Claude Code configuration

Personal global configuration for Claude Code.

Structure

.claude/
├── CLAUDE.md           # Global instructions (orchestrator behavior, style, constraints)
├── settings.json       # Permissions, plugins, and Claude Code settings
├── agents/             # Specialized sub-agent definitions
├── commands/           # Slash command definitions (/command-name)
└── skills/             # Auto-triggered skills based on context

Note: MCP servers are not tracked here as a mcp-servers.json file; see
Setup → MCP servers for the claude mcp add-json commands.

Components

CLAUDE.md

Global instructions that apply to all Claude Code sessions. Defines:

• Orchestrator role and behavior
• Decision-making constraints
• Interaction style
• Project-specific notes

Agents

Specialized sub-agents invoked via the Task tool for domain-specific work:

Commands

Slash commands triggered by /command-name:

Hooks

settings.json registers a PreToolUse hook on Bash that shells out to rtk hook claude
(see RTK). rtk must be on PATH.

Skills

Context-aware capabilities auto-triggered when relevant:

Many skills from:

• https://skills.sh/
• https://github.com/Im5tu/dotnet-skills

Vendored skills carry a metadata.sourceUrl in their frontmatter pointing at the
upstream file, plus the sourceCommit they were taken at, so they can be refreshed:

curl -sL "$(sed -n 's/^  rawUrl: //p' skills/unslop/SKILL.md)" -o /tmp/upstream.md
diff <(sed '1,/^---$/d;1,/^---$/d' skills/unslop/SKILL.md) \
     <(sed '1,/^---$/d;1,/^---$/d' /tmp/upstream.md)

unslop is taken from cursor/plugins at 99559f2f.
That repo publishes no LICENSE file, so it carries no explicit grant of reuse.

Setup

MCP servers

The mcp-servers.json file contains MCP server definitions that need to be merged into your ~/.claude.json. To install them, run each server via the Claude CLI:

claude mcp add-json -s user sequential-thinking "{\"command\":\"cmd\",\"args\":[\"/c\",\"npx\",\"-y\",\"@modelcontextprotocol/server-sequential-thinking\"]}"
claude mcp add-json -s user playwright "{\"command\":\"cmd\",\"args\":[\"/c\",\"npx\",\"-y\",\"@playwright/mcp@latest\"]}"
claude mcp add-json -s user aws-knowledge-mcp-server "{\"type\":\"http\",\"url\":\"https://knowledge-mcp.global.api.aws\"}"
claude mcp add-json -s user aws-documentation-mcp-server "{\"command\":\"cmd\",\"args\":[\"/c\",\"uvx\",\"awslabs.aws-documentation-mcp-server@latest\"],\"env\":{\"FASTMCP_LOG_LEVEL\":\"ERROR\",\"AWS_DOCUMENTATION_PARTITION\":\"aws\"}}"

Prerequisites:

• Node.js / npm (for npx-based servers)
• Python 3.10+ / uv (for uvx-based servers)

Verify with:

claude mcp list

Usage

This configuration is loaded automatically by Claude Code from ~/.claude/
(%USERPROFILE%\.claude\ on Windows).

Files are applied globally across all projects unless overridden by project-specific CLAUDE.md files.

Keeping a checkout in sync

Rather than copying files back and forth, symlink the tracked directories from ~/.claude
into a checkout of this repo:

git clone https://github.com/Im5tu/claude ~/projects/.claude
mv ~/.claude/skills ~/.claude/skills.bak && ln -s ~/projects/.claude/skills ~/.claude/skills
mv ~/.claude/agents ~/.claude/agents.bak && ln -s ~/projects/.claude/agents ~/.claude/agents

Caveat: the symlink itself survives git checkout, but the contents do not. Switching
branches rewrites your live skills and agents, and any tracked file absent from the target branch
is deleted from the working tree. Files matched by .gitignore (e.g. the skills/ads/ bundle)
are left untouched. Commit or stash before switching branches.

CLAUDE.md, settings.json, and commands/ are not symlinked and must be copied
manually after pulling.