The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill architecture relies on ingesting translation data from external JSON files which are then parsed and rendered. While the instructions explicitly warn about potential injection risks in ICU parsers and recommend sanitization, the provided code snippets do not implement specific sanitization logic. This creates a potential surface for indirect prompt injection if translation files are sourced from untrusted environments.
Ingestion points: Translation files loaded in SKILL.md (Step 4, Step 5) and references/i18n-architecture.md (Node.js/TypeScript Setup).
Boundary markers: None present in the provided code templates to delimit untrusted translation content.
Capability inventory: The skill generates code for API message delivery, server-side rendering, and automated translation workflows.
Sanitization: The 'Security' section in SKILL.md identifies the risk and mandates validation/sanitization, but the implementation is left to the developer.
[COMMAND_EXECUTION]: The skill provides examples for using standard development tools including npx i18next-scanner and the crowdin CLI. These commands are integral to the described i18n workflow and are used for extracting and synchronizing translation keys.
[EXTERNAL_DOWNLOADS]: The skill references and provides integration patterns for well-known translation management platforms such as Crowdin and Lokalise. These are standard industry services and the references are informative.

backend-internationalization