'fast inference' framing carries minor overlap risk with vLLM-oriented skills.	4 / 5
	Total	17 / 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: OpenLAIR/dr-claw
Path: skills/inference-serving/sglang/SKILL.md
Commit: 4b35756

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.