agent-context-isolation

To prevent agent output from flooding the main context window, configure background agents to write their output directly to designated cache files instead of returning it to the main conversation thread.

Using TaskOutput for retrieving agent results pulls large outputs directly into the main context window, which causes context pollution and overwrites earlier conversation context, leading to lost information.

File-based coordination for background agents is a pattern where agents read and write to specific files in a cache directory to communicate, ensuring the main conversation thread remains concise and focused.

To configure agents to write research reports to a cache directory, direct their output to files in the `.maestro/cache/agents/` directory so subsequent agents or the main process can read the results without polluting the context.

You need context isolation for AI agent execution when an agent performs extensive research or generates a large report, ensuring the main conversation context is preserved by directing outputs to designated cache files.