translate-site
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external site content (HTML and Markdown) from locale JSON files and content directories to generate translation tasks. This creates a surface where malicious instructions could be embedded in the source text to influence the agent's behavior during the translation phase.
- Ingestion points: The scripts
prepare-translation.mjsandprepare-content-translation.mjsread data from files located inrosey/locales/andsrc/content/. - Boundary markers: The skill lacks explicit delimiters or instructions telling the AI to ignore any embedded commands within the text it is translating, although it does use structured JSON for the task manifest.
- Capability inventory: The skill includes utility scripts that perform file system operations, including reading, writing (
writeFileSync), and deleting (unlinkSync) files within the project directory. - Sanitization: The input content is extracted and passed to the agent without sanitization, filtering, or escaping, relying on the agent's own safety guardrails.
Audit Metadata