An agreement between you, your team, and your agents. Without naming, agent is blind.
{project} {2026-03-04} topic breaks sorting and agent parsing.
Option + -. Windows: Alt+0150. Max 80 characters total.
– Claude Code, – Cursor, – ChatGPT at the end is deprecated since 2026-02-20. Filename ends with date.
Click any example to load it into the builder.
Start with 5-7 types. Add more as you need them.
2-8 characters. Unique. Stable. Every new project gets its own code.
{AIM} {research} market analysis -- project + type. Maximum 2 codes before description.
An agreement between you, your team, and your AI agents. Not a technical constraint -- a shared language. When everyone (humans and agents) follows the same pattern, files become searchable, parseable, and navigable without opening them.
An AI agent searches your vault by patterns: glob("*{research}*"), grep("{AIM}"). Without structured naming, the agent is blind -- it cannot distinguish a research doc from a meeting transcript without reading every file. The name IS the API.
The format is {project} {type} description – YYYY-MM-DD.md. Date after the en dash, at the very end. This ensures chronological sorting within any folder. Putting the date in the middle ({project} {2026-03-04} {type}) breaks everything.
Since 2026-02-20, suffixes like – Claude Code, – Cursor, – ChatGPT are no longer added. The filename ends with the date. Old files with suffixes are NOT renamed -- only new files follow the new convention.
Use – (U+2013, EN DASH) as the separator before date. Not em dash (—), not hyphen (-), not minus. Mac: Option + -. Windows: Alt+0150.
Keep filenames under 80 characters for terminal compatibility, Obsidian sidebar readability, and cross-platform safety. If the description is too long, shorten it -- the details live inside the file, not in the name.
In the AI-Native model, your Obsidian vault is not a note-taking app -- it is the data layer of your Personal Operating System. Every file is a resource that agents can query, process, and act on. Naming convention is the schema of that data layer.
Put your naming convention into CLAUDE.md at the vault root. This is the instruction file that Claude Code, Codex, and other agents read on startup. Your convention becomes an enforceable contract -- agents will follow it automatically.
A well-named file is findable by glob: *{research}*2026-03* finds all research from March 2026 across the entire vault. No need to know which folder it is in. The name carries context -- project, type, topic, date -- in a machine-parseable format.
Markdown file with all types, projects, and rules