translate
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content from external translation resources, creating a surface for indirect prompt injection attacks.
- Ingestion points: The agent reads local translation catalogs (e.g., PO, JSON, XML) and downloads external glossary, translation memory (TM), and style guide files into the
.crowdin/resources/directory (Step 3). - Boundary markers: The instructions do not provide explicit delimiters for external content, although they require strict mechanical matching for placeholders and structural components to prevent accidental instruction execution.
- Capability inventory: The skill can execute shell commands (
crowdin,lingui,jq,git), modify local project resource files (Step 4), and perform network uploads to the Crowdin platform (Step 6). - Sanitization: Significant sanitization is implemented via
scripts/check-drafts.sh, which usesjqto validate that placeholders and plural categories remain byte-identical and consistent with target language rules. It also utilizes standard format validators (msgfmt,xmllint,plutil) and requires user confirmation before performing large operations or uploads to the Crowdin service.
Audit Metadata