Big Idea

The argument in one line.

Claude Code and Hermes can execute multi-week goals by decomposing them into AI-managed sprints and human-only tasks tracked in a unified dashboard, turning single-session goal loops into repeatable systems that blend agent autonomy with human handoffs.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You're running Claude Code or Hermes and have complex projects spanning multiple weeks that need decomposition into discrete AI sprints plus human-only checkpoints.
A solo founder or small team lead using AI agents for business tasks and wants a dashboard system to track and orchestrate goals across sprints without manual context-switching.
You've hit the limits of single-turn prompts and need a framework for maintaining goal coherence across long-term workflows with built-in human decision gates.

SKIP IF…

You're using Claude or Hermes for one-off tasks, customer support, or content creation—this system is built for multi-week project orchestration, not single-session work.
You haven't installed Claude Code or Hermes yet, or you're on a different LLM platform—the /goal feature and agentic OS architecture are specific to these two environments.
You're already running a mature AI workflow system with established sprint tracking and human handoff protocols—this covers foundational setup, not optimization for advanced teams.

TL;DR

The full version, fast.

Claude Code's /goal feature keeps an agent looping until a defined outcome is met, but it only works when goals are measurable, scoped to about twenty turns, and self-serve, and even then it caps out at single sprints too small for anything that actually moves a business. The fix is to wrap /goal in a longer-horizon system that decomposes a three-to-four-week binary deliverable into mini-goals, then splits each one into AI sprints and real-world handshakes only you can perform, like recording a video or recording a Loom. Before planning, the system interrogates you for course topic, deadline, existing assets, audience, connected tools, and what requires you personally, then tracks progress in a mission-control dashboard that tags you back in when human work is due.

Members feature

Chat with this breakdown.

Modern Creator members can chat with any breakdown — ask for the hook, quote a framework, find the exact transcript moment. Unlocks at T2: refer 3 friends + add your own API key.

Create a free account →

Chapters

Where the time goes.

00:00 – 01:09

01 · /goal feature explained

What /goal is: a goal that survives across conversation turns. The REPL loop rebranded as Chief Wiggum.

01:09 – 02:13

02 · Architecture: worker/judge/loop

The auto-loop broken into three parts. Released across Codex, Claude, and Hermes.

02:13 – 02:58

03 · Installing the goal skill

Copy skill from description, claude update in terminal, trust the desktop environment.

02:58 – 03:52

04 · The Three Rules nobody tells you

Goals must be measurable, scoped to 20 turns or fewer, and self-serve. Parchment graphic with worked examples.

03:52 – 05:28

05 · Live demo: short-term goals

Simultaneously runs a goal in Claude and Hermes. Agent self-heals an expired API key.

05:28 – 06:22

06 · Why /goal alone breaks down

Scope too narrow for anything meaningful. Leads into the real unlock.

06:22 – 07:45

07 · Midterm goals: the real unlock

Every goal demo is one sprint. Real goals span weeks and require human involvement.

07:45 – 08:45

08 · Chief Wiggum: AI sprints plus human handshakes

Framework: binary deliverable, 3-4 week window, split between what AI does and what only you can do.

08:45 – 09:50

09 · Agentic OS plus course upsell

Full Claude Code masterclass CTA, download the dashboard, track usage and spend.

09:50 – 11:27

10 · Live build: 500 signups in 4 weeks

Drops the Chief Wiggum prompt into Hermes. Agent asks 8 clarifying questions.

11:27 – 12:36

11 · Six mini-goals generated

Jack answers questions, Hermes generates a six-part mission with AI and human jobs broken out.

12:36 – 13:38

12 · Mini-goals become prompts

Each mini-goal auto-generates an optimized prompt. Copy-paste to Hermes/Claude to execute each sprint.

13:38 – 14:39

13 · Human handshakes surface explicitly

Action 4 tells Jack: record the promo videos. Agent waits, then runs campaigns when content is ready.

14:39 – 15:26

14 · Mission Control plus CTA

Green pulsing indicator for active missions. Mark complete, replenish, repeat.

Atomic Insights

Lines worth screenshotting.

The /goal feature makes Claude Code goals survive across turns — instead of a single-shot prompt, you have a running conversation until the defined outcome is achieved.
The RALPH loop — specify what you want to be true, run the agent until it hits the criteria — is the foundation that /goal is built on.
The failure mode of pure RALPH loops is that they exclude human judgment; the fix is bringing time and a human into the equation at defined checkpoints.
Chief Wiggum is RALPH 2.0 — a multi-week agentic OS that decomposes large goals into AI sprints and human-only handshake tasks orchestrated through a mission-control dashboard.
An agentic OS with a worker, a judge, and a loop is the minimum viable architecture for any autonomous AI system that needs to run to completion rather than run forever.
The /goal feature shipped to Codex in late April, then to Claude, then to Hermes — the timing gap is a rough indicator of which platforms are in the best-supported tier.
Long-term goals that span multiple weeks require connecting the AI's memory system to the goal tracker so the agent knows what was completed before the current session started.
Turning skills into long-term multi-week capabilities is not a feature request — it is a system design decision that requires explicit handoff and state management infrastructure.

Takeaway

Steal the human handshake frame.

Chief Wiggum playbook

Split every project into AI sprints and human handshakes, and make the handshakes explicit, time-boxed, and tagged.

The Three Rules (measurable, scoped, self-serve) are a portable prompt-quality check to use in JoeFlow sessions panel.
The live Hermes demo where the agent self-heals an expired API key is the strongest social proof moment. Replicate this for JoeFlow tutorials.
Chief Wiggum is smart positioning: it gives the REPL loop a personality and a progression story.
Mid-video product pitch works because the product IS the demo. Structure JoeFlow tutorials the same way.
The mission-control dashboard is the visual anchor. Every tutorial needs one clear UI moment that makes the system feel real.

Glossary

Terms worth knowing.

/goal: A slash command in Claude Code and Hermes that creates a persistent goal which survives across turns, looping the agent until defined success criteria are met rather than completing a single one-shot prompt.
Ralph loop: An agentic pattern where you specify exactly what should be true at the end, then let the agent run in a loop, self-checking against those criteria until they are satisfied.
Chief Wiggum: The speaker's nickname for an upgraded version of the Ralph loop that stretches goals over multiple weeks and adds human-handshake steps alongside AI sprints.
Hermes: A web-based agentic AI assistant used here as an alternative to Claude Code, capable of running the same slash-goal workflows and connecting to outside tools.
Claude Code: Anthropic's command-line coding agent that runs locally in a terminal, executes multi-step tasks, edits files, and supports slash commands and skills.
Codex: OpenAI's coding agent product, mentioned as having shipped its own equivalent of the long-running goal feature before Claude and Hermes.
Worker-judge-loop: A three-part agent architecture where one component does the work, another evaluates the output against criteria, and a loop repeats until the judge accepts the result.
Skill: A reusable prompt or instruction bundle you install into an agent like Claude Code or Hermes so it can be triggered repeatedly with consistent behavior.
Turn: A single back-and-forth exchange between a user and an AI agent; goals are often scoped to be completable within a small budget of turns.
Firecrawl: A web-scraping API that strips a page down to its meaningful content and returns clean text, making it cheaper and more reliable than feeding raw HTML to an agent.
Sprint: A short, narrowly scoped unit of work the agent can finish in one focused run, used here as the building block that adds up to a larger multi-week goal.
Mini goal: A single deliverable inside a larger mission, small enough for an agent (or a tagged-in human) to complete on its own before the next one starts.
Binary deliverable: A goal phrased so the answer to "is it done?" is unambiguously yes or no, such as "500 people opted in," rather than a vague aspiration.
Human handshake: A step in an automated workflow that only a person can perform, such as recording a video or making a judgment call, where the agent pauses and tags you in.
Agentic operating system: A dashboard layer that sits over agents like Claude Code and Hermes, tracking goals, spend, file activity, and mission status from one mission-control interface.
Mission control: The dashboard view inside the agentic OS that displays active long-term missions, their mini-goals, and which steps are waiting on the agent versus the human.
Opt-in: A sign-up where someone gives their email in exchange for content, used here as the binary success metric for the example email-course mission.

Resources Mentioned

Things they pointed at.

00:00toolFirecrawl ↗

08:45productClaude Code Full Course / Agentic OS ↗

00:00toolHermes Agent (Nous Research) ↗

02:13linkFree resources / skill prompts ↗

Quotables

Lines you could clip.

06:54

“Every goal demo is essentially one sprint.”

Tight, reframeable, sets up the problem in one line→ TikTok hook↗ Tweet quote

08:03

“Real goals require two things right now. They need technical sprints dominated by AI, and they need what I call real world handshakes.”

Introduces the framework cleanly→ IG reel cold open↗ Tweet quote

13:38

“It is that harmony between the human and the agent that will actually drive all the value.”

Philosophical close, works as pull-quote→ newsletter pull-quote↗ Tweet quote

The Script

Word for word.

metaphoranalogy

00:00Claude Code Agenetic Systems just got 10 times better. Claude Code and Hermes just released the goal feature, and it is game changing, but only if you know how to use it correctly. This enables them to solve complex problems.

00:12And in this video, I'll show you exactly how to use the goals feature and a brand new system I found that enables to build goals over multiple weeks that will help you save time and grow your business, unlocking brand new capabilities. And if you're new, my name's Jack. I built an software last tech startup with a gazillion customers.

00:31Now I'm running my own AI startups, and I just share here the stuff that works. So if you haven't already, grab that beautiful coffee, and let's dive straight in.

00:38Now these skills are incredible, but when we turn them into long term skills, which I call chief Wiggin two point o, as you're gonna see, it works for Hermes, it also works for Claude Code, it connects to our entire Agencik memory system, and it does unlock new things. And I can't wait to share with you because I've been working on this for a while and I think you're gonna love it.

00:56So we have to start with what is r slash goal and what does it mean? So r slash goal, if you're not familiar, essentially is a goal that survives across turns. So instead of saying, hey, do this thing.

01:07You effectively begin to have a conversation, a rapport if you like, with Hermes, with Open Claw, with Claude Code until the defined goal is actually achieved.

01:19Okay. Now typically speaking, this was known as the Ralph loop where you basically specify exactly what you want to be true, and then the agent would run-in a loop until it hits the criteria.

01:29Okay. So you know Ralph, but what I'm gonna show you here blends the realities because I've actually used these skills. I can tell you where they fall down, and I'm sure you've seen this too, how we can actually improve them and make them actually effective by bringing in time and also you into the equation.

01:44I call this chief Wiggum. It's kind of Ralph two point out. So the way this is gonna work, if you think about an auto loop, we have a worker, a judge, and a loop.

01:51That's effectively what's happening with the system. Now this works and has been released across different platforms. Codex released at the end of April, then Claude, and then our beautiful Hermes agent has also come to the table.

02:03Now before we go into the long term goals, the real juice, we have to understand the foundations of these skills and the kind of capabilities it unlocks in Claude and Hermes.

02:12The first I wanna do is get a short term goal skill like so. I'm gonna come over here and grab it. This will be available for you down below in description.

02:18This is really cool. I just want you to go ahead and literally copy this.

02:22Now we can do this in Hermes. We can also do it in Claude code. I'll show you exactly what I mean.

02:27Awesome. So now we're learning Claude, and you're gonna come down here, and you're gonna do forward slash g for goal, and goal will appear. Now if you don't see this yet, I want you to come over to the terminal.

02:35So first thing I'm gonna do is a quick yum. And what you might need to do, if you haven't already, is just basically update your Claude, which if you're not familiar, you come down here, you click on terminal. All you're gonna do is space Claude update, and this will make sure that you've downloaded the very latest version of Claude.

02:50Also, it needs to trust the environment you're in. So if you haven't just run this code right here in your terminal, and basically, it will make sure that it trusts your desktop. Okay?

02:57Now, we've got this here. So I'm gonna do r slash goal to begin the exercise. Now, the thing is like goals, if you don't configure them properly, are actually worthless.

03:06Essentially, for our goals to be worth anything at all, they need to do three things. They need to be measurable, they need to be scoped, and they need to be self serve. In the context of short term goals, think about it from this point of view.

03:15Okay. You'd say something that is quantifiably measurable.

03:18For example, create for me five slides. That's grim. You would say build a sign up form rather than build a SaaS.

03:24One day, we'll do that. But unlike the goals that make it New Year's Eve, these are gonna be ones that can actually be accomplished. And by self serve, it means something like, send 10 outreach emails versus close three sponsorships.

03:35See what I'm saying. Need to be measurable. Need to be scoped.

03:37So something realistically done within 20 turns until I show you the v two in this video, which you're make sure you buckled in and ready for, and they're self serves. So for example, think about it like this. If you put garbage in, you give it a good goal.

03:49You'll get good stuff out. If you give it a bad goal, you will get garbage out. So when we're ready to rock and roll, we can literally come down here and do slash goal and begin our beautiful journey.

03:57Now, of course, if you want to, you can also create a short term goal, create a skill. This basically explains everything it should be. You can literally paste this into your agent and it will basically spout for you the perfect short form skill generator.

04:08Okay. Key thing is it's measurable. It's something that's realistically cheap on 20 turns until I show you the v two in a second and it's self serve.

04:15So for example, I could say, hey, dude, do me a favor. I want you to go over on YouTube. I want to find me the top five videos on Hermes agent in the last two weeks.

04:24And I want you to tell me based on the comments and the topics, what are the ideas and philosophies that are going viral right now for that audience. And I'd like to create for me a specific HTML presentation to show what that looks like.

04:36Awesome. It's scoped. It hits all by three criteria, and then we can send that one off.

04:40And in tandem, we can do the exact same thing with Hermes agents. And then on Hermes agent, I'm gonna forward slash go. And I'm gonna say, hey, though.

04:47I would like you to go ahead and find for me the largest untapped niche based on the products on Product Hunt. I'd like you to use Firecrawl to do fantastic research. And the other part I'd like back from you is a detailed list of the best, second best, and third best products that you could be creating in 2026 based on market demand.

05:06Do me a h one presentation with some of your findings, please. Send that one off. And now this will just start to iterate and do everything it needs to in the background.

05:14I'm gonna say, if you don't have the 5.1 integration, just let me know what you need. And then we're gonna send this one off and then just let it come back with all of its key information there. And what's really cool here, you can see that it's kind of solving its own problems.

05:25Right? So for example, it tried to use a data API key that had expired. Now just on that, it's actually gone ahead and found something else, a different alternative route.

05:33It's not coming back and asking me a million things. It's just cracking on in its own way until it solves the problem, which is super handy. And guys, one of the reasons why we're using Firecrawl is because it is profoundly effective at actually getting the right information and it's significantly cheaper.

05:48It has the ability to read web pages. Because basically, web pages have loads of HTML. When we use something like Fycrawl, effectively, it's able to navigate and only extract the relevant stuff that it physically needs.

05:59So it's fantastic for that sort of stuff and what I use to build these kind of things. If you don't have it, can literally just grab your API key and put it in your credentials. And just like that, Hermes agent has gone back and got the presentation for us, launch clusters in the latest 50 posts, and it's given all that detail, and it's pretty much done this for us in one go, which is fantastic.

06:15And just like that, it's basically analyze the videos, found the top trending, had a look at the comments, and produced this incredible resource for us to look at. And this is very cool, but the goals function has one huge limitation and drawback. The scope is so small, and it has one big problem that we can actually solve with a ClaudeCode agentic operating system.

06:35And essentially, what we're gonna be able to do guys is build out longer term goals using our agentic operating system, which we can trigger with Claude or we can trigger with Hermes, and you're gonna see exactly what I mean because this is freaking epic. So let me explain what I mean. Okay?

06:49So think about this. Every goal demo is essentially one sprint.

06:53Right? For example, you have one thing that an AI agent is doing. It has to achieve that task text narrowly.

06:59But the problem is that in reality, any goal that you wanna achieve that is of any significance is gonna require you to do things. Right? Any big goal, what I would call midterm goals.

07:08So the way that I built this midterm skill, call it, achieve Wiggum, effectively is that you are gonna give it a larger task to do. For example, go ahead and launch a five day email course and get me 500 sign ups in four weeks. This is the kind of goal that you can build an agentic operating system.

07:23Right? And what it'll do is come back and ask you questions. It's based on the system that I've built.

07:27But also great. What's the course topic and the exact outcome promise? Right?

07:31What assets already exist? What's your current audience? Which platforms are you signed up to?

07:36Can Hermes access or write to any these tools directly? What channels are you willing to do on it? And only when you have that level of context can you realistically build an actual strategy.

07:47And this system that I built will work in Claude, it will work in Hermes, or any of the systems that you're building. It won't guess, it will map to the real world, and it will write this back. And crucially, it has one huge unlock that I want you to take a board with this goals thing.

08:00Because real goals require two things right now. They need technical sprints that are gonna be managed and dominated by AI, and they need what I call real world handshakes. It's stuff that only you can do, like record a three minute Loom video.

08:14Right? You know, things like that you actually need to be involved in the process of creating. So great, Jack.

08:19What is an excellent long term goal? Well, typically speaking, we want something shippable in around three to four weeks. We want a binary deliverable.

08:27So for example, it needs to be something that we can be specific has happened. And what we're gonna do here, as you can see, is break down the delivery into roles. So AI will do one thing, you'll do another, and essentially, we break every mini goal down in between what you're doing and what AI is doing at the same time.

08:43And so to do this, I'm gonna be using the agentic operating system. And by the way, if you haven't already seen it, I've got a full Claude code masterclass starting from foundation setups, building websites, power features. All this is stuff that I have not included on YouTube.

08:57It goes very, very deep design systems compliance, turning it into dollars, and everything. You can just go ahead and grab by clicking the link down below. And also, I have to call this out, it's really important, is that you can literally grab all this stuff by coming down, clawed code on her mesogenic operating system, and you can download it.

09:12I've got a full explainer guide of how you get everything. So literally, you can build all this stuff or you can literally come over and just grab this entire dashboard and then connect it to everything. Because in the home section of this, it will track your spendage, your usage.

09:24It will dream and analyze all the files on your laptop and give you specific recommendations of how exactly you can improve. And the coolest thing that I've just added is this mission control.

09:35And I'll show you exactly how it works, and if you wanna build your own, I'll put a link as well so you can create this prompt, which is down here in the long goal skill. And all you're gonna do is pretty much grab this here, and this explains the whole flow in the prompt. So you can just give it and start working around to build it, which is pretty cool.

09:50So how does this actually physically well, let's go ahead. Okay. And let's go to the Hermes agent and try this whole long term goal skill up.

09:57So what we're do is gonna come down here. It's true. Every hero does need a goal.

10:00We're gonna come down here and we're gonna copy this prompt. Then I'm going to head over to your Hermes agent or if you wanna do Claude, again, just go to the Claude section and copy that one. Okay.

10:09Beautiful. So now on my Hermes agent, let me go ahead and drop in that prompt like so. And this will cover everything.

10:15I'll explain what goes into this, but essentially we've been really specific on exactly what it needs to do in terms of creating this long term goal. Now the way this will work is we're using the r slash goal feature, but we're breaking it down into manageable chunks. And when you're required to make an action, it will actually tag you in and let you know what it needs to be.

10:33So the first thing that Hermes or Claude, based on where you entered this, is gonna come back and ask you is, what's the great goal you want me to help you ship? Give me a sentence or short paragraph, and I will turn it into a structured mission. So it comes down as asked these wonderful eight questions.

10:47What is the course topic? What is a yes, no check deadline? What already exists right now?

10:51Do you have a warm pool already? What accounts and platforms and assets do you have on hand that I need to connect to? What parts require you personally outside of chat?

10:58So you can see without this context, we wouldn't possibly be able to create a mid term goal. So I came back, I answered those questions, and again, it's asking further clarification questions. Your YouTube one, Paul Matt, has lot here.

11:09Roughly, how many subscribers or viewers do you have now? This is all built into the Agencik operating system dashboard. And all you do is you stack these mid term goals up and you keep crushing them.

11:18And then these Hermes and Claud just run away and crush it. So we just answer these follow-up questions here. So we're gonna come down and answer these.

11:24So I have roughly, call it, 1,500,000 views a month. Since there's no site list yet, I'm happy for you to, yeah, create an opt in system.

11:32That sounds fantastic. I'm happy to record four videos. That is no problem at all.

11:36I'll do short form, long form. Obviously, you just tell me what's gonna be best. And in terms of home age agent best course, whatever you think, do the research and let me know how you think that'd be decent to do that.

11:45And there are no hard constraints. Probably no paid ads. I'd like to do this organically.

11:49Let me know what you think. So again, you just give it your own and yap your own thoughts, and we send that one off and let Hermes do the magic in the background. And just like that, it's actually gone ahead and set the mission with six mini goals.

11:59So all you're gonna do now is come back over to dashboard, and you can check these in Hermes agent, alternatively, you can check it in Claude. But if I scroll down now, you're gonna see we have everything here. So look, launch Homeless course to 500 sign ups, 500 opted into free email course.

12:13Also, what you can do, and I built into this for you as well, if you come over to your home section, you can see when you scroll down, you now have this one here as well. And the idea with this in my mind is that effectively you can either chat with Hermes on the go when you're about, so you can do that or you can do it in Claude.

12:30So these are a one to one match connected to the same system. And what this will basically do is explain what your jobs are and what its job is.

12:39So for example, what this needs you to do is record the promo videos. Awesome. But then what it's going to do is shape the cost promise, build the opt in engine, and it's basically broken these down into individual chunks.

12:49So if you wanna learn more about this, what do we do? Well, we click into this, like so. And what it's done is automatically built this into a beautiful optimized prompt.

12:57So in this case, all we do is copy this one like so. We would then head over to our wonderful Hermes agent, and then we literally come down here. And all we're gonna do is come down, copy and paste like so, and send that one off.

13:06And you can see the goal here. Okay? So look, launch Hermes course to 500 sign ups, mini goal one, binary outcome.

13:12500 opted into free Hermes email course. Research and shape the strongest free five day Hermes agent email course Promise for a Creator. So what's really cool here is that it's really broken down, and this probably took a lot of time when I was building this out to really get this specific.

13:26But breaking it down into okay. Like, if this is the midterm goal, like, specifically, what many goals would we need to achieve? So what we're doing here is leveraging the power of the r slash goal, but we're also recognizing that there's a human aspect into it.

13:39So when we wanna hit those midterm goals, like actual meaningful things, not just like running around circles not really accomplishing much, We can really leverage both things together, the human and the agent, and it's that harmony, I think, that will actually drive all the value, and I found to be super effective. Oh, look.

13:55Now it's running around. It's doing all the research and will just work for us in the background. And the cool thing guys, is once this has actually completed this mission, all you literally do is you come down and you mark that off.

14:06And then we go to the next section, and again, any point you can come down and see. So for example, here on action four, it's told me you need to go ahead, Jack, and actually record the videos, then I can work on the other campaigns. So if I click on this, for example, I can see I've got the headline action that I need to do.

14:20And I can go through this and say, okay, this is really cool. This is what I need to do. And look at this, use the scripts and hooks in this location, time box ninety minutes, and include a quick note on which take is best.

14:29That's it. So now we're actually leveraging your physical work with the agents. And then, obviously, after this, it's got one final thing here for pushing the final sign ups, and this can go ahead.

14:38And look, it's even referencing all the assets that we let so, like, it's told us, hey, Jack, record this stuff, put it in this folder, and then it's told itself, action five, look, Jack's got this cool stuff there. Go ahead and push that live for us.

14:50How freaking cool is that? And then, obviously, whenever you have an active mission control, we've gotta show that, guys. We're gonna have a little bit of a green a green buzzing button to show that it's actually working.

14:58But this is really cool. And then when you're done with it, you can literally hit drop and then replenish a new one. And, course, you can go down and grab this prompt.

15:04Go ahead and it builds you up those individual sections and build it into any dashboard you want to. Or if wanna go grab this one, you click the link down below and I'll be ready to rock and roll. Now building super goals is one thing.

15:13But if you don't have an agentic operating system that's connecting everything you're doing on your computer, you don't have the visibility over cost and where you can improve. So the next thing we're gonna do is slot this into your system by watching this video right

The Hook

The bait, then the rug-pull.

The cold open is the product payoff: Mission Control dashboard on screen, superhero mascot lit up green. Then Jack cuts to the mic and drops the promise. The first 26 seconds buy him the entire 15 minutes.

Frameworks