RPA modules — Node.js code inside your flows

A script is a graph of blocks on the visual canvas. A module is a Node.js function — pure code — that you wrap with a form schema and call from inside a script via the executeModule block. Scripts give you orchestration; modules give you imperative logic when blocks aren't enough.

Call create_module via API, MCP or UI — Afina scaffolds a folder with index.js, utils.js, package.json and settings.json. Edit index.js in your editor, add npm dependencies, then call resign_module so the executor accepts the new files. The new module shows up in the executeModule block picker.

To stop the executor from running tampered or swapped code. Each save updates an Ed25519 signature over an md5 manifest of every file in the module folder. If the signature doesn't match the manifest, the executor refuses to start — surfacing a clear modules.error.signature_invalid.

Almost anything. Afina runs npm install in the module folder, so your package.json's dependencies live there in isolation. The runtime's allowedFunctions list controls which Node built-ins are exposed; you can broaden or narrow it per module.

Yes — through Afina's module helpers. The module receives a context that lets it call into the same primitives flows use: navigate, click, eval, set/get account variables, hit the local API. It runs in the same task as the parent flow.

Yes. The MCP server ships create_module/update_module/resign_module tools, and the LLM uses host filesystem tools (Read/Edit/Write) to edit module files under moduleDirAbs. The signing step is what guarantees what the LLM ships is what runs.