Latest / AI Fire Daily / #439 Neil: Save A Full Weekend Every Month With These 7 Claude Commands
Transcript
- 0:00You open your laptop and you type a question.
- 0:02You get a quick answer back. Then you just close
- 0:04the tab. Right. It's the standard workflow. Most
- 0:07of us use AI exactly like this. But that basic
- 0:10habit leaves massive value on the table. It really
- 0:13does. Welcome to the deep dive. We are doing
- 0:16something a bit different today. We're exploring
- 0:18a hidden layer beneath the surface. Yeah, we're
- 0:21looking at seven specific slash commands. And
- 0:24these aren't just minor software shortcuts. They
- 0:27transform Claude completely. Exactly. It goes
- 0:30from a simple chatbot to a high -performance
- 0:32operating system. That transformation is really
- 0:35the core idea here. True productivity isn't about
- 0:37writing longer prompts anymore. That era is essentially
- 0:41over. Right. I mean, it's about choosing the
- 0:43precise command for the task. We're going on
- 0:45the specific journey today. We're moving from
- 0:48heavy reasoning into token efficiency. Yeah.
- 0:50Then we'll explore focus management. Finally,
- 0:53we'll step into totally hands -free automation.
- 0:55I love that progression because typing out endless
- 0:58context is just exhausting. Let's start with
- 1:01a heavy reasoning aspect. Sometimes you just
- 1:04need the model to think harder. That is where
- 1:06our first tool comes in. It's the slash ultra
- 1:09think command slash ultra think. Right. Typing
- 1:12this triggers a massive maximum reasoning mode.
- 1:15It gives Claude a thirty one thousand nine hundred
- 1:18ninety nine token budget, which Just for context,
- 1:22a token budget is the mental scratchpad an AI
- 1:25uses to process text. Exactly right. And this
- 1:28specific budget is huge. It's about three times
- 1:31the standard processing limit. Wow. You actually
- 1:34see a rainbow terminal font when it activates.
- 1:36Wait, a rainbow font? Yeah, it physically looks
- 1:39different on your screen. That sounds kind of
- 1:40playful. But the underlying utility is deeply
- 1:44serious. It is. You're forcing the AI to catch
- 1:47complex edge cases. You're demanding it way difficult
- 1:51multi -variable trade -offs. Right, because it's
- 1:53not just generating text anymore, it's using
- 1:55those tokens to debate itself internally. It
- 1:58runs dozens of hypothetical scenarios in the
- 2:00background. It actively moves past the easy surface
- 2:04-level answers. Let's ground this with a real
- 2:06-world example. Imagine you are analyzing a Q3
- 2:10newsletter drop. Okay. Your open rate is holding
- 2:13steady at 28 percent. That is a very healthy
- 2:17open rate. It is. But let's say your click rate
- 2:19is dropping. It fell off a cliff from 4 .1 to
- 2:232 .7 percent. Ouch. And this strange drop happened
- 2:26over the last eight weeks. Standard Claude usually
- 2:29gives very generic advice here. Right. It might
- 2:31tell you to write better subject lines, or it
- 2:34suggests tweaking the button colors. Because
- 2:37standard Claude is optimizing for a fast response.
- 2:40Exactly. But slash UltraThink fundamentally acts
- 2:43differently. With that massive scratch pad, it
- 2:46diagnoses the actual edge cases. It looks at
- 2:49your historic sending patterns. Yeah, and it
- 2:51ranks the five most likely causes by actual probability.
- 2:55And it doesn't just stop at a list. It provides
- 2:57one specific A -B test for each cause. You can
- 3:00literally run these tests starting in week one.
- 3:02It functions as a real diagnostic tool. It's
- 3:05like bringing in a specialist consultant instead
- 3:07of just asking an intern. That's a great way
- 3:08to frame it. It's beat. But this brings up an
- 3:11important boundary question. Is this massive
- 3:14reasoning budget overkill for my daily email
- 3:17drag? Oh, yes. It is absolute overkill. It will
- 3:21waste your usage limits instantly. So we shouldn't
- 3:24use it for everything? No. We follow the two
- 3:26-hour rule for this specific command. If you've
- 3:29been stuck on a coding bug for two hours, use
- 3:32it. Okay. If you're doing a high stakes legal
- 3:34document review, use it. Do not use it for daily
- 3:37maintenance or basic formatting. Exactly. Save
- 3:39the heavy artillery for high stakes planning,
- 3:42not daily typos. Precisely. You want to match
- 3:45the compute power to the problem's gravity. Which
- 3:47leads to a very natural friction point. Since
- 3:50slash UltraThink burns massive tokens, we're
- 3:52going to hit our limits fast. Very fast. How
- 3:55do we actually balance the budget across a busy
- 3:57day? Well, we have to actively trim the fat.
- 4:00We need tools that pull in the opposite direction.
- 4:02Right. There's an open source skill called slash
- 4:04caveman. You just install it once from the community
- 4:07skill registry. Slash caveman. Yeah. Its entire
- 4:11job is aggressively cutting your output tokens.
- 4:13It does this by delivering raw ultra dense answers.
- 4:18It strips out absolutely. all the polite fluff.
- 4:21Right. It brutally removes those three -line
- 4:24friendly introductions. It completely skips the
- 4:26repetitive closing summaries. The name is obviously
- 4:29a joke about talking like a caveman. Just grunt
- 4:31the answer. Yeah. But the useful compression
- 4:33is very real and highly efficient. It is. I find
- 4:37this great for the messy middle of brainstorming.
- 4:40Let's say you need 10 one -sentence LinkedIn
- 4:43hooks. OK. The topic is why AI agents fail in
- 4:47production. You don't need a greeting. You just
- 4:49want the meat of the answer. Exactly. You just
- 4:52get a tight, bulleted list. Now you skip it when
- 4:54you need polished writing for a client. Sure.
- 4:56But for your own daily brainstorming, it is an
- 4:59absolute lifesaver. This is especially true for
- 5:01our pro plan users. Because that $20 tier hits
- 5:05usage limits surprisingly fast. Exactly. So slash
- 5:08caveman trims the output. But what about the
- 5:10input? Even if the AI's answers are short. My
- 5:13sessions still get incredibly bloated. Yeah,
- 5:15that happens. I still wrestle with prompt drift
- 5:18myself. Two secs silence, I'll be three hours
- 5:20in, and suddenly Claude thinks we're coding in
- 5:23Python. When you started out writing ad copy.
- 5:25Right. It gets messy. It almost always leads
- 5:28to strange AI hallucinations. We all fall into
- 5:32that trap eventually. Research shows AI models
- 5:35heavily lose accuracy as context fills up. This
- 5:39is a classic case of context bloat. Which basically
- 5:42when a full memory makes the AI forget things.
- 5:45Exactly. Under the hood... Claude is forced to
- 5:48reread every previous turn. Oh, wow. Yeah, it
- 5:51rereads the whole transcript with every new request
- 5:53you make. It's incredibly inefficient. It causes
- 5:55a major drop in reasoning quality. It also causes
- 5:58huge hidden token cost spikes. Right, which is
- 6:02why the slash clear command is so essential.
- 6:04Typing slash clear simply wipes the current session
- 6:07context. It instantly cures that context bloat.
- 6:09Consolidating your tasks and periodically clearing
- 6:12the slate helps massively. It actually saves
- 6:14about 15 percent on your context loading costs.
- 6:17The usage rule feels pretty intuitive here. You
- 6:20finish a YouTube script, then draft an Instagram
- 6:23caption. You clear it. Yep. You analyze med ads,
- 6:27them a plan, a newsletter. You clear it. You're
- 6:30isolating the variables. So clearing the slate.
- 6:33stops the model from mixing up old, irrelevant
- 6:37data. Yes, absolutely. If the new task needs
- 6:40no previous context, wipe the memory. Right.
- 6:43It keeps the processing performance incredibly
- 6:45sharp. It stops hallucinations. And it keeps
- 6:48your daily operating costs way down. Right. A
- 6:51clean slate guarantees faster, sharper, and much
- 6:53cheaper answers. It really is the most valuable
- 6:56daily habit you can build. OK, so we have deeply
- 6:58clean, efficient sessions now. but human focus
- 7:01still drifts constantly. Oh, for sure. What happens
- 7:04when you get interrupted mid -task? Anthropic
- 7:05actually solved this gracefully back in March
- 7:082026. They released a specific command called
- 7:11slash B2W. Like by the way. Literally stands
- 7:13for by the way. It handles human interruptions
- 7:15beautifully. Because focus drift is a huge operational
- 7:18problem. Normally you ask a random question during
- 7:21a massive task. And the model switches its focus
- 7:24entirely. It totally breaks the whole workflow.
- 7:27Right. but slash be to do allows safe side inquiries.
- 7:32It fundamentally does not break the main tasks
- 7:34flow. Let's say Claude is writing a 5 ,000 word
- 7:37corporate strategy document. You're pacing around
- 7:41and you suddenly remember something important.
- 7:43You can just type slash BTW. What is the standard
- 7:46image size for a LinkedIn carousel in 2026? Exactly.
- 7:50The AI opens a tiny side window execution thread.
- 7:54Claude gives you a one -line answer instantly,
- 7:57but the main engine continues running safely
- 7:59in the background. So it keeps holding the context
- 8:01for your strategy document. You maintain your
- 8:03complete context integrity. That is legitimately
- 8:06amazing. Or if Claude is deeply researching three
- 8:08different competitors, you just type slash BTW,
- 8:12add the founding year to the final table. Claude
- 8:15takes the quick note and keeps researching. No
- 8:17hard restart is ever needed. There is zero lost
- 8:19context, but there is a very critical constraint
- 8:22here. You must never use this for major project
- 8:25pivots. Wait, so I couldn't use slash BTW to
- 8:29spin up a huge secondary research project? No,
- 8:31absolutely not. If your side question is a complex
- 8:3410 -minute task... Open a new session. OK. Combining
- 8:38two heavy reasoning tasks under one hood dilutes
- 8:41the quality. It degrades the accuracy of both
- 8:44outputs. Now, keep it to quick questions. Heavy
- 8:46tasks need a new session. Precisely. It is a
- 8:49quick side window, not a new engine. Mid -roll
- 8:52sponsor or read placeholder. So we are managing
- 8:54our active open sessions really efficiently now.
- 8:57We really are. We're thinking deep. We're cutting
- 8:59fluff. We're handling interruptions. Now, how
- 9:02do we get the AI to work? when we step away entirely.
- 9:06We have to move into the automators. We are finally
- 9:08breaking free from the keyboard entirely. Let's
- 9:10explore that. The command for this is slash loop.
- 9:13This specifically handles local system automation.
- 9:16It executes tasks repeatedly at very specific
- 9:19intervals. It functions sort of like a highly
- 9:21intelligent smart timer. Right. Claude converts
- 9:23your plain text timing into a P06 cron expression.
- 9:27Which is a standard system for scheduling recurring
- 9:29computer tasks. Exactly. It assigns your loop
- 9:33a unique 8 -character tracking ID. It's remarkably
- 9:36flexible. It supports scheduling by seconds,
- 9:39minutes, hours, and days. You can actually run
- 9:41up to 50 scheduled tasks per active session,
- 9:44though there is a built -in fairness offset.
- 9:47What does that fairness offset mean exactly?
- 9:49It adds a slight deterministic delay to the trigger
- 9:52time. It's usually up to 10 % of the scheduled
- 9:55interval. OK. So an hourly task might fire a
- 9:58few minutes past the exact hour. Let's see. It's
- 10:01like a staggered exit at a stadium. Yeah. If
- 10:03everyone leaves at once, the exit's completely
- 10:05jammed. This prevents global server spikes on
- 10:08the hour mark. That's a perfect analogy. Let's
- 10:10look at a highly practical example for this.
- 10:12Let's do it. Automating the dreaded blank page
- 10:15problem for writers. You simply tell Claude to
- 10:18check your local drafts folder every four hours.
- 10:20Exactly. It quietly scans your files and picks
- 10:24the oldest unfinished piece. Let's say a rough
- 10:27draft has been untouched for 11 days. It automatically
- 10:30suggests three fresh headline rewrites. And it
- 10:33writes one noticeably cleaner opening line. It
- 10:36uses psychologically driven headlines instead
- 10:38of repetitive social media hooks. It saves the
- 10:40entire output as a neat comment block. It even
- 10:43suggests a specific 400 word cut. Wow. It diagnoses
- 10:48your structural flaws perfectly. But there are
- 10:51some strict hardware limitations here, right?
- 10:53This relies entirely on your local machine running.
- 10:55Yes. Shutting your laptop completely kills the
- 10:58active loop. Unfortunately, yes. Closing the
- 11:01lid pauses the cron job entirely. It is also
- 11:05permanently bound to your current active session.
- 11:07Finally, it has a stripped seven day expiration
- 11:11date built in. So it's meant for short term sprints.
- 11:14It is definitely not permanent. So what is the
- 11:16alternative for longer? background tasks. We
- 11:20use the slash schedule command. This handles
- 11:22pure cloud -based automation. It essentially
- 11:25clones your current project state to anthropic
- 11:28servers. It runs purely on their remote infrastructure.
- 11:31So it fires accurately even if my machine is
- 11:33completely turned off. Exactly. It elegantly
- 11:36overcomes the main limitation of the local loop
- 11:38command. It does. But you have to remember, it
- 11:41removes access to your local file system. It
- 11:43can't read your hard drive. However, integrations
- 11:46with Slack, Notion, and Gmail remain fully functional.
- 11:50The daily run limits totally depend on your specific
- 11:53subscription plan. Pro users get about five runs
- 11:56per day. Max users get around 15 runs. Team tiers
- 12:01get roughly 25. Let's discuss setting up a daily
- 12:03morning brief. This is arguably the perfect use
- 12:07case for slash schedule. You schedule it for
- 12:09exactly 7 a .m. daily. It pulls yesterday's top
- 12:12news from five different sources. It checks the
- 12:15Anthropic Developer Blog and Hacker News. It
- 12:18checks the top posts on the AI subreddit and
- 12:21TechCrunch. Right. It intelligently picks the
- 12:23three most important, relevant stories. It writes
- 12:26two concise sentences analyzing each one. Then
- 12:29it sends the finalized result to a private Slack
- 12:31channel. Whoa, imagine waking up. Your laptop
- 12:34was asleep all night, but the cloud just curated
- 12:36your entire industry while you slept. Two -sec
- 12:38silence. That is a wild feeling. It is a profound
- 12:42structural shift in how we work. You read a totally
- 12:46clean, personalized brief over your morning coffee.
- 12:49The cloud simply does the heavy lifting for you.
- 12:51Okay, so if I'm building a system right now,
- 12:54how do I instantly decide whether a task needs
- 12:57a loop or a schedule? You use a very simple decision
- 13:01matrix. Does the task need local files or very
- 13:05short intervals? You choose loop. Right. Does
- 13:07it run offline or need to run for months? You
- 13:09choose schedule. loop for local, short intervals,
- 13:13schedule for offline, long -term cloud runs.
- 13:16It really is that straightforward. You can even
- 13:18build advanced cloud routines later. You can
- 13:20actually trigger them dynamically with GitHub
- 13:22push events. With all these complex commands
- 13:25firing off constantly, how do we track our health?
- 13:27That's a great point. We might just be creating
- 13:29massive chaos instead of actually saving time.
- 13:31That brings us to our final and arguably most
- 13:34important command. It's called slash insights.
- 13:38Yeah, it functions as your personal AI auditor.
- 13:41It securely analyzes your actual session history
- 13:43from the last 30 days. It delivers a really comprehensive
- 13:47HTML report right in your browser. It visually
- 13:50acts as a digital personal coach. Exactly. It
- 13:53shows a visual heat map revealing the tasks you
- 13:56perform most frequently. Very cool. It flags
- 13:58the exact conversational patterns where you are
- 14:01wasting tokens unnecessarily. It identifies completely
- 14:05unused integrations matching your work style.
- 14:07And it doesn't just complain. It suggests highly
- 14:10actionable fixes. It might flag that opening
- 14:14too many short sessions is getting expensive.
- 14:16It will show you how consolidating related tasks
- 14:19actually saves you tokens. You really should
- 14:22run this on the first Monday of every single
- 14:23month. Spend just 10 minutes reviewing the HTML
- 14:26report. Identify one small adjustment for your
- 14:29daily workflow. But reading a list of problems
- 14:31is easy. Changing how you actually work day to
- 14:34day is incredibly hard. That is exactly why we
- 14:36recommend the seven day mastery plan. Okay. You
- 14:39intentionally adopt only one command per day.
- 14:41Day one is just getting used to slash clear.
- 14:44Use it faithfully between unrelated tasks. Right.
- 14:47Day two is slash ultra think. Try applying it
- 14:49on one genuinely hard task. Day three is slash
- 14:53insights. Read your personal performance report.
- 14:55Day four is slash BTW. Use it seamlessly during
- 15:00a long writing task. Right. Day five is slash
- 15:02caveman. Try it on a messy brainstorming session.
- 15:05Day six is slash loop. Automate one simple local
- 15:09system check. And day seven is slash schedule.
- 15:12Set up your personalized morning brief. I know
- 15:14some people are listening right now thinking,
- 15:16why not just install and launch all seven of
- 15:18these commands this afternoon? Because changing
- 15:20your fundamental workflow is psychologically
- 15:23very hard. Deploying one new command per day
- 15:26prevents it from feeling like massive homework.
- 15:29Yeah. It actively prevents complete system overwhelm.
- 15:32Avoid overwhelm. Slowly stacking these commands
- 15:35one by one builds lasting habits. Let's briefly
- 15:38summarize the core philosophy we really uncover
- 15:41today. Please do. True productivity in the modern
- 15:43AI era is evolving fast. It is no longer about
- 15:46writing massive paragraphs along prompts. No.
- 15:48It's about deploying the exact right command.
- 15:50You match the specific command to the specific
- 15:52friction in your workflow. You optimize absolutely
- 15:55every second and every token spent. Beat. Thank
- 16:00you for joining us on this deep dive. Thanks
- 16:02for having me. I want to leave you with a final
- 16:04thought to mull over. If Slash Insights can already
- 16:07accurately audit our 30 -day habits, how long
- 16:10until Claude simply auto -triggers these commands
- 16:12for us? That's the real frontier. Imagine it
- 16:14realizing we need Slash Caveman or Slash UltraThink
- 16:18before we even finish typing the question. AT
- 16:20Row music.