Learn this skill and you drastically improve the quality and speed of development.
Please share this with anyone who struggles and is constantly stuck in a debugging loop 🙏
Prompt: "As this is a UI task, let Claude Opus 5 do the spec and planning. Grok will do the implementation and Claude will do the validation and review in parallel with yourself. Incorporate the results accordingly. Use the new harness sub-agent skill. All harness calls from sub-agent skill are allowed. Make sure each harness sub-agent uses the appropriate superpowers. Report to me in a CTO-level bullet point style."
How does this work? Cursor-agent in terminal is the orchestrator. I do not use cursor's autorouting for this. Only autorouting is for what cursor does itself. I stored a /harness-subagent skill with the exact commands on how to execute the different harnesses (Claude Code, Grok Build, Codex) on my system.
Of course the prompt needs some adaption based on what you try to achieve. Here are my recommendations:
- Cursor Agents CLI as Orchestrator
- Claude Code Opus 5 Everything UI
- Grok 4.6 for Review and Implementation
- Codex for adverserial Review (currently out of usage limits for me)
Why is this faster you may ask? Because that way it is spot on the first time. No more debugging loop getting stuck on a single problem for hours. None of the harnesses are 100% correct. But using two or more gets them really close to perfect. The devil is in the detail. /superpowers gives them all a routined way to work. Running different reviewers after implementation finds implementation bugs and inconsistencies most of the time.
Finally the CTO bullet point reporting keeps outputs brief and efficient.
@zilvestro And that's ignoring the obfuscation coming from the division into employee and employer taxes.
Which, from an economical viewpoint, is total BS.
Only regarding salary: what the employer pays in total and what arrives at your bank account, is way less than 50%.
@zilvestro That's an absurd view and I don't know how anyone can defend this.
Income tax is only one part. There are many more. Social security, Soli,... Easily > 50 %.
Ignoring taxes you have to pay for buying anything (VAT etc.)
Just look at the government share of the total GDP.
@tibor_tee@cursor_ai Local Cursor too or only cloud?
I'd really love some more native ways (plugin?) to spawn harness CLIs like cursor-agent CLI (win & WSL) 🙏
Any chance that's on your list? 🤔☺️
@tibor_tee Once I had a really great social media employee called Ricarda.
She was really great.
I named my Social Media Bot after her. Should I tell her? 😄
Hey, looking for a new monitor setup.
I currently have an AOC 25" + 34" Ultrawide AOC + some cheapish Dell 24" in portrait.
Thinking about switching to either
- 2x ultra-wides or
- 2x 32"-ish normal aspect ratio
Any recommendations? Needs decent color. Mostly coding.
Either Grok Bot is unfair towards Cursor Ultra and Supergrok heavy users. Or the usage limits frustration will be enormous.
Still love @bot though. My hopes for Bot were to be my main interface to the computer. But the usage limits quickly made me readapt back to CLI.
Hot take: opening Grok Bot to entry-level tiers will actually be a bad experience.
A simple office task, like searching for a new monitor, burned 3% of my weekly usage limit. Given linear scaling, this would be about 15 to 20% of the entry-level usage tier.
So you could perform 5 to 6 medium-demanding office tasks per week with Grok Bot. This feels like a frustrating experience.
And if you are really at the level of five to six medium business tasks a week, you are probably better off using the web interface.
What a story 😄
Imagine being a normal user. You'd be completely lost.
On the flip side, when done correctly it works incredibly well.
Email the Supergrok Cursor Ultra request. Got auto-forwarded from grok team to cursor team and got an - at least a heavily AI-assisted - answer within minutes.
We're making Grok Bot more widely available.
All SuperGrok Plus, Cursor Pro+, and Cursor Teams subscribers now have access.
We're also offering a free trial with limited usage for all other users.
@opencode The circle is now obvious.
1. Our models are too mighty and dangerous.
2. Government regulation.
3. Cheap mighty Chinese model on the horizon.
4. Yolo Release US Frontier Models.
5. Danger Hype Cycle Restarts