BBufCommunity·74 Agent Skills Included

AI-Infra-Auto-Driven-SKILLS

Agent playbooks for LLM serving benchmarks, profiling, and optimization

Automates fair LLM serving benchmarks across SGLang, vLLM, TensorRT-LLM, and TokenSpeed, then profiles and patches the slowest framework. Explains GPU memory and capacity from startup logs, triages production incidents from replayable evidence, and reviews SGLang PRs like a human maintainer. Includes 72 model PR history dossiers and architecture diagrams so agents optimize with real source evidence instead of guesses.

783|67

↗

npx skills add BBuf/AI-Infra-Auto-Driven-SKILLS --all -g -y

All Skills in This Repository (74)

Pure Emerald Level Indicators

📦 In Repo

BBuf

h100

SSH into h100_sglang to attach to the sglang_bbuf container for GPU-enabled SGLang development.

h100-sglang-diffusion

Automate diffusion smoke tests and kernel validation on the H100 SGLang remote environment.

llm-serving-auto-benchmark

Benchmark LLM serving frameworks across SGLang, vLLM, and TensorRT-LLM.

model-architecture-diagram

Return public original architecture diagram URLs for specified models.

sglang-prod-incident-triage

Convert live SGLang incidents into replayable debugging workflows.

llm-torch-profiler-analysis

Analyze LLM torch-profiler traces to produce kernel, overlap, and fuse tables.

sglang-sota-performance

Benchmark models across SGLang, vLLM, and TensorRT-LLM to identify performance gaps.

model-pr-diff-dossier

Create diff-reviewed PR cards for model optimization PR history documents.

vllm-nemotron-super-optimization

Identify diff-driven optimizations for Nemotron models in vLLM.

vllm-deepseek-v4-optimization

Audits and documents DeepSeek V4 PRs (#40760, #40811, #40806 progress in vLLM.

vllm-glm45-optimization

Audit PR diffs and generate release-ready dossiers for GLM-4.5 series in vLLM.

vllm-glm-vlm-ocr-optimization

Document production-level GLM VLM OCR optimization changes from vLLM PR histories.

Community

Advanced

Frequently Asked Questions

FAQPage Schema

How to install AI-Infra-Auto-Driven-SKILLS?▼

Run `npx skills add BBuf/AI-Infra-Auto-Driven-SKILLS --all -g -y` in your terminal to install all skills in this suite globally.

How to benchmark SGLang vs vLLM fairly?▼

The llm-serving-auto-benchmark skill runs a bounded, config-driven search for each framework under the same model, GPU, workload, and SLA, then reports the best SLA-passing deployment command for each.

Can AI agents optimize LLM serving performance automatically?▼

Yes. The SGLang and vLLM SOTA humanize loops benchmark against competitor frameworks, profile the bottleneck, patch the serving code, and revalidate until the target framework matches or beats the best result.

How to debug SGLang production incidents with AI?▼

The sglang-prod-incident-triage skill collects a read-only server bundle, saves the failing request, replays it on a clean target, and only then escalates to profiling or distributed-hang debugging.

Does this work with Claude Code and Codex?▼

Yes. All skills follow the standard SKILL.md format and run in Claude Code, Codex, and other compatible agent runtimes.