agent-context-isolation

Prevent background agent output from flooding the main context by mandating file-based output and coordination, directing agent results to persistent storage instead of returning them directly to the main conversation.

The best way to preserve tokens when running background agents is to isolate agent execution context by avoiding direct output returns, instead writing processed results to designated files for later access.

Coordinate file-based handoffs by having background agents write their results to files in the cache directory, allowing the main instance to read the data without consuming conversational context tokens.

Background agent output degrades conversational context quality because returning large results directly to the main context window causes token overflow, which wastes tokens and can overwrite critical conversation history.

Isolate agent execution context using file-based output whenever an agent performs complex research or data processing, ensuring its output is saved to a file rather than degrading the main conversational context.

Agent context isolation requires no external dependencies, relying solely on directing agents to read and write data via files in a designated cache directory to maintain a clean main conversation.