Memory Layer Skill

To give an agent persistent memory across sessions, you use a SQLite-backed memory store to save observations, decisions, and context. This allows the agent to retain useful knowledge between runs instead of relearning information.

You load relevant context at the start of a new agent session by searching the memory store for prior decisions or patterns. The skill finds relevant memories and injects them to improve decision quality for your current project.

The best way to index code and documentation for agent retrieval is to use a doc-indexer capability. This indexes code symbols and Markdown docs so lookups are faster and more targeted than keyword-only recall.

You need a local SQLite database to save agent learnings and observations. This memory-store capability retains useful bugfixes, patterns, and session context locally so the agent can reuse the knowledge later.

You can search for prior bugfixes and constraints by asking the memory layer to load the most relevant memories for your bug. It retrieves stored session context and summarizes key constraints to assist your current work.

An agent forgets project patterns between runs because useful decisions and context usually get lost without a persistent memory layer. Storing these observations in a local SQLite memory store prevents this loss and strengthens long-horizon planning.