The problem nobody talks about
If you freelance on Upwork, you know the feeling. You find a job that looks perfect. You spend forty minutes on a proposal. You pay the Connects. The I choose a Claude skill to give the final verdict.
After applying in the job, I noticed: the client has posted 30 jobs and hired for two. There are already 50+ proposals. The post is three weeks old. Every warning sign was right there. I just didn’t check, because I was excited about the work.
I did this more times than I want to admit. It wasn’t a skills problem. It was a filtering problem. So I decided to fix it. I wrote a small skill for Claude called Bid or Skip.
What checks the Claude skill?
In the Claude Chat, you send it a job link, and it tells you whether the job is worth your time. It works on Upwork, Indeed, with their Connections in the Claude.
It gives a verdict, Bid, Maybe or Skip, and a score out of 100. The score comes from four things I used to check by hand, when I remembered to:
| What it checks | Points | What it looks at |
|---|---|---|
| The client | 40 | Is payment verified? Do they actually hire? How much have they spent? What do other freelancers say about them? |
| The competition | 25 | How many people already applied? Is the client still active? |
| The fit | 25 | Does this match my skills, my niche and my experience? |
| The pay | 10 | Is it at or above my minimum rate? |
If the verdict is Bid, it suggests an angle and an opening line for the proposal. Not a full proposal. I still write those myself, and I think that matters.
The first time I used it on myself, it told me to skip a full-time role I was about to spend an evening applying for. It was right. The job needed me in an office three days a week, and my week was already full. I would have figured that out eventually. It just got me there in a minute instead of an hour.
The post in LinkedIn
I shared it on LinkedIn without expecting much. A few likes from friends, maybe.
It reached more than 50,000 people. It got 150+ comments, and most of them said the same thing: “I want to try it”.
Honestly, that surprised me more than the numbers. People weren’t just being polite. They wanted to use it. It turns out a lot of freelancers have the same quiet problem I had. We all know we should check the client before we bid. We just don’t, because it’s boring, and because hope is louder than data.
So I cleaned it up. I took out my personal details, added a short guide on how to install it, and started sending it out.
How does the Claude skill work?
The best part wasn’t the reach. It was what people told me once they started using it.
“It flagged posts that were a month old.” One freelancer wrote to me with really kind feedback, and one honest catch: the skill didn’t care how old a post was. It was suggesting jobs from weeks ago. They were completely right. So I fixed it. Version 3.1 now skips anything older than 14 days when searching, takes points off old posts, and treats a month-old post with a quiet client as an automatic Skip. You can change the 14 days to whatever suits you.
“Your weights are just guesses.” Another person pointed out that 40, 25, 25 and 10 are my assumptions, not something tested against real results. That’s fair, and I said so. Right now it’s a structured version of my own instincts. It becomes a real decision tool only when it learns from what actually happens after the bid.
“Where does it get its information?” From two places. The job itself, and a short profile you fill in once with your skills, your rate and your experience. That second part is why two people can get different verdicts on the same job. Which is exactly how it should be.
“Can AI make better decisions for you?” My honest answer: not better, more consistent. It does the boring checks I skip when I’m tired or excited. It can’t tell me whether a client might become a long-term relationship, or whether I want to stretch into something new. The final call is still mine, every time.
What’s next?
I am going to test the weights properly. From now on I am logging every job the skill scores: the verdict, whether I bid, and what happened. A reply, an interview, a contract, or silence. After 50 or so bids, I’ll know which signals actually predict a win, and I’ll adjust the scoring. Then I’ll share what I found, including the parts where I was wrong.
If you are using it too, I would love your results. More data makes it better for everyone.
Can I use Claude skills for free?
It’s free. Here’s how:
- Download here the file
- In Claude, go to Chat box, Choose + button, then Skills, Manage Skills, and click “Add skill”.
- Upload the zip file because that is the Claude skill itself.
- Fill in your Upwork profile, or let Claude ask you the first time.
- Send it a job link, or just type a keyword, using the skill
A quick tip: if you use Upwork, connect the Upwork connector in Claude so it can see the full client history. And if it saves you even one wasted proposal, tell me. That’s the whole point.
