markdown-memory

You can persist AI decisions and context across sessions by writing them into a structured memory root with context, decisions, and insights directories. This enforces append-only decision files and frontmatter metadata so agents retrieve prior session knowledge.

Organizing persistent AI memory inside a repo involves creating a memory root layout with dedicated directories for decisions, insights, and context. Per-domain and per-agent insights enable collaborative knowledge sharing across tasks.

Frontmatter metadata is required to drive tooling for a rolling decisions log. It structures append-only decision files so AI agents can read and apply prior decisions across sessions and tasks.

Repository memory should not be used for async agent-to-agent messaging. It is designed to store decisions, insights, and session context, while asynchronous messaging between agents requires a separate message creation mechanism.

Per-domain insight sharing works by storing known codebase information into domain-specific files within the insights directory. This allows multiple agents to access and apply shared knowledge across different tasks and sessions.

Accumulated decisions should be distilled into a summary file to maintain readability as the append-only decisions log grows. This distillation process condenses past decisions into a concise overview while preserving the original entries.