double6-pdf-translation
Pass
Audited by Gen Agent Trust Hub on Oct 19, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill manages complex workflows by executing local binaries such as
xelatex,lualatex, andpdflatexfor document compilation, andpdftotextfor text extraction. These commands are executed within a monitoring framework (scripts/bounded_execution.py) that enforces CPU and memory limits. - [EXTERNAL_DOWNLOADS]: The skill interacts with external LLM providers (OpenAI-compatible and DeepSeek) for translation services. It also includes utility scripts (
scripts/external_link_check.py) that usecurlto verify the availability of URLs found within the documents being processed. - [DYNAMIC_EXECUTION]: Several components use dynamic module loading (
__import__) and script execution (runpy.run_module) to manage internal project paths and interact with the vendored translation backends (BabelDOC and PDFMathTranslate-next). - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text from input PDF and LaTeX files and incorporates it into LLM prompts. To mitigate risks, it utilizes a sophisticated tagging system (PTPROTECTED, PTC, PTS markers) to isolate structural elements and includes multiple QA gates to verify the integrity and modality of the translated content.
Audit Metadata