Client Folders Structure
Organize client assets under a standard structure. This prevents context confusion between running Claude Code sessions. Click folders to inspect files and configurations.
Why We Structure Client Workspaces for AI Agents
Prevent Context Leakage
Bounding agent workspace contexts to clients/{client_name}/ subdirectories prevents the agent from reading other client folders. This eliminates cross-account data exposure and brand styling leakage.
Deterministic Paths
Standardizing paths (e.g., outputs/ads/) ensures CLI scripts run relative to predictable locations. AI agents lack human intuition; rigid structures prevent script execution and path navigation errors.
No Self-Reference Loops
Separating static inputs (knowledge/, leads/) from generated assets (outputs/) keeps guidelines clean. This prevents the agent from reading its own draft outputs as rules, stopping self-reference loops.
Credential Isolation
Private API tokens belong in a local .env file inside scripts/analytics/. By adding it to .gitignore, the AI agent is physically blocked from staging or committing secrets to git.
CLAUDE.md Instruction Boundaries
Placing a client-specific CLAUDE.md at the client root loads custom run commands and instructions. The local Claude Code agent parses this automatically on boot to establish contextual bounds.
Workspace Tree View
clients/
The root directory mapping all client campaigns. Keep individual accounts separated under unique subfolders so that your running LLM agents don't bleed context between different clients.
- Expected Contents: Folder nodes for each client account.
- Agent Actions: Read-only checks on active folder namespaces.
mkdir -p clients/{client_name}/{outputs/{ads,blog-posts,directory-listings,email-copy,press-releases,prompts,social},knowledge,leads,scripts/analytics,assets}