filesystem-context

To manage agent context beyond token limits, you can offload large tool outputs to the filesystem, storing only a summary and reference in the active context. This allows agents to handle vast amounts of information without token bloat.

Dynamic context discovery works by loading relevant information on-demand from files rather than including everything statically. Agents utilize file I/O operations, glob, and grep to retrieve specific data points as needed, keeping the active context lean and efficient.

Yes, you can use the filesystem for agent memory and inter-agent communication by using files as scratch pads. This enables plan persistence and allows sub-agents to share state and communicate effectively through file I/O operations.

The best way to process large datasets without hitting token limits is writing intermediate analysis results to files. Agents can then use grep to retrieve specific data points dynamically, keeping the active context lean and efficient.

No specific dependencies are required to enable dynamic skill loading from the filesystem. The skill utilizes standard file I/O operations, glob, and grep for context discovery and management without relying on external packages.