agent-memory

You maintain long-term project context by storing decisions, architecture notes, and user preferences in a structured memory directory. This prevents context loss by consolidating episodic, semantic, procedural, and preference data for recall during planning and coding.

Setting up a persistent memory store requires creating a .github/memory/ directory containing four files: episodic.md, semantic.md, procedural.md, and preferences.md. These files capture decisions, architectural maps, and user preferences.

Update project memory incrementally after reaching development milestones. Following safe update rules and logging context at these checkpoints prevents drift and noise, ensuring stored decisions and architecture notes remain accurate.

Yes, the memory lifecycle supports temporal chaining. It operates through a stepwise process of recall, consolidation, and temporal chaining to connect past episodic decisions with current architectural context during development.

The best way to preserve preferences and procedural knowledge is logging them into dedicated markdown files within a .github/memory/ directory. This centralizes long-term memory, capturing both user preferences and procedural steps.

No, initializing the .github/memory/ directory is required. This directory houses the episodic, semantic, procedural, and preferences markdown files needed to store and recall project context safely.