The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides tools that allow for arbitrary code and command execution within the context of the Tauri application.
webview_execute_js: Executes arbitrary JavaScript strings within the application's webview.
ipc_execute_command: Allows the agent to invoke backend Tauri commands directly.
webview_interact and webview_keyboard: Enable programmatic interaction with the UI, including clicking and typing.
[EXTERNAL_DOWNLOADS]: The installation process involves downloading and executing packages from external registries.
Uses npx -y install-mcp to download and run @hypothesi/tauri-mcp-server.
Recommends global installation of @hypothesi/tauri-mcp-cli via npm.
Requires adding tauri-plugin-mcp-bridge as a Rust dependency from crates.io.
[DATA_EXFILTRATION]: Multiple tools provide access to potentially sensitive data that could be exfiltrated from the user's environment or application.
read_logs: Can stream console logs, system logs, Android logcat, and iOS logs, which frequently contain sensitive tokens, PII, or internal application state.
webview_screenshot: Captures the visual state of the application, which may display private user information.
ipc_get_captured: Intercepts and reads IPC traffic between the frontend and backend, potentially exposing sensitive data payloads.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8).
Ingestion points: Reads untrusted data from read_logs (console/system logs), webview_dom_snapshot (UI text), and ipc_get_captured (external IPC events).
Boundary markers: None observed in the instructions to help the agent distinguish between data and commands.
Capability inventory: Significant capabilities including webview_execute_js, ipc_execute_command, and file-writing via CLI screenshot tools.
Sanitization: No explicit sanitization of log or UI content is mentioned before the agent processes it.

tauri-mcp-server-development