ascend-transformer-boost
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes automated hooks (PreToolUse, PostToolUse, and Stop) to execute local shell commands like grep and echo. These commands perform sanity checks, such as verifying environment variables like CANN_PATH and ATB_REPO_PATH, and monitoring workflow progress through internal plan files. These are benign utility operations for workflow management.
- [DYNAMIC_EXECUTION]: The workflow directs the agent to compile local source code and run tests using standard Bash commands (e.g., script-based builds for the test framework). These activities are essential for NPU-based software development and rely on files present within the user's local repository.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests documentation links to help generate design specifications. To mitigate the risk of untrusted content influencing agent behavior, the skill enforces mandatory 'Human-in-the-Loop' (HIL) gates, requiring explicit user approval of designs and test cases before proceeding to any automated implementation steps.
Audit Metadata