AIBlueprint applies safety defaults so you can move fast without exposing yourself to destructive commands or leaked secrets.
Permission defaults
Setup updates ~/.claude/settings.json to set a default permission mode and add
deny rules that block dangerous operations - even when permission prompts are
skipped by the cc shortcut.
Deny rules
| Pattern | Blocks |
|---|---|
Bash(rm -rf *) | Destructive recursive deletion. |
Bash(sudo *) | Privilege escalation. |
Bash(curl * | bash) | Piping remote scripts into a shell. |
Bash(wget * | bash) | Piping remote downloads into a shell. |
Read(./.env) | Reading the .env file. |
Read(./.env.*) | Reading .env variants. |
Command validation hook
For Codex, a PreToolUse hook runs before any Bash command and blocks dangerous
ones via a deny list:
TOML
[[hooks.PreToolUse]]
matcher = "^Bash$"
[[hooks.PreToolUse.hooks]]
type = "command"
command = "/usr/bin/env node \"$HOME/.codex/hooks/command-deny-list.ts\""
timeout = 5
statusMessage = "Checking command safety"
Backups
Setup and sync create snapshots before updating existing configuration. Use
configs backups to restore these snapshots. configs undo
reverses the most recent configuration load; agents backup
reads the separate legacy backup store.