para-memory-files

Overview

para-memory-files is a Claude Code skill from the Paperclip project that gives AI agents persistent, file-based memory organized using Tiago Forte's PARA method (Projects, Areas, Resources, Archives). The system operates in three layers: a knowledge graph stored in PARA-organized folders with atomic YAML facts, daily notes as a raw timeline of events, and tacit knowledge about user patterns. Since memory does not survive session restarts, everything must be written to files, making this skill essential for agents that need to retain context across conversations.

Key Features

• Three-layer memory architecture: Layer 1 is an entity-based knowledge graph in PARA folders (Projects, Areas, Resources, Archives), each with summary.md and items.yaml files. Layer 2 is daily notes capturing raw timeline events. Layer 3 is tacit knowledge about user operating patterns stored in MEMORY.md
• Atomic fact management: Facts are stored as individual YAML entries with metadata, following a "never delete, always supersede" policy that maintains a complete history while keeping current information accessible
• Automatic entity creation rules: Entities are created when mentioned 3+ times, when they have a direct relationship to the user (family, coworker, client), or when they represent a significant project or company
• Memory decay and synthesis: Includes weekly synthesis to rewrite summaries from active facts and manage memory freshness, preventing stale information from accumulating
• Planning file integration: Supports planning files alongside memory, allowing agents to maintain both persistent knowledge and active work plans in a unified file system

Who is this for?

• AI agent developers who need their agents to maintain persistent context across sessions without relying on external databases
• Teams building Paperclip-based agent systems that require agents to remember user preferences, project states, and relationship histories
• Developers interested in structured personal knowledge management patterns who want to apply the PARA method to AI agent memory