ai-agent-rules 0.18.0

Manage user-level AI agent configurations

AI Rules

Manage AI agent configurations through symlinks. Keep all your configs in one git-tracked location.

Overview

Consolidates config files for AI coding agents (Claude Code, Goose) into a single source of truth via symlinks:

• Git-tracked configs synced across machines
• Edit once, apply everywhere
• Declarative plugin management (Claude Code)
• Exclude specific files (e.g., company-managed)
• Per-agent customizations

Supported: Claude Code (settings, agents, commands, plugins), Goose (hints, config), Shared (AGENTS.md)

Installation

Requirements: Python 3.10+, uv

From PyPI (Recommended)

One-command setup from PyPI:

uvx --from ai-agent-rules ai-rules setup

This will:

1. Install AI agent configuration symlinks
2. Make ai-rules available system-wide
3. Auto-install optional tools (claude-code-statusline)

After setup, you can run ai-rules from any directory.

From GitHub (Development)

Install from GitHub to get the latest development code:

uvx --from ai-agent-rules ai-rules setup --github

This installs from the main branch and auto-detects the GitHub source for future updates.

Local Development

For contributing or local development:

git clone https://github.com/wpfleger96/ai-rules.git
cd ai-rules
uv run ai-rules install

Use uv run ai-rules <command> to test local changes. The global ai-rules command continues to run your installed version (PyPI/GitHub).

Updating

Check for and install updates:

ai-rules upgrade              # Check and install updates
ai-rules upgrade --check      # Only check for updates

Auto-update checks run weekly by default. Configure in ~/.ai-rules/update_config.yaml:

enabled: true
frequency: weekly  # daily, weekly, never
notify_only: false

Usage

User-Level Configuration

ai-rules setup                      # One-time setup: install symlinks + make available system-wide
ai-rules setup --github             # Install from GitHub (pre-release)
ai-rules setup --profile work       # Setup with a specific profile
ai-rules upgrade                    # Upgrade to latest version
ai-rules upgrade --check            # Check for updates without installing

ai-rules install                    # Install all agent configs + optional tools
ai-rules install --agents claude    # Install specific agents
ai-rules install --dry-run          # Preview changes
ai-rules install --force            # Skip confirmations
ai-rules install --rebuild-cache    # Rebuild merged settings cache

ai-rules status                     # Check symlink status + optional tools + active profile (✓✗⚠○)
ai-rules diff                       # Show config differences
ai-rules validate                   # Verify source files exist
ai-rules install                    # Re-sync after adding files
ai-rules uninstall                  # Remove all symlinks
ai-rules list-agents                # Show available agents

Configuration Management

# Interactive wizard for first-time setup
ai-rules config init                # Start configuration wizard

# View configuration
ai-rules config show                # Show raw config files
ai-rules config show --merged       # Show merged settings with overrides
ai-rules config show --agent claude # Show config for specific agent
ai-rules config edit                # Edit user config in $EDITOR

# Manage exclusions
ai-rules exclude add "~/.claude/*.json"      # Add exclusion pattern (supports globs)
ai-rules exclude remove "~/.claude/*.json"   # Remove exclusion pattern
ai-rules exclude list                        # List all exclusions

# Manage settings overrides (for machine-specific settings)
ai-rules override set claude.model "claude-sonnet-4-5-20250929"  # Set simple override
ai-rules override set claude.hooks.SubagentStop[0].hooks[0].command "script.py"  # Array notation
ai-rules override unset claude.model         # Remove override
ai-rules override list                       # List all overrides

Configuration

Quick Start with Config Wizard

Run the interactive configuration wizard for guided setup:

ai-rules config init

This will walk you through:

1. Selecting common exclusions
2. Adding custom exclusion patterns (with glob support)
3. Setting up machine-specific settings overrides

User-Level Config

Create ~/.ai-rules-config.yaml for user-level settings:

version: 1

# Global exclusions (apply to all contexts)
# Supports glob patterns: *.json, **/*.yaml, etc.
exclude_symlinks:
  - "~/.config/goose/config.yaml"
  - "~/.claude/*.log"              # Glob: exclude all log files
  - "~/.claude/agents/debug-*.md"  # Glob: exclude debug agents

# Machine-specific settings overrides
# Keeps repo settings.json synced via git, but allows local overrides
settings_overrides:
  claude:
    model: "claude-sonnet-4-5-20250929"  # Override model on personal laptop
    # Other settings inherited from base config/claude/settings.json
  goose:
    provider: "anthropic"

Config File Location:

• ~/.ai-rules-config.yaml - User-specific config (exclusions and overrides)
• ~/.ai-rules/state.yaml - Active profile and last install timestamp (auto-managed)
• ~/.ai-rules/cache/ - Merged settings cache (auto-generated)
• ~/.ai-rules/update_config.yaml - Update check configuration

Settings Overrides - Syncing Configs Across Machines

Problem: You want to sync your settings.json via git, but need different settings on different machines (e.g., different model access on work vs personal laptop).

Solution: Use settings_overrides in your user config:

# ~/.ai-rules-config.yaml on personal laptop
settings_overrides:
  claude:
    model: "claude-sonnet-4-5-20250929"  # No Opus access

# ~/.ai-rules-config.yaml on work laptop
settings_overrides:
  claude:
    model: "claude-opus-4-20250514"  # Has Opus access

Both machines sync the same config/claude/settings.json via git, but each has different local overrides. The system merges them at install time:

1. Base settings from config/claude/settings.json (git-tracked)
2. Merged with overrides from ~/.ai-rules-config.yaml (local only)
3. Cached in ~/.ai-rules/cache/claude/settings.json
4. Symlinked to ~/.claude/settings.json

After changing overrides, run:

ai-rules install --rebuild-cache

Array Notation for Nested Settings

Override commands support array index notation for complex nested structures:

# Override nested array elements (e.g., hooks)
ai-rules override set claude.hooks.SubagentStop[0].hooks[0].command "uv run ~/my-hook.py"

# Override environment variables
ai-rules override set claude.env.MY_VAR "value"

# The system validates paths and provides helpful suggestions
ai-rules override set claude.modle "sonnet"
# Error: Key 'modle' not found at 'modle'
# Available options: model, env, hooks, statusLine, ...

Path validation ensures you only set valid overrides that exist in the base settings, preventing typos and configuration errors.

Profiles - Machine-Specific Configuration

Profiles let you group configuration overrides into named presets. Instead of manually maintaining different ~/.ai-rules-config.yaml files across machines, define profiles once and select them at install time.

# List available profiles
ai-rules profile list

# View profile details
ai-rules profile show work
ai-rules profile show work --resolved  # Show with inheritance

# Check which profile is active
ai-rules profile current

# Switch to a different profile
ai-rules profile switch work

# Install with a specific profile
ai-rules install --profile work

Profiles are stored in src/ai_rules/config/profiles/ and support inheritance:

# profiles/work.yaml
name: work
description: Work laptop with extended context model
extends: null
settings_overrides:
  claude:
    env:
      ANTHROPIC_DEFAULT_SONNET_MODEL: "claude-sonnet-4-5-20250929[1m]"
    model: opusplan

Configuration layers (lowest to highest priority):

1. Profile overrides
2. Local ~/.ai-rules-config.yaml overrides

Your local config always wins, so you can use a profile as a base and tweak specific settings per-machine. Profiles are git-tracked and can be shared across your team.

The active profile is tracked in ~/.ai-rules/state.yaml and persists across sessions. Use profile current to see which profile is active, or profile switch to quickly change profiles without re-running the full install.

Plugin Management - Declarative Claude Code Plugins

Specify Claude Code plugins in your profile or user config, and ai-rules automatically installs them via claude plugin install:

# profiles/work.yaml or ~/.ai-rules-config.yaml
plugins:
  - name: feature-dev
    marketplace: claude-plugins-official
  - name: python-expert
    marketplace: cc-marketplace

marketplaces:
  - name: cc-marketplace
    source: ananddtyagi/cc-marketplace

When you run ai-rules install:

1. Adds missing marketplaces via claude plugin marketplace add
2. Installs missing plugins via claude plugin install <name>@<marketplace>
3. Warns about extra plugins (doesn't auto-uninstall)

Check plugin status:

ai-rules status  # Shows Plugins section with installed/pending/extra

Plugins are tracked in ~/.claude/plugins/installed_plugins.json and synced across machines via your git-tracked profile configuration.

Structure

config/
├── AGENTS.md              # User-level rules → ~/AGENTS.md, ~/.CLAUDE.md, ~/.config/goose/.goosehints
├── claude/
│   ├── settings.json      # → ~/.claude/settings.json
│   ├── agents/*.md        # → ~/.claude/agents/*.md (dynamic)
│   ├── commands/*.md      # → ~/.claude/commands/*.md (dynamic)
│   ├── hooks/*.py         # → ~/.claude/hooks/*.py (dynamic)
│   └── skills/*/SKILL.md  # → ~/.claude/skills/*/SKILL.md (dynamic)
└── goose/
    └── config.yaml        # → ~/.config/goose/config.yaml

Optional Tools

AI Rules automatically installs optional tools that enhance functionality:

• claude-code-statusline - Custom status line for Claude Code showing token usage, git info, time, and workspace details

These tools are installed automatically during setup and install commands. Check installation status:

ai-rules status  # Shows Optional Tools section

If a tool fails to install, ai-rules continues normally (fail-open behavior).

Extending

Add Claude agent/command:

1. Create config/claude/agents/my-agent.md or config/claude/commands/my-cmd.md
2. Run ai-rules install

Add Claude skill with auto-activation:

1. Create config/claude/skills/my-skill/SKILL.md with frontmatter:

---
name: my-skill
description: "Brief description"
metadata:
  trigger-keywords: "keyword1, keyword2"
  trigger-patterns: "pattern1, pattern2"
---

2. Run ai-rules install
3. Skill auto-suggests when user prompt matches keywords/patterns

Add Claude hook:

1. Create config/claude/hooks/my-hook.py
2. Run ai-rules install
3. Configure in settings or overrides (e.g., UserPromptSubmit, SubagentStop)

Add new AI tool:

1. Add configs to config/<tool>/
2. Implement ai_rules/agents/<tool>.py
3. Register in ai_rules/cli.py::get_agents()

Safety

• First-run warnings
• Timestamped backups (*.ai-rules-backup.YYYYMMDD-HHMMSS)
• Interactive prompts and dry-run mode
• Only manages symlinks (never deletes real files)
• Contextual error messages with tips

Development

Quick Start with Just

This project uses just for task automation.

Install just:

# macOS
brew install just

# Linux
cargo install just
# or: sudo apt install just (Ubuntu 23.04+)

# Windows
choco install just

Common commands:

just              # Run quick quality checks (sync, type-check, lint-check, format-check)
just --list       # List all available recipes

# Setup
just setup        # First-time setup: sync deps + install git hooks
just sync         # Sync dependencies only

# Code Quality
just check        # Quick quality checks (no tests)
just check-all    # All checks including tests
just lint         # Fix linting issues
just format       # Auto-format code
just type-check   # Run mypy type checking

# Testing
just test         # Run all tests (default config)
just test-unit    # Unit tests only
just test-integration  # Integration tests only
just test-cov     # Tests with coverage report

# Benchmarking
just benchmark-save     # Run and save baseline
just benchmark-compare  # Compare against baseline
just benchmark-record   # Compare and save
just benchmark-list     # List saved benchmarks
just benchmark-clean    # Remove all benchmarks

# Build
just build        # Build package
just rebuild      # Clean and build

Running Tests

The test suite includes both unit tests and integration tests.

Using just (recommended):

just test         # Run all tests with default config
just test-unit    # Only unit tests
just test-integration  # Only integration tests
just test-cov     # Tests with coverage report

Using uv directly:

uv run pytest [--cov=src --cov-report=term-missing]  # All tests
uv run pytest -m unit      # Unit tests only
uv run pytest -m integration  # Integration tests only

Troubleshooting

Wrong target: ai-rules status then ai-rules install --force

Restore backup:

ls -la ~/.CLAUDE.md.ai-rules-backup.*
mv ~/.CLAUDE.md.ai-rules-backup.20250104-143022 ~/.CLAUDE.md

Disable symlink: Use the exclude command or add to config manually:

ai-rules exclude add "~/.claude/settings.json"
# Or edit manually: ai-rules config edit

Override not applying: Rebuild the merged settings cache:

ai-rules install --rebuild-cache

View merged settings: Check what's actually being applied:

ai-rules config show --merged
ai-rules config show --merged --agent claude

License

MIT