adding-model-support
Pass
Audited by Gen Agent Trust Hub on May 25, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill workflow involves fetching model configurations (config.json) and implementation source code (modeling_*.py) from HuggingFace, which is a well-known and standard service for machine learning model distribution.- [COMMAND_EXECUTION]: Documentation provides shell command templates for executing conversion, inference, and functional test scripts using standard Python execution environments like
uvandtorch.distributed.run.- [INDIRECT_PROMPT_INJECTION]: The skill requires the agent to ingest and analyze external, user-provided or remote files (model configs and code) to generate bridge logic. This represents a potential surface for indirect prompt injection via the analyzed model assets.- [DATA_EXPOSURE]: Instructions include patterns for dequantizing weights during model import (FP8/FP4 to BF16) to ensure numerical stability and model integrity during the conversion process.
Audit Metadata