The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands defined in the project's configuration file for building the application.
Evidence: In scripts/drive.mjs, the start function uses spawnSync(config.build, { cwd: PROJECT_ROOT, stdio: 'inherit', shell: true }) to run the build command specified in drive.config.json.
[DYNAMIC_EXECUTION]: The skill enables the execution of arbitrary JavaScript code in both the renderer and main processes of the Electron application.
Evidence: In scripts/daemon.mjs, the main command handler uses new AsyncFunction('electron', body)(electron) to execute code in the main process, providing full access to Node.js APIs within the app's context. The eval command uses page.evaluate for renderer-side execution.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection as it processes application UI state and logs.
Ingestion points: The snapshot command in scripts/daemon.mjs retrieves the accessibility tree (DOM structure), and the logs command retrieves console output and log files.
Boundary markers: Absent. There are no explicit instructions or delimiters provided to help the agent distinguish between application data and potential malicious instructions embedded in the UI or logs.
Capability inventory: The skill can execute shell commands (spawnSync in drive.mjs) and arbitrary JavaScript in the main process (AsyncFunction in daemon.mjs), which includes file system and network access.
Sanitization: Absent. Content retrieved from the application is passed to the agent without filtering or sanitization.

electron-drive-skill