Project - Persistent AI Context System

The problem: AI assistants are stateless by default. Every conversation starts from scratch, requiring the user to re-establish context: what they're working on, what tools they use, what's outstanding. Across a varied workload with multiple active projects and responsibilities, this overhead compounds quickly. Standard workarounds, such as pasting notes into prompts and maintaining separate reference documents, are fragile and easy to neglect.

What was done: A persistent context layer was built connecting Claude Code to an Obsidian vault via a custom CLI skill. A vault interaction layer exposes the full Obsidian API to Claude, allowing it to read and write notes, search content, manage tasks, and update metadata, all through a set of CLI commands the agent can call at runtime. A context management layer sits on top: a structured folder of agent-maintained knowledge files covering active projects, tasks, preferences, and working context. These files are read at the start of every session and updated silently throughout, without any action required from the user. A task management system was designed alongside it - a project-task-subtask hierarchy serving as a single source of truth for all outstanding work, with daily notes used for scheduling rather than tracking.

The result: An AI assistant that carries a full working context across every conversation. Projects, tasks, and preferences are known from the first message without any re-briefing. Completed work is logged and cleared automatically. The system is robust enough for daily use and extensible enough to accommodate new tools and workflows as they are adopted.