itechmeat/open-second-brain ? reverse-engineered prompt

Build me a local first memory layer for an AI coding agent that lives inside my Obsidian vault and reads and writes plain Markdown files in a Brain folder.

I want it to remember project context, preferences, evidence, and audit trails in a way I can open, search, edit by hand, and version with git, with no hidden state outside the vault. It should work with Hermes Agent through deterministic CLI and MCP tools, and have adapters for Claude Code, Codex, and OpenClaw, plus an MCP server so other tools can connect too.

Please make it feel like a real product, with install and setup flow, clear commands, and a simple way to inspect what the agent remembered. If there are any current docs or best practices that would help, look them up online while building. Make sure the memory files are easy to understand in Obsidian and that the whole thing stays local first and transparent.