bailian-finetune
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided datasets which may contain malicious instructions that could manipulate the agent's behavior during the fine-tuning workflow.\n
- Ingestion points: The skill ingests local files via
bl dataset upload --file(reference/dataset.md) andbl finetune create --datasets(reference/finetune.md).\n - Boundary markers: There are no documented instructions for the agent to use specific delimiters or to ignore instructions embedded within the training data content.\n
- Capability inventory: The
blCLI is used to read local files, perform network operations to Aliyun services, and initiate model training and deployment tasks.\n - Sanitization: While the skill performs structural validation (e.g.,
bl dataset validate), it does not specify sanitization or filtering of the natural language content within the datasets to prevent prompt injection.
Audit Metadata