“Rather than putting orchestrator content in CLAUDE.md, write everything in a me.md and have CLAUDE.md simply point to it — so any future harness can be aimed at the same me.md and its sub-files.” — Scott Delinger, 2026-06-11
Session context: Scott Delinger shared an architectural pattern he had implemented — a second-brain approach designed to survive a platform move. Lou immediately connected it to his own reasoning for keeping a vendor-independent global skills and agents repo.
Core Idea
CLAUDE.md is a harness declaration. It’s tool-specific config — the instructions Claude Code reads when it boots a session. It belongs to the platform. me.md is the identity and context layer — who you are, how you think, what you care about, your operational directives. It belongs to you.
When you collapse both into CLAUDE.md, you’ve tied your identity to a platform. Swap the harness — move from Claude Code to Codex CLI to Gemini — and you have to rebuild your context from scratch in whatever format the new harness expects.
The me.md pattern separates them. CLAUDE.md becomes a pointer file: “load me.md and its sub-files.” The sub-files hold the real content: cognitive profile, voice preferences, operational rules, project context. When a better harness emerges, you point it at the same me.md. Your identity travels with you unchanged; only the pointer format changes.
This is exactly the same reasoning behind keeping a global skills/agents repo neutral and vendor-independent: the implementation resolves against the harness, not the other way around. Your expertise is the asset; the platform is the pipe.
Practical Application
Structure your second brain with a clear two-layer separation:
Harness layer (tool-specific, expect to rewrite):
CLAUDE.md— project-level config for Claude CodeAGENTS.md— Codex CLI equivalent- Any future harness entrypoint
Identity layer (yours, expect to keep):
me.md— top-level: who you are, how you work, key preferencesme/cognitive-profile.md— thinking style, ADHD accommodations, decision patternsme/operational-directives.md— how you want AI to behave with youme/voice.md— writing voice and style guidanceme/projects/— active project context, scoped per project
Each harness declaration is a single include: load ~/me.md and its sub-files. That’s the entire migration path to any future platform.
The test for whether you’ve implemented this correctly: can you point a brand-new AI harness at your me.md and have it understand you within one session? If yes, you own your intelligence. If no, you’re tenant-farming your context in someone else’s format.
Evolution Across Sessions
The Resolver Pattern insight established that CLAUDE.md should be a pointer, not a store. This insight sharpens the separation into an explicit architectural pattern with a named file and a concrete two-layer model. Where Resolver is about what CLAUDE.md should NOT contain, me.md is about what you should build instead — and why that distinction protects your compounding work against platform risk.