An executive coach built a self-maintaining knowledge wiki for AI agents in days

MindValley: 2026 AI Mastery Program Submitted Artifact Title: Self-Improving Agentic Workspace Capstone Final Project Category The Applied AI Excellence Award Description Recognizing outstanding AI projects that don’t fall into the existing award categories. This award honors high-quality, well-executed applications of AI that demonstrate clear value, thoughtful implementation, and meaningful results. Date Submitted Aug 8, 2026 Submitted By Dr. LJ (Linda Johnson) ljohnsondba@gmail.com PROJECT DESCRIPTION Dr. LJ Executive Coaching produces content for YouTube and social media.  A repository for the source material behind that content was required, one that was relational in nature and accessible to both humans and AI agents working as a team. Excel and AirTable were both considered: Excel's static structure couldn't represent relationships between sources, and while AirtTble is dynamic and relational, it lacks the visual fluidity that lets a person or an agent see how ideas connect. Ease of use and long-term maintenance were the deciding factors, and a wiki, built in Obsidian on Callum Wanderloots’s framework, met all three requirements with the additional advantage of future-forward adaptability. This wiki is the first step in a larger project: a self-improving agentic workspace that ingests, organizes, and maintains its own content. Part one is complete and working as designed. 3 Attachments Cover-Self-Improving Agentic Workspace-DrLJ.png Graph-Files ingested ito RAW-DrLJ..png Graph-Filtered-5 Topic groups-DrLJ.png   ACHIEVEMENTS •    Built a working, self-maintaining knowledge system in a matter of days: a searchable, linked wiki that ingests, organizes, and validates its own content. •    Established a clear Files versus Sources sorting model, so every piece of research has a defined home and a traceable path back to its origin. •    Designed and implemented a five-topic classification structure (Coaching, Neuroscience, Leadership, Conversation, Mindset) that the graph view now visibly demonstrates through real note connections. •    Created a repeatable Daily Loop: a single prompt that ingests new material, runs maintenance, and commits changes, turning ongoing upkeep into a habit rather than a chore. •    Resolved hardware instability (an unreliable SSD on the Mac Mini) that could have stalled the entire build, and completed the setup on a replacement drive.   TOOLS •      MindValley AI Mastery 2026 lessons •      Claude.ai, Code •      Obsidian, with Kepano's obsidian-skills (obsidian-markdown, obsidian-bases, JSON Canvas, Obsidian-CLI, Defuddle) •      YouTube •      NotebookLM (NBLM) •      Google Gemini Mac •      Mini M4 (local build environment) •      MacBook M2   METHODOLOGY The workspace was built as a sequence of setup phases, followed by an ongoing Daily Loop that keeps the wiki current without manual upkeep:   Phase Step Action Start 00 Initialize empty vault, git, and .gitignore Structure 01–02 Create RAW / Wiki / Schema folders and agent skills Templates 03 Create note templates: concept, source, entity, project, topic, log Tooling 04 Add wiki_tool.py, command docs, hook setup, and public audit First Ingest 05 Add one RAW source and compile it into focused Wiki notes Validate 06 Query and lint, build catalog, update source manifest Daily Loop ongoing Single prompt: ingest new RAW, run maintenance, commit to git. Triggers search, compile/update notes, build/lint, and query.    LEARNINGS •       MindValley's foundational lessons provided the base that made this build possible, enabling a real jump to a new personal level. •       AI tools became noticeably more accessible over the course of the program (May through July): some coding literacy helps with understanding what is happening under the hood, but coding expertise is no longer required to build something like this. •       Having Claude.ai and Google Gemini available in parallel compressed the build into a matter of days. •       Going in with a clear vision of the target repository structure made the first build function as intended, rather than requiring a redo. •       The build itself was smooth; the real friction was entirely hardware related (boot delays, an SSD issue). No amount of AI assistance speeds up equipment booting up. •       YouTube videos offered helpful insights and different perspectives. •       NotebookLM became an indispensable tool for converting transcripts into a document that could then be discussed directly with Claude.   RELEVANT URLS • Kepano's Obsidian-flavored skills for agents: https://github.com/kepano/obsidian-skills • Callum (Wanderloots), source video and walkthrough: https://www.youtube.com/watch?v=QbjAQFJJyt0&t=535s Relevant URLs • Kepano's Obsidian-flavored skills for agents: https://github.com/kepano/obsidian-skills • Callum (Wanderloots), source video and walkthrough: https://www.youtube.com/watch?v=QbjAQFJJyt0&t=535s