MLMartin Lepageinpharos-ai.hashnode.dev·May 7 · 6 min readInside an AI-native Obsidian vault: the architecture of persistent agent memoryInside an AI-native Obsidian vault: the architecture of persistent agent memory Most AI agent memory systems fall into one of two failure modes. Too flat: A single CLAUDE.md or AGENTS.md that grows into a wall of text. The agent reads all of it on ev...00
MLMartin Lepageinpharos-ai.hashnode.dev·May 7 · 7 min readThe Active Constraints Note: The One File That Stops AI from Violating Your Project RulesEvery project has rules the AI doesn't know about. Not requirements — those are usually in the spec. Not goals — those are in the brief. Rules: the non-obvious constraints that have accumulated through decisions, negotiations, discoveries, and hard l...00
MLMartin Lepageinpharos-ai.hashnode.dev·May 7 · 6 min readThe Open Questions Protocol: How to Stop AI from Filling In What It Doesn't KnowThere's a failure mode in AI-assisted work that's almost invisible until you know to look for it. Questions that haven't been answered don't stay open in the AI's working model. They get filled in. If you're working on an API design and it's genuinel...00
MLMartin Lepageinpharos-ai.hashnode.dev·May 7 · 6 min readWriting Notes for AI, Not for Humans — The Structural Differences That MatterWhen you write a note for yourself, you're writing for a reader who shares your context. You know what project this belongs to. You know what decisions led here. You know which parts are settled and which are still open. You can afford shorthand, par...00
MLMartin Lepageinpharos-ai.hashnode.dev·May 7 · 8 min readHow to Write AI Skill Guides That Replace Prompt EngineeringPrompt engineering has a maintenance problem. You write a careful prompt for a task you do repeatedly — synthesizing research notes, auditing link hygiene, archiving stale material. It works. You save it in a prompts folder. Three months later you ca...00
