The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands to set up the environment and run training scripts.
Evidence: bash test/train_performance_8p_base_fp32.sh is executed to start the training process.
Evidence: Uses git clone, git apply, ln -sf, and mkdir for filesystem and repository management.
[EXTERNAL_DOWNLOADS]: The skill downloads code and resources from external Git repositories.
Evidence: Clones the Ascend DrivingSDK from https://gitcode.com/Ascend/DrivingSDK (vendor resource).
Evidence: Clones the BEVFormer repository from https://github.com/fundamentalvision/BEVFormer.git.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where instructions could be embedded in the external repositories it clones or the user-provided paths it processes.
Ingestion points: External code cloned via git clone from GitHub/GitCode and user-supplied filesystem paths for datasets and weights.
Boundary markers: None identified; external content is processed directly.
Capability inventory: File writing (mkdir, cp), symbolic link creation (ln), and shell script execution (bash).
Sanitization: No specific sanitization or validation of the cloned code or user paths is performed before execution.

model-migration