Memory Skill — Claude

To connect a long-term markdown vault to Claude, use a retrieval pipeline that reads the vault index, selects starting nodes, and performs bounded link traversal. This extracts constraints and active decisions into a working context for your tasks.

Writing updates back into a knowledge graph requires applying a write gate to update canonical docs or create new ones via templates. This enforces mandatory links, updates the vault index following shared rules, and performs consolidation checks.

Bounded link traversal retrieves memory by reading the vault index and navigating linked entities up to depth 2 and a maximum of 10 nodes. This ensures the agent stays within a fixed document budget while gathering contextual constraints.

Yes, Claude can manage active decisions and open questions by retrieving them from your memory vault before a task. Post-task, it converts ambiguity into new notes rather than aggressively merging existing documents.

Memory retrieval requires a vault index formatted as mem-index-vault.md. The retrieval pipeline reads this index to select appropriate starting nodes by type and tags before performing bounded link traversal.

A fixed traversal and document budget limits the retrieved context to a maximum of 10 nodes at depth 2. This prevents context overflow but requires precise starting node selection from the vault index to avoid missing relevant constraints.