omamori

Safety guard for AI CLI tools. Blocks dangerous commands — and resists being disabled.

When AI tools like Claude Code, Codex, or Cursor run shell commands, omamori intercepts destructive operations and replaces them with safe alternatives. It also defends itself against AI agents attempting to disable or bypass its protection (#22).

macOS only. Terminal commands are never affected — omamori only activates when it detects an AI tool's environment variable.

Quick Start

# Install (macOS)
brew install yottayoshida/tap/omamori

# Setup (shims + hooks + config — all in one)
omamori install --hooks

# Add to your shell profile (~/.zshrc or ~/.bashrc)
export PATH="$HOME/.omamori/shim:$PATH"

That's it. After brew upgrade, hooks and shims are auto-updated on the next command — no manual re-install needed.

What It Blocks

--delete, --del, --delete-before, --delete-during, --delete-after, --delete-excluded, --delete-delay, --remove-source-files

All rules are customizable via TOML config. See Configuration below.

How It Works

AI CLI tool → CLAUDECODE=1 → rm -rf src/
                                ↓
                          [omamori shim]
                                ↓
                        blocked (protected path)

Terminal → rm -rf src/
                ↓
          [/usr/bin/rm]
                ↓
          deleted normally

Layer 1 — PATH shim: Symlinks for rm, git, chmod, find, rsync point to omamori. Rules apply only when an AI environment variable is detected.

Layer 2 — Hooks: Recursively unwraps shell wrappers (sudo env bash -c "..." → extracts inner command) and evaluates against the same rules as Layer 1. Also blocks pipe-to-shell (curl | bash) and dynamic generation (bash -c "$(cmd)"). Available for Claude Code and Cursor.

Self-defense (#22): AI agents cannot config disable, uninstall, or edit config.toml while detected. Hooks block env var unsetting and direct config file editing. This is a key differentiator from other CLI guards — omamori assumes adversarial AI behavior and defends against it.

Auto-sync: After brew upgrade, the shim detects version mismatch and auto-regenerates hooks on the next invocation. No manual intervention needed.

Core policy: The 7 built-in rules cannot be disabled via config.toml — an AI agent setting enabled = false is silently ignored. For legitimate overrides, see omamori override in CLI Reference.

Integrity monitoring (omamori status): Verifies all defense layers are intact — shims, hooks, config, core policy, PATH order. Detects tampering including subtle hook edits where the version comment is preserved but the body is rewritten.

Supported Tools

Context-Aware Evaluation

omamori can adjust actions based on what the command targets:

Opt-in: Add [context] to ~/.config/omamori/config.toml. Built-in lists for regenerable (target/, node_modules/, etc.) and protected (src/, .git/, .env, etc.) paths activate automatically.

[context]
# Built-in defaults activate with just [context].
# To customize, specify your own lists (replaces built-in defaults):
# regenerable_paths = ["target/", "node_modules/", "my-cache/"]
# protected_paths = ["src/", "lib/", ".git/", ".env", ".ssh/", "secrets/"]

Note: Specifying regenerable_paths or protected_paths replaces the built-in defaults (not appends). Include the built-in entries you want to keep.

Security features: symlink defense via canonicalize(), path traversal normalization, NEVER_REGENERABLE hardcoded list, fail-close on errors.

Configuration

Built-in rules are always inherited. Only write the rules you want to change:

omamori config list                          # show all rules
omamori config disable git-push-force-block  # disable a rule
omamori config enable git-push-force-block   # restore default
omamori test                                 # verify policy

Or edit ~/.config/omamori/config.toml directly. Config is auto-created by install --hooks. See omamori init --stdout for the full template.

Disable a rule:

[[rules]]
name = "git-push-force-block"
enabled = false

Move files to a custom directory:

[[rules]]
name = "rm-to-backup"
command = "rm"
action = "move-to"
destination = "/tmp/omamori-quarantine/"
match_any = ["-r", "-rf", "-fr", "--recursive"]

Override an existing rule:

[[rules]]
name = "rm-recursive-to-trash"
action = "move-to"
destination = "/tmp/omamori-quarantine/"

Notes: Config requires chmod 600. Destinations must be absolute paths on the same volume. System directories and symlinks are rejected.

CLI Reference

omamori install [--hooks]                # Setup shims + hooks + config
omamori test [--config PATH]             # Verify policy rules
omamori status [--refresh]               # Health check all defense layers
omamori exec [--config PATH] -- CMD      # Run command through policy engine

omamori config list                      # Show rules with status
omamori config disable <rule>            # Disable a rule
omamori config enable <rule>             # Re-enable a rule
omamori override disable <rule>          # Override a core safety rule
omamori override enable <rule>           # Restore a core safety rule

omamori init [--force] [--stdout]        # Create/reset config
omamori uninstall                        # Remove shims + hooks
omamori hook-check [--provider NAME]     # Hook detection engine (used internally by hooks)
omamori cursor-hook                      # Cursor hook handler
omamori --version                        # Show version

Structural Limitations

These are inherent to the PATH shim approach and documented honestly:

• Full-path execution (/bin/rm) bypasses the shim — mitigated by Layer 2 hooks
• sudo changes PATH — omamori blocks when it detects elevated execution
• Interpreter commands (python -c "shutil.rmtree(...)") — not detected (bash/sh/zsh/dash/ksh only)
• Obfuscated commands (base64, variable indirection) — cannot be detected by static analysis
• AI self-bypass — config disable/uninstall are blocked; direct file editing blocked by hooks (Claude Code only)

For the full security model, bypass corpus, and known limitations, see SECURITY.md.

Related

• nanika — explains what AI commands will do. Complementary to omamori.

License

Licensed under either of Apache License, Version 2.0 or MIT license at your option.