The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): In references/ipc-patterns.md, the skill provides IPC handlers for fs:readTextFile and fs:writeTextFile that execute file system operations based on parameters provided by the renderer process. The example code lacks any path validation or sanitization, allowing a compromised renderer or a malicious prompt to read or write arbitrary files on the host system.
[DATA_EXFILTRATION] (HIGH): The fs:readTextFile capability, as documented, allows the retrieval of any file content. When used in an agentic context, this allows an agent to access sensitive configuration files (~/.ssh, .env) and potentially exfiltrate them through other available tools or its own output.
[INDIRECT_PROMPT_INJECTION] (HIGH): The skill exposes an attack surface where untrusted data (renderer-side content) can influence file system write operations.
Ingestion points: IPC handlers fs:writeTextFile and save-file in references/ipc-patterns.md.
Boundary markers: Absent in the provided code snippets.
Capability inventory: fs.promises.writeFile, fs.writeFileSync, fs.promises.readFile.
Sanitization: None. The validateSender function checks the source frame but does not validate the content or the target file path.
[EXTERNAL_DOWNLOADS] (LOW): references/packaging.md configures electron-builder and electron-updater to download dependencies and updates from GitHub. These are considered trusted sources per [TRUST-SCOPE-RULE], though they represent a remote code dependency.

electron-app-dev