dotfiles

depends on chezmoi

How to use(for me)

1. install chezmoi
   • sh -c "$(curl -fsLS get.chezmoi.io)" -- -b $HOME/.local/bin
   • winget install twpayne.chezmoi
2. if you are in PowerShell, Set-ExecutionPolicy -ExecutionPolicy ByPass -Scope Process
3. chezmoi init --apply berlysia

note

• ~/.local/bin will be added to $PATH
• ~/.local/.bin will be added to $PATH , and overwrite this directory with symlink

Local CI

Local workflow verification uses actrun, managed by this project's mise config.

mise install
pnpm ci:local:typescript
pnpm ci:local:codex
pnpm ci:local:shell

If your shell is not activated by mise, run commands through mise exec -- ... instead.

GitHub Environments

Claude workflows use environment-scoped secrets instead of repository-wide secrets.

• claude-manual: used by .github/workflows/claude.yml
• claude-autofix: used by .github/workflows/auto-fix-dependencies.yml

Add CLAUDE_CODE_OAUTH_TOKEN to both environments. If you want unattended automation, leave approvals and wait timers disabled.

list of something will be set up

• homebrew (only mac)

• mise

• direnv

• fzf

• ripgrep

• bat

• WSL2 ssh-agent (only WSL)

• my custom zsh prompt

Claude Decision Log Analysis

This dotfiles project includes tools for analyzing Claude Code hook decision logs:

home/dot_claude/scripts/update-auto-approve.ts

Automated permission pattern analysis and management tool:

# Interactive review of all patterns
bun home/dot_claude/scripts/update-auto-approve.ts

# Preview changes without applying
bun home/dot_claude/scripts/update-auto-approve.ts --dry-run

# Analyze last 7 days only
bun home/dot_claude/scripts/update-auto-approve.ts --since 7d

# Auto-approve safe patterns without interaction
bun home/dot_claude/scripts/update-auto-approve.ts --auto-approve-safe

# Show detailed analysis information
bun home/dot_claude/scripts/update-auto-approve.ts --verbose

Features:

• Analyzes ~/.claude/logs/decisions.jsonl for permission patterns
• Provides risk scoring and confidence levels
• Interactive review interface for pattern approval
• Automatic backup of permission files
• Supports time-based filtering of logs

Claude Code Configuration

This repository includes comprehensive Claude Code configuration with:

• Custom skills (see .skills/ directory)
• Global development guidelines (CLAUDE.md)
• Development rules (debugging, external-review, TypeScript standards)

Settings.json Split Management

~/.claude/settings.json is managed by splitting it into multiple files:

• .settings.base.json: Core settings (model, language, statusLine, alwaysThinkingEnabled, etc.)
• .settings.permissions.json: Permission settings (allow/deny)
• .settings.hooks.json.tmpl: Hooks configuration (dynamically generated with chezmoi variables)
• .settings.plugins.json: Plugin configuration (enabledPlugins)

These files are automatically merged by run_onchange_update-settings-json.sh.tmpl during chezmoi apply.

Important notes:

• To modify hooks configuration, edit .settings.hooks.json.tmpl and run chezmoi apply
• Existing enabledPlugins settings are preserved (user manual changes are not overwritten)

~/.claude.json Automatic Management

~/.claude.json (MCP servers configuration) is automatically managed by run_onchange_update-claude-json.sh.tmpl:

How it works:

• Reads MCP server versions from package.json dependencies
• Merges template content with existing ~/.claude.json
• Auto-updates mcpServers, preferredNotifChannel, and defaultMode
• Creates backup before changes and shows diff

Managed MCP servers:

• @mizchi/readability - Web page content extraction
• chrome-devtools-mcp - Chrome DevTools automation
• @playwright/mcp - Browser automation
• @upstash/context7-mcp - Documentation search

To update versions: Edit package.json dependencies and run chezmoi apply

Dual-Layer Skills Management

This project uses a two-layer approach for managing Claude Code skills:

1. Handcrafted Skills (.skills/)

• Stored in ${projectRoot}/.skills/ for unified management (at repo root, outside home/)
• Synced to both ~/.claude/skills/ and ~/.codex/skills/ via run_after_sync-skills.sh.tmpl
• Used by both Claude Code and Codex
• Always preserved during updates

2. External Skills (home/dot_apm/apm.yml)

• Declaratively managed in APM manifest (apm.yml)
• Installed via apm install -g from GitHub repositories
• Tracked via apm.lock.yaml
• Auto-removed when deleted from manifest (handcrafted skills are protected)

Both layers coexist in ~/.claude/skills/ directory.

Plugin Management

Plugins are managed declaratively through .settings.plugins.json:

After chezmoi apply:

• show-missing-plugins.sh detects missing plugins
• Displays marketplace registration commands (if needed)
• Shows plugin installation commands to run in Claude Code IDE

Syncing plugin changes back to dotfiles:

~/.claude/scripts/sync-enabled-plugins.sh

This exports current enabledPlugins from ~/.claude/settings.json to dotfiles for version control.

Using in Other Projects

To use these skills in another project:

cd /path/to/your/project
curl -fsSL https://raw.githubusercontent.com/berlysia/dotfiles/master/scripts/setup-claude-skills.sh | bash

This creates a simple .claude/settings.json with inline SessionStart command - no hook files needed!

Skills are installed to ~/.claude/ and auto-update on Claude Code startup.

See docs/external-usage.md for detailed documentation.

Skills Included

See .skills/ directory for available skills. Key skills include setup-claude-skills-for-web, commit-conventions, react-hooks, and logic-validation.