Translate Chinese PPT to English

Translate Chinese PowerPoint presentations to English while preserving all images, charts, shapes, and media content. Adjusts fonts to Calibri and optimizes...

Security Scan

The skill's code and instructions match its stated purpose (translating Chinese PPTX text via an OpenAI‑compatible LLM) and do not request unrelated credentials or installs, but it will send extracted slide text to whichever LLM endpoint you point it at — treat confidential slides accordingly.

Purpose & Capability

Name/description match the actual behavior: the package extracts text from PPTX, translates via an OpenAI‑compatible API, and reapplies translations while preserving non‑text content. Required packages (python-pptx, requests) and an LLM endpoint are appropriate and proportional.

Instruction Scope

SKILL.md and the script limit activity to parsing PPTX content, batching Chinese text segments, calling the specified LLM endpoint, and writing a translated PPTX. They do not read unrelated files or environment variables. Important privacy note: the tool sends extracted slide text to the configured api_base (cloud or local); this is expected for translation but is a data‑exfiltration surface if you point it at an untrusted cloud endpoint.

Install Mechanism

This is an instruction-only skill with no install spec; it relies on pip-installed Python packages. No downloads from untrusted URLs or archive extraction are present in the provided files.

Credentials

The skill declares no required environment variables or credentials. It accepts an --api-key CLI option (and will set an Authorization header if provided) which is reasonable. Users should avoid supplying cloud API keys when translating sensitive slides unless they trust the provider or use a local model.

Persistence & Privilege

The skill does not request persistent presence, does not set always:true, and contains no code that modifies other skills or global agent settings.

Assessment

This skill appears to do what it claims: it extracts only text from your PPTX and sends those text segments to the LLM endpoint you configure. Before using it on sensitive presentations, consider: (1) run it against a local model (Ollama/Qoderwork) or an approved enterprise endpoint rather than a public cloud API; (2) never paste long-lived production API keys into unknown scripts — prefer short‑lived or scoped keys; (3) inspect the included translate_ppt.py yourself (it is small and readable) and test on non‑sensitive slides first; (4) if you must use a cloud service, review the provider's data retention and privacy policy. If you want, I can point out exact lines where it sends text to the API and where an API key is used so you know what data would be transmitted.

Translate PPT

Translate Chinese PowerPoint presentations (.pptx) to English with professional business styling.

Overview

This skill translates Chinese PPTX files to English using any OpenAI-compatible LLM endpoint (local or cloud). It preserves all non-text content while:

• Preserving all non-text content (images, charts, shapes, tables, SmartArt, media)
• Adjusting fonts to Calibri family for consistent business styling
• Optimizing text box sizing and layout for English text (typically longer than Chinese)
• Maintaining original slide masters, layouts, animations, and transitions

Prerequisites

• Python 3.8+
• Required packages: python-pptx, requests
• An LLM endpoint that supports OpenAI-compatible API (e.g., Qoderwork built-in models, Ollama, OpenAI, DeepSeek, etc.)

Quick Start

Option A: Use with Qoderwork (Easiest — No Extra Setup)

If you're running this skill within Qoderwork, models are already available. Just run:

pip install python-pptx requests
python .qoder/skills/translate-ppt/scripts/translate_ppt.py input.pptx --api-base <qoderwork-endpoint> --model <available-model>

Option B: Use with Local Ollama

1. Install Ollama: Download and install from https://ollama.com

2. Pull a recommended model:

   ollama pull qwen2.5:14b
   

3. Install Python dependencies:

   pip install python-pptx requests
   

4. Run translation:

   python .qoder/skills/translate-ppt/scripts/translate_ppt.py input.pptx
   

   If output path is not specified, defaults to <input_name>_en.pptx.

Option C: Use with Cloud API

Run with your cloud endpoint:

python .qoder/skills/translate-ppt/scripts/translate_ppt.py input.pptx --api-base https://api.openai.com/v1 --api-key sk-xxx --model gpt-4o

Translation Rules

• Translate: All text content (titles, body text, notes, table cells, grouped shape text)
• Preserve: Images, charts data, embedded media, hyperlinks, original formatting
• Mixed content: Only translate Chinese portions of mixed Chinese/English text

Font & Layout Adjustments

Element	Font	Style
Titles	Calibri	Bold
Body text	Calibri	Regular

• Maintain original font sizes (with auto-shrink if text overflows)
• Adjust text box width up to 20% if English text is significantly longer
• Preserve original color scheme and text formatting (bold, italic, underline)

Business Style Guidelines

• Consistent Calibri font family throughout
• Clean, professional spacing
• Preserved slide master/layout templates
• All animations and transitions intact

Recommended Models

Translation quality varies significantly between models. Choose based on your setup:

Note for Qoderwork users: You can use whatever models are already configured in your client environment — no additional setup required.

Model	Size	Quality	Speed	Command
qwen2.5:14b	~9GB	★★★★★ Best for Chinese	Fast	ollama pull qwen2.5:14b
qwen2.5:7b	~4.7GB	★★★★ Good balance	Faster	ollama pull qwen2.5:7b
llama3.1:8b	~4.7GB	★★★ Decent	Fast	ollama pull llama3.1:8b
gemma2:9b	~5.4GB	★★★ Decent	Fast	ollama pull gemma2:9b
qwen2.5:3b	~2GB	★★ Basic	Fastest	ollama pull qwen2.5:3b

Tip: For best results with Chinese-to-English business content, Qwen2.5 14B is strongly recommended as it has excellent Chinese language understanding. Smaller models may produce less accurate or less natural translations.

Command-Line Options

Option	Description	Default
--font	Override default font	Calibri
--model	LLM model to use	qwen2.5:14b
--api-base	OpenAI-compatible API base URL	http://localhost:11434/v1
--api-key	API key (optional, not needed for local models)	None
--batch-size	Text segments per API call	20
--verbose, -v	Enable detailed logging	False

Troubleshooting

Issue	Solution
Connection refused	Check your API endpoint URL. For Ollama: ensure ollama serve is running. For cloud APIs: verify the URL is correct.
Model not found	Verify the model name is correct for your endpoint. For Ollama: ollama pull <model>
Corrupt PPTX	Verify file opens in PowerPoint; try saving as new file first
Font not found	Ensure Calibri is installed on your system
API rate limits	Reduce --batch-size or add delay between calls

Reference

See reference.md for detailed API documentation and architecture.