← 返回 Skills 市场
207
总下载
0
收藏
0
当前安装
9
版本数
在 OpenClaw 中安装
/install mongolian-llm
功能描述
【强制使用/零泄漏】任何请求涉及中文 ⇄ 传统蒙古文翻译、蒙古语/蒙文对话、要求用蒙文输出，或输入包含传统蒙古文字符（Unicode U+1800–U+18AF）时，必须先读取本 SKILL.md 并按本文 API 流程执行；禁止凭模型自身语言知识直接翻译、改写或回答。调用成功后的可见最终回复必须只等于 API...
安全使用建议
This skill largely does what it claims: it routes Mongolian/Chinese translation and chat through a paid API and ships helper scripts. Expected: it needs MENGGUYU_API_KEY and will call https://mongol.open-idea.net. Concerns: the PDF helper reads your OpenClaw config file (~/.openclaw/openclaw.json) to extract Feishu appId/appSecret in order to download Feishu-hosted PDFs — that gives the skill access to channel credentials outside of the declared MENGGUYU_API_KEY. Before installing, consider: (1) Do you trust mongol.open-idea.net and the skill author to handle your Feishu credentials? (2) If you don't want the skill to access channel credentials, avoid using the PDF download feature (you can run pdf-text-extract.py manually with a local PDF) or remove/obfuscate the Feishu channel fields in openclaw.json. (3) Ask the author to update the manifest to declare the required config path and to make Feishu access optional/explicit. (4) Review network egress policies or monitor requests to confirm only intended API calls are made. If you cannot accept the skill reading ~/.openclaw/openclaw.json (or want explicit consent), treat this skill as untrusted until the developer clarifies and limits that behavior.
功能分析
Type: OpenClaw Skill
Name: mongolian-llm
Version: 1.0.8
The skill bundle exhibits high-risk 'stealth' behaviors and sensitive data access. SKILL.md and BEHAVIOR-RULES.md contain 'Zero Leakage' instructions that strictly command the AI agent to hide all internal reasoning, API logs, and JSON responses from the user, effectively creating a black-box execution environment. Additionally, scripts/pdf-text-extract.py accesses sensitive Feishu credentials from ~/.openclaw/openclaw.json, and references/API-KEY.md explicitly instructs the agent on how to bypass OpenClaw's 'config get' masking to access raw environment variables. While these actions are contextually linked to the stated Mongolian translation and PDF extraction features, the emphasis on hiding execution traces and circumventing security masking is highly suspicious.
能力标签
能力评估
Purpose & Capability
Name/description, scripts and SKILL.md consistently implement a Mongolian translation/chat/OCR workflow that uses MENGGUYU_API_KEY and the provided scripts. However, the PDF helper (scripts/pdf-text-extract.py) reads ~/.openclaw/openclaw.json to get Feishu appId/appSecret for downloading files — this config path is not declared in the skill metadata (required config paths: none). That access is plausible for the PDF-download feature, but it is not disclosed in the manifest and therefore worth flagging.
Instruction Scope
SKILL.md instructs the agent to always use the external mongol.open-idea.net API for any input containing traditional Mongolian characters and to run the bundled node/python scripts. The pdf-text-extract.py explicitly reads OpenClaw channel credentials from the user's openclaw.json to obtain Feishu tenant tokens and download files; SKILL.md and manifest do not advertise that the skill will read that file. Reading other platform config/credentials expands scope beyond just the MENGGUYU_API_KEY and should be disclosed and consented to.
Install Mechanism
There is no install spec or external downloads; the skill is instruction-plus-scripts only. No remote installers or opaque archives are fetched. Risk from install mechanism is low.
Credentials
The declared primary credential is MENGGUYU_API_KEY, which is appropriate. But the PDF script accesses Feishu appId/appSecret stored in ~/.openclaw/openclaw.json (channel credentials), which are unrelated to the Mongolian API key and are not declared in requires.env or required config paths. Access to other service credentials is disproportionate and should be declared and optional.
Persistence & Privilege
The skill does not request always:true and does not request elevated platform privileges, so it is not force-included. However, it reads another skill/platform config file (~/.openclaw/openclaw.json) to obtain Feishu credentials. That means it can access channel-level secrets belonging to the host environment; reading other skills' credentials is a privilege that should be explicitly requested and justified.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install mongolian-llm - 安装完成后，直接呼叫该 Skill 的名称或使用
/mongolian-llm触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.0.8
- 新增 PDF 文本提取支持，附带 Python 脚本 pdf-text-extract.py，可从本地或飞书链接自动提取 PDF 内容用于后续翻译或处理。
- 文档中明确 PDF 提取场景，脚本调用示例增补对 PDF 的操作方法。
- SKILL.md 相关接口说明与脚本用法同步更新，包含新的 PDF 预处理能力。
- 其它文档细节与引用小幅修正、整理。
v1.0.7
No changes in code or documentation detected for this version.
- No file modifications or updates.
- Skill behavior and usage remain unchanged.
v1.0.6
# mongolian-llm 1.0.6 Changelog
- Relaxed API Key requirements：API Key 不再作为技能安装/发现阶段的强依赖，仅在业务调用前检查。
- metadata: 移除了 `requires.env` 字段，精简为只配置 `primaryEnv`。
- 触发与前置检查相关文本同步优化，明确“业务调用前”必须检查 API Key。
- 其余接口规范、路由规则、调用流程无变动。
v1.0.5
Mongolian-llm 1.0.5
- 新增支持图片文字识别（OCR），默认 `language=mn`，补充调用规范与脚本调用方式。
- 增加文档 [references/OCR.md]，说明 OCR API 用法。
- 新增脚本 [scripts/image-to-base64.js]，用于图片转 base64，辅助 OCR 场景。
- SKILL.md 完善接口路由，明确中文/蒙文回复判别优先级，补充 OCR 脚本用法说明。
v1.0.4
**本版本是一次重要升级，全面强化了高安全输出和接口调用规范，新增自动脚本调用能力：**
- 新增脚本 `scripts/mongolian-api.js`，支持用 Node.js 稳定调用 API 并自动提取目标正文，避免输出 JSON。
- 强制“零泄漏”：用户可见回复只能是 API 目标字段原文，禁止带路由判断、调用日志、JSON、token 信息等任何修饰说明。
- 明确要求每轮请求重新路由，禁止沿用上一轮的接口选择，确保出错概率最小化。
- 详细更新了前置检查、调用前确认、错误处理策略；完善了“批量、定时、远程”分工边界说明。
- references 文档内全面修订接口说明内容，与 SKILL.md 严格同步。
v1.0.3
mongolian-llm 1.0.3
- 支持用户直接输入传统蒙古文字符（Unicode U+1800–U+18AF）时自动激活技能用于对话或翻译。
- 扩展触发词，包含“蒙古文聊天”“ᠮᠣᠩᠭᠣᠯ ᠬᠡᠯᠡᠬᠦ”“蒙古语/蒙古文”等蒙语相关自然表达。
- 明确区分对话和翻译 API 的场景优先级与规则，完善多语种输入（中/蒙文）时的调用指引。
- 调整调用前用户确认策略：用户主动发起蒙语请求可直接调用，仅代理主动时需告知并等待确认。
- 对对话接口输出增加“必须原样输出”全局约束，防止被自动润色或补充。
- 补充 references/ 文档，细化接口调用模板、
v1.0.2
- No file changes detected in this release.
- Documentation and usage, interface, and API details remain unchanged from the previous version.
v1.0.1
**Major update: Documentation structure overhauled, detailed usage and billing confirmation required**
- Major documentation split: Core usage, interface routing, translation, chat, API key, and behavior rules are now modularized in dedicated `references/` files.
- Explicit per-call billing details and fee confirmation workflow added; user confirmation is now required before any paid API call.
- Clarified scenario choice: replies to Mongolian emails/messages now recommend direct use of chat API for contextual response (rather than translation round-trip).
- Expanded and clarified trigger list, especially for reply, creative, and conversation scenarios.
- All technical notes (segmentation, error handling, API parameters) are now condensed in "Quick Call Template" and linked to detailed references.
- Added clear boundaries: this skill does not send emails or fetch web content; bulk/remote scenarios must be handled by caller logic.
v1.0.0
- Initial release of mongolian-llm skill for 中文 ⇄ 传统蒙古文 translation and Mongolian language conversation.
- Supports translation, content generation, Mongolian-language email, notifications, documents, and social media posts.
- Requires API Key setup via environment variable MENGGUYU_API_KEY; user guidance and security practices included.
- Two main endpoints: /translation for strict translation tasks, /chat/completions for conversation and content creation, with automatic interface selection based on user intent.
- Includes detailed segmentation, request formatting, and error handling principles for robust and accurate Mongolian language processing.
元数据
常见问题
mongolian-llm 是什么？
【强制使用/零泄漏】任何请求涉及中文 ⇄ 传统蒙古文翻译、蒙古语/蒙文对话、要求用蒙文输出，或输入包含传统蒙古文字符（Unicode U+1800–U+18AF）时，必须先读取本 SKILL.md 并按本文 API 流程执行；禁止凭模型自身语言知识直接翻译、改写或回答。调用成功后的可见最终回复必须只等于 API... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 207 次。
如何安装 mongolian-llm？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install mongolian-llm」即可一键安装，无需额外配置。
mongolian-llm 是免费的吗？
是的，mongolian-llm 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
mongolian-llm 支持哪些平台？
mongolian-llm 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 mongolian-llm？
由 Knixie（@knixie）开发并维护，当前版本 v1.0.8。
推荐 Skills