The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes project content (JSON, MD, MDX) which could contain adversarial instructions targeting the AI agent during the translation phase.
Ingestion points: scripts/prepare-translation.mjs reads locale strings from rosey/locales/, and scripts/prepare-content-translation.mjs extracts frontmatter and body text from src/content/ files.
Boundary markers: The task manifests (.translation-task-*.json) do not use specific boundary markers or 'ignore' instructions for the content data being translated.
Capability inventory: The skill possesses file-write capabilities via scripts/merge-translation.mjs and scripts/merge-content-translation.mjs which update the project's source files.
Sanitization: While scripts/merge-translation.mjs performs basic HTML tag count validation to ensure structural integrity, there is no sanitization or filtering of the text content to prevent the agent from executing instructions embedded in the source strings.

translate-site