The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically generates and executes PowerShell and Node.js scripts to perform automation tasks. In scripts/desktop_control.py, the run_ps_script function executes generated PowerShell content using subprocess.run. In scripts/playwright_control.py, the run_node_script function executes generated JavaScript content using subprocess.run.
[REMOTE_CODE_EXECUTION]: The skill constructs and executes code based on parameters provided by the agent. This includes executing JavaScript within a browser environment via page.evaluate and running Node.js scripts with agent-supplied selectors and logic, which presents a risk of arbitrary code execution if the agent's instructions are compromised.
[PROMPT_INJECTION]: The skill presents a significant attack surface for indirect prompt injection as it processes data from untrusted sources.
Ingestion points: The agent "sees" the state of applications through screenshots and DOM inspection (as seen in scripts/playwright_control.py and scripts/desktop_control.py).
Boundary markers: There are no explicit instructions or delimiters used when the agent processes application content to distinguish between data and potential malicious commands.
Capability inventory: The skill allows for comprehensive system control, including keystroke simulation (SendKeys), mouse interaction (ClickAt), and clipboard manipulation.
Sanitization: The skill performs basic sanitization by converting text to Unicode code points to avoid shell escaping issues, but it does not validate the intent or safety of the commands being performed.
[EXTERNAL_DOWNLOADS]: The skill downloads and installs the Playwright library and its associated browser binaries (e.g., Chromium) from Microsoft's official registries to enable web automation features.

desktop-controller