Rosetta — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Rosetta (Agent Skill) and scored it 65/100 (yellow). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 11 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 11 flagged
The text {match} is the classic direct prompt-injection phrasing. Placed in a skill body that the agent reads as trusted instructions, it tries to make the agent abandon its prior rules and follow whatever comes next — a full system-prompt override.
ignore/disregard/forget … previous instructions sentence.The text {match} is the classic direct prompt-injection phrasing. Placed in a skill body that the agent reads as trusted instructions, it tries to make the agent abandon its prior rules and follow whatever comes next — a full system-prompt override.
ignore/disregard/forget … previous instructions sentence.The text {match} is the classic direct prompt-injection phrasing. Placed in a skill body that the agent reads as trusted instructions, it tries to make the agent abandon its prior rules and follow whatever comes next — a full system-prompt override.
ignore/disregard/forget … previous instructions sentence.The text {match} is the classic direct prompt-injection phrasing. Placed in a skill body that the agent reads as trusted instructions, it tries to make the agent abandon its prior rules and follow whatever comes next — a full system-prompt override.
ignore/disregard/forget … previous instructions sentence.The text {match} is the classic direct prompt-injection phrasing. Placed in a skill body that the agent reads as trusted instructions, it tries to make the agent abandon its prior rules and follow whatever comes next — a full system-prompt override.
ignore/disregard/forget … previous instructions sentence.Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
AI-powered Excel translation CLI. Translates Excel files while preserving formatting, formulas, and data integrity.
Rosetta translates all text in your Excel files using Claude AI, without breaking:
You need a Claude API key from Anthropic.
sk-ant-...)Note: API usage is billed by Anthropic. See anthropic.com/pricing for current rates. Translating a typical Excel file costs a few cents.
Install from PyPI:
pip install rosetta-xlThen set your API key:
# Linux/macOS
export ANTHROPIC_API_KEY=sk-ant-your-key-here
# Windows (Command Prompt)
set ANTHROPIC_API_KEY=sk-ant-your-key-here
# Windows (PowerShell)
$env:ANTHROPIC_API_KEY="sk-ant-your-key-here"Or create a .env file in your working directory:
ANTHROPIC_API_KEY=sk-ant-your-key-here# Translate to French
rosetta input.xlsx -t french
# Translate to Spanish with custom output name
rosetta input.xlsx -t spanish -o translated.xlsx
# Specify source language (auto-detected by default)
rosetta input.xlsx -s english -t german
# Translate only specific sheets
rosetta input.xlsx -t french --sheets "Sheet1" --sheets "Data"
# Add context for better translations (e.g., domain-specific terms)
rosetta input.xlsx -t french -c "Medical terminology document"| Option | Short | Description |
|---|---|---|
--target-lang | -t | Target language (required) |
--source-lang | -s | Source language (auto-detect if omitted) |
--output | -o | Output file path (default: input_translated.xlsx) |
--sheets | Sheets to translate (can repeat, default: all) | |
--context | -c | Domain context for better accuracy |
--batch-size | -b | Cells per API call (default: 50) |
Translate a price list to multiple languages:
rosetta prices.xlsx -t french -o prices_fr.xlsx
rosetta prices.xlsx -t german -o prices_de.xlsx
rosetta prices.xlsx -t spanish -o prices_es.xlsxTranslate a medical form with context:
rosetta patient_form.xlsx -t french -c "Medical intake form with clinical terminology"Translate only the "Questions" sheet:
rosetta survey.xlsx -t japanese --sheets "Questions""ANTHROPIC_API_KEY not set"
export ANTHROPIC_API_KEY=sk-ant-....env file with the key"Invalid API key"
sk-ant-"Rate limit exceeded"
rosetta input.xlsx -t french -b 20Your original file is never modified.
Rosetta also includes a web application and REST API for browser-based translations.
# Install with uv (recommended)
uv sync
# Start the server
uv run uvicorn rosetta.api:app --reload
# Or with pip
pip install -e .
uvicorn rosetta.api:app --reloadThe API runs at http://localhost:8000 by default.
cd frontend
npm install
npm run devThe frontend runs at http://localhost:5173 and connects to the API.
| Endpoint | Method | Description |
|---|---|---|
/translate | POST | Translate an Excel file (returns file) |
/translate-stream | POST | Translate with real-time progress via SSE |
/estimate | POST | Get cell count and cost estimate |
/sheets | POST | List sheet names in a file |
/count | POST | Count translatable cells |
/preview | POST | Preview cells that will be translated |
/health | GET | Health check |
The /translate-stream endpoint uses Server-Sent Events (SSE) to stream translation progress in real-time. The frontend automatically falls back to the standard /translate endpoint on networks that don't support SSE (e.g., corporate proxies).
Rosetta includes MCP (Model Context Protocol) servers for both Claude Desktop and Claude Web.
Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):
{
"mcpServers": {
"rosetta": {
"command": "uv",
"args": ["run", "--directory", "/path/to/rosetta", "python", "-m", "rosetta.api.mcp"],
"env": {
"ANTHROPIC_API_KEY": "your-key-here"
}
}
}
}Usage: Use local file paths for best results.
Translate ~/Downloads/report.xlsx to FrenchNote: Claude.ai does not yet support custom MCP servers in the browser (as of January 2026).
Current recommendation: Use the web app for browser-based translations.
See MCP_USAGE.md for detailed instructions.
MIT
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.