The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of external dataset configuration files (e.g., data.yaml). This presents an indirect prompt injection surface where a malicious configuration file could attempt to influence the agent's behavior during the training setup.
Ingestion points: Dataset configuration files (data.yaml) referenced in SKILL.md.
Boundary markers: Absent; the instructions do not specify any delimiters or warnings for the agent when parsing user-provided configuration data.
Capability inventory: Execution of shell commands (yolo) and Python library calls (model.train()).
Sanitization: Not explicitly addressed in the provided training instructions.
[DYNAMIC_EXECUTION]: Multi-GPU training using Distributed Data Parallel (DDP) involves the generation of scripts and the use of cloudpickle to serialize and reconstruct the training state (trainer, model, and callbacks) across processes. This dynamic code generation and serialization is a standard and necessary component for distributed training in the PyTorch ecosystem.
[COMMAND_EXECUTION]: The skill provides numerous examples of shell commands using the yolo CLI for tasks such as training (yolo detect train) and validation (yolo val). These commands are standard operations for the tool and are executed within the user's environment.
[EXTERNAL_DOWNLOADS]: The skill utilizes the ul:// URI scheme to fetch datasets from the Ultralytics Platform and downloads pretrained model weights (e.g., yolo26n.pt) from official Ultralytics sources. It also references documentation on the ultralytics.com domain. These are all recognized as legitimate vendor resources associated with the skill's author.

yolo-training