One line of product context. Twelve testable opening hooks, ranked. The same module our buyers use every Monday — yours, no upsell.
HOOK-1 is one of the six modules behind the Klik pipeline. Tagging feeds it. Brief-tree feeds it. The copy engine consumes its output. We pulled it out, simplified the I/O, and ship it as a standalone skill so you can see the system the way we see it.
What it does, in one sentence: takes a one-line product description, returns 12 opening hooks across four angles — pain, curiosity, authority, outcome — each scored 1–10 against a hit-rate prior we keep updated from 90 days of public Meta ad data.
Three of the twelve come back starred — that's the cut. Everything below 7 gets dropped before the file is even returned, so you don't waste time rejecting bad ones. You're reading hooks that already cleared one filter.
You can run it as-is on a brand brief. You can fork it. You can wire its output into a downstream script that prompts an image generator. The point of shipping it as a Claude Code skill — not a SaaS — is that the source is yours. Read it. Tweak the angles. Swap the scoring prior. Use it ten times and forget it.
Easier to play than to read about, but here's what comes out the other end. Two different products — one DTC, one B2B SaaS — same skill, same scoring. We're showing the top six of twelve per run; the full set ships with each invocation.
$ claude run hook-1 \ "kettle & fire bone broth" → 12 hooks · 6 cut · 6 returned → showing top 6 of 12 below ★ PAIN · 9 Your gut hates your morning coffee. Here's what to drink instead. ★ CURIOSITY · 9 The 9-hour simmer that turned a $4 cup into a wellness habit. ★ OUTCOME · 9 Drank this for 14 days. My rings fit again. AUTHORITY · 8 What a functional med doc keeps in the fridge for gut-health clients. PAIN · 8 Why your protein shake is bloating you — it's not the whey. CURIOSITY · 7 Three things bone broth fixes that protein powder can't. ─ cut below 7 ─
$ claude run hook-1 \ "linear, the project mgmt tool" → 12 hooks · 5 cut · 7 returned → showing top 6 of 12 below ★ AUTHORITY · 9 Every Series A I spoke with last month had quietly switched to Linear. ★ OUTCOME · 9 Closed a sprint in 11 keystrokes. Didn't touch the mouse. ★ PAIN · 9 Notion turned your project into 47 nested docs. Antidote inside. CURIOSITY · 8 What if your tracker felt like Spotlight instead of a CRM? OUTCOME · 8 Our backlog shrunk 60% the week we left Jira. PAIN · 7 Jira is a tax. You're paying it in hours every week. ─ cut below 7 ─
The skill is one markdown file. Frontmatter on top, instructions below, scoring rule at the bottom. Read it before you install it. Tweak any part of it. The model does the heavy lifting; the file just tells it what game to play.
Most “free AI tools” are landing pages for a $29/mo subscription. We didn't want that — partly because we don't run a SaaS, partly because a skill you can read in 90 seconds is a better demo of how we think than any video.
Open it. Read it. The angles are obvious. The scoring prior is opinionated. If you disagree with the cut threshold, change it. If you want a fifth angle (we cut “social proof” — too easy to fake), add it back.
The version on the right is the file you'll receive. No telemetry. No call-home. Nothing imported.
--- name: hook-1 description: brief → 12 ranked hooks version: 1.0.0 module: 3 of 6 · klik pipeline license: MIT --- # Hook generator ## Input One line of product context. Brand + what it does. No more, no less. ## Output 12 hooks across 4 angles, each ≤ 18 words: · pain (problem the buyer feels) · curiosity (unexpected mechanic) · authority (who else uses it) · outcome (the after-state) ## Scoring Score 1–10 vs. the angle's hit-rate prior from 90d of public Meta data. Cut below 7. Star the top 3 across the whole set. ## Constraints · no emoji · no “imagine if…” openers · no rhetorical Qs longer than 8 words · each hook must work as scroll-stop in 1.5s $ claude run hook-1 "your product"
If Claude Code is already on your machine, you're 30 seconds away. If it isn't, install it first — it's free, official, takes a minute. After that, drop the file in the skills folder and run.
The CLI we built this for. One install, runs on your own Anthropic API key. You'll use it for everything downstream too.
$npm install -g @anthropic-ai/claude-code// or: brew install claude-code
Save the file we emailed you to your local skills folder. Claude Code auto-discovers it on next run — no registration step.
$mv ~/Downloads/hook-1.md \ ~/.claude/skills/// Windows: %USERPROFILE%\.claude\skills\
Pass one line of context. Twelve scored hooks come back in about nine seconds. Re-run with a tweaked brief to A/B the angle yourself.
$claude run hook-1 \ "your brand, one sentence"// → 12 hooks · top 3 starred ★
One email arrives within 60 seconds. It has the .md file attached and a 2-minute Loom showing a real run. That's the whole transaction.
HOOK-1 is module 03 of the six that make up the Klik pipeline. The other five aren't shipped as standalone skills (yet) — they live inside our retainer engagements and inside the install package for teams who buy the system. Here's what they do.
Vision-tags every ad in your library + competitors. Surfaces what's working in your category in 24 hours.
Expands a one-line brand brief into a structured brief: audience, angle, mechanic, proof, call. Feeds HOOK-1.
One line of context → 12 ranked hooks. The cartridge you're getting today.
Takes a starred hook + brief, spins ten 30-second scripts. Each scored, top three returned, ready to shoot.
Script → image prompts for Midjourney / fal.ai / Runway. Three styles per shot, format-locked for 1:1, 4:5, 9:16.
Ranks the week's finished concepts, uploads them to your Drive folder by Monday 9am with naming convention applied.
Hire the shop — we run it weekly and ship you the output. Buy the system — we install all six modules into your stack, full source, you ship it yourselves. Same engine, different operator.