c-drive-cleaner

Warn

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPERSISTENCEPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill constructs and executes command-line strings for system tools using variable interpolation.
  • File _common.ps1: The Remove-Directory function starts a command process using cmd /c rmdir /s /q "$Path".
  • File _common.ps1: The Get-PathLogicalMeasurement function uses robocopy $Path $probe /L to calculate sizes.
  • File clean-deep.ps1: Executes various system binaries including Dism, vssadmin, and powercfg.exe with user-supplied or calculated arguments.
  • [PERSISTENCE]: The skill establishes persistent execution mechanisms on the host system to perform recurring tasks.
  • File scheduled/weekly-cleanup.xml: Defines a Windows Scheduled Task named C-Drive-Weekly-Cleanup that runs every Sunday at 02:00 with highest available privileges.
  • File scheduled/daily-monitor.ps1: Implements a background monitoring script designed for recurring disk capacity checks.
  • [PRIVILEGE_ESCALATION]: The skill requires and leverages elevated permissions to perform core functions.
  • File analyze.ps1 and cleaners/clean-apps.ps1: Include logic to detect administrative privileges and instruct the user to relaunch with elevated rights.
  • File scheduled/weekly-cleanup.xml: Explicitly sets the task run level to HighestAvailable and invokes PowerShell with -ExecutionPolicy Bypass to circumvent script execution restrictions.
  • [DYNAMIC_EXECUTION]: The skill uses runtime compilation to execute high-performance file system operations.
  • File _common.ps1: The Initialize-NativeFileScanner function uses Add-Type -TypeDefinition to compile the C# source code found in scripts/NativeFileScanner.cs during execution.
  • File _common.ps1: The Initialize-NtfsAllocationProbe function contains an additional inline C# definition compiled at runtime to probe NTFS-specific file attributes via P/Invoke.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes arbitrary metadata from the host filesystem which serves as an ingestion surface for potentially malicious content.
  • Ingestion points: scanners/scan-duplicate-runtimes.ps1 (reads product names from package.json files), scanners/scan-unused-software.ps1 (reads DisplayName and Publisher strings from the Windows Registry), memory/log-conversation.ps1 (stores user dialogue history in local JSON files).
  • Boundary markers: Absent; untrusted data is not explicitly delimited when processed.
  • Capability inventory: High; the skill can delete directories (Remove-Directory) and files (Remove-SafeFile) based on its analysis logic.
  • Sanitization: Absent; the skill relies on standard JSON parsing (ConvertFrom-Json) but does not sanitize the semantic content of the metadata before using it to generate recommendations.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 5, 2026, 03:13 AM