The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install the open-computer-use package globally from the NPM registry and fetches additional skill configurations from the GitHub repository iFurySt/open-codex-computer-use.
[COMMAND_EXECUTION]: Provides tools to simulate user input, including click_element, type_text, and press_key. These tools allow an AI agent to programmatically control applications and the host operating system.
[DATA_EXFILTRATION]: The take_screenshot and get_app_state tools enable the agent to capture and read everything displayed on the screen, including potentially sensitive information in open applications (e.g., passwords, private messages, or financial data).
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from the user interface of third-party applications. Content displayed in an app (e.g., a browser or document) could contain instructions that influence the agent's behavior.
Ingestion points: UI element hierarchy and attributes (titles, values) processed by get_app_state in SKILL.md.
Boundary markers: None identified in the provided documentation or tool definitions.
Capability inventory: Full UI control (clicks, typing, keyboard shortcuts) and screen capture capabilities as defined in SKILL.md.
Sanitization: No explicit sanitization or filtering of the UI content is described.

open-computer-use-automation