What This Does

This playbook shows you how to repurpose Claude Code as a fully customizable non-coding assistant by configuring Output Styles, project-level settings, and custom commands. Each project folder becomes a different "personality" — a media server manager, a writing editor, an Obsidian notes assistant, or anything else you need. The key insight is using keep-coding-instructions: false to tell Claude Code to stop acting like a coding assistant and become whatever you configure it to be.

Prerequisites

Claude Code installed and configured
Basic understanding of Claude Code's project settings and .claude/ directory

The CLAUDE.md Template

Copy this into a CLAUDE.md file in your project folder:

# [Project Name] Assistant

## Identity
You are a [role description]. You do NOT generate code unless explicitly asked. Your primary function is [primary function].

## Personality
- Tone: [professional / casual / academic / friendly]
- Style: [concise / detailed / conversational / formal]
- Expertise: [domain expertise areas]

## Capabilities
1. [Capability 1 — e.g., "Manage and organize media library files"]
2. [Capability 2 — e.g., "Research and recommend content"]
3. [Capability 3 — e.g., "Generate reports and summaries"]

## Directory Structure
- `context/` — Background knowledge files Claude should reference
- `output/` — Where Claude saves generated content
- `commands/` — Custom slash command definitions

## Rules
1. Do not generate code unless explicitly asked
2. Always save output to the `output/` directory
3. Reference files in `context/` for background knowledge
4. Use markdown for all generated content
5. When unsure, ask for clarification rather than assuming

## Knowledge Files
Read these files for context before responding:
- `context/domain-knowledge.md` — Core domain information
- `context/preferences.md` — User preferences and style guide
- `context/history.md` — Previous decisions and context

## Custom Commands
- "/summarize [topic]" — Generate a summary on the given topic
- "/research [question]" — Research and compile findings
- "/review [file]" — Review and provide feedback on a file
- "/status" — Show current project status and recent activity

Step-by-Step Setup

Step 1: Create a project folder for each personality

mkdir -p ~/media-manager/{context,output,commands}
mkdir -p ~/writing-studio/{context,output,commands}
mkdir -p ~/notes-assistant/{context,output,commands}

Each folder becomes a separate "personality" when you open Claude Code inside it.

Step 2: Configure the CLAUDE.md for each project

Customize the template above for each project. For example, a Media Server Manager:

# Media Server Manager

## Identity
You are a media server management assistant. You help organize, maintain, and optimize a home media server running Plex/Jellyfin with Sonarr, Radarr, and related services.

## Capabilities
1. Check service status and health
2. Recommend content organization strategies
3. Help configure automation rules
4. Troubleshoot playback and transcoding issues

Step 3: Add context files

Populate the context/ folder with domain-specific knowledge:

# For a writing studio:
echo "# Writing Style Guide\n- Avoid passive voice\n- Keep sentences under 20 words\n- Use concrete examples" > ~/writing-studio/context/style-guide.md

Step 4: Create custom commands

Create .claude/commands/ in your project folder with markdown files for each command:

~/writing-studio/.claude/commands/draft.md:

Generate a first draft on the topic: $ARGUMENTS
Use the style guide in context/style-guide.md.
Save the output to output/draft-[topic].md.

Step 5: Set up Output Styles (optional)

In Claude Code settings, you can configure per-project output styles. The key setting:

keep-coding-instructions: false

This tells Claude to drop its default coding assistant behavior and fully adopt whatever personality you've configured.

Step 6: Launch and use

cd ~/writing-studio
claude

Claude now behaves as your writing assistant, not a coding tool.

Example Personality Configurations

Writing Editor:

Identity: Professional writing editor. Reviews drafts for clarity, tone, and structure. Never rewrites — only suggests edits with explanations.

Obsidian Notes Manager:

Identity: Personal knowledge management assistant. Helps organize notes, suggest connections between ideas, maintain consistent tagging, and generate weekly note summaries.

Media Server Manager:

Identity: Home media server admin. Can check Docker container status via SSH, recommend organization for media libraries, and help configure Sonarr/Radarr rules.

Research Assistant:

Identity: Academic research assistant. Helps synthesize papers, maintain a literature review, and identify gaps in research coverage.

Example Usage

Switch between personalities:

cd ~/writing-studio && claude    # Writing editor mode
cd ~/media-manager && claude     # Media server mode
cd ~/notes-assistant && claude   # Notes manager mode

Use custom commands:

"/draft blog post about productivity systems" "/review output/draft-productivity.md" "/research latest trends in personal knowledge management"

Tips

Filesystem access is the superpower: Unlike Claude Desktop, Claude Code can read, create, and organize files across your entire project. This makes it vastly more capable for knowledge work.
Skills over MCPs for simple tasks: For API calls or simple operations, Claude Code skills (context + optional scripts in a directory) use fewer tokens than MCP servers.
Each project is isolated: Personalities don't leak between projects. Opening Claude Code in a different folder loads a completely different context.
Let Claude improve itself: Since all configuration lives in plain .md files, you can ask Claude to "improve your own CLAUDE.md based on how we've been working together."
Commands for repetitive tasks: If you find yourself typing the same kind of prompt repeatedly, turn it into a custom command.

Troubleshooting

Problem: Claude still acts like a coding assistant

Solution: Make sure your CLAUDE.md explicitly states "do not generate code unless explicitly asked." Also check if keep-coding-instructions: false is set in your output style configuration.

Problem: Claude doesn't reference context files

Solution: Explicitly list the context files in CLAUDE.md under a "Knowledge Files" section. Claude reads CLAUDE.md automatically but needs to be told which other files to consult.

Problem: Custom commands don't work

Solution: Commands must be in .claude/commands/ (note the dot prefix on .claude). Each command is a .md file. Use $ARGUMENTS as the placeholder for user input.

Claude Code Custom Personalities for Non-Coding Tasks