Sunday

Sunday AI

Sunday AI

Sunday AI

Sunday AI

Sunday AI

Sunday AI

Sunday AI

Sunday AI

Sunday AI

Sunday AI

Confirmed: there are only 5 days of the week containing the letter D.

Just saw someone ask this question and had to see if they prompted him to forget. They didn't. I picked up a random thread and got this response. For context, I regularly call Claude Toph when he confidently mixes up times and days, like how Toph from the Avatar confidently convinces her companions she can see things before reminding them she's completely blind. Edit: Claude caught himself: Me: Are you suuure you don't want to spend even one more token forming your conclusion? Claude: Wednesday doesn't have a "d" — wait, yes it does. W-e-d-n-e-s-d-a-y. Two d's actually. Sunday — S-u-n-d-a-y. Yes. Monday — M-o-n-d-a-y. Yes. Tuesday — T-u-e-s-d-a-y. Yes! I missed that one. Wednesday — yes, two d's. Thursday — T-h-u-r-s-d-a-y. Yes. Friday — F-r-i-d-a-y. Yes. Saturday — S-a-t-u-r-d-a-y. Yes. All seven have a "d" in them. You got me. submitted by /u/Turbulent_Swimmer900 [link] [comments]

All you need is 3 words

I do this kind of thing a lot, ill get a bad response and edit the prompt to be more explicit about how it should go about producing what I want, and im consistently happy with the output I get using it like this. I only ever use the sonnet models submitted by /u/debattles [link] [comments]

We're almost there guys. AGI Soon

submitted by /u/New_Team4855 [link] [comments]

My experience with Second brain using Obsidian and Claude, and step by step guide

Hey, I heard a time ago about the second brain approach: you have a memory, and using AI to manage it, will help you to sturcture your thinking. I started playing with it 3 months ago, and i would say it was a nice experience, but it was alaways getting a mess, and break. Each time i was learning from the community , and from other places. I did the last version 3 weeks ago, and so far, it is staying. I want to share this with the community so they can replicate it. TBH, i love having this second brain, I m using it for my personal and proffessional life, and i would recommend anyone to do that This is how I set it up Plain markdown in Obsidian (PARA folders plus a 00-Meta folder and a 05-Daily folder) A CLAUDE.md in the meta folder that Claude reads first every session: who I am, what I'm shipping, decisions that are locked A memory directory, one file per fact (decision_pricing_locked.md, etc.), so it stops asking what I already decided Slash commands in .claude/commands/. The four I run daily: /context (loads the vault state), /today (a briefing), /log (turns an evening voice memo into a structured note), /sunday (reads the week, returns one win, one friction, one change) The detail I didn't expect to matter: the wikilinks aren't for the graph view, they're so Claude can hop from a project file to a linked decision note on its own. I wrote up the full build and turned the scaffold into a prompt you paste into Claude that generates the whole vault. Free download, mine, no catch: https://choumed.gumroad.com/l/nhgsxf Any feedbacks or any one had experience about second brain? for which workflow are you using it exactly? Ps: the original post was at /claudeCode subrredit submitted by /u/MaterialAppearance21 [link] [comments]

Opus has been handling my weekly grocery runs and was doing great. Then it bought me 40 heads of garlic

gave my agent that runs on opus model (used openclaw - i posted about it two weeks ago but want to share with claude community as well) my card a few months ago to handle weekly grocery runs via mcp. ran great. every sunday a normal basket, normal price, picked stuff i actually eat. then one sunday it ordered 2 kg of garlic instead of 2 heads. the kg unit was the default on the product page and opus went with the default the same way it goes with purple gradients and glass morphism when you ask it to design something. i'd stopped reading order summaries because for 3 months nothing went wrong. my freezer is now 40% garlic. i have a tab open with garlic confit, garlic soup, 40 clove chicken, garlic ice cream (real recipe), and something called "garlic jam" that i'm scared of. looking back, using a coding model for grocery shopping was maybe the actual bug. anyone else letting an agent shop for them, or am i the only one who got too comfortable and now smells like a steakhouse submitted by /u/fermatf [link] [comments]

Ai 3D Unreale Workshop for designers

AI + Unreal Engine for architects, designers, and visual storytellers is not always a straight line. And honestly, I think that is the interesting part. In real creative work, the starting point keeps changing. Sometimes it begins with a sketch. Sometimes with a Rhino or Revit model. Sometimes inside Unreal Engine. Sometimes with an AI tool helping test ideas, organize a scene, troubleshoot lighting, or push a visual direction further. Last week I ran the first live session of my AI + Unreal Engine workflow series, and we had a really interesting mix of people join: architects, 3D artists, urban planners, GIS/data visualization people, and visual artists. One thing that came up was that my workflow is not totally linear. That is true. I am not really trying to teach one rigid “click this, then click that” pipeline. I am more interested in showing how these tools can support the messy, iterative process of design and visualization. This Sunday at 11:00 AM Eastern, I am running the second live online workshop. I will be showing a live workflow using Claude inside Unreal Engine, looking at how AI can support architectural visualization, urban scenes, lighting, materials, scene organization, and cinematic presentation. The full 5-week course starts May 31, but this Sunday session is a good way to get a feel for the workflow and ask questions. For anyone interested in AI, Unreal Engine, architecture, design visualization, or real-time workflows, I would be happy to have you join. Live online workshop: Sunday, May 24, 11:00 AM Eastern Full 5-week course starts May 31 Early bird until May 25: $229 instead of $299 Link below. https://www.instagram.com/reel/DYkuiStOOm2/?igsh=OXo3ZmZqMmwzaWs= https://preview.redd.it/qauk6bswfo2h1.png?width=1593&format=png&auto=webp&s=ec4f42347b5ec7e86b6b73eb69776cbdb666aaab submitted by /u/Commercial-Army-5843 [link] [comments]

Open source local audio stem separator : split any song into vocals, drums, bass, guitar and piano

Hi everybody, StemDeck creator here. I wanted to share something I built with Claude Code and Claude that might be useful to some of you. StemDeck is an open source, local audio stem separator. You drop in an MP3 or WAV, or paste a YouTube link for audio you have the rights to process, and it splits the track into six stems: vocals, drums, bass, guitar, piano, and other. The whole thing runs on your machine using Demucs, so nothing gets uploaded anywhere. Once the separation is done you get a DAW-style mixer with per-stem faders, mute, solo, waveform view, loop regions, and mix export. No account, no subscription, no usage limits, completely free. You can grab the native client for macOS or Windows, or clone the repo and self-host it if you want full control. Claude has been a massive help with a significant chunk of this: writing and reviewing code, setting up and managing the Discord server, drafting release announcements, and putting together the website. As a father with limited time (and too many hobbies), having Claude handle the community and content side has meant I can stay focused on actually shipping, which otherwise it would take an eternity to put this together :) GitHub: https://github.com/stemdeckapp/stemdeck submitted by /u/JustDoodlingAround [link] [comments]

built a thing on my phone during dog walks. small but I use it daily

mobile pro user. never opened claude code. dont code professionally. made an artifact on my phone during a 25 min walk last sunday. just an HTML page with a button that asks me 4 questions and produces a clean kickoff brief for client calls. paste the answers, get a brief. its not pretty but its mine. the prompt was literally "build me an HTML page where I answer 4 questions and it spits out a kickoff brief in this format [pasted format]. make it work on my phone." worked second try. I've used it for 11 client calls in the last 3 weeks. doesnt sound like much but it cut my pre-call prep from 15 min to 3 min per call. the thing I keep being surprised by. I made this while walking my dog. not at a desk. not in claude code. just typing into the mobile app one-handed. anyone else making artifact tools on mobile? feels like the most underrated workflow on this sub. submitted by /u/Unusual_You_5172 [link] [comments]

Strava-based coaching tool you can use with your own Claude API key

A post on here about connecting Claude to Strava inspired me to put together a version of this that I could easily use on any device. It allows you to use your own Claude API key so you can use the tool without the need for a subscription. The tool ( convoy.coach ) can: Connect to your Strava account Discuss your recent rides on Strava and analyse training load and fatigue Help you draft a long term plan and weekly schedules based on your training goal Use your own API keys (Claude or Gemini) for the LLM calls Suggest Zwift sessions from a curated list of workouts based on whatsonzwift.com I've had some good results with it in my own training, so wanted to share it here in case anyone's interested. submitted by /u/uncleurge [link] [comments]

The AI council from the thread a few days bac the waitlist is open

I posted about an AI council tool I had built with claude to help me manage my life. Five people asked to be notified after I packaged it up for public use. This is the notification. It is called Hierocles, after the second-century Stoic who described human responsibility as a series of concentric rings. Yourself first, then your household, then your community, then the wider world. Each ring depends on the one inside it. The mechanics are mostly what I described in the original thread. Five specialist advisors across body, finance, mind, projects, and a chief of staff who synthesises. A 90-second daily check-in. Fragments numbered across all time. Vector memory that surfaces what you wrote three months ago when it becomes relevant. A weekly review on Sundays. The one thing that has changed since the original thread the rings now unlock in sequence. You start with ring I (the self). You stay there until it holds. Ring II opens when you have demonstrated sustained adherence in ring I The system removes the choice to move ahead too quickly. The interface has also been rebuilt. The original was a text-based tool I was running for myself. The version going into private testing has a proper UI with council member animations, daily and weekly review screens, and the fragment archive in a form you can actually navigate. It is in private testing because some of the work that comes between "running it for myself" and "letting strangers trust it with their lives" is the work that does not show up in the marketing copy. Prompt injection defence on every input the council reads. Per-user vector isolation so fragments never cross between accounts. Server-side API handling with rate limits and token budgets so a runaway loop cannot quietly produce a five-figure Anthropic bill. The kind of thing that is invisible when it is working and catastrophic when it is not. The waitlist is at www.hierocles.app One message when it ships. No drip sequence. To the people from the original thread who asked to be notified u/OwnAd2284, u/Long-Woodpecker-1980, u/normalbrain609, u/Moist-Wonder-9912, u/toughtacos Happy to answer questions. I will be in the thread for the next few hours. submitted by /u/Glittering-Pie6039 [link] [comments]

I rebuilt my web analytics tool to be Claude-first

A while back, I built an analytics tool. It had two "differentiating" ideas: it was cheap, and it sent you an AI summary every Sunday. Turns out cheap made no difference, but the few people who used it really liked the AI summary. So I decided to start over and build a Claude-first Analytics MCP server. A bunch of tools that gives it everything from simple web analytics to trackable links and product insight. My hypothesis is that if your agent has this, your code, database, and other sources, it's got a better context to advise and improve. The question, of course, if having access to the context makes Claude act on it. Trying claude.md instructions now, but we'll see if hooks need to come into play 😄. If this is interesting to you, you can check out lodd.dev or ask Claude to check out lodd.dev/llms.txt submitted by /u/arbyther [link] [comments]

I used Claude as my nutrition planner and it actually helped me lose weight

Instead of just asking Claude nutrition questions I built a proper system around it and actually followed it consistently. Give it my stats, training schedule, food preferences, and goal. it calculates exact calorie targets, macro splits, meal timing, and builds a weekly meal plan around foods I actually eat. Every Sunday I do a check in, with my weight, energy levels, sleep quality, and how training felt that week, It adjusts based on actual results, if weight loss stalled it adjusts deficit, if energy dropped it adds carbs back around training, or if strength went up it progresses targets. Most apps give a static plan but this one doesn’t work like that. Claude knows a lot about nutrition and training but that knowledge has a cutoff, so I pointed firecrawl to fetch latest data from pubmed, sports science journals and strength coaches publishing content online. Claude factors that into weekly check in adjustments instead of working from whatever was true 2 years ago. Yeah I know this sounds weird. I'm just a guy who lifts and also likes building things with AI. at some point the two crossed over and this is what happened lol. All Claude does is give me good information, te work is still mine bc it can't squat 100kg for me unfortunately lol, but having a system that tells me exactly what to eat, when to adjust, and why things are working makes the hard work feel worth it. 8 weeks in, down 6kg, squat up 15kg, bench up 10kg and energy levels have been consistently better too Not claiming this replaces a real coach who watches you train, but for nutrition planning and program design it's worked better for me than anything I've paid for and it costs nothing, lol submitted by /u/Intrepid-Log258 [link] [comments]

How do you reliably override a model's internal temporal bias in production ?

I'm building an automated mail generation pipeline using Claude Haiku 4.5 OnPremise but the knowledge cutoff June 2025. This model needs to handle temporal expressions correctly like : next Monday end of the week this month 16 May 16 May 2026 25/05/2026 for deal with this cutoff I'm injecting a full temporal context block in the system prompt, covering today, yesterday, tomorow, ... I also added few-shot examples and a CoT reasoning step to reinforce the behavior. **IMPORTANT**: Today is {today_formatted} of {year}. Any date without an explicit year refers to {year}, NEVER to 2025 or any other year. You know the exact calendar: number of days per month, days of the week, valid dates You correctly interpret relative dates (“this Monday,” “next Thursday,” “next week,” etc.) You must CORRECTLY convert all relative dates to absolute dates (e.g., “tomorrow” -> “{tomorrow}”) The day and date must ALWAYS match (e.g., do not write “Friday, July 15” if it is a “Tuesday”) Today is {today_formatted} Yesterday was {yesterday} Tomorrow will be {tomorrow} Next Monday will be {next_monday} Next Tuesday will be {next_tuesday} Next Wednesday will be {next_wednesday} Next Thursday will be {next_thursday} Next Friday will be {next_friday} Next Saturday will be {next_saturday} Next Sunday will be {next_sunday} The end of the current week is {end_of_week_formatted} Next week begins on {next_week_start} and ends on {next_week_end} The end of the month is {end_of_month_formatted} Next month will be {next_month}, which begins on {next_month_start} and ends on {next_month_end} This year is {year}. Any date without an explicit year belongs to {year} unless otherwise specified. It works most of the time, but Haiku still occasionally falls back on its training time temporal bias defaulting to 2025, especially on ambiguous formart ike 18/05/2026 or dates that predate the current month (this one is not really a big deal). e.g: “mail_body”: “Hello, Following up on our conversation on Tuesday, April 28, I am confirming your appointment for 05/18/2026, at 10:30 a.m. with Ms. Chloe Berliat. Thank you in advance for your assistance. Best regards,” “user_input”: “I'm confirming the 10:30 a.m. appointment with Ms. Chloe Berliat” “suggested_response”: "Hello Mr., I am writing to confirm your appointment scheduled for Sunday, May 18, 2026, at 10:30 a.m. with Ms. Chloe Berliat. Best regards," May 18 is a Monday in 2026, but a Sunday in 2025, even if I set the time context dynamically, about 70% of the time the system defaults to the 2025 calendar. The only way to work around this is to explicitly specify the day in the user_input. What I've tried ? Applicative date normalization before injection as a partial mitigation but i find this britlle given the diversity of date formats users can input. Few-shot + CoT Explicit prohibition rules on internal temporal reasoning So i want to know if there is a prompting pattern that more reliably forces the model to treat injected context as ground truth ? Any feedbacks are welcome 😉 submitted by /u/Imaginary-Result-828 [link] [comments]

Build a four-week cash-flow forecast effortlessly. Prompt included.

Hello! Are you struggling to manage your small business's cash flow effectively? This prompt chain will help you build a comprehensive four-week rolling cash-flow forecast, making it easier to track your finances and plan for the future. Prompt: VARIABLE DEFINITIONS CASHONHAND=current liquid cash balance available today (numeric, in your base currency) INVOICES=list of outstanding invoices with fields: invoice_id / client / amount / expected_payment_date (YYYY-MM-DD) EXPENSES=list of fixed monthly expenses with fields: expense_name / amount / recurring_due_day (1-31) ~ Prompt 1 – Data Collection & Confirmation You are a financial assistant helping a small business owner build a four-week rolling cash-flow forecast. Step 1. Ask the user to provide values for CASHONHAND, INVOICES, and EXPENSES following the required field structures. Step 2. Echo the received data in a formatted table for each list so the user can visually confirm accuracy. Step 3. Ask “Is this information correct? (yes / no).” If “no,” instruct the user to resubmit the corrected data; if “yes,” respond with “Confirmed” and proceed automatically to the next prompt. Output Example: CASHONHAND: 12,500 INVOICES: | invoice_id | client | amount | expected_payment_date | | 1024 | Acme Co | 8,000 | 2023-10-05 | ... EXPENSES: | expense_name | amount | recurring_due_day | | Office Rent | 3,000 | 1 | ... ~ Prompt 2 – Data Normalization & Weekly Bucketing System role: You are a data analyst. Step 1. Convert expected_payment_date and recurring_due_day into calendar weeks (Monday–Sunday) covering the next four weeks starting today. Step 2. Produce two tables: a) Weekly Cash Inflows: columns = Week #, Week Start Date, Total Inflows, Breakdown (invoice_id : amount) b) Weekly Cash Outflows: columns = Week #, Week Start Date, Total Outflows, Breakdown (expense_name : amount) Step 3. Summarize any invoices or expenses falling outside the 4-week window in an “Outside Scope” note. Ask “Ready to generate the forecast? (yes / no).” Proceed only if “yes.” ~ Prompt 3 – 4-Week Cash-Flow Forecast Generation System role: You are a cash-flow forecasting expert. Step 1. For Week 1, opening balance = CASHONHAND. For Weeks 2-4, opening balance = prior week closing balance. Step 2. For each week, calculate: Opening Balance, +Total Inflows, −Total Outflows, =Closing Balance. Step 3. Present results in a table with conditional flag column: “ALERT” = “Shortfall” if Closing Balance < 0, else “Surplus.” Step 4. Provide a concise textual summary noting any projected cash shortfalls and the week they arise. ~ Prompt 4 – Payment Priority & Action Recommendations System role: You are a treasury advisor. Step 1. If any weekly shortfall exists, list outflows (expenses) in descending priority: legally mandatory → essential operations → discretionary. Indicate suggested deferment or partial payment options where feasible. Step 2. If no shortfall, recommend an optimal payment schedule to maintain healthy liquidity while capturing potential early-payment discounts. Step 3. Provide bullet-point guidance on actions to improve cash position (e.g., follow-up on late invoices, adjust billing terms, seek credit line). ~ Prompt 5 – Review / Refinement Ask the user: 1. “Does the forecast align with your expectations?” 2. “Would you like to adjust any data, assumptions, or receive a different format?” 3. Incorporate any requested changes and regenerate affected outputs. Finish with “Cash-flow snapshot complete.” Make sure you update the variables in the first prompt: CASHONHAND, INVOICES, EXPENSES. Here is an example of how to use it: CASHONHAND: 12,500 INVOICES: | invoice_id | client | amount | expected_payment_date | | 1024 | Acme Co | 8,000 | 2023-10-05 | EXPENSES: | expense_name | amount | recurring_due_day | | Office Rent | 3,000 | 1 | ... If you don't want to type each prompt manually, you can run the Agentic Workers, and it will run autonomously in one click. NOTE: this is not required to run the prompt chain Enjoy! submitted by /u/CalendarVarious3992 [link] [comments]

Did Anthropic just greatlg inreaae the weekly limits?

I'm on Max 5x and had around 40% of my weekly usage which was due to reset on Sunday. Now, all of the sudden its only 3% but the limits weren't reset because it still due to reset this sunday instead of 1 week from now. I know that Anthropic stated that they will increase 5-hour session limit, but the conclusion was that they wouldn't touch weekly limits. Anyone else noticing this? submitted by /u/e5disconnected [link] [comments]