In partnership with
1,000+ Claude Prompts Top Professionals Actually Use at Work
Claude can be your analyst, editor, and strategist.
But most professionals are using it to fix grammar.
These 1,000+ Claude prompts take it from grammar tool to your most powerful AI work assistant.
Sign up for Superhuman AI and get:
1,000+ ready-to-use Claude prompts to get real work done in minutes — researched, tested, and used by professionals at Google, Microsoft, and NASA
Superhuman AI newsletter (4 min daily) so you keep learning new AI tools and skills to stay ahead in your career — the prompts are just the beginning
FROM OUR PARTNER
ADS Placement Here
TODAY AI BRIEF
3 things about agents that work outside the terminal
1 Vercel released an Electron skill that lets agents drive desktop apps.
The skill for its agent-browser tool lets an agent click, type and pull data from apps such as Discord, Figma, Notion, Spotify and VS Code. It works through the Chrome DevTools Protocol, which The Code describes as turning an Electron app into an API. Directory listings show the skill exists, but the release itself is reported through a social media post. Read The Code's report.
2 Cursor's review agent now fixes the bugs it finds.
Bugbot Autofix launches cloud agents in virtual machines to test code and suggest patches. The Code reports that over 35% of proposed fixes get merged, and that the bug resolution rate rose from 52% to 76%. These are Cursor's figures as reported, and it is available to all users. Read the Bugbot details.
3 Cursor's CEO describes three stages of AI coding.
Michael Truell lists tab autocomplete, synchronous agents, and autonomous cloud agents that submit tested pull requests. He says 35% of the pull requests Cursor merges internally now come from cloud agents. That is a company figure. Read the three stages.
AI FORWORD
Agents are leaving the terminal, and the review is what remains
The shift. An agent that can click through Notion or Figma is no longer limited to code. Vercel's skill reaches into apps built on Electron, which covers a large share of desktop software. Read the report.
The loop. Cursor's Bugbot Autofix tests the patch in a virtual machine before proposing it, and the merge rate is the number that decides whether anyone trusts it. See the autofix figures.
The gap. Most figures in the coverage come from company posts and social media. The desktop skill is reported through a post, and the autofix numbers come from Cursor's own blog. Check the sources.
The reframe. An agent that can reach your desktop can also reach your mistakes. The question is no longer whether it can click, but which apps it may touch without you watching.
PRESENTED BY :
Create work your clients can rely on
With Lovable, you'll have a working version of your client's site ready for the first meeting. Plus:
Launch on your client's own domain
Build sites that get found on Google
Catch bugs and errors with project monitoring
Run every client site on one plan
Discount of $5 applies to first month of any paid plan. You need to sign up via this webpage in order to claim the discount. The discount applies to new and returning users without an active paid plan. Terms apply.
SOCIAL SNAPSHOT
What to check before you give an agent your desktop
Apps named in the report. Discord, Figma, Notion, Spotify and VS Code are the examples given. Check each app's own terms before you automate it. Read the list.
The install command. The report gives the command npx skills add vercel-labs/agent-browser --skill electron. Read the skill's files before running any install command. See the skill listing.
Fixes need a test. Cursor's autofix runs its tests in a virtual machine. A proposed patch still needs a human reviewer before it merges. Read the autofix note.
PROMPT STATION
Plan a desktop test before an agent clicks anything
An agent that drives a real app can change data you did not mean to touch. This prompt works in Claude, ChatGPT or Gemini and writes a test plan with safe steps and a stop rule before the first click.
I want an agent to test this desktop app: [APP NAME AND VERSION]. The task is [DESCRIBE THE TASK IN ONE SENTENCE], and the test account is [TEST ACCOUNT, NOT A REAL ONE].
Act as a careful QA lead. Write a test plan with numbered steps. Mark each step as read-only or as changing data, and for each changing step, name the data it touches and how to undo it.
Finish with the stop rule: the three situations in which the agent must halt and ask me, and the one check I should run on screen before I trust the result.Tip: Swap the brackets for a real app, such as a notes app or a design file in a test workspace. Never point the test at a live account, and keep a copy of any file the agent may edit.