The Free Skill That Fixes Fable 5's Biggest Problem
A former art director breaks a single sentence into a seven-piece "goal prompt" and gets Fable 5 to autonomously build a playable game, a 3D website, and a cinematic video in one session.
A structured "goal prompt" — built from seven specific pieces like stated stakes, a vivid one-sentence quality bar, and an explicit autonomy directive — turns a single sentence into a fully autonomous AI build session that ships a game, a website, and a video without further instruction.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
Someone experimenting with autonomous AI coding agents who keeps getting mediocre or generic output from short prompts.
A builder who wants one AI session to produce multiple finished deliverables (a site, an app, a video) without babysitting each step.
Someone used to writing detailed step-by-step instructions for AI tools and wants to understand why a goal-based approach can outperform that.
SKIP IF…
You're looking for a review or benchmark of the underlying AI agent's raw capabilities rather than prompting technique.
You want a technical walkthrough of the specific tools referenced (scroll-world repo, Higgs Field MCP) — they're named as examples, not explained.
TL;DR
The full version, fast.
The video argues that autonomous AI coding agents produce their best work when given a goal — a stated success state — instead of step-by-step instructions, then breaks the ideal goal prompt into seven parts: desire and stakes, a vivid one-sentence quality bar, a tool inventory plus a directive to hunt for missing resources, explicit creative freedom, a multi-pass verification loop adapted to the deliverable's medium, a delivery spec, and a closing autonomy directive. Demonstrated live, one sentence turned into a fully playable game, an infinite-scroll 3D website, and a cinematic video, all built and self-checked without further input. A free skill referenced in the video is said to convert a rambled idea into this seven-piece structure automatically.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Cold open claims a free skill fixes Fable 5's biggest weakness; Duncan introduces himself as a former art director for Apple, PlayStation, and Nissan who now runs a Claude Code community.
00:44 – 01:04
02 · Fable builds a playable game
Screen recording of "Emberguard," a browser dungeon-crawler with real-world physics (gems get pulled toward the player) and a level-up power selection screen — none of it specified in the prompt.
01:04 – 01:30
03 · Fable builds a 3D website
An infinite-scroll website with an underwater theme, particle effects, jellyfish, and original copy ("The surface forgets you quickly") — again, none of it was specified.
01:30 – 01:52
04 · Fable builds a cinematic video
A rendered video scene using Duncan's likeness from a single AI reference photo; he calls it his least favorite of the three but notes the likeness held up well.
01:52 – 02:09
05 · One prompt built it all
Reveals the entire input was one sentence using the /fable-goal skill: "Using the plugins and skills available to you, I want to create a 3D website, a game, and a video scene with myself in it."
02:09 – 02:50
06 · Best way to use Fable
Contrasts the old step-by-step instruction style (Sonnet/Opus era) with giving Fable 5 a stated success state instead, introducing the /goal prompt concept.
02:50 – 03:20
07 · How goal prompts actually work
Explains the loop: give Fable a goal, it executes, then checks its own work until success is achieved — and previews the seven pieces that make a goal prompt work.
03:20 – 03:41
08 · Piece 1: Desire and stakes
The concrete deliverable, quantity, and why it matters — e.g. five landing pages for a 200,000-person audience. Real numbers beat vague requests.
03:41 – 04:00
09 · Piece 2: Setting the quality bar
What "excellent" looks like in one sentence, using vivid adjectives ("otherworldly beautiful animations") rather than a requirements list.
04:00 – 04:59
10 · Piece 3: Tool inventory and discovery
List the specific tools, MCPs, paths, and keys already available, then tell the model to hunt the web for references, libraries, and assets it doesn't have.
04:59 – 05:18
11 · Piece 4: Give the model freedom
Grant creative freedom and decision authority — explicit permission to deviate from the plan and swap in better tools.
05:18 – 05:55
12 · Piece 5: The verification loop
Specify how many iteration passes and adapt the check to the medium: watch the rendered video, click through the site, play the actual game.
05:55 – 06:12
13 · Pieces 6 and 7: Delivery and the goal
Piece 6 states where results should land (file paths, URLs). Piece 7 closes by restating the deliverable as one sentence plus an explicit autonomy directive, with a sub-agent nudge for big jobs.
06:12 – 09:04
14 · Using the Fable Goal skill
Walks through the actual /fable-goal output: how one rambled sentence was expanded into the full seven-piece structured prompt, shown scrolling on screen.
09:04 – 09:05
15 · Outro: get the free skill
Points to the free GitHub repo for the skill and the paid Claude Code Club community.
Atomic Insights
Lines worth screenshotting.
Goal prompts replace step-by-step instructions with a stated success state, letting the model decide the how.
Real numbers in a prompt, like a stated audience size, produce sharper output than a vague request for quality.
Vivid adjectives such as "otherworldly beautiful" can outperform an itemized requirement list because they let the model interpret intent.
Naming the exact tools, MCPs, and repos available in a session turns a guessing session into one that finds what it needs.
Explicitly telling a model to search the web and fetch missing libraries or assets beats letting it guess from memory alone.
Giving an AI agent permission to deviate from your original plan can produce results outside what you would have specified yourself.
A verification loop of multiple iteration passes, adapted to the medium being built, catches problems a single pass misses.
Ending a goal prompt with an explicit autonomy directive is what lets one prompt produce several finished deliverables unattended.
For large, parallelizable jobs, explicitly nudging a model to use sub-agents can change both completion speed and token efficiency.
A single well-structured sentence can direct an autonomous AI session to produce a playable game, a full website, and a rendered video in one pass.
Takeaway
A stated success state beats a list of instructions.
PROMPT STRUCTURE
The seven-piece goal prompt works because each piece closes off a different way an autonomous AI agent guesses wrong — vague stakes, a vague quality bar, missing tools, over-constrained creativity, no self-check, no delivery spec, and no permission to finish unsupervised.
Stating a concrete deliverable with a real number and a named audience gives an AI agent more to work with than an abstract request.
A one-sentence, adjective-driven quality bar lets a model interpret "excellent" more richly than a checklist of requirements would.
Explicitly listing which tools, MCPs, and file paths already exist in a session prevents an agent from reinventing something already available.
Telling an agent to search the web and fetch missing libraries or assets turns a session that guesses from memory into one that finds what it actually needs.
Giving an agent permission to deviate from your original idea, rather than constraining it to your exact vision, can produce results outside what you would have specified yourself.
A verification pass should be adapted to the medium being built — watching a rendered video, clicking through a live site, and playing a game are different checks that catch different problems.
Specifying exactly where results should land, such as file paths and URLs, removes ambiguity about what "done" looks like.
An explicit closing instruction to work autonomously without stopping for confirmation is what allows a single prompt to produce multiple unattended, finished deliverables.
For large jobs, explicitly asking an agent to parallelize work across sub-agents can meaningfully change both completion speed and token usage.
Glossary
Terms worth knowing.
Goal prompt
A prompt that states a desired end-state and lets the model determine and execute the steps to reach it, rather than a numbered instruction list.
Success state
The measurable outcome a goal prompt targets, such as an award-winning result at a stated value, used instead of task-by-task instructions.
Verification loop
A structured self-check pass an AI agent runs against its own output, such as multiple iteration passes, before declaring the work complete.
Autonomy directive
An explicit closing instruction telling the model to keep working without stopping to ask for confirmation until it is fully done.
Tool inventory
An upfront list of the specific tools, MCPs, file paths, and access keys available in a session, given so the model doesn't guess at what already exists.
MCP
Model Context Protocol — a way for an AI agent to connect to external tools and services, referenced here as how the session reached an image/video generation service.
Counterintuitive claim about prompt quality bars.→ newsletter pull-quote↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
metaphor
00:00A free Cloud Code skill just dropped and it solves Fable five's biggest problem. I used it to make an incredibly fun game, a beautiful infinite scroll website, and a cinematic video with myself as the hero, all of it from a single prompt.
00:16I'm gonna walk you through what Fable five built, the best way to use Fable and why that's also its biggest problem, and the free skill that unlocks everything. And if we haven't met yet, my name is Duncan Rogoff.
00:27I'm a former art director for brands like Apple, PlayStation, and Nissan, I now run one of the top communities for learning Cloud Code and building income. By the end, you'll have the tools and knowledge to make anything you want with Fable five so you can save time, earn more money, stay ahead of everyone else. So focus in, close all your open tabs, and let's build.
00:48So this is the first thing that Fable built, and it's really an incredibly fun game. Like, you move around as this little fire demon, like, killing these ghouls with your little fireballs, and, like, it has, like, real world physics where, like, as you get closer to these gems, they actually, like, suck into you so you can collect them and power up.
01:04It even includes these upgrades, and I literally prompted none of this. Fable decided what the game should be, built the whole thing, and checked its own work to make sure that the game is actually playable. And it's also pretty fun.
01:18The next thing Fable built is this infinite scroll website. You can see I can scroll all the way down, and the background takes you through this incredible underwater world with these particles and jellyfish, and it's completely seamless.
01:31Again, I didn't tell Fable what to build, I didn't tell it to make it an underwater theme, I didn't write any of this copy, I did absolutely none of it. Fable came up with all of this on its own. And the last thing that Fable built is this cinematic video with myself as the hero.
01:44Honestly, it's arguably my least favorite of the three, but it did a pretty good job with my likeness, even though I gave it an AI photo of myself as the reference. This is the entire prompt that I gave it to create all three of those things, and you can see it's using the Fable Goal skill, which you can get absolutely for free at the link in the description.
02:02All I said was using the plugins and skills available to you, I want to create a three d website, a game, and a video scene with myself in it. That's it. But now you're probably wondering how I got such incredible inputs from such a simple prompt.
02:15Well, brings me to number two, the best way to use Fable five and why it's also its biggest problem. So before Fable with models like Sonnet and Opus, you would give it these step by step instructions like research this, write that, post it over here. But Fable five changed everything.
02:31Instead of giving it these instructions, now all you have to do is give it a success state. I want an award winning three d website worth $10,000. But that's not enough to get websites like the one you saw.
02:42So the best way to use Fable is by giving it a slash goal prompt. This will work for any model inside of Cloud Code, but Fable is particularly good at this. The way it works is you give it a goal, Fable will execute the task, and then it will check its own work until success is achieved.
02:57But how does a goal prompt even work? I've actually broken it down into seven pieces to make sure that you get incredible results every single time. And the whole reason that I created the fable goal skill was so that I could take a really simple prompt like this, or I could just ramble to Claude, and it would break this out into the seven core pieces to make a highly optimized prompt for Fable five.
03:20So the first thing that makes a good goal prompt is desire and stakes. The concrete deliverable and quantity and why it matters. So something like five landing pages seen by my 200,000 plus audience.
03:32Giving the model stakes makes it try harder and real numbers will beat anything vague. Number two is the quality bar. So what does excellent look like in a single sentence?
03:41Vivid adjectives beat something specific, so say something like otherworldly beautiful animations or exceptional color palettes. This is going to outperform a list of requirements because you're actually going to let Fable interpret what all of these things mean. And remember, with the new skill, you don't have to come up with any of this yourself.
03:59Number three is super important. It's tool inventory and discovery. What this means is let Fable know what specific tools, MCP's, paths, plugins, like environment keys that already exist that you want it to have access to.
04:11That way if there is something specific that you want it to use, it knows where to find it. Like in my case, I gave it a link to this other free GitHub repo called scroll world, which creates these beautiful infinite scrolling three d animations. I also gave it access to the Higgs field MCP, so it was able to create my video.
04:27And I also gave it access to a couple of resources we built exclusively for the Cloud Code Club community. The site kit, the game kit, and the scene cast kit. But whatever you have access to is fine because tool inventory is just one part of it.
04:39The other part is discovery. Tell the model to go hunt. Search the web for references and best practices.
04:45Pull libraries and repos. Fetch assets. This is going to turn a session that guesses from memory into one that finds what it needs on autopilot.
04:54Number four is the thing that makes a goal prompt different than everything else and why Fable is so powerful. You want to stay out of the model's way. You want to give it creative freedom and decision authority.
05:05Give Fable permission to deviate, swap in better tools, show what you're capable of. Let the model do its job, get out of its way. So number five is the verification loop, how it checks its own work before calling anything done.
05:18So like three iteration passes, like this means go back through the finished output with a fine tooth comb for problems and improvements. And then in my case, make sure you adapt it to the medium. So I actually gave it three things to look at, like render and watch the full video, load and click through the page, and run the script or the game on real input.
05:35So it actually played the game and checked its own work. Number six is delivery, where you want all the results to land, the file path, the URL, etcetera. And number seven is the last part.
05:45This is the part where you actually give Fable your goal. So close by restating the deliverable as one sentence. A game that's fun to play and is optimized for mobile is your slash goal.
05:55That's the command. Work completely autonomously and do not ask me for anything until you are all done. So for big jobs, add a nudge to use sub agents, so that way you can maximize your tokens and speed.
06:05So that's it. That's how the Fable goal skill works. You just ramble your idea out loud and you get back one exceptional goal prompt.
06:12You can paste straight into a fresh Fable session. So let me just show you real quick how I used it. If you remember, what I did is I went into Claude code, and you can actually do this part with Sonnet or Opus.
06:22This part doesn't matter. You just use the Fable goal slash command. I said using the plugins and skills available to you, I want to create a three d website, a game, and a video scene with myself in it.
06:32Now here's the Fable five prompt that the skill created. It turned that single sentence into this. I want you to build three showcase pieces that demonstrate what's possible when I build with AI.
06:43A three d website, a browser game, and a video scene with me in it. These will be shown to my audience of 200,000 followers and the 5,000 members of Quad Code Club as proof of what a single autonomous session can produce, so the bar is high on all three. Otherworldly beautiful three d interactions, a game that's actually fun to play for more than thirty seconds, and a video scene that looks cinematic, not like a deep fake demo.
07:05Now here's the section with the tools. I have some resources sitting around like all of the kits that I showed you. Unzip and inspect each one.
07:12Here's exactly what they are. Here's the scroll world repo that I showed you with the infinite scrolling header, and I let it know that it has access to the Higgs field MCP for any asset generation. And then it says this, beyond named resources, you have total creative freedom on the concept, subject matter, and aesthetic for all three pieces.
07:28That's crazy. Don't default to any brand system of mine. Pick whatever visual direction and theme you genuinely think is the strongest for each.
07:36They don't need to relate to each other or to me beyond the video featuring my likeness. Aim for range. I wanna see what different directions look like when given a completely open hand.
07:45The website and game should both run locally. Before you start, take stock of every tool, plugin, and an MCP you actually have and go find or fetch any additional references, libraries, or assets you need along the way. The Internet is available to you.
07:59Make every judgment call yourself. Concept, aesthetic, mechanics, copy, which resources to use. Don't defer decisions back to me.
08:05Do at least three iteration passes on each of the three pieces before you call any of them done. For the website and game, load them at the local URLs and click and play through them as a real user would. Hunt for bugs, rough edges, and opportunities to make it sharper.
08:20For the video, watch the rendered scene back and check it holds up. The likeness reads correctly and the motion feels intentional. Paralyze the three builds across sub agents where you can, and when done, give me the local project path and the local URLs for the website and the game, plus a finished video file.
08:35And then this is the goal piece, a locally served three d website, a locally served playable game, and a finished video scene featuring me, each with its own independently chosen aesthetic and three iteration passes is your slash goal. Work completely autonomously, and do not ask me for anything until you are all done. If you wanna get access to the free skill or join the Claude Code Club for all of my resources, just check the links in the description.
08:59If you wanna see how I use Claude Code for all of my content creation, just check out this video right here. I'll see you over.
The Hook
The bait, then the rug-pull.
Duncan Rogoff opens by claiming a free skill fixes Fable 5's biggest weakness, then proves it: one sentence, fed through a seven-piece prompt structure, produced a playable game, an award-caliber 3D website, and a cinematic video in a single autonomous session.
Frameworks
Named ideas worth stealing.
02:50list
The seven-piece goal prompt
Desire and stakes
Quality bar
Tool inventory and discovery
Creative freedom and decision authority
Verification loop
Delivery
The goal line and autonomy directive
A seven-part structure for writing prompts that give an autonomous AI agent a success state instead of step-by-step instructions.
Steal forAny prompt handed to an autonomous coding agent expected to complete a multi-part deliverable unattended.
CTA Breakdown
How they asked for the click.
VERBAL ASK
08:55product
“If you wanna get access to the free skill or join the Claude Code Club for all of my resources, just check the links in the description.”
Low-key verbal CTA delivered after the full lesson is already given, pointing to both a free GitHub skill and a paid $9 community — soft-sell, not gated behind the value.
A screen-recorded walkthrough of building a custom Claude Code skill that watches any viral video, breaks it into timestamped beats, and recreates it with Higgsfield's Seedance 2.0.
A 12-minute tutorial that reverse-engineers a faceless YouTube channel earning $12K/month and rebuilds its entire production pipeline inside Claude Code.
A 15-minute walkthrough of Anthropic's open-source skill that interviews you, writes your agent definition, and deploys a self-improving automation to the cloud — with an honest post-mortem on a $12 first run.
00:13
00:16
00:25
00:31
00:40
00:54
01:00
01:12
01:20
01:26
01:41
01:48
01:59
02:06
02:19
02:26
02:32
02:39
02:46
02:57
03:06
03:16
03:30
03:37
03:48
03:54
04:11
04:15
04:23
04:27
04:35
04:42
04:51
04:54
05:10
05:12
05:29
05:35
05:43
05:48
06:03
06:09
06:23
06:31
06:39
06:47
06:55
07:02
07:10
07:18
07:26
07:34
07:41
07:49
07:57
08:05
08:13
08:21
08:28
08:36
08:44
08:55
08:59
09:04
09:04
10:58
11:05
12:53
13:05
12:41
15:07