c-drive-cleaner

Warn

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATIONDATA_EXFILTRATION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill uses the Add-Type cmdlet in _common.ps1 and measure-space.ps1 to dynamically compile C# source code provided in scripts/NativeFileScanner.cs and as embedded string literals. This code interfaces with low-level Win32 APIs via P/Invoke (targeting kernel32.dll) to perform filesystem operations. While this is used for performance, dynamic compilation of local source files can be utilized to execute arbitrary low-level code.
  • [PERSISTENCE]: A Windows Scheduled Task definition is provided in scheduled/weekly-cleanup.xml. This XML allows the cleanup process to be registered to run weekly at the HighestAvailable privilege level. Although presented as an optional automation feature, scheduled tasks are a primary persistence vector.
  • [INDIRECT_PROMPT_INJECTION]: The analysis logic in analyze.ps1 and various scanners (e.g., scan-large-files.ps1) retrieve directory names and file paths from the host system and interpolate them into a Markdown report for the AI agent. These filesystem objects represent untrusted data that is not comprehensively sanitized, which could allow specifically crafted directory names to inject instructions or manipulate the agent's interpretation of the report.
  • [PRIVILEGE_ESCALATION]: Many components of the skill, such as clean-apps.ps1 and scan-admin-deep-accounting.ps1, explicitly check for and request Administrator privileges to perform deep system analysis and file deletion. The provided scheduled task is also configured to run with elevated privileges.
  • [DATA_EXFILTRATION]: The memory/ module (including build-user-profile.ps1 and log-conversation.ps1) systematically collects and stores metadata about the user's local environment, including installed software, IDE configurations, and browser usage, as well as a history of interactions. Although this data is stored locally in JSON files within the skill directory, it represents an extensive collection of environment fingerprinting data.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 10, 2026, 06:35 AM