filesystem-context

To manage agent context beyond fixed window limits, you can offload massive tool outputs like search results or logs to the filesystem. This allows the agent to retrieve specific details on demand without exceeding context limits.

State persistence for multi-step plans is achieved by saving and reloading plans to the filesystem. This maintains task continuity across long interactions by storing structured data in files for later retrieval.

Yes, sub-agent communication is enabled by utilizing a shared filesystem workspace. Agents can share information dynamically by writing to and reading from common files within the workspace.

To reduce static context bloat, load skill instructions dynamically only when needed. By utilizing the filesystem for context discovery, agents retrieve specific instructions on demand rather than holding them in memory.

Yes, when web searches or logs return thousands of tokens, the skill saves the output to a file and provides a summary and path. Agents then use file-based operations like grep and read_file to retrieve specific details on demand.

Using the filesystem for dynamic context discovery requires available local storage and file access permissions. While it overcomes context window bloat, agents must execute file operations like grep and read_file to retrieve information, adding operational overhead.