filesystem-context

Context persistence via the filesystem allows agents to maintain state and intermediate results across complex tasks. It uses durable scratchpads and tool-output offloading to manage large context beyond the prompt window limits.

Tool-output offloading is a context management technique that saves intermediate tool results to the filesystem. You need it when task outputs exceed prompt window limits, ensuring durable storage for complex multi-step operations.

Dynamic context discovery is set up by activating the filesystem-context skill to handle file-backed operations. It optimizes token usage by encouraging on-demand context loading rather than loading everything at once.

Yes, you need filesystem access and appropriate Python libraries for file operations to manage context persistence. The setup specifically requires Python dependencies like os, json, yaml, datetime, and pathlib.

Yes, filesystem context management is ideal for multi-agent coordination. It enables agents to share durable scratchpads and handle complex tasks requiring persistence of state and intermediate results across multiple processes.

The main limitation of using the filesystem for durable agent scratchpads is the strict requirement for filesystem access. You must ensure your environment supports the required file operations and Python libraries before implementation.