dwg-extraction
Fail
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill clones source code from a remote repository (https://github.com/LibreDWG/libredwg.git) within scripts/compile_libredwg.sh and SKILL.md to obtain the libredwg library.\n- [REMOTE_CODE_EXECUTION]: By downloading external code, compiling it, and then running the produced dwg2dxf binary, the skill establishes a path for remote code execution from a source not recognized as trusted.\n- [DYNAMIC_EXECUTION]: The skill performs runtime compilation of C source files using gcc in scripts/compile_libredwg.sh and SKILL.md. This includes manual creation of config.h and linking multiple object files to create a functional executable on the fly.\n- [COMMAND_EXECUTION]: Executes shell commands to perform environment setup, repository cloning, and code compilation.\n- [INDIRECT_PROMPT_INJECTION]:\n
- Ingestion points: Processes external .dwg and .dxf files in scripts/parse_dxf.py and the dwg2dxf binary.\n
- Boundary markers: Lacks explicit markers to distinguish between CAD data and potential embedded instructions.\n
- Capability inventory: Possesses the ability to execute shell commands, compile code, and read/write to the file system.\n
- Sanitization: While the Python script uses basic error handling for file reading, there is no robust validation to prevent exploitation of the parsing logic by malicious CAD files.
Recommendations
- AI detected serious security threats
Audit Metadata