rclone-cli
Fail
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: CRITICALCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The documentation details several high-risk commands such as
sync,purge,delete, andmovewhich can irreversibly modify or delete large amounts of data across cloud storage providers. The skill instructions attempt to mitigate this risk by requiring the agent to use--dry-runand--interactiveflags for these operations and by stubbing the installer to prevent unauthorized binary deployment. - [INDIRECT_PROMPT_INJECTION]: The skill involves commands that list remote files and retrieve metadata (e.g.,
lsjson,lsf), which introduces a surface where untrusted data from a remote storage provider enters the agent's context. - Ingestion points: Remote file listings and metadata retrieval documented in
references/commands/rclone_lsjson.mdandrclone_ls.md. - Capability inventory: The agent has the capability to execute the
rcloneCLI, which can perform destructive actions or data transfers based on listed content. - Boundary markers: High-level warnings in
SKILL.mdemphasize caution but do not provide specific technical delimiters for handling remote content, and there are no explicit instructions for the agent to sanitize these inputs before further processing. - [EXTERNAL_DOWNLOADS]: The documentation refers to numerous cloud storage domains used as examples for configuration. Automated security scans flagged domains such as
linodeobjects.com,wasabisys.com, andliara.spacedue to their presence on reputation-based blacklists. However, these are established cloud service providers used for object storage, and they appear in the skill solely as configuration examples (e.g., inreferences/providers/filescom.md). - [DATA_EXFILTRATION]: Rclone is a tool designed for data transfer. The documentation explains how to read and transfer files and configuration data (e.g.,
rclone.conf) to remote destinations. While the skill includes warnings against committing private keys or tokens, the inherent capability of the tool facilitates data exfiltration if the agent is directed to transfer sensitive local paths to a user-controlled or attacker-controlled remote.
Recommendations
- Contains 5 malicious URL(s) - DO NOT USE
Audit Metadata