senior-computer-vision
Warn
Audited by Gen Agent Trust Hub on Apr 2, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The script
inference_optimizer.pyutilizestorch.load()to process model files. This function uses thepicklemodule internally, which can be exploited to execute arbitrary code if a maliciously crafted model file is loaded by the user.\n- [DATA_EXFILTRATION]: Thedataset_pipeline_builder.pyscript parses Pascal VOC annotations using the standardxml.etree.ElementTreelibrary. This parser is vulnerable to XML External Entity (XXE) attacks, allowing an attacker to potentially read local files or conduct server-side request forgery (SSRF) via malicious XML input.\n- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection through its data ingestion workflows.\n - Ingestion points: Reading and parsing XML, JSON, and YAML files in
dataset_pipeline_builder.pyandvision_model_trainer.py.\n - Boundary markers: No delimiters or instructions are used to separate untrusted data from processing logic.\n
- Capability inventory: The skill has the ability to write files, suggest shell command execution, and contains an unsafe deserialization path through
torch.load.\n - Sanitization: The scripts use standard parsers without security hardening or validation of the data content.
Audit Metadata