Claude Skill

vllm-rfc-generator

Generate a vLLM-style RFC (Request for Comments) document based on user input. Use when the user wants to create an RFC for a major architectural change or design decision in vLLM. Triggered by requests like "帮我生成一个vLLM RFC", "create a vLLM RFC for ...", "写一个RFC关于...", "生成RFC文档".

LLM Mart · 0 points · 2 views 0 listing impressions 0 install-command copies
Download shen-shanshan-vllm-dev-skills-skills_vllm-rfc-generator-f687af6.zip · 8 KB

Install

skills CLI npx skills add https://github.com/shen-shanshan/vllm-dev-skills/tree/master/skills/vllm-rfc-generator
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install shen-shanshan-vllm-dev-skills@llmmart
Git git clone https://github.com/shen-shanshan/vllm-dev-skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole shen-shanshan/vllm-dev-skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

vLLM RFC Generator

Generate a vLLM-style RFC Markdown document from user-provided information and save it to ./outputs/.

Inputs

Collect the following from the user (ask if not provided):

Field Required Description
RFC Title Yes Short, descriptive title (will be prefixed with [RFC]:)
Main Content Yes Description of the motivation and proposed changes
Related PRs No List of PR numbers or URLs; classify them by category
Feedback Period No Duration for feedback (default: "At least 1 week")
CC List No GitHub usernames to CC (e.g., @user1, @user2)
Other Notes No Any additional context or notes

Workflow

  1. Collect inputs — gather all necessary information from the user
  2. Classify PRs — if PRs are provided, group them by category (see PR Classification below)
  3. Generate RFC content — write each section following the template
  4. Save output — write to ./outputs/rfc-<slug>.md
  5. Confirm — tell the user the output path

PR Classification

When the user provides related PRs, classify them into appropriate categories such as:

  • Background / Motivation PRs — existing issues, bugs, or prior discussions that motivate this RFC
  • Implementation PRs — concrete code changes implementing the proposed design
  • Dependency PRs — prerequisite changes that must land first
  • Related / Follow-up PRs — tangential improvements or future work inspired by this RFC

Format each category as a bullet list:

### Related PRs

**Background / Motivation:**
- [#XXXX](https://github.com/vllm-project/vllm/pull/XXXX): Brief description

**Implementation:**
- [#YYYY](https://github.com/vllm-project/vllm/pull/YYYY): Brief description

**Dependencies:**
- [#ZZZZ](https://github.com/vllm-project/vllm/pull/ZZZZ): Brief description

If a PR number has no description, fetch its title via the GitHub API or gh CLI:

gh pr view <NUMBER> --repo vllm-project/vllm --json title,url -q '"\(.url): \(.title)"'

RFC Document Template

Save to ~/.claude/skills/vllm-rfc-generator/outputs/rfc-<slug>.md where <slug> is the title converted to lowercase with hyphens (e.g., rfc-async-engine-redesign.md).

# [RFC]: <Title>

> **Labels**: RFC
> **Status**: Open for Feedback
> **Feedback Period**: <period>

---

## Motivation

<2–4 paragraphs explaining WHY this change is needed>

- What problem does it solve?
- What are the current limitations or pain points?
- Why is this the right time to address it?

## Proposed Change

<3–6 paragraphs or structured subsections describing the proposed design>

Key points to cover:
- What is being changed / added / removed?
- High-level design or architecture
- API surface changes (if any)
- Migration path or backward compatibility

### Key Design Decisions

- **Decision 1**: Rationale
- **Decision 2**: Rationale

## Related PRs

<Classified PR list — see PR Classification section>

## Feedback Period

<Feedback period statement, e.g.: "We plan to keep this RFC open for at least 2 weeks (until YYYY-MM-DD). Please share your thoughts in the comments.">

## CC List

<@username1>, <@username2>

## Any Other Things

<Optional: open questions, alternatives considered, links to design docs, benchmarks, etc.>

---

> Please take a look at previous [RFCs](https://github.com/vllm-project/vllm/issues?q=label%3ARFC+sort%3Aupdated-desc) for reference.

Writing Guidelines

  • Motivation: Be concrete. Reference real pain points, GitHub issues, or user feedback. 2–4 short paragraphs.
  • Proposed Change: Be specific enough that reviewers can evaluate the design, but avoid implementation minutiae. Use subsections for complex proposals.
  • Tone: Professional, neutral, open to feedback. This is a proposal, not a decree.
  • Length: Aim for 400–800 words total across Motivation and Proposed Change. Quality over quantity.
  • Language: Write in English by default. If the user requests Chinese, write in Chinese.

Example Output Path

~/.claude/skills/vllm-rfc-generator/outputs/rfc-async-engine-redesign.md
Files (vllm-dev-skills)
  • outputs
    • .gitkeep 0 B · in bundle
    • rfc-deepseek-v4-rocm-decode-performance-optimization.md 6.2 KB
      Loading…
    • rfc-support-vit-full-cuda-graph.md 6 KB
      Loading…
  • references
    • rfc-template.yml 1.6 KB
      Loading…
  • SKILL.md 4.5 KB
    ---
    name: vllm-rfc-generator
    description: Generate a vLLM-style RFC (Request for Comments) document based on user input. Use when the user wants to create an RFC for a major architectural change or design decision in vLLM. Triggered by requests like "帮我生成一个vLLM RFC", "create a vLLM RFC for ...", "写一个RFC关于...", "生成RFC文档".
    ---
    
    # vLLM RFC Generator
    
    Generate a vLLM-style RFC Markdown document from user-provided information and save it to `./outputs/`.
    
    ## Inputs
    
    Collect the following from the user (ask if not provided):
    
    | Field | Required | Description |
    |-------|----------|-------------|
    | **RFC Title** | Yes | Short, descriptive title (will be prefixed with `[RFC]:`) |
    | **Main Content** | Yes | Description of the motivation and proposed changes |
    | **Related PRs** | No | List of PR numbers or URLs; classify them by category |
    | **Feedback Period** | No | Duration for feedback (default: "At least 1 week") |
    | **CC List** | No | GitHub usernames to CC (e.g., `@user1`, `@user2`) |
    | **Other Notes** | No | Any additional context or notes |
    
    ## Workflow
    
    1. **Collect inputs** — gather all necessary information from the user
    2. **Classify PRs** — if PRs are provided, group them by category (see PR Classification below)
    3. **Generate RFC content** — write each section following the template
    4. **Save output** — write to `./outputs/rfc-<slug>.md`
    5. **Confirm** — tell the user the output path
    
    ## PR Classification
    
    When the user provides related PRs, classify them into appropriate categories such as:
    
    - **Background / Motivation PRs** — existing issues, bugs, or prior discussions that motivate this RFC
    - **Implementation PRs** — concrete code changes implementing the proposed design
    - **Dependency PRs** — prerequisite changes that must land first
    - **Related / Follow-up PRs** — tangential improvements or future work inspired by this RFC
    
    Format each category as a bullet list:
    
    ```markdown
    ### Related PRs
    
    **Background / Motivation:**
    - [#XXXX](https://github.com/vllm-project/vllm/pull/XXXX): Brief description
    
    **Implementation:**
    - [#YYYY](https://github.com/vllm-project/vllm/pull/YYYY): Brief description
    
    **Dependencies:**
    - [#ZZZZ](https://github.com/vllm-project/vllm/pull/ZZZZ): Brief description
    ```
    
    If a PR number has no description, fetch its title via the GitHub API or `gh` CLI:
    ```bash
    gh pr view <NUMBER> --repo vllm-project/vllm --json title,url -q '"\(.url): \(.title)"'
    ```
    
    ## RFC Document Template
    
    Save to `~/.claude/skills/vllm-rfc-generator/outputs/rfc-<slug>.md`
    where `<slug>` is the title converted to lowercase with hyphens (e.g., `rfc-async-engine-redesign.md`).
    
    ```markdown
    # [RFC]: <Title>
    
    > **Labels**: RFC
    > **Status**: Open for Feedback
    > **Feedback Period**: <period>
    
    ---
    
    ## Motivation
    
    <2–4 paragraphs explaining WHY this change is needed>
    
    - What problem does it solve?
    - What are the current limitations or pain points?
    - Why is this the right time to address it?
    
    ## Proposed Change
    
    <3–6 paragraphs or structured subsections describing the proposed design>
    
    Key points to cover:
    - What is being changed / added / removed?
    - High-level design or architecture
    - API surface changes (if any)
    - Migration path or backward compatibility
    
    ### Key Design Decisions
    
    - **Decision 1**: Rationale
    - **Decision 2**: Rationale
    
    ## Related PRs
    
    <Classified PR list — see PR Classification section>
    
    ## Feedback Period
    
    <Feedback period statement, e.g.: "We plan to keep this RFC open for at least 2 weeks (until YYYY-MM-DD). Please share your thoughts in the comments.">
    
    ## CC List
    
    <@username1>, <@username2>
    
    ## Any Other Things
    
    <Optional: open questions, alternatives considered, links to design docs, benchmarks, etc.>
    
    ---
    
    > Please take a look at previous [RFCs](https://github.com/vllm-project/vllm/issues?q=label%3ARFC+sort%3Aupdated-desc) for reference.
    ```
    
    ## Writing Guidelines
    
    - **Motivation**: Be concrete. Reference real pain points, GitHub issues, or user feedback. 2–4 short paragraphs.
    - **Proposed Change**: Be specific enough that reviewers can evaluate the design, but avoid implementation minutiae. Use subsections for complex proposals.
    - **Tone**: Professional, neutral, open to feedback. This is a proposal, not a decree.
    - **Length**: Aim for 400–800 words total across Motivation and Proposed Change. Quality over quantity.
    - **Language**: Write in **English** by default. If the user requests Chinese, write in Chinese.
    
    ## Example Output Path
    
    ```
    ~/.claude/skills/vllm-rfc-generator/outputs/rfc-async-engine-redesign.md
    ```
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.