YOLO Mode

YOLO mode lets your agent execute commands without asking for permission at each step. It skips Claude Code’s built-in approval prompts while installing safety hooks that intercept genuinely destructive operations.

What YOLO mode does

When you launch YOLO mode, two things happen:

1. Approval gates are skipped — Claude Code runs with --dangerously-skip-permissions, so the agent executes shell commands, file writes, and tool calls without pausing for confirmation.
2. Safety hooks are installed — The yolo-safety hook pack intercepts destructive commands before they execute.

The result: fast, autonomous execution with a safety net.

Launching YOLO mode

npx @soleri/cli yolo

This installs the yolo-safety hook pack (if not already installed) and launches Claude Code with permissions skipped.

Options

Flag	Effect
--dry-run	Show what would happen without launching Claude
--project	Install safety hooks to project .claude/ instead of ~/.claude/

Dry run

npx @soleri/cli yolo --dry-run

This verifies the safety pack is available, installs hooks if needed, and prints the command that would run — without actually launching Claude.

Safety guardrails

The yolo-safety hook pack composes the safety pack (see Customizing Your Agent for how hooks work), which installs an anti-deletion hook that runs as a PreToolUse lifecycle hook on every Bash command. It intercepts:

Blocked command	Why
rm -rf	Prevents accidental file/directory deletion
git push --force	Prevents force-pushing over remote history
git reset --hard	Prevents discarding uncommitted work
git clean -f	Prevents removing untracked files
DROP TABLE / DROP DATABASE	Prevents database destruction
docker rm	Prevents container removal

When a destructive command is detected, the hook blocks execution and warns the agent. Files are staged before deletion when possible.

When to use YOLO mode

Good fit:

• Rapid prototyping where you’re iterating fast
• Solo work on a personal branch
• Batch operations like migrations or refactors where you trust the agent’s judgment
• Environments with version control as a safety net (you can always git checkout)

Not recommended:

• Production environments
• Shared branches where mistakes affect others
• Tasks involving secrets, credentials, or infrastructure
• When you want to review each step before it happens

How it interacts with agent planning

YOLO mode only affects Claude Code’s permission system — it does not change the agent’s planning behavior. The agent still creates plans, searches the vault, and captures knowledge. The difference is that shell commands and file operations execute without the “allow/deny” prompt.

If you also want the agent to skip plan approval gates (the two-gate approve_plan / plan_split cycle), use the YOLO mode skill in conversation:

You: “Go YOLO on this task”

Agent: YOLO mode activated. Skipping approval gates, safety invariants preserved.

The skill activates autonomous execution at the agent level, while soleri yolo activates it at the Claude Code level. They complement each other.

---

Next: Validation Loops — let the agent iterate toward quality targets automatically. See also Customizing Your Agent for hooks and governance, and the CLI Reference for the full soleri yolo command details.