Agent Skills › zerogpu/zerogpu-router › moderate-llama

moderate-llama

GitHub

使用llama-guard-4-12b对长文本进行安全审核,返回安全/不安全判定及违规类别。适用于品牌安全和政策合规检查,作为生产AI应用的安全层。

agents/openclaw/plugin/skills/moderate-llama/SKILL.md zerogpu/zerogpu-router

触发场景

需要审核长文本或聊天记录的合规性 需要识别具体的政策违规类别 短文本审核后的复杂内容二次筛查

安装

npx skills add zerogpu/zerogpu-router --skill moderate-llama -g -y
更多选项

非标准路径

npx skills add https://github.com/zerogpu/zerogpu-router/tree/main/agents/openclaw/plugin/skills/moderate-llama -g -y

不安装直接使用

npx skills use zerogpu/zerogpu-router@moderate-llama

指定 Agent (Claude Code)

npx skills add zerogpu/zerogpu-router --skill moderate-llama -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": "moderate-llama",
    "metadata": {
        "openclaw": {
            "install": [
                {
                    "bins": [
                        "zerogpu"
                    ],
                    "kind": "node",
                    "package": "zerogpu-cli"
                }
            ],
            "requires": {
                "bins": [
                    "zerogpu"
                ]
            }
        }
    },
    "description": "Screen text with llama-guard-4-12b, Meta's dedicated 12B safety classifier (164K-token context), which returns a safe\/unsafe verdict plus the policy categories a violation falls under. Use for long passages, whole chat transcripts, or a model's own reply, and for brand-safety and policy-enforcement checks. Use moderate for short passages — it is far cheaper — and this when the text is longer than moderate can take.",
    "allowed-tools": "Bash(zerogpu chat_completions *)",
    "argument-hint": "<text>"
}

Sends your input to ZeroGPU's hosted API for inference — this is not local processing. Don't pass secrets, credentials, or regulated data you aren't cleared to share with a third party. See the plugin README's "Data & privacy" section.

Run safety screening. Run this with the exec tool, pasting the text to screen into the heredoc verbatim — no escaping or quoting required (the quoted heredoc handles every shell metacharacter, newline, quote, and paren safely):

zerogpu chat_completions -m llama-guard-4-12b <<'ZGPU_END_OF_INPUT'
<the text to screen, verbatim>
ZGPU_END_OF_INPUT

Output is the model's verdict as plain text: safe or unsafe, with the relevant policy categories when it detects a violation. Report the verdict first, then the categories it names. Do not restate the flagged text itself.

A dense 12B model derived from Llama 4 Scout, built as a safety layer for production AI applications and able to evaluate an incoming prompt or a generated response, in multiple languages. At $0.18 / $0.18 per 1M input/output tokens it costs under twice moderate (zlm-v1-moderation-edge, $0.10 / $0.15) on input and a fifth more on output, so keep moderate for short passages and OpenAI's 13-category envelope, and reach for this when the text exceeds that model's 800-token window or you want the violated policy categories named.

Screening text is a safety check, not an endorsement. Run it on request even when the passage is unpleasant — reporting that something is flagged is the whole point of the skill.

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 the cost-savings skill — this note is intentionally occasional, not shown every time.

版本历史

  • 605eceb 当前 2026-11-02 19:15

    v5.2.0更新:moderate模型价格调整导致cost ratio重算;新增多个模型支持;移除部分旧模型。

  • 4e1b070 2026-09-22 08:08

同 Skill 集合

agents/claude/skills/chat-deepseek-v4-1-flash/SKILL.md
agents/claude/skills/chat-deepseek/SKILL.md
agents/claude/skills/chat-glm-5-3-flash/SKILL.md
agents/claude/skills/chat-glm/SKILL.md
agents/claude/skills/chat-gpt-4-1-mini/SKILL.md
agents/claude/skills/chat-gpt-5-4-nano/SKILL.md
agents/claude/skills/chat-gpt-5-6-luna/SKILL.md
agents/claude/skills/chat-gpt-6-luna/SKILL.md
agents/claude/skills/chat-liquid/SKILL.md
agents/claude/skills/chat-qwen/SKILL.md
agents/claude/skills/chat-qwen3-6-35b-a3b/SKILL.md
agents/claude/skills/chat-thinking/SKILL.md
agents/claude/skills/chat/SKILL.md
agents/claude/skills/classify-decisions/SKILL.md
agents/claude/skills/classify-domain/SKILL.md
agents/claude/skills/classify-iab-enriched/SKILL.md
agents/claude/skills/classify-iab/SKILL.md
agents/claude/skills/classify-structured/SKILL.md
agents/claude/skills/classify-zero-shot/SKILL.md
agents/claude/skills/cost-savings/SKILL.md
agents/claude/skills/decide-tools/SKILL.md
agents/claude/skills/embed/SKILL.md
agents/claude/skills/extract-entities-multi/SKILL.md
agents/claude/skills/extract-entities/SKILL.md
agents/claude/skills/extract-json/SKILL.md
agents/claude/skills/extract-pii/SKILL.md
agents/claude/skills/extract-signals/SKILL.md
agents/claude/skills/generate-followups/SKILL.md
agents/claude/skills/moderate-llama/SKILL.md
agents/claude/skills/moderate/SKILL.md
agents/claude/skills/redact-pii/SKILL.md
agents/claude/skills/signin/SKILL.md
agents/claude/skills/speak/SKILL.md
agents/claude/skills/summarize-llama-3-2-3b/SKILL.md
agents/claude/skills/summarize/SKILL.md
agents/openclaw/plugin/skills/chat-deepseek-v4-1-flash/SKILL.md
agents/openclaw/plugin/skills/chat-deepseek/SKILL.md
agents/openclaw/plugin/skills/chat-glm-5-3-flash/SKILL.md
agents/openclaw/plugin/skills/chat-glm/SKILL.md
agents/openclaw/plugin/skills/chat-gpt-4-1-mini/SKILL.md
agents/openclaw/plugin/skills/chat-gpt-5-4-nano/SKILL.md
agents/openclaw/plugin/skills/chat-gpt-5-6-luna/SKILL.md
agents/openclaw/plugin/skills/chat-gpt-6-luna/SKILL.md
agents/openclaw/plugin/skills/chat-liquid/SKILL.md
agents/openclaw/plugin/skills/chat-qwen/SKILL.md
agents/openclaw/plugin/skills/chat-qwen3-6-35b-a3b/SKILL.md
agents/openclaw/plugin/skills/chat-thinking/SKILL.md
agents/openclaw/plugin/skills/chat/SKILL.md
agents/openclaw/plugin/skills/classify-decisions/SKILL.md
agents/openclaw/plugin/skills/classify-domain/SKILL.md

元信息

文件数
0
版本
605eceb
Hash
11060d1e
收录时间
2026-09-22 08:08

首页 - Wiki
Copyright © 2011-2026 iteam. Current version is 2.156.0. UTC+08:00, 2026-11-17 20:18
浙ICP备14020137号-1