Agent Skills › vllm-project/llm-compressor › contribute-research

contribute-research

GitHub

基于实验结果、日志或讨论,在LLM Compressor研究库中创建、更新或审查文档条目。

.agents/skills/contribute-research/SKILL.md vllm-project/llm-compressor

Trigger Scenarios

提交新的实验研究成果 更新已有的实验数据 审查研究条目准确性

Install

npx skills add vllm-project/llm-compressor --skill contribute-research -g -y
More Options

Non-standard path

npx skills add https://github.com/vllm-project/llm-compressor/tree/main/.agents/skills/contribute-research -g -y

Use without installing

npx skills use vllm-project/llm-compressor@contribute-research

指定 Agent (Claude Code)

npx skills add vllm-project/llm-compressor --skill contribute-research -a claude-code -g -y

安装 repo 全部 skill

npx skills add vllm-project/llm-compressor --all -g -y

预览 repo 内 skill

npx skills add vllm-project/llm-compressor --list

SKILL.md

Frontmatter
{
    "name": "contribute-research",
    "description": "Create, update, or review LLM Compressor Open Research Hub entries in docs\/research from experiment results, logs, or issue discussions."
}

Contribute to the Open Research Hub

Read docs/research/index.md before editing. It is the single source of truth for submission, formatting, and attribution. Follow it if this skill, its template, or an existing entry differs.

  1. Read the supplied evidence and search docs/research/ for related experiments. If the contributor has already posted an experiment and has related results to add, update their existing post. Otherwise, create a new entry.
  2. For new entries, copy the Markdown block in the entry template, keeping its block-style tags, into docs/research/<short-title>-<username>.md. Start the filename with the algorithm or scheme and end with the contributor's GitHub handle.
  3. Fill the template from the evidence. Use the experiment date and the index's byline and author-consent rules. State hardware and calibration setup. Resolve missing information without inventing details; distinguish unmeasured values from measured results. Negative results are valuable and should be included.
  4. Add reproduction details or source links when useful and available; these do not require additional experiments. For updates, retain original credits and experiment dates, explain corrections or new findings, and apply the index's updated-date guidance.
  5. Check the entry against the index and source evidence, including numbers, links, and unfilled placeholders. New entries appear automatically; do not edit navigation files.

For rendering checks, follow docs/DEVELOPMENT.md. Keep generated build output out of the contribution. Report checks performed and unresolved gaps; a docs build does not reproduce an experiment.

When asked to submit, follow CONTRIBUTING.md and .github/PULL_REQUEST_TEMPLATE.md. Otherwise, hand off the draft and its remaining gaps.

Version History

  • 0.15.0 Current 2026-11-02 19:49

Same Skill Collection

.agents/skills/create-tiny-model/SKILL.md
.claude/skills/create-tiny-model/SKILL.md
examples/.agents/skills/fp8/SKILL.md
examples/.agents/skills/nvfp4/SKILL.md
examples/.claude/skills/fp8/SKILL.md
examples/.claude/skills/nvfp4/SKILL.md

Metadata

Files
0
Version
0.15.0
Hash
319b0b4d
Indexed
2026-11-02 19:49

Home - Wiki
Copyright © 2011-2026 iteam. Current version is 2.155.2. UTC+08:00, 2026-11-02 20:49
浙ICP备14020137号-1