remote
Fail
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill stores server passwords in plaintext within the
~/.local/share/remote/servers.jsonfile, exposing sensitive credentials to any process with read access to the local state directory. Additionally,remote.shcreates a temporary executable script containing the server password to facilitateSSH_ASKPASSoperations. The skill also disables host key verification (StrictHostKeyChecking=no), which increases vulnerability to man-in-the-middle attacks. - [PRIVILEGE_ESCALATION]: The
setup.shscript utilizessudoto install thesshpassdependency on Linux systems through package managers such asapt-get,dnf, oryum. - [EXTERNAL_DOWNLOADS]: On macOS, the skill attempts to install
sshpassfrom a third-party Homebrew tap (hudochenkov/sshpass/sshpass), which is an untrusted external repository. - [COMMAND_EXECUTION]: The skill provides the ability to execute arbitrary commands on remote servers via SSH, using command strings provided by the user or agent at runtime.
- [INDIRECT_PROMPT_INJECTION]: The skill presents a vulnerability surface for indirect prompt injection. 1. Ingestion points: Remote command arguments and outputs processed in
remote.shandremote.ps1. 2. Boundary markers: Absent. 3. Capability inventory: SSH network access, sudo command execution, and local file writes. 4. Sanitization: Limited to base64 encoding of command strings for local script-to-script transport.
Recommendations
- AI detected serious security threats
Audit Metadata