The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to configure the environment via set_env.sh, perform automated testing using pytest, and run the resulting precision report scripts. These actions are fundamental to its operation as a testing utility.
[DYNAMIC_EXECUTION]: The skill implements a template-based code generation workflow, creating Python test scripts that are subsequently executed. These scripts contain logic to dynamically locate and load compiled shared object libraries (.so) using torch.ops.load_library, which is standard for testing custom NPU kernels.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from an external file, csrc/ops/<op_name>/test/<op_name>-test-cases.md, to determine test parameters and code expressions.
Ingestion points: Technical specifications, including NPU call expressions and supported data types, are read from the markdown file located in csrc/ops/<op_name>/test/ during Phase 1.
Boundary markers: The skill does not employ explicit boundary markers or "ignore instructions" delimiters when extracting content from the test-cases file for code generation.
Capability inventory: The skill possesses capabilities for file system modification, shell command execution (pytest, python), and dynamic loading of shared libraries.
Sanitization: There is no evidence of sanitization or validation of the technical strings extracted from the markdown file before they are interpolated into the executable Python scripts.

ascendc-operator-precision-eval