notis-desktop-use
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill requires the agent to switch from a sandboxed environment to the
local_shellto execute thepeekabooCLI. This allows the agent to interact directly with the user's host operating system, bypassing standard sandbox restrictions. - Evidence: Instructions to use
set_shell_modewithmode: "local_shell"and the use of theLOCAL_NOTIS_RUN_DESKTOP_SHELLtool for cloud-based agents. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves reading UI elements, static labels, and text from native applications, which constitutes a significant surface for indirect prompt injection if an application displays malicious instructions.
- Ingestion points: The
peekaboo seeandpeekaboo see --treecommands inSKILL.mdandreferences/command-and-capture.mdingest arbitrary UI text. - Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potentially malicious content found within the UI data.
- Capability inventory: The agent has extensive capabilities including clicking, typing, and window management (
peekaboo click,peekaboo type,peekaboo app). - Sanitization: No sanitization or validation of the ingested UI text is performed before processing.
Audit Metadata