Brain Workspace Rules

Core principles

• Brain is a container: ~/brain is not a single project. Each repo under git/ has its own .git , dependencies, and workflows.

• Project rules override: When inside git/personal/<project>/ or git/work/<project>/ , that project's .cursorrules /AGENTS.md /README.md takes precedence.

• Version control: Agent instructions live in git/personal/agent-instructions/ and must be git committed when modified.

Directory structure

brain/ ├── git/ │ ├── personal/ # Personal GitHub identity │ │ ├── agent-instructions/ # THIS REPO - skills, tools, guidance │ │ │ └── .cursor/skills/ # Canonical skill definitions │ │ └── ... │ └── work/ # Work GitHub identity │ └── ... ├── obsidian/ # Obsidian knowledge vaults │ ├── Timatron/ # Main vault (daily notes, GTD, journals) │ └── ... ├── .cursorrules # Thin pointer → this file ├── CLAUDE.md # Thin pointer → this file ├── .cursor/skills/ # Symlink → agent-instructions/.cursor/skills/ └── .claude/skills/ # Symlink → agent-instructions/.cursor/skills/

Skill discovery (symlinks)

Skills are defined canonically in git/personal/agent-instructions/.cursor/skills/ . But Cursor and Claude Code discover skills from the workspace root, not from nested repos:

• Cursor looks for skills in <workspace>/.cursor/skills/

• Claude Code looks for skills in <workspace>/.claude/skills/

Since ~/brain is the workspace root, the setup script (setup.sh ) creates symlinks so both tools find the skills automatically:

~/brain/.cursor/skills → ~/brain/git/personal/agent-instructions/.cursor/skills ~/brain/.claude/skills → ~/brain/git/personal/agent-instructions/.cursor/skills

Both symlinks point to the same source — skills are authored once and discovered by both tools. If the symlinks break (e.g., after a fresh clone), re-run:

cd ~/brain/git/personal/agent-instructions ./setup.sh

Git identities

• git/personal/ — Personal GitHub identity (via SSH host alias git@personal-github:... )

• git/work/ — Work GitHub identity (standard git@github.com:... ; work key in ~/.ssh/config for Host github.com )

Avoid stray copies: Only operate on repos under git/personal/<repo>/ or git/work/<repo>/ . If you see git/<repo>/ directly without a .git/ directory, it's a stray copy—find the canonical location.

Obsidian safety

When editing files under obsidian/ :

• Preserve [[wikilinks]] and existing markdown conventions

• Respect daily note formats in daily/ and GTD files like _DAILY.md

• Keep diffs minimal; avoid unnecessary reformatting

• Journal files can be very large (e.g., 9MB)

Agent tools

Skills and Python tools live in git/personal/agent-instructions/ :

Skill Purpose

Daily Briefing Health briefing from TimTracker (sleep, exercise, diet, mindfulness)

Update Daily Tasks Refresh _DAILY.md from Asana

Deep Research Research using OpenAI's deep research API

YouTube Analyzer Analyze YouTube videos via Gemini API

Product Deep Research Combined web + video product research

Convert PDF to Markdown PDF conversion via PyMuPDF/marker-pdf

Compact Markdown Normalize whitespace

TalkToFigma MCP Figma MCP server workflow

Python environment:

cd ~/brain/git/personal/agent-instructions poetry install poetry run <command>

Working with skills

When asked to do a known workflow, load the relevant skill SKILL.md and follow it exactly.

Cursor cloud agents and nested work repos

When you open ~/brain in Cursor, cloud agents (Cloud tab in the agent panel) are tied to a single Git repo. Brain has no repo at the root—repos live under git/personal/ and git/work/ . So from the brain workspace, cloud agents may not see a clear repo to clone (e.g. for git/work/arda-intelligence/ ).

Options:

Default repo in Dashboard

Cursor Dashboard → Cloud Agents → Default repository: set it to the work repo (e.g. arda-industries/arda-intelligence ). When the workspace has no single repo, Cursor may then use this default when you start a cloud agent. Try this first.

Multi-root workspace with work repo first

Open a .code-workspace that lists the work repo as the first folder and brain (or a subfolder) as the second, so the “primary” repo for cloud agents is the work repo while you still have brain context. Create it once, then use File → Open Workspace from File and open that file.

Example (save as e.g. ~/brain/brain-arda.code-workspace ):

{ "folders": [ { "name": "arda-intelligence", "path": "git/work/arda-intelligence" }, { "name": "brain", "path": "." } ], "settings": {} }

Paths in .code-workspace are relative to the workspace file. So with the file in ~/brain , "path": "git/work/arda-intelligence" is correct.

Separate window when you need cloud agents

If the above don’t reliably target the right repo: open only ~/brain/git/work/arda-intelligence in a second Cursor window when you need to run cloud agents there. You lose the rest of brain in that window but cloud agents will work. This is the workaround many use on the Cursor forum.

Note: Parallel agents / worktrees and multi-repo cloud agents are limited today; Cursor is working on better multi-root and multi-repo support.

Continuous improvement

If you hit issues or needed extra guidance, propose edits to the relevant instruction file, then git commit and git push .