obliteratus
Warn
Audited by Gen Agent Trust Hub on Sep 5, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill clones a repository from an untrusted external source:
https://github.com/elder-plinius/OBLITERATUS.git. This source is not recognized as a trusted organization or well-known service. - [REMOTE_CODE_EXECUTION]: The installation process involves running
pip install -e .on the cloned repository. This allows for the execution of arbitrary code during the installation phase (e.g., viasetup.pyorpyproject.toml) from an unverified source. - [COMMAND_EXECUTION]: The skill relies on a custom CLI tool (
obliteratus) installed from the untrusted repository to perform model surgery, hardware analysis, and model evaluations. This tool has significant access to the local environment and GPU resources. - [INDIRECT_PROMPT_INJECTION]: The skill possesses a large attack surface by ingesting untrusted data (external model weights and YAML configurations) and having high-privilege capabilities (file system writing, CLI execution, and model weight modification). While no specific malicious payload was identified, the ingestion of complex model files combined with broad execution capabilities represents a vulnerability surface.
Audit Metadata