Suggestion: Add Support for Batch File Operations in Built-in File System (e.g., read/write multiple files)

[shtse8]

Description:

With the growing trend of models supporting extremely large context windows (e.g., Claude 3.5 with 200K, Gemini 1.5 Pro with 1M), it's becoming increasingly practical to provide a large amount of data in a single prompt. This trend is expected to continue.

In this context, limiting file read operations by line count becomes counterproductive. If we limit the number of lines AI can read at a time (e.g., 200 lines), it will attempt to read the file in multiple steps, resulting in more requests and significantly more output tokens. Since output tokens are often 5x more expensive than input tokens, this leads to higher costs and latency.

To address this, I’ve increased the read line limit from 200 to 5000 per file read, essentially removing the limit.

However, I still notice a significant bottleneck: reading files sequentially.

In my current setup, I rely on memory banks, and every time I initialize a session, 7–8 files are read. But since the built-in read file command only supports one file at a time, the process is unnecessarily slow.

To work around this, I was forced to install the mcp filesystem, which supports batch file operations. However:

• MCP introduces extra complexity.
• Since roo code already provides a built-in read file functionality, the AI may not always leverage MCP's batch capabilities.

Suggestion

Enhance the built-in file system in roo code with batch file operation support:

• read files: Read multiple files in one request
• write files: Write multiple files at once
• list folders: List multiple folders
• create folders: Create multiple folders
• delete files: Delete multiple files
• delete folders: Delete multiple folders

Currently, the AI sometimes attempts to do batch file operations via command-line workarounds. But as I’m on Windows, most of these rely on Linux-based commands, which often fail. Since Windows doesn't provide a simple CLI syntax for batch actions, the AI falls back to doing them one by one, again creating inefficiency.

I believe that supporting these batch operations natively in roo code will greatly improve performance, reduce token usage, and simplify workflows, especially for Windows users.

Thanks for considering!