classify-decisions

Agent Skills › zerogpu/zerogpu-router › classify-decisions

classify-decisions

GitHub

利用 gliner2.5-decide-v1 模型进行动态标签决策与路由，支持意图分类、工单分流、内容审核及人工介入判断。适用于高并发场景下的多标签分类和批量路由任务。

agents/claude/skills/classify-decisions/SKILL.md zerogpu/zerogpu-router

Trigger Scenarios

需要对用户输入进行意图识别或分类根据结果决定后续处理流程或工具调用需要判断是否将请求转交人工处理执行大规模数据分流或工单分级

Install

npx skills add zerogpu/zerogpu-router --skill classify-decisions -g -y

More Options

Non-standard path

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

Use without installing

npx skills use zerogpu/zerogpu-router@classify-decisions

指定 Agent (Claude Code)

npx skills add zerogpu/zerogpu-router --skill classify-decisions -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-decisions",
    "description": "Dynamic-label decisions and routing with gliner2.5-decide-v1 (486M) — supply the labels or options at request time to classify intent, route agents and tools, triage support requests, make moderation decisions, and decide when a human handoff is needed. Single-label, multi-label, and several decision heads in one call. Use for high-volume routing and triage — it is the cheapest classifier on the platform and bills nothing on output.",
    "allowed-tools": "Bash(zerogpu chat_completions *)",
    "argument-hint": "<text> -s '<json labels>'"
}

Run a dynamic-label decision. The request below holds the text and the decision heads — each head with its allowed options — given as -s '<json>' or in plain words.

Run this with the Bash tool, putting the heads in --metadata 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 gliner2.5-decide-v1 --metadata '{"usecase":"decide","labels":{"intent":["payment_issue","refund","account_access","general_question"],"handoff":["yes","no"]}}' <<'ZGPU_END_OF_INPUT'
<the source text, verbatim>
ZGPU_END_OF_INPUT

"usecase":"decide" is required exactly as written; the API rejects the request without it.
labels maps each decision head to its allowed options, and several heads may be decided in one call. If the request gives heads without options, pick sensible option sets and say which you used.
For plain single- or multi-label classification against a schema, "usecase":"classification" with a schema works too — the same shape /zerogpu-router:classify-structured uses.
--metadata is single-quoted JSON, so write any ' inside an option as '\''.

Output is a JSON object holding the chosen option per decision head.

At $0.004 per 1M input tokens, billing nothing on output, this is the cheapest classifier on the platform — a fifth of /zerogpu-router:classify-structured on input. Prefer classify-structured when you want a schema-shaped classification, and this skill when you are routing or triaging at volume.

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