winapp-ui-automation
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- UI Interaction and Input Injection: The skill provides tools like
invoke,click,send-keys, andset-valueto simulate user input. This allows an agent to automate workflows within Windows applications, including the ability to fire system-level hotkeys via the--allow-system-keysflag when using thesend-inputtransport. These features are standard for UI automation tools and allow the agent to drive application logic as intended. - Application Data and Visual Capture: The
screenshot,record, andget-valuecommands enable the agent to capture the visual state of the desktop or read text content from UI elements. Thescreenshotandrecordtools include a--capture-screenoption that can capture overlays and popups that are not part of the primary application window, which is helpful for full-context automation but allows the agent to see all content currently displayed on the desktop. - Indirect Prompt Injection Surface: The skill exposes a potential surface for indirect prompt injection due to how it processes data from external applications.
- Ingestion points: The agent receives untrusted data from the UI of running applications through the
inspect,search,get-value,get-property, andwait-forcommands inSKILL.md. - Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are documented for the content read from UI elements.
- Capability inventory: The skill allows the agent to perform actions based on this data using the
invoke,click,send-keys,set-value,drag,touch, andpencommands inSKILL.md. - Sanitization: The documentation does not specify any automated sanitization, escaping, or filtering of the content retrieved from application UI elements before it is processed by the agent.
- Sandbox Environment Support: The skill includes explicit support for running and automating applications within an isolated Windows Sandbox using the
--on sandboxflag. This allows for testing and automation in a controlled environment, which can mitigate risks to the host system during development.
Audit Metadata