The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface where untrusted data from the renderer process is ingested into the main process via IPC handlers (ipcMain.handle, ipcMain.on). This data could potentially influence downstream logic or prompts if not properly sanitized. Ingestion points: Found in SKILL.md (Pattern 4) and examples/core.md (Typed Main Process Handlers). Boundary markers: The skill mandates the use of contextBridge and typed wrappers to restrict API exposure. Capability inventory: Includes file system access (fs.readFile, fs.writeFile) and utility process management (utilityProcess.fork) in examples/core.md and examples/message-ports.md. Sanitization: Patterns emphasize runtime validation using path.resolve boundary checks and zod schema validation in examples/electron-trpc.md.
[COMMAND_EXECUTION]: The main process handlers are designed to perform operations like reading/writing files and forking processes based on renderer requests. While the skill explicitly teaches safe implementation via whitelisting and validation, the underlying capability to execute operations based on external input is present.

desktop-ipc-electron