Brunogama AI Crap Manager AI Crap Manager

AI Crap Manager

Centralized management for AI-generated configuration files across all your projects.

Problem

AI coding tools (Claude Code, Cursor, Windsurf, Copilot, etc.) generate config directories and files at project roots:

• .claude/, CLAUDE.md, .claude/settings.json
• .cursor/, .cursorules
• .windsurf/, .windsurfrules
• .github/copilot-instructions.md
• .aider*, .cline/, .roo/, .codex/, .bolt/, .devin/, .replit/
• Agent configs, hooks, skills, subagents, MCP configs

These pollute git history, create merge conflicts, and clutter repos.

Solution

~/.agent-crap/ is the single source of truth. Each project gets a subdirectory. Files are symlinked back to project roots.

~/.agent-crap/
  projects/
    my-app/                    # mirrors ~/Code/my-app AI configs
      .claude/
      CLAUDE.md
      .cursor/
    another-repo/
      .claude/
      CLAUDE.md
  templates/                   # reusable config templates
  snapshots/                   # named point-in-time snapshots
  hooks/                       # auto-sync watcher scripts
  backups/                     # automatic backups before operations
  logs/                        # hook execution logs
  registry.json                # maps project names -> original paths

Setup Script

ACM="bash ~/.claude/skills/ai-crap-manager/scripts/setup.sh"
# Or after alias-install:
# acm <command>

Quick Start

$ACM init                          # Create ~/.agent-crap/
$ACM register ~/Code/my-app       # Move configs to store, symlink back
$ACM alias-install                 # Add 'acm' alias to shell

Command Reference

Core Commands

$ACM init                    # Initialize ~/.agent-crap/ store
$ACM register [path]         # Register project (default: cwd)
$ACM unregister [path]       # Restore files and remove from registry
$ACM sync [path]             # Verify/fix symlinks, absorb new AI files
$ACM sync-all                # Sync ALL registered projects
$ACM status [path]           # Show tracked files, symlink health, snapshots
$ACM list                    # List all registered projects with metadata
$ACM detect [path]           # Scan which AI tools are in use
$ACM gitignore [path]        # Update .gitignore only

Batch Commands

$ACM batch-register ~/Code/* # Register multiple projects at once
$ACM scan ~/Code             # Scan directory for repos with AI configs
$ACM adopt ~/Code            # Scan + register all found projects

scan shows what exists without changing anything. adopt does scan + register in one step.

Template Commands

Templates let you save and reuse AI configs across projects.

$ACM template list                 # List available templates
$ACM template save <name> [path]   # Save project configs as template
$ACM template apply <name> [path]  # Apply template to a project
$ACM template delete <name>        # Delete a template
$ACM template show <name>          # Inspect template contents

Use cases:

• Save your Claude Code setup (CLAUDE.md, hooks, agents) as a template
• Apply it to every new project
• Maintain "starter" templates per project type (swift-app, react-app, python-lib)

# Save your best setup
$ACM template save swift-pro ~/Code/my-polished-app

# Start a new project with it
$ACM template apply swift-pro ~/Code/new-app

Snapshot Commands

Snapshots capture the exact state of a project's AI configs at a point in time. Unlike backups (automatic, time-based), snapshots are named and intentional.

$ACM snapshot <name> [path]            # Create named snapshot
$ACM snapshot-list [path]              # List snapshots for project
$ACM snapshot-restore <name> [path]    # Restore a snapshot

Use cases:

• Snapshot before experimenting with new agent configs
• Snapshot "working" state before a major CLAUDE.md rewrite
• Quick rollback when something breaks

$ACM snapshot before-refactor
# ... experiment with configs ...
$ACM snapshot-restore before-refactor  # oops, go back

Diff & Inspect

$ACM diff [path]             # Show changes between store and project
$ACM tree [path]             # Tree view of managed files
$ACM stats                   # Global statistics across all projects

diff catches files that were modified directly in the project instead of through symlinks (e.g., after a git checkout that replaced a symlink with a real file).

stats shows: project count, agent popularity, disk usage breakdown.

Custom File Tracking

Track ANY file or directory beyond the built-in AI patterns. Useful for project-specific configs, prompt libraries, AI training data, or anything else you want centralized.

$ACM track <file-or-dir> [path]      # Move to store + symlink + gitignore
$ACM untrack <file-or-dir> [path]    # Restore to project + remove from store
$ACM tracked [path]                  # List all tracked files (built-in + custom)

Features:

• Comma-separated multiple files: $ACM track "file1,file2,dir1"
• Glob patterns: $ACM track "*.prompts"
• Nested paths: $ACM track configs/ai/prompts.yaml
• Auto-registers project if not yet registered
• Auto-adds to .gitignore
• Auto-removes from .gitignore on untrack

Examples:

# Track a custom prompt library
$ACM track prompts/

# Track project-specific AI training data
$ACM track .ai-data,.ai-prompts

# Track a Makefile.ai and custom rules
$ACM track "Makefile.ai,rules.yaml"

# Stop tracking .cursorules (restore to project)
$ACM untrack .cursorules

# See what's tracked (split by built-in vs custom)
$ACM tracked

Auto-Sync Hooks

Install hooks so syncing happens automatically.

$ACM hook-install [path]     # Install git hooks + optional fswatch/LaunchAgent
$ACM hook-remove [path]      # Remove all installed hooks

What gets installed:

Maintenance

$ACM doctor                  # Health check: registry, symlinks, orphans
$ACM gc                      # Remove old backups (>30 days), orphaned dirs
$ACM export <path.tar.gz>    # Export entire store as portable archive
$ACM import <archive.tar.gz> # Import store from archive
$ACM alias-install           # Add 'acm' alias to .zshrc/.bashrc

Options

--dry-run       # Show what would happen without doing it
--force         # Skip confirmation prompts
--verbose       # Detailed output
--only <agent>  # Only track specific agent (claude, cursor, etc.)

Tracked File Patterns

Workflows

First time: register everything at once

$ACM init
$ACM adopt ~/Code            # Finds and registers all repos with AI configs
$ACM alias-install           # 'acm' shortcut

New project with template

cd ~/Code/new-project
acm template apply swift-pro # Apply your standard AI config
acm hook-install             # Auto-sync on git operations

Before risky config changes

acm snapshot stable          # Save current working state
# ... edit CLAUDE.md, add hooks, change agents ...
acm snapshot-restore stable  # Nope, go back

Moving to a new machine

# Old machine:
acm export ~/Desktop/ai-configs.tar.gz

# New machine:
acm import ~/Desktop/ai-configs.tar.gz
acm sync-all                 # Recreate all symlinks

Periodic maintenance

acm doctor                   # Check health
acm gc                       # Clean old backups
acm stats                    # See usage overview

Registry Format

~/.agent-crap/registry.json:

{
  "version": 2,
  "projects": {
    "my-app": {
      "path": "/Users/bruno/Code/my-app",
      "registered": "2026-03-12T10:00:00+00:00",
      "last_sync": "2026-03-12T15:30:00+00:00",
      "tracked_files": [".claude/", "CLAUDE.md", ".cursor/"],
      "agents": ["claude-code", "cursor"],
      "tags": []
    }
  }
}

Safety

• All destructive operations create automatic backups
• unregister restores files to their original locations
• snapshot-restore backs up current state before restoring
• import backs up existing store before overwriting
• Symlink verification before any destructive operation
• Registry is the single source of truth for mappings
• doctor catches corruption, orphans, and broken links
• gc only removes backups older than 30 days

Source: brunogama/ai-crap-manager — distributed by TomeVault.