The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of the bl command-line interface for managing the entire model fine-tuning lifecycle, including data validation, upload, training job creation, and service deployment.
[CREDENTIALS_UNSAFE]: The documentation references flags for sensitive authentication credentials, including --api-key, --access-key-id, and --access-key-secret. These are standard authentication mechanisms for the platform's CLI, and the skill correctly notes that an API key is required for the pipeline.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill processes external data in the form of training datasets (.jsonl and .zip files) through the bl dataset upload and bl dataset validate commands.
Boundary markers: The instructions do not define specific prompt boundaries for the content of the datasets, but they mandate the use of bl dataset validate as a pre-flight structural check.
Capability inventory: The skill has the capability to read local files, perform network uploads to Alibaba Cloud, and manage remote model training and deployment processes.
Sanitization: Local validation is performed using the bl dataset validate command, which checks records against specific schemas (e.g., chatml, dpo, cpt) before the data is uploaded or used for training.

bailian-finetune