Toggle Danger

Arguments: $ARGUMENTS. The first is the harness; if it's missing, use the one you are running in (claude in Claude Code, cursor in Cursor, antigravity in agy). The second is the action: on, off, toggle (the default), or status. status on its own reports every harness.

1. Run bash "99 - System/scripts/toggle-danger.sh" <harness> <action> from the vault root. It needs jq.
   • on checks the config first, then saves a timestamped backup (*.bak-YYYYMMDD-HHMMSS, never overwritten) and a restore record in ~/.local/state/second-brain/toggle-danger/<harness>.json (user-only permissions, shared across vaults), and only then edits the config. The file's permissions are preserved.
   • off undoes only those changes, so rules or modes you had before stay intact.
   • If danger mode was enabled some other way, or the config can't be read, off refuses and explains.
2. Relay the script's output verbatim. Don't claim a state the script didn't print.
3. Explain the scope in one line:
   • Claude Code: it changes the default mode for new sessions (~/.claude/settings.json). The current session keeps its mode; Shift+Tab cycles it right now.
   • Cursor: it changes Cursor CLI rules only. The IDE's auto-run toggle is in Settings > Agents.
   • Antigravity: it edits the allow-list in ~/.gemini/antigravity-cli/settings.json. For a one-off session, use agy --dangerously-skip-permissions instead.
4. Use status to report the state without changing anything: enabled (all wildcard rules present), partial (some), disabled, or unknown. It exits 2 if a config is unreadable or has unexpected types, or if jq is missing.

Never run this on your own initiative, even to get past a permission prompt.