Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add shen-shanshan/vllm-dev-skills --skill vllm-feature-tutorialgit clone --depth 1 https://github.com/shen-shanshan/vllm-dev-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/shen-shanshan/vllm-dev-skills/vllm-feature-tutorial)<a href="https://agentmods.dev/skills/shen-shanshan/vllm-dev-skills/vllm-feature-tutorial"><img src="https://agentmods.dev/badge/skills/shen-shanshan/vllm-dev-skills/vllm-feature-tutorial/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/shen-shanshan/vllm-dev-skills/vllm-feature-tutorial"><img src="https://agentmods.dev/badge/skills/shen-shanshan/vllm-dev-skills/vllm-feature-tutorial.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00152 | $0.01064 |
| Opus 5.5 | $0.00061 | $0.00426 |
| Sonnet 5.5 | $0.00030 | $0.00213 |
| Haiku 4.5 | $0.00015 | $0.00106 |
Grade A, and why
vllm-feature-tutorial scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 101 lines — stays where its author put it; the contents beside it link to each section on GitHub.
vLLM Feature Tutorial Generator
Generate a comprehensive code walkthrough tutorial document for a given vLLM feature or module.
Workflow
Step 1: Identify the Feature
Extract the feature/module name from the user's request. Examples:
- "spec decode" / "speculative decoding"
- "chunked prefill"
- "automatic prefix caching"
- "tensor parallelism"
- "KV cache management"
- "continuous batching"
- "LoRA"
- "multimodal"
Step 2: Research the Feature
Gather information from multiple sources. This is the most important step — thorough research determines document quality.
2a. vLLM Official Documentation
- Fetch relevant pages from
https://docs.vllm.ai/en/latest/using WebFetch - Look for design docs, API references, usage guides
2b. vLLM Source Code
- Use
gh apior WebFetch to browse the vLLM GitHub repo (vllm-project/vllm) - Identify core source files for the feature (use GitHub code search or browse directory structure)
- Read key implementation files to understand:
- Core classes and their responsibilities
- Key interfaces and method signatures
- Data flow and control flow
- Important algorithms and data structures
2c. Related Resources
- Search for relevant blog posts, papers, or design documents
- Check vLLM GitHub discussions/issues for design rationale
Step 3: Generate the Tutorial Document
Read references/style-guide.md for the complete document structure and formatting conventions.
Key requirements:
- Write in Chinese (简体中文), keeping English for technical terms
- Follow the multi-part structure defined in the style guide
- Include rich visual elements:
- Mermaid flowcharts for workflows and data flow
- Mermaid architecture diagrams (flowchart TB with subgraph) for system overview
- Mermaid sequence diagrams for component interactions
- Mermaid class diagrams for class hierarchies
- Tables for parameter references, method comparisons, performance metrics
- Code snippets with file path annotations from actual vLLM source
- LaTeX formulas for algorithm analysis (where applicable)
- Include a document header with version info and date
- Include a 文档概述 section with target audience and reading guide
- End with appendices: code location index and glossary
What ships with it
14 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- outputs/amd_aiter_in_vllm.md 33 KB
- outputs/breakable_cudagraph.md 26 KB
- outputs/dcp.md 62 KB
- outputs/deepseek_v4_kv_cache_multistream.md 30 KB
- outputs/disaggregated_encoder_epd.md 37 KB
- outputs/dspark.md 77 KB
- outputs/dual_batch_overlap.md 49 KB
- outputs/jit_kernel_warmup.md 38 KB
- outputs/model_runner_v2.md 27 KB
- outputs/multi_stream_overlap.md 24 KB
- outputs/prefill_dbo.md 38 KB
- outputs/vit_full_cuda_graph.md 86 KB
- outputs/workspace_manager.md 34 KB
- references/style-guide.md 3.4 KB
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 101 lines · 152 tokens per session scan A 0d6d6a38fb58
vllm-feature-tutorial is a skill published in the GitHub repository shen-shanshan/vllm-dev-skills (22 stars, last pushed yesterday), licensed Apache-2.0. It adds 152 tokens to every session and 1,064 once invoked, about $0.0006 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
fs-notebook-tabs
A computer-science capstone: an on-device ML keyboard that predicts next words privately — problem, method, evaluation, and defense answers. Built as a decision-grade coursework defense deck for professor, defense committee.
implementing-llms-litgpt
Implements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained architectures (Llama, Gemma, Phi, Qwen, Mistral). Use when need clean model implementations, educational understanding of architectures, or production fine-tuning with LoRA/QLoRA. Single-file implementations, no abstraction layers.
rwkv-architecture
RNN+Transformer hybrid with O(n) inference. Linear time, infinite context, no KV cache. Train like GPT (parallel), infer like RNN (sequential). Linux Foundation AI project. Production at Windows, Office, NeMo. RWKV-7 (March 2025). Models up to 14B parameters.
context7-docs
Fetch up-to-date documentation and code examples for any library, framework, SDK, CLI tool, or cloud service. Use whenever the user asks about a specific library — even well-known ones like React, Next.js, Prisma, Express, Tailwind, Django, or Spring Boot — because training data may not reflect recent API changes or…
openrlhf-training
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2× faster than DeepSpeedChat with distributed architecture and GPU resource sharing.
learning-notes-automation
A workflow for turning videos, podcasts, and articles into structured learning notes. It extracts key ideas and creates flashcards that can be imported into Anki, a spaced-repetition study app.