Claude Code and the /goal Command: Annoyingly Persistent
How the new /goal command changes Claude Code from giving up too easily to sticking with tough tasks.
How the new /goal command changes Claude Code from giving up too easily to sticking with tough tasks.
I recently tried out the new /goal command in Claude Code and I am in love. There are so many times that I need Claude to just go for it and keep trying until it gets a task right. Some of these tasks are risky, like making a fragile build work. Others are low risk, like ensuring there is no corner radius larger than 16 points. Consistently, Claude would just give up.
Oddly, I do not have this problem with Grok Code in Cursor. That can of worms comes with its own risks. When using OpenClaw, the agent is overly persistent and will act recklessly to achieve its original goal. I sometimes wish that Claude Code would have the same level of persistence. Until now, that was rarely possible. With /goal, Claude Code finally has an "annoyingly persistent" option.
When you use /goal, a couple of things are really important:
If you follow these guidelines, Claude Code can achieve a lot more with a hands-off approach than it could before. As a simple example, I have been having Claude Code with /goal localize some of my older apps into all iOS supported languages. It ends up using Google Translate and then falls back to using Sonnet for anything that Google Translate fails to translate. While imperfect, it makes the availability of some of my apps go from a handful of languages up to over 70. This greatly increases the potential user base.
Prior to /goal, I would spin up dozens of parallel agents to take care of this. With this command I can now write a single prompt, walk away from my computer, and come back to 70 plus locales fully supported. It is wonderful.
There are some limitations to keep in mind:
/goal.--dangerously-skip-permissions to keep things moving. You might choose to stay cautious./goal those mistakes can compound for longer. Be prepared.On the up side, using WOZ can pass some of what Claude was wasting tokens on back onto the computer. This can really reduce the usage of Claude, increase reliability, and cut down on costs.