Is this repo supposed to be global or repo level scoped?

[dpatchigolla]

In the readme Quick Setup, following is recommended:

These subagents are automatically available when placed in the ~/.claude/agents/ directory. Claude Code will automatically detect and load them on startup.

# Clone the repository to your Claude agents directory
# Documents are base on the scaffold from https://github.com/wshobson/agents.git
cd ~/.claude
git clone https://github.com/lst97/claude-code-sub-agents.git agents

# Or if the directory already exists, pull the latest updates
cd ~/.claude/agents
git pull origin main

But within the claude.md, it says the following:

⚠️ IMPORTANT: This file should be placed in your PROJECT ROOT DIRECTORY, not globally.

# ✅ Recommended: Project-specific usage
cp CLAUDE.md /path/to/your/project/CLAUDE.md

# ❌ Not recommended: Global scope
# cp CLAUDE.md ~/.claude/CLAUDE.md

Both seem contradictory. For the folks who have used these roles successfully, can you suggest what's the workflow that has worked well for you?

[lst97]

Hi @dpatchigolla , my apologies for the confusion.

TLDR: Place CLAUDE.md in your project root for better context management and to potentially manage token use from sub-agents. However, recent tests suggest global placement might also be fine as sub-agents aren't called too aggressively.

---

Once the command is executed, all agents become available in Claude Code. Claude Code reads CLAUDE.md at the start of each new session then agent-organizer should be more likely to be involved . The agent-organizer may modify this file to include necessary context. Therefore, I recommend placing CLAUDE.md in your project root directory. This allows Claude Code to trigger the agent-organizer, which can then modify the project-scoped CLAUDE.md by designed.

Another reason for this recommendation is that the organizer typically uses 3-5 sub-agents per task, which consumes tokens.. By requiring you to manually move CLAUDE.md to your project's root or global scope, it acts as a conscious safeguard. This ensures you're deliberately choosing to enable the agent-organizer within that specific project scope, helping to prevent unintended or overly aggressive token usage

However, based on my experiments, even with this crafted CLAUDE.md in the project root, Claude Code doesn't call sub-agents too aggressively. So, placing it in the global scope and project scope might also be acceptable.

[dpatchigolla]

Thanks for a quick response. This makes sense. I'll try with it in my project scope.

[lst97]

You can append this CLAUDE.md into your existing CLAUDE.md if any