pikvm
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exposes powerful Human Interface Device (HID) capabilities, including arbitrary keyboard typing and mouse movement on a remote host. If the agent processes untrusted data (e.g., from a website or email) while this skill is active, an attacker could attempt to inject commands that the agent then translates into physical keystrokes on the managed server.\n
- Ingestion points: Any data processed by the agent which may influence the use of PiKVM commands.\n
- Boundary markers: Absent; there are no instructions to ignore embedded commands in processed text.\n
- Capability inventory: Full keyboard emulation (send_key, print), mouse control (send_mouse_move, send_mouse_button), and power state management (atx/power).\n
- Sanitization: Absent; the skill directly pipes input strings into the curl command data payloads.\n- [COMMAND_EXECUTION]: The provided commands use the
curl -k(or--insecure) flag. This allows the agent to communicate with the PiKVM API without verifying SSL certificates, making the connection vulnerable to Man-in-the-Middle (MitM) attacks where authentication tokens could be intercepted.
Audit Metadata