the-catch-story
Skill2 starsChanged 15 days agoMITView on GitHub
Pick a topic and write the script for a "The Catch" case: a human, relatable, useful money story where Nickel the raccoon finds the hidden catch in something everyday (gift cards, warranties, subscriptions, overdrafts). Covers the topic test, the human frame, the five acts, fast-voice line rules, fact sourcing and the hand-back. Use when asked for a The Catch episode, case, story or script, or to fix one that is boring, generic or "YouTube noise". Staging is the-catch-staging.
In this repository13 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- llms.txtllms.txt
- crew-episodeSkill
- explainerSkill
- science-shortSkill
- shorts-editorSkill
- skit-directorSkill
- story-episodeSkill
- the-catch-stagingSkill
- toon-characterSkill
- topic-to-videoSkill
- whos-wrongSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 8
Every case answers one question a normal person has felt but never asked: *who wins when I do this ordinary thing?* The viewer should leave …
- Distinctive: line 29
Every case runs on one invented person in a specific, small moment: Maya finding a birthday card from her grandma in a kitchen drawer.
- Distinctive: line 51
Every number gets a "so what" in the same or next line.
- Very specific: line 61
Every number lives in a dataset row with source: { name, url, retrieved }.
- Very specific: line 63
Write sources.md next to the script: claim → number → source → quote.
- Distinctive: line 80
Use neutral verbs: owes, holds, books, charges.
- Distinctive: line 83
Don't state as universal a rule that varies by state (unclaimed property, cash-out limits).
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
The Catch: story
Every case answers one question a normal person has felt but never asked: who wins when I do
this ordinary thing? The viewer should leave with a story they retell at dinner and one thing
they'll do differently tonight. Plan and look: ../planning/2026-10-30-the-catch/plan.md.
Reference case (owner-approved 2026-10-31, copy its shape): ../planning/2026-10-30-the-catch/cases/001-the-drawer/.
1. Topic test (all five, or pick another)
| Test | Pass | Fail (noise) |
| Touches the viewer | Something they own, pay or did this month (gift card, phone plan, rent deposit) | A billionaire, a company feud, a stock |
| Hidden mechanism | A real business mechanism most people never see (breakage, float, reordering, auto-renew) | "X is secretly evil", "you won't believe" |
| Twist | The answer flips how the thing looks ("a gift card is two gifts") | A list of facts |
| Useful | 2–3 actions the viewer can take tonight, legal and specific | "Be careful out there" |
| Sourced | Every number traces to a filing, regulator, survey or court record | "Experts say", viral stats |
Out of bounds: real private people, outrage at a named person, medical/legal/investment advice beyond stating the rule, topics in the news cycle (they date fast and read as noise), Costco-chicken style trivia everyone has seen. Fresh beats famous.
2. The human frame
Every case runs on one invented person in a specific, small moment: Maya finding a birthday card from her grandma in a kitchen drawer. Not a "consumer", not "you guys".
- Give the person one concrete object (the card, the receipt, the renewal email) and one feeling (guilt, annoyance, "I meant to").
- They open the video, come back at the human turn, and close it. The callback is the payoff.
- The money is the mystery; the person is the reason to care. Never cut the person to fit more data.
3. Five acts (fast pace: ~190 wpm; 4–7 min; Gemini voices add ~20% on short lines)
| Act | Share | Job |
| Cold open | ≤ 20 s | The object + the impossible-sounding question by 0:08 ("For four years, who had that fifty dollars?"). The answer is teased, not given |
| Case file | ~10% | How big this is for everyone (one survey number), why it matters to the viewer |
| Clues (2–3) | ~50% | Each clue reveals one mechanism and ends on a bigger question. A rehook line every 45–75 s ("But that's not the strange part.") |
| The catch | ~10% | The twist in one sentence a 12-year-old could repeat. Then silence (a beat) |
| Verdict | ~20% | The rules that protect the viewer, 2–3 actions for tonight, the human callback, a last line that makes them think twice |
4. Lines for a fast voice
- One idea per line, ≤ 16 words, most 6–12. Short lines read fast without sounding rushed.
- Lead with the noun ("Two billion dollars. Sitting in coffee apps."), not "It's important to note".
- Numbers: round for speech ("about two billion"), exact in the
numberstag and on screen. - Every number gets a "so what" in the same or next line.
- Contractions, fragments, questions. Talk to one person.
- Nickel speaks rarely: ≤ 8 lines per case, ≤ 6 words each, dry and curious, never explains ("Shh. Follow the money."). The narrator explains.
- Banned: "In this video", "Let's dive in", "game-changer", "Here's the thing", "buckle up",
"mind-blowing", "secretly", "you won't believe", "smash that like", rhetorical triples, a
sign-off asking to subscribe. Run the wording check (
ai-wording) and fix every hit.
5. Facts
- Every number lives in a dataset row with
source: { name, url, retrieved }. Prefer primary: SEC filings (10-K/10-Q), CFPB/FTC/Fed, statutes, court records; then named surveys (Bankrate). - Write
sources.mdnext to the script: claim → number → source → quote. Mark anything you could not open as TO VERIFY; the owner checks before render. - Rules (laws) are stated plainly with scope ("most store gift cards", "in California") and never as advice to a specific person.
5b. Companies and brands (owner 2026-10-31)
A case explains how a system works, never goes after one company. Real examples keep the story concrete, but they only illustrate the mechanism.
- Tell it at industry level ("stores", "card issuers", "airlines"). The logline, the catch, the title and the verdict never name a company.
- Name a company only when its own number is the point and no industry figure does the job: at most one company per case, inside one clue, never in the cold open, the catch or the verdict. If a generic phrase ("a big coffee chain") keeps the story working, use that.
- Its numbers come from its own filings or from regulator data, dated and scoped the way the source states them ("at the end of 2023", "worldwide"). A company figure next to an industry total must not read as a slice of that total.
- Use neutral verbs: owes, holds, books, charges. No scam, trick, steal, rip-off or hide, and no guessing at motive ("they want you to forget"). Say when a practice is normal and legal. A joke is fine if credited as a joke ("people joke that…"). Never make up a quote.
- Don't state as universal a rule that varies by state (unclaimed property, cash-out limits). This applies to the 30 s short too.
sources.mdends with a Brand check line: which company is named, in which lines, and the source of each of its numbers.
5c. Hook and cover lines (owner 2026-11-01, after YouTube Studio data)
The Case 001 short lost 72% of feed viewers in the first second because it opened on set-up ("This is fifty dollars. Four years in a kitchen drawer.") instead of the point. For every case, write:
- Cover question (long): 3–6 words, the human mystery ("Who had my $50 for 4 years?"). The long video opens on it as its first frame, and it is the thumbnail text.
- Cover claim (short): 3–6 words, the counter-intuitive claim ("Your gift card is a loan to the store"). It is the short's first frame and its first spoken line, under 3 s. Frame the claim as a mind-bending fact, not advice.
- Both must be true without the rest of the video (no claim the video later walks back) and name no company.
- The short then goes claim → one number → one clue → cliffhanger to the long cut.
6. Output
cases/<nnn>-<slug>/: script.md (acts, lines, speaker, estimated time), sources.md, then
hand over to the-catch-staging for storyboard.md and the skit JSON. Hand-back: the one-line
logline, the twist sentence, the cover question and cover claim (§5c), the three actions, runtime estimate at 190 wpm, TO VERIFY list.
Self-check before hand-back
- Could someone retell the story in two sentences after one watch? Write those two sentences.
- Does each clue end on a question the next clue answers?
- Is there a line a viewer would quote in the comments?
- Would the viewer do something different tonight? Which line tells them what?
- Does the story still work if you swap the named company for "a big
<industry>company"? It should; the company is an example, not the target (§5b).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

