white-stone-mem_skill

Overview

This skill is a personal memory system that organizes four memory types: common knowledge, project notes, error logs, and daily reviews. It uses on-demand loading to avoid memory pollution and automatic loading rules for shared, global memories. Vector search is optional and can be enabled when embedding services and a vector store are configured.

How this skill works

On startup the agent automatically loads common knowledge and the global error log so shared heuristics and past mistakes are available. Project memory files are only loaded when the user explicitly requests them to prevent accidental leakage. Daily review files are created automatically each morning and can be edited later. Optionally, a vector search index can be built to enable semantic lookups across the memory files when configured with an embedding provider and a vector database.

When to use it

• Keep consistent working habits and product thinking available via common knowledge at agent start.
• Store and consult project-specific details only when actively working on or discussing that project.
• Log bugs, pitfalls, and lessons learned in the global error book for all agents to reference.
• Create and refine a daily review each morning to capture progress and items to distill.
• Enable vector search when you need semantic search across many notes and have an embedding provider and vector DB configured.

Best practices

• Follow the directory layout: knowledge/, projects/, errors/, daily/ for predictable behavior.
• Never rely on project files to be loaded implicitly; always request project memory explicitly to avoid contamination.
• Keep the error log concise: problem, cause, and solution entries to maximize reuse by agents.
• Automate daily review creation and use it to capture raw observations for later synthesis.
• If enabling vector search, ensure embeddings and the index are rebuilt after changes to memory files.

Example use cases

• Agent starts with product and process heuristics from common.md to guide interactions.
• A user asks about Project X; the agent loads memory/projects/ProjectX.md on demand to provide context.
• Team-wide error patterns are aggregated in memory/errors/ to prevent repeat mistakes across agents.
• Daily reviews capture daily accomplishments and to-dos, later distilled into project memory or knowledge.
• Enable /memory build-index and /memory search when you need semantic retrieval across historical notes.