Wire failsafe into Claude Code¶

Put failsafe in front of every Bash command Claude Code runs, so destructive infra commands are blocked before they execute.

1. Install the binary¶

brew install undermountaincc/tap/failsafe   # Homebrew (recommended)
# or, from source (needs Go 1.22+):
go install github.com/UndermountainCC/failsafe/cmd/failsafe@latest

Make sure failsafe is on the PATH of the shell Claude Code launches. The binary needs no configuration to start guarding against the bundled default policies.

2. Register the PreToolUse hook¶

Add a PreToolUse hook to ~/.claude/settings.json that matches the Bash tool:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [{ "type": "command", "command": "failsafe hook" }]
      }
    ]
  }
}

Claude Code streams a JSON envelope on stdin; failsafe replies on stdout with allow, deny, or ask. The bare binary defaults to hook, so "command": "failsafe" also works.

Agent-agnostic by design

Anything with a pre-exec or Bash hook works the same way. The hook reads the agent's command on stdin and writes a decision on stdout. There is nothing Claude Code-specific in the protocol.

3. Confirm it's live¶

In a fresh pane (read-only by default), ask Claude Code to run a mutating command, or dry-run it yourself:

failsafe explain "kubectl delete ns payments"

A blocked command prints a BLOCK decision with the matching reason. If it does, the chain is wired correctly.

4. Flip a pane to write when you mean to¶

failsafe stays read-only until you deliberately flip a pane:

failsafe toggle        # flip the current pane

Bind a one-keystroke toggle in your terminal so you never type the command. See WezTerm, iTerm2, or tmux.