persistent-memory

To preserve task context across sessions, maintain a memory log using a task.md file or .memory/ directory to record decisions, observations, and summaries. This ensures traceability and coherent continuous workflows without guessing prior history.

Observation logging for multi-step planning works by automatically recording every tool usage as a standardized Observation. This builds a verifiable timeline of actions and decisions, ensuring full traceability of context injection throughout the task lifecycle.

The best way to retrieve knowledge artifacts from prior sessions is to read the saved memory logs directly. Injecting this historical context into your current decisions prevents rework and eliminates the need to guess previous task history.

Yes, enabling persistent memory imposes requirements for standardized memory logs, typically utilizing a task.md file or a .memory/ directory. You must also provide a clear end-of-session summary to ensure traceability within your workflow.

You need context injection for long-running tasks when working through multi-step planning scenarios that span multiple sessions. It maintains an ongoing memory log per task, enabling pre-session review and in-turn consistency checks to reduce rework.