senior-computer-vision
Warn
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The
scripts/inference_optimizer.pyutility utilizestorch.load()to ingest PyTorch models for analysis and benchmarking. This function is built on thepickleserialization format, which is fundamentally insecure as it allows for the execution of arbitrary code during the loading process if the source file is untrusted. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external, potentially untrusted data including image datasets and their corresponding annotations in multiple formats.
- Ingestion points: Untrusted content enters the agent's context through
scripts/dataset_pipeline_builder.py(which parses JSON, XML, and TXT annotation files) andscripts/inference_optimizer.py(which loads model files). - Boundary markers: The instructions lack boundary markers or explicit safety guidelines to prevent the agent from being influenced by adversarial content hidden within the processed files.
- Capability inventory: The skill possesses significant system-level capabilities, such as directory deletion (
shutil.rmtree), file copying (shutil.copy2), and the execution of complex machine learning framework operations. - Sanitization: The implementation focuses on format compliance but does not perform sanitization of the values or metadata within the ingested data to mitigate potential injection attacks.
Audit Metadata