The prompt-fixer skill: find out why your prompt failed
TL;DR
prompt-fixer is the free skill for the night the output came back nothing like what you wanted. You paste the prompt that gave you bad output, it names the one thing that did the most damage, and it writes you a fixed prompt saved to ~/briefs/. Then you run the fix in a fresh session, not on top of the bad one.
You asked for a landing page and got a purple gradient with three five-star reviews from people who don't exist. It feels like the model let you down. Almost every time, it's a thin prompt instead of a bad model, and the fix is one pass over the prompt you already have.
prompt-fixer reads the prompt you actually used, checks it against the usual causes of bad output, and names the one that did the most damage in a single sentence tied to what you got back. Then it rewrites the prompt in a copy-ready block and shows you in plain words what changed and why the result will be different.
That last part is the real payoff. It's teaching you to write the prompt right the first time, which is why it's one of the 55 free skills in our freeskills repo and one of the earliest ones worth installing.
Names why your prompt failed, then hands back a rewritten one you can run tonight.
Install the prompt-fixer skill
Three lines. Grab the repo, make the skills folder if you don't have one, then drop this folder into it.
bashgit clone https://github.com/duncan-buildroom/freeskills.git
mkdir -p ~/.claude/skills
cp -r freeskills/prompt-fixer ~/.claude/skills/Restart Claude Code or start a new session and the skill is live. That's all of it. No API keys and nothing to configure.
Run it the first time
Type /prompt-fixer. It asks for three things: the prompt you used word for word, what came back, and one sentence on what you actually wanted instead. If you've lost the original prompt, it can work from what you got back, because the shape of a wrong answer shows what was missing.
It leads with the reassuring line once, that this is a prompt problem and nothing was wasted, then it gets to work. You'll get the primary cause in your own language, a couple of secondary ones as short bullets, the rewritten prompt in a block you can copy, any assumptions it had to make, and two or three lines on what changed.
Everything saves to ~/briefs/ as a file. Then start a fresh Claude Code session and paste the rewrite there. Running a fix on top of the conversation that produced the bad answer drags the wrong answer along with it.
Fix the prompt that gave you a generic page
This is the most common first run. Give it the prompt, the result, and the target, and let it do the diagnosis.
/prompt-fixer
The prompt I used, word for word: [paste it]
What I got back: [describe it, e.g. a purple gradient page with fake testimonials and pricing I never mentioned]
What I actually wanted: [one sentence]Fix a prompt when you can't remember what you typed
You don't need the original. Describe the output honestly and it works backwards from there.
/prompt-fixer
I don't have the prompt anymore. Here's what it produced: [describe what you got, including the bits that felt wrong]. What I wanted was [one sentence]. Work out what was missing and write me the prompt I should have used.Split a prompt that asked for three things at once
If your prompt has two or three deliverables in it, everything comes out at half depth. Ask for it broken up.
/prompt-fixer
Here's my prompt: [paste it]. I think I asked for too much in one go. Rewrite just the first job properly, and list the others as separate prompts I run afterwards, in the order I should run them.The 60-second version
Here is what the skill does with a prompt that failed, and why the answer is worth knowing.
- 1Hook: there's a brand new free Claude skill that takes you from staring at output that's nothing like what you wanted, to knowing exactly why it happened and having the fixed prompt in your hands.
- 2Mechanism: you hand it the prompt word for word, whatever came back, and one sentence on what you actually wanted. It runs seven checks against your wording and names the one that did the damage, usually missing context or nothing telling it what not to do. Then it rewrites your prompt in five labeled parts and tells you in plain words what it added and why that changes the result.
- 3Transformation: bad output stops feeling like proof that you're bad at this. You get a reason and a rewrite the same night, and after a few rounds you're writing the good version yourself.
- 4Close: it's free and it's one of 55. This one fixes tonight's prompt. If you want the path where you stop needing to fix them, Claude Code Club is $9 a month. Link below.
Tonight, take the one prompt that disappointed you this week, run it through the skill, and paste the rewrite into a brand new session. That's the whole job.
Common questions
Do I need to know how to code?
No, and the skill never puts code in the rewrite. It writes prompts that describe what you want to see or be able to do, which is the point.
What if I don't have the original prompt?
Work from the output instead. A wrong answer usually shows what was missing from the ask, and the skill diagnoses from that.
What if my prompt was actually fine?
It'll tell you. Sometimes the output was just a first draft, and the answer is to react in plain words and run it again. Five to ten rounds on one build is normal, not a sign anything is broken.
What if I'm stuck on an error message instead?
That's a different problem, and this skill will hand you off to the error-decoder skill in the same repo. If you're starting something new rather than repairing a prompt, brief-builder is the one to use.
Why does it say to start a fresh session?
Because the old conversation still contains the wrong answer, and it pulls the new attempt back towards it. Paste the rewrite somewhere clean.
Keep going
Want the whole prompting system, not just this one fix?
Get 650+ plug-and-play skills, MCPs & prompts, plus 7,000+ members - $9/mo, cancel anytime.
Join the Club