readme-i18n
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting untrusted data from
README.mdfiles, which could contain adversarial instructions designed to influence the agent's behavior during the translation process. - Ingestion points: The skill reads the root
README.mdand any existing localized siblings (e.g.,README.zh.md) as source material for translation and structural auditing. - Boundary markers: The instructions lack explicit delimiters or sandboxing prompts to wrap the untrusted content being translated, relying instead on natural language guidance to differentiate between prose and code.
- Capability inventory: The skill is authorized to perform file-write operations to create new localized variants and update existing README files with language selectors.
- Sanitization: The instructions include a comprehensive "Preservation Checklist" and specific "Do not translate" rules for shell commands, environment variables (e.g.,
OPENAI_API_KEY), and code fences, which provides a level of protection against malicious payloads embedded in technical snippets.
Audit Metadata