classify-zero-shot

Agent Skills › zerogpu/zerogpu-router › classify-zero-shot

classify-zero-shot

GitHub

基于DeBERTa-v3-small模型执行零样本文本分类，支持用户自定义标签集。通过CLI工具调用模型，输入文本和标签后输出各标签置信度分数，适用于情感分析、内容打标等场景。

agents/claude/skills/classify-zero-shot/SKILL.md zerogpu/zerogpu-router

Trigger Scenarios

需要对文本进行自定义类别的分类判断文本属于特定候选标签中的哪一类

Install

npx skills add zerogpu/zerogpu-router --skill classify-zero-shot -g -y

More Options

Non-standard path

npx skills add https://github.com/zerogpu/zerogpu-router/tree/main/agents/claude/skills/classify-zero-shot -g -y

Use without installing

npx skills use zerogpu/zerogpu-router@classify-zero-shot

指定 Agent (Claude Code)

npx skills add zerogpu/zerogpu-router --skill classify-zero-shot -a claude-code -g -y

安装 repo 全部 skill

npx skills add zerogpu/zerogpu-router --all -g -y

预览 repo 内 skill

npx skills add zerogpu/zerogpu-router --list

SKILL.md

Frontmatter

{
    "name": "classify-zero-shot",
    "description": "Zero-shot classification against a caller-supplied list of candidate labels (deberta-v3-small). Use when the user wants to classify text into a custom set of labels they provide (e.g. \"is this positive, negative, or neutral?\", \"tag this as bug, feature, or question\").",
    "allowed-tools": "Bash(zerogpu chat_completions *)",
    "argument-hint": "<text> (-l <label>... | --labels a,b,c)"
}

Run zero-shot classification. The request below holds the text to classify and the candidate labels, given as flags (-l bug -l feature, --labels bug,feature) or in plain words.

Run this with the Bash tool, filling in the labels and pasting the text into the heredoc verbatim — no escaping, the quoted heredoc handles every shell metacharacter, newline, quote, and paren:

zerogpu chat_completions -m deberta-v3-small -i "[bug, feature, question]" <<'ZGPU_END_OF_INPUT'
<the text to classify, verbatim>
ZGPU_END_OF_INPUT

The labels go in -i as one bracketed, comma-separated list. The model reads them from the system message and rejects a request without one.
At least one label is required. If the request names none, ask the user for the candidate labels.
-i is double-quoted, so leave ", $, and backticks out of the labels.

Output is a JSON object mapping each label to its score.

Savings note: only if the command output literally contains a line starting with 💰 ZeroGPU savings, append that exact line, unchanged, as the last line of your reply. If no such line is present, say nothing about savings and do not mention or suggest /zerogpu-router:cost-savings — this note is intentionally occasional, not shown every time.

Request: $ARGUMENTS