by 安全扫描
OpenClaw
安全
high confidenceThe skill's code and instructions align with its stated purpose (layout-preserving PPTX translation using LibreOffice rendering + vision checks); there are no hidden endpoints or credential requests, but the runtime design relies on the agent calling external LLM/Vision services (privacy consideration) and there is a small usability bug to be aware of.
评估建议
This skill appears to be what it claims: a local utility + agent loop that rewrites PPTX and uses LibreOffice to render for overflow checks. Before installing/use: (1) Ensure LibreOffice and python-pptx are available locally; the script calls the libreoffice binary (subprocess) and will write output files in the specified directory. (2) Provide an explicit --output path when invoking the script (args.output is optional in code but passing None may cause errors). (3) Understand privacy implicatio...详细分析 ▾
✓ 用途与能力
Name/description (translate PPTX while preserving layout) match the included script and SKILL.md. The script extracts text, rewrites slides with scaled fonts, and can invoke LibreOffice to render PNGs — all coherent with the stated purpose. Required binaries listed in _meta.json (libreoffice, python-pptx) are appropriate.
ℹ 指令范围
SKILL.md instructs an agent loop that uses an LLM for translation and a Vision model for overflow detection. The shipped script only performs extraction, rewriting, and local LibreOffice rendering; it does not itself call external APIs. However, the agent-level instructions implicitly require sending slide content to an LLM and to a Vision model for verification — this is expected for the feature but is a privacy/data-flow consideration because endpoints/credentials are not specified.
✓ 安装机制
No install spec is provided (instruction-only with a small helper script). That is low-risk: nothing is automatically downloaded or written to disk by an installer. Runtime requires system LibreOffice and Python dependency python-pptx, which is reasonable and declared.
ℹ 凭证需求
The skill declares no required environment variables or credentials (correct). However, SKILL.md assumes use of an LLM and a Vision model; depending on your agent configuration these calls may transmit slide content to external services and may require API keys not declared here. This is a privacy/operational consideration, not a hidden credential request by the skill itself.
✓ 持久化与权限
Skill is not marked always:true and is user-invocable. It does not request persistent system-wide changes or other skills' configs. It will read and write files provided by the user (input/output paths) which is expected for its purpose.
安全有层次，运行前请审查代码。
运行时依赖
无特殊依赖
版本
latestv1.0.02026/2/25
Initial release: layout-preserving PPT translation with LibreOffice render + Vision agent loop
● 可疑
安装命令 点击复制
官方npx clawhub@latest install ppt-translator
镜像加速npx clawhub@latest install ppt-translator --registry https://cn.clawhub-mirror.com
技能文档
保持排版翻译 PPT，支持任意语言对。
核心原理
Agent Loop（渲染验证）：
- 提取 PPT 文字 + 样式
- LLM 翻译（保证准确）
- 写回 PPT（python-pptx）
- LibreOffice 渲染成真实 PNG
- Vision 模型检测文字溢出
- 有溢出 → 缩字号 15% → 回步骤 3
- 通过 → 输出最终 PPTX
依赖
python-pptx（pip）libreoffice（yum/apt）
使用方法
python3 scripts/translate.py \
--input /path/to/file.pptx \
--output /path/to/output.pptx \
--translations '{"原文": "translation", ...}' \
--max-iter 5
Agent 调用流程
- 用户发送 PPTX 文件
- Agent 提取所有文字 → 翻译（自己翻译或调用 LLM）
- 调用
scripts/translate.py执行渲染验证循环 - 循环完成后将输出 PPTX 发回用户
输出
- 最终 PPTX 文件（可编辑）
- 渲染验证 PNG（可选，用于确认）
- 迭代日志（几轮收敛、最终字号缩放比）
注意事项
- LibreOffice 首次启动较慢（~5s），之后正常
- 字号缩放是全局的（所有 shape 等比缩），未来可优化为 per-shape
- 不支持 .ppt（老格式），仅支持 .pptx
数据来源：ClawHub ↗ · 中文优化：龙虾技能库
OpenClaw 技能定制 / 插件定制 / 私有工作流定制
免费技能或插件可能存在安全风险，如需更匹配、更安全的方案，建议联系付费定制