What problem does it solve? Everything an agent learns disappears at the end of a turn unless it is written down. This Skill provides the operational discipline for persisting facts about users, projects, and decisions into a structured memory map, finding them again reliably, and keeping the map from degrading into noisy, duplicated clutter. ## Core Features & Use Cases - Structured fact storage: Write one fact per call into dotted, snake_case node ids (e.g., projects.laruche), with absolute dates and sources, using memory_write after checking for existing entries with memory_search and memory_suggest_nodes. - Retrieval workflows: Orient before any task with memory_search, read whole nodes with memory_read_node, find literal strings with memory_grep, and inspect the map shape with memory_tree. - Map curation: Diagnose overload with memory_doctor, merge bloated nodes with memory_consolidate, move or delete stale items with reasons, and review pending proposals via memory_list_proposed and memory_review. - Use Case: A user mentions they chose the MPL-2.0 license for their project. The agent searches memory first, finds no existing entry, writes the decision with its date and rationale to decisions.licence_mpl, and confirms the write by re-reading the node. ## Quick Start Ask the agent to remember a specific fact about your project or preferences, and it will search for duplicates, pick the right memory node, and store it permanently.
