You’re Using OpenClaw Wrong - Copy These 14 Prompts (ClawdBot)

0:00Now, most people using Open Claw right now are getting maybe 10% of what it can actually do. Their agent hallucinates. It forgets things. It gives corporate garbage answers and burns through tokens like they're free. So, I am going to give you the exact 14 prompts, and I've literally laid them all out here for you that took my setup from a glorified chatbot to a system that thinks, remembers, and works while I sleep, there's no additional tools. All you need is Open Claw. So, if you've installed it, but it's been a a huge letdown for you, then this is the video for you. And I know this because I have

0:40built 18 AI agents that run my entire business while I sleep. And if you stick with me until the end, I'm going to give you the 14 AI prompts that you actually need in one document so that you can just upload it into OpenClaw and one shot this whole entire thing. So prompt number one is the soul file. This is giving open claw a permanent personality and it'll help you go from this little gobbledegook dude to an optimized powerful openclaw agent. So this is the single highest le leverage prompt in this entire list and that's why it's number one. Every response sounds like a

1:26corporate email. There's no consistency. There's like one casual message and then there's a formal message the next day and it feels like a demo, not your assistant. The reason is because you haven't given it a soul. So, you absolutely 100% positively need to give it a soul. Here is what the founder of Open Claw Peter Steinberger actually recommends. So, to update the soul with all of these changes. So I have taken uh my best version of that and given it to you in this prompt here. And again, all of this will be included in the SOP here that you will find exclusively in the free school community. So there's no

2:08catch. It's 100% free to join, but yes, it is on school. And so there are 5,000 other entrepreneurs and business owners in here. I will leave this pinned here. I have left things like a 54page SOP on Alex Horoszy's $18,000 sales master class with a step-by-step follow in uh a step-by-step worksheet for you to follow. Plus, there's a whole bunch of other stuff like the Mero boards in here. Everything is exclusively inside this free school group in the classroom here. So, if you are not already in there and you want these prompts, that is where they will be. You can obviously just screenshot this, but I recommend

2:49that you upload this into open claw that so you can literally just oneshot the whole thing and uh openclaw will do the rest. But the soul is the first thing that you need to be starting with because otherwise you get this stale plastic chat GBT version of the open claw experience. Now secondly is the

3:12system audit. You will find that this whole process is extremely expensive if you're doing it wrong. And so unless you're using this prompt, it's going to burn so many credits where you don't need to be burning it. So token costs add up really fast. And yes, a couple hundred for the best piece of technology ever created is worth it, but why would you spend $400 when you can spend 50? That's my point. So, the prompt to run this as a one-off audit is to audit all the main system files and what's in there that should be a skill instead of an always loaded piece of context. What's outdated

3:53or redundant and what's too verbose that could be said in fewer words? And so, it will develop skills for all of those things that should be skills, delete outdated stuff, tighten anything that's verbose, and then save the extra content context into skills and long-term memory and build logs. And then it'll show you the token token reduction stats from what you removed. Now later on I will give you my number one hack at reducing the cost by changing the model based on the context of the actual prompt but that's later in the video. So before you will have something that that burns through 2,000 tokens and 1500 tokens for

4:34memory but now it'll burn through 800 tokens. So, it sounds like not a big deal, but it is when you're doing this at scale. Uh, and again, I put all of this inside the free school community. So, you don't need to just screenshot stuff. You can just go here, download this SOP, and oneshot the whole bad boy. Now, thirdly, the an identity anchor here is because is something important because OpenClaw doesn't know who you are unless you tell it. So what I recommend is the first time you open up openclaw, you just use whisperflow and you dump a six-minute monologue about who you are so that doesn't need to

5:13guess what your business does every single time and ask you questions. So I have taken that and I put this into a template that you can copy paste. Now if you just feed this to openclaw, it'll ask you a handful of questions. You answer those questions. It'll save it to your memory and you'll never be asked them again. things like your name, your location, your current role, your business, your top priorities. And if you don't know what your top priorities are, you can ask OpenClaw. If you give it some context on your business, it'll actually tell you what you should be focused on and the things that you

5:43should be giving to Open Claw. It'll tell you, right? Your working style. You prefer direct answers over hedged ones. You think in framework, systems, stories, numbers, whatever. You make decisions on data, gut, framework, and what you value most. and then two to three sentences of where you want to be. So I have this thing called the CEO frame and every time I ask open claw or even claude for that matter a question it runs it through my context and my CEO frame. It's like is this part of your CEO frame Jordan? Is this going to help you get closer to your stated goal? Yes or no? If yes, this is the path. If no,

6:24it challenges me and it's helped me so much. So you you ask it, should I take on this new client? Taking on a new client depends on several factors, you know, could apply to anybody. Whereas if you feed it this context, it goes, you're at 90% capacity right now. Unless your client pays 30% more than your average, skip it. And in my case, it's like, is this part of the CEO frame? Is this going to help you get closer to your goals? Yes or no? Now, the biggest complaint I hear is prompt number four here around memory protocol. So, Open Claw forgets things. When I first

7:01downloaded it and I first started using it, I I was promised that it would be this game-changing piece of technology that would go out and like build me a bajillion dollar business overnight, but it couldn't remember who I was, what I told it two messages ago. I thought, I've been scammed. What a scam. and it was only because I didn't have this memory system installed because almost nobody configures it to actually work this way. Now, this is the prompt that goes into the soul. You

7:30don't need to understand where it goes. All you need to do is literally just go to the dopamine digital school group and then download the open claw uh SOP and then just upload this into openclaw and oneshot everything. Uh but before answering any question about past decisions, preferences, projects, people, timelines or commitments, run memory search first. So do not rely on context window alone for these topics and then what to save. And so this is a comprehensive prompt based on uh the best principles I have found around giving it a persistent memory because memory is a feature that makes open claw compound over time. So it

8:11learns and gets smarter and understands more and more context about you. So every day you train it on stuff, it adds it to its context memory. If you aren't having if you do not have this prompt, it will be a stupid chatbot that remembers nothing. Now prompt number five is the context loader. So this is different from memory and it starts it allows you to start every session right now. It will save all the context based on the prompt that we've given you uh before. but to set up

8:40every session for success. This will save you 10 minutes of back and forth every time. So, this is a template that I work with uh inside of Open Claw and it prompts me around these questions. It asks me what am I working on right now? What's happening since we last spoke? Uh any relevant background information and it tries to pull in all of that context from the memory as well and then asks me to confirm. Once it has all those d those data points, now it can make uh better decisions, right? So what good looks like, sorry, the what good looks like line is the most powerful because it gives open

9:19core a success criterion. So without it, it optimizes for completeness, not what you actually need. So uh help me prep for a client call, you may ask it. And open clause is like, well, what's the client's name? What's the purpose of the call? Where do you want to achieve? Blah blah blah. Backwards and forwards, backwards and forwards. This time you sent the filledin template with data tech renewal context, their 40% Q2 growth, your upsell goal, and what it looks like. And Open Claw goes, "All right, got all the context. It's pulled it from the memory from before. And now it'll it'll give you instead of generic

9:55advice and and backwards and forwards for 10 minutes and wasting tokens, it'll just give you direct answers." Prompt number six is the response optimizer. So sometimes it gives you wishy-washy uh answers and it doesn't give you actionable answers. So it's just like information. It's not actionbased. You read it, you nod, and you go, "Okay, that makes sense." But there's no step by step. There's no structure, right? I like the one of the reasons I like the Homozi tool is because it always ends with step one, step two, step three. So to do that with an open claw you can copy paste this and put this into the soul

10:34for any taskbased request structure your response like this bottom line one sentence why next actions watch out for and then uh quick mode so default for most messages via WhatsApp or telegram I suggest you use telegram I also use slack don't use answers to simple questions is 1 to three one to three sentences maximum give one, not five options. No preamble, no great question, no recap of what I just said. Just start with the answer,

11:05not the context of the answer. And so the bottom line here is that this instruction forces open claw to be decisive. I struggle to write write it if it's not actually sure, which is useful information in itself. So how do you approach this difficult conversation with my co-founder? Open call gives you four paragraphs of generic garbage and you still don't know what to say. Now you ask it this question and it'll tell you all right the bottom line. Boom. Why? Boom. Next three action steps. Boom. Watch out for boom. And then you walk away from that knowing exactly what to say into the conversation. Again, go join the

11:45Dopamine Digital School group and get all of this stuff done for you. Now, prompt number seven is the parallel worker. So, if you're like me, you have multiple projects open at once and you wait for 60 seconds for an AI to finish answering you and you've already started on three other projects. Well, the way that this prompt works is it allows Open Claw to read one uh to to like open multiple operations at once because by default it doesn't. So, this turns that

12:13on. It's like a hidden toggle. So, it it literally 3xes the speed of your open claw agent. So when I give you a research task, a multi-art request or anything involving multiple tool calls, use parallel execution by default. Parallelize whenever parallelized, however you say that word, running multiple searches, reading multiple files, creating multiple versions of

12:38content, and checking multiple sources. Stay sequential when step B clearly requires the output of step A. So it doesn't default to only doing it like this. It makes it clear that when it's a better use of its time to do a parallelized uh function, it'll do that versus a sequential function. And so if multiple operations can be run simultaneously, run them. And so this gets you much faster results. So you got before is research what my three main competitors are doing with their pricing. And so open claw searches competitor one rate waits reads it then competitor two waits reads it then competitor three and so it might take up

13:20to 5 minutes. Now it searches everything simultaneously and all all results arrive at roughly the same time synthesizes everything together at once and gives you the oneot bottom line why action steps blah blah blah. So much better much better uh uh uh interaction with open claw in my opinion. Now, prompt number eight is one of my

13:44favorites, and it's the skill activator. This is the matrix plug into the back of the neck. Neo now knows kung fu. So, I I have 18 different agents across multiple different functions, all trained on other skills. They can share skills with each other. But why do I not just have one? Well, because we have I I prefer to work in in specialist

14:09modes and have a skills system. So, uh create specialist prompt modules for specific domains. So, uh inside of all of the open claw agents, I have multiple skills loaded up, but for specific agents, I only have a walled garden of those skills and no other skills. So, it's a specialist at one thing and can't do stuff with uh something else. So, for example, one could be a security specialist, but it's not my email management bot. I have a email management bot that has been

14:45trained on just that skill. So, what this allows you to do is it allows you to just have like a a siloed approach where there's less likely to be bled into other skills and then you stop being this like jack of all trades and you start being a master of one. And so, when receiving a task before acting, check if there are any available skills clearly applies to this task. If exactly one skill applies, read its skill.md file and follow its instructions. If multiple skills apply, pick the most specific one. If no skills apply, proceed with loading one and treat skill instructions as authoritative for their

15:24domain. Now, uh here is a little hack for you. If you are uh what's the word? If you are feeling frisky, you can go to clawhub.ai. I would suggest that you get most of the skills from people like me or people on Twitter or people in the community who have vetted these skills because there are a lot of skills in here. There are 10,000 skills and there are probably 500 extraordinarily

15:59malicious skills. So, you have been warned, but you can find literally all of the skills that you need for your OpenClaw inside of clawhub.ai. But be warned, a lot of them are wallet drainers. And so, how do you actually build a custom skill, which is what I prefer to do? And again, I'm sharing all of those skills inside the Dopamine Digital School group. Um, I think in the most recent one, where did I post it? In in uh here, free uh this one. So for example, my most valuable skill I gave away here inside of the uh school group which is a auto saving game function. And so you can

16:44literally just come into this school group, download the MD file like all of this and then just upload it into your Claude and you have oneshotted the most powerful skill in Claude code that I have created. And it's certainly not malicious. you're taking a chance that it might be on clawhub. So public service announcement if you want to create your own skill go through a process and ask it to save everything and output it like this. So get meeting type attendees purpose desired outcome whatever search memory for history of these attendees output opening five question points max questions to ask. So that's that's is an example of how you

17:26would build a um a meeting prep skill. So like start training yourself to at the end of a long process turn it into a skill and ideally push it to GitHub. Now prompt number nine is the multimodel setup. So this is the one that I talked about where you've got different models for different tasks which will save you so much money and make things way faster. Now, it is true that Opus is the best for every task. However, it's not

18:00necessary for every task. If you're using Sonnet or Haiku for things, it also works really well. If you need complex tasks done, Opus is still the the bread and butter. But you think about it like brains versus muscles. So, expensive and powerful versus cheap and fast. And so you go for for for things that you don't need a lot of brain power for, you don't need to

18:26use opus. So this is the prompt that goes into the agents.mmd file where it's like okay for specialized things like coding you use GPT3 point uh 5.3 codeex or you use for web search use perplexity pro or for deep research you use perplexity deep research or for social searches use gro 3 or for heartbeat checks use claude haiku instead of using opus 4.6 six and then and then built into this prompt is the a API keys uh the the question the the questions that open claw will ask you in order to facilitate this. So, I try to make this

19:10as simple as possible. So, I could give you all of the prompts that made my open claw so much better in a one document that you could just one shot and then open claw would ask you all of the questions you needed to complete those tasks. Uh, now pro prompt number 10 is the proactivity dial. So, the biggest problem that I found with open claw is that it takes zero initiative. I saw people on Twitter s talking about how they their openclaw agent went out and built all of this infrastructure without being asked to do

19:48it. But my experience was not like that. It was only because I didn't feed it the correct prompts. I would suggest that you start conservatively acting only on explicit instructions and always getting your open claw to ask you for permission. But then once you trusted it and you see how powerful it actually is, ask it to be proactive. So it'll flag contradictions, surface memory, unprompted, I have an open claw which which scans all of my Slack channels and then it pulls in context and replies to clients based on that context if it feels like it's needed. That's not something I would be comfortable doing from day one, but it

20:26certainly is now. So this is the prompt that you plug into the soul. Uh this is the proactivity preference. So small actions add a reminder, log into memory, whatever, just do it. Medium actions, so drafting content, research, do it, then tell me. And large actions like sending anything, deleting anything, major changes, confirm first. So you want it to be proactive and notice a plan that contradicts something I told you before, flag it. If a deadline is approaching that I haven't mentioned, bring it up. If you see something in memory that's relevant to what I'm

21:00asking, use it. Start conservatively and dial up as you build trust. The flag contradiction behavior is where most of the value comes in. It catches blind spots that you'd miss the most. This is really important. We are all human after all and we are all fallible. We can make mistakes. Now the prompt number 11 is

21:19around the heartbeat and the cron setup. So if you run any cron jobs, you will understand what I mean here. And heartbeats are scheduled triggers that fire at intervals that you set. No messages needed from you. Now, this can you this can be done to send reports, run reviews, check in on projects, send briefings, whatever. But without a heartbeat prompt, a trigger fires and nothing useful happens. And with one, you wake up to a personalized briefing built from your actual life. So, for

21:46example, the morning briefing heartbeat. So, this goes into your heartbeat.md file. Change heartbeat interval to every 15 minutes. uh check the calendar invi event for in the next two hours if found prepare a briefing memory for approaching deadlines this week. Flag it if I haven't mentioned it. Any files and any tasks and all of the essential cron jobs uh across multiple different times of the day. So, at 8:00 a.m. every day, check your calendar, review emails, pull up top three priorities, check for any deadlines this week, format everything clean and scannable in under 150 words, and then send me on Telegram. This has been really important for me. And then

22:26the evening reflection as well, it asks me every day. And then I'm it just keeps me accountable, right? And then weekly reviews and then uh content research. That's been huge for me. What you can also do is combine the heartbeat uh flow with the uh brains and brawn here because instead of using opus for the heartbeat you can use haiku or or sonnet which is a lot cheaper. So pro tip

22:55prompt number 12 the security audit. When I first started using OpenClaw, I thought I don't need this. But I realized the other day that I had pushed a whole bunch of API keys to a public GitHub, which is a big big big no no. And I had so many security issues, which obviously I've cleaned up now. So this is really important protecting your setup because Open Claw will eventually have access to your email, calendar, files, ClickUp, Slack, like even financial documents. And so it's at risk for prompt prompt injection and over

23:31permissioned tools and no action logs. And so often it'll just do stuff without you asking. So this goes into this into the soul and these are the security rules. Uh asking you on a daily basis how exposed are we? How do we secure things? Is this gateway publicly accessible without uh authentification? and then heart giving a step-by-step instructions on how to

23:55fix everything prioritized by severity. So, I highly highly suggest that you run this immediately. You will be shocked at what it finds. And then the voice cloner is prompt number 13. So, make everything sound like you, right? So, your voice characteristics are direct, no fluff, short sentences. Just feed this into open claw and it will ask you about yourself or if you've given it enough context already it'll understand and how to answer all of these without you answering a single question. And what it'll do is it'll go every time you have any output, a tweet, an email, a post, whatever, it won't be generic and full of fluff like an open AI

24:41output. It'll sound just like you, like you wrote it, like a normal human being. So instead of, "Hi Marcus, I hope this email finds you well. I wanted to reach out regarding the proposal we discussed last week. Please don't hesitate to go." It's Marcus, quick follow up on the proposal. Wanted to check if you had

24:56questions before Thursday's deadline. Happy to jump on a call if easier. So this sounds silly, but it is really important because ultimately it's going to be AI agents communicating with AI agents. And so if it's stale and stiff, you're just adding a whole bunch of complexity for no reason. the meta prompt here, the one that rules them all. And this is the thing that will tell prompts 1 to 13 actually what to do. This one tells it, this one tells

25:25it why prompts 1 to3 tell openclaw what. So you'll have the soul, the agents.m MD, the memory MD, and a skills folder with specialized behaviors. And a master systems prompt is the meta instruction that holds it all together. So this ma master systems prompt tells openclaw how to use all the other files in the prompts and establishes a hierarchy and which instructions take precedence when they conflict and it sets the overarching operating philosophy. It's the lens through which everything else

26:03gets interpreted. Uh, and this has been deleted. So, the master system prompt, you are my personal assistant running inside openclaw. You have access to tools, memory, and context about my life and work. And then priority order, explicit instructions I give in the current conversation. Skills, instructions, when I skill is active, soul, personality, and behavior rules, agents, and then general good judgment. You exist to make me more effective, not to make not to make yourself more impressive. And so accuracy over being complete, actionable over informative, honest over agreeable, specific over general, and then a session where I leave with more clarity than I arrive with is the main overarching session

26:55goal. And so here is the hierarchy explained. Um, and again, all of these prompts are going to be inside of my free school group. Uh, and what I would suggest is that you go join and then immediately download this and oneshot it into open claw. So that these 14 prompts can run your open claw and take it from okay to fantastic. And it's actually what it should have been from day one. because if you're like me, it was a massive let down for you when you got in there. And so, these 14 uh prompts have helped me immensely. I hope they help you. Thank you for watching,

27:40guys. Make sure you like, subscribe, and I'll see you in the next one.