windows-rescue-from-linux
Fail
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads software and configuration scripts from external repositories, including NodeSource (deb.nodesource.com) and GitHub (raw.githubusercontent.com/nvm-sh/nvm). These are well-known services used to set up the execution environment and install the required rescue tools.
- [REMOTE_CODE_EXECUTION]: Several scripts, such as
scripts/install-rescue-tools.shand the instructions inreferences/00-rescue-usb-preparation.md, download installation scripts for Node.js and NVM and pipe them directly into the shell for execution (curl | bash). While these sources are legitimate and standard for these tools, this pattern involves executing code fetched from a remote server. - [COMMAND_EXECUTION]: The skill makes extensive use of
sudoto perform high-privilege system tasks. These include mounting and unmounting block devices, interacting with EFI variables viaefibootmgr, and modifying Windows system files and Registry hives. These actions are necessary and expected for a tool designed for low-level system repair. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local environment and has access to high-privilege system capabilities, creating a surface for indirect prompt injection.
- Ingestion points: The skill reads Windows Event Logs (.evtx) and Registry Hives (SAM, SYSTEM, SOFTWARE, NTUSER.DAT) using utilities like
python-evtx,chntpw, andhivexshas documented inscripts/boot-diagnostic.shandscripts/malware-quick-scan.sh. - Boundary markers: The skill does not employ explicit delimiters or instructions to the agent to ignore potentially malicious commands embedded within the system data it processes.
- Capability inventory: The skill possesses the ability to execute shell commands with root privileges, modify the Windows Registry, and delete or move system files.
- Sanitization: The scripts use basic text filtering (e.g., grep) for reporting but lack specific validation or sanitization logic to handle adversarial content within the ingested log files.
Recommendations
- HIGH: Downloads and executes remote code from: https://deb.nodesource.com/setup_20.x - DO NOT USE without thorough review
Audit Metadata