rmarji/nitpick

Your harshest reviewer, before your audience is.
One command that tears apart anything you are about to ship, ranks every problem by how much it hurts, and tells you which fixes are worth the time.
AI is great at telling you your work looks good. That is the problem.
/nitpick does the opposite. It opens the real thing the way your audience will meet it (stills from every scene of your video, your landing page at phone width, your post as a stranger skims it), hunts for what breaks, and hands you a ranked list with a fix for every item. Then it picks the 3 to 5 fixes worth making and tells you what they cost.
It never edits your work. It just makes sure you know before everyone else does.
## Install
bash
npx skills add rmarji/nitpick
Or in Claude Code as a plugin:
/plugin marketplace add rmarji/nitpick
/plugin install nitpick@nitpick
Or copy skills/nitpick/SKILL.md into your agent's skills folder (for Claude Code: ~/.claude/skills/nitpick/SKILL.md).
## Use it
/nitpick
/nitpick launch-video.mp4 --for "X launch, strangers scrolling on a phone"
/nitpick the landing page --depth quick
/nitpick this PR
No target means it nitpicks the last thing you made together. Plain language works too: "nitpick this", "tear this apart", "roast this", "what's wrong with this?"
## What you get
A real run on a 60 second launch video, trimmed:
> Biggest problem: it claims benefits it never shows. "Act the moment you arrive" is said, but nothing is ever clicked, finished or celebrated. Confidence 4/5. What would change my mind: if your audience mostly wants the idea, not the product.
> Hurts the message
> 1. No action on screen. Nobody presses Done, no confetti, no level-up. The payoff moment is the benefit and it is missing.
> 2. "Example:" on every label reads as a mockup. Use realistic fictional names in the video, put the honesty note in the post.
> 3. The first benefit lands at 0:08. The aerial has a cloud covering a third of the frame and labels unreadable on a phone.
> Weakens trust
> 4. "70 AI agents" in the hook, about 10 on screen. Viewers notice. Say "my agents" or fill the scene.
> 5. Same location shown twice, so it looks like there is only one project.
> Polish
> 6. Clunky caption at 0:50. Try "Remember by place, not by list."
> My pick: fix 1, 2, 3, 5 and 6 now (4 comes almost free). About 2 hours of re-render.
> Next action: reply "fix my pick" or "skip".
A hypothetical review of a draft social post:
> Biggest problem: the opening never names the product. Confidence 4/5. What would change my mind: if this is a reply where the product is already named.
> Hurts the message
> 1. "Meet the future" says nothing specific. Name the product and what it does in the opening line.
> My pick: fix 1 now. One opening-line rewrite, no new assets.
> Next action: reply "fix my pick" or "skip".
## How it judges
Every nitpick runs nine lenses and skips the ones that do not apply:
| Lens | It asks |
| Promise vs proof | Is every claim shown, or only stated? Do the numbers match? |