ssh-manager

Warn

Audited by Gen Agent Trust Hub on Mar 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides scripts (connect.sh, exec.sh, tunnel.sh) that use the local shell to launch SSH sessions and execute commands on remote systems.
  • [COMMAND_EXECUTION]: All SSH implementations in the skill explicitly disable host key verification using -o StrictHostKeyChecking=no and -o UserKnownHostsFile=/dev/null. This bypasses a fundamental security feature of SSH designed to prevent Man-in-the-Middle (MITM) attacks.
  • [REMOTE_CODE_EXECUTION]: The skill includes a dedicated script (exec.sh) and instructions for executing arbitrary commands on remote hosts, which constitutes a significant remote code execution surface.
  • [DATA_EXFILTRATION]: The skill discloses internal network configuration details, including specific Tailscale IP addresses and hostnames, exposing the structure of the user's private infrastructure.
  • [DATA_EXFILTRATION]: The documentation references sensitive file paths such as ~/.ssh/config and ~/.ssh/known_hosts, which contain metadata about existing remote connections and trust relationships.
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface.
  • Ingestion points: Data is ingested from tailscale status output across SKILL.md, connect.sh, and check-host.sh.
  • Boundary markers: None present; the agent parses the command output directly without explicit delimiters or instructions to ignore embedded content.
  • Capability inventory: Includes local subprocess execution (ping, tailscale, lsof) and remote execution via SSH (connect.sh, exec.sh).
  • Sanitization: No input validation or escaping is performed on hostnames or IP addresses retrieved from external commands before they are used in shell commands.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 17, 2026, 01:18 AM