catlass-operator-dev
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads the Catlass library from a repository on GitCode (
https://gitcode.com/cann/catlass.git). This is a legitimate vendor resource associated with the Ascend (CANN) ecosystem. - [COMMAND_EXECUTION]: The skill involves the execution of various shell commands to manage the environment and build process, including
git clone,source set_env.sh,conda activate, and executing a localbuild.shscript. - [DYNAMIC_EXECUTION]: The workflow involves compiling C++ source code via CMake (Phase 3) and installing the resulting Python wheel package using
pip install. This is standard practice for developing hardware-accelerated kernels. - [INDIRECT_PROMPT_INJECTION]: The skill exhibits a potential vulnerability surface by reading and processing local project files which could be influenced by external input.
- Ingestion points: The skill reads contents from
design.md, source code files (register.cpp,ops.h,op_host), and tool-generated performance logs (op_statistic.csv). - Boundary markers: No explicit boundary markers or "ignore instructions" delimiters are defined for the processed file content within this orchestrator skill.
- Capability inventory: The agent has capabilities to execute shell scripts (
build.sh), install packages (pip install), and perform network operations (git clone). - Sanitization: There is no explicit sanitization or validation logic described for the content extracted from the design or source files before it is used to influence subsequent agent actions.
Audit Metadata