bdistill-knowledge-extraction
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONMETADATA_POISONING
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill requires installing the
bdistillpackage via pip and adding a third-party MCP tool. These components are external to the skill and the platform, introducing a dependency on unverified software. - [METADATA_POISONING]: The author listed in the skill metadata ('FrancyJGLisboa') does not match the system-attributed author ('sickn33'), which may indicate deceptive or inconsistent metadata.
- [INDIRECT_PROMPT_INJECTION]: The skill processes data generated by other AI models, which could contain malicious instructions intended to manipulate the agent during the distillation process.
- Ingestion points: Data is sourced from model responses in subscription sessions or local Ollama instances.
- Boundary markers: There are no specific delimiters or instructions provided to the agent to distinguish source data from executable instructions.
- Capability inventory: The agent uses CLI tools and shell-based MCP commands to manage and export distilled data.
- Sanitization: The documentation provides no evidence of data sanitization or validation of the model-generated content.
Audit Metadata