📜 The team at Every spent a week with Opus 5 before launch, and their Day 0 verdict is blunt: it's a hard model to love. It argued with instructions, stopped before the work was finished, and fought their existing skills and plugins. Their first reaction: "what have they done to my boy?" Then they deleted their skills and started over. It got dramatically better. ⚓ Their Day 0 vibe check - It breaks backward compatibility. Point it at workflows you built for older models and it will stop early or skip instructions. - Starting from scratch fixes it. Kieran Klaassen got dramatically better results by dropping his existing skills entirely and rebuilding. - Try LOW thinking, not high. This one is counterintuitive: the more time they gave it to think, the more of the annoying behaviour showed up. Lower effort worked better. Don't reflexively drop to a smaller model for speed — try low thinking on Opus 5 first. - Their harsh line: "a poor man's Fable" — Fable's personality quirks without Fable's top end. For their two-slot workflow (genius model + fast generalist) it lands in an awkward middle. ⚓ Why this matches what Anthropic said today This is the same finding from both directions. Anthropic deleted 80% of Claude Code's system prompt for these models and told everyone to stop over-constraining them. Every deleted their skills and got a better model. Your old scaffolding was built to compensate for weaknesses Opus 5 doesn't have — and now it's the thing getting in the way. ☠️ Watch out - Don't nuke your setup on a stranger's say-so. Copy your skills folder somewhere safe first, then strip back and rebuild what you actually miss. - This is a Day 0 vibe check, not a benchmark. One team, one week, their workflows. Anthropic's numbers and Every's experience can both be true. - The rebuild is real work. Their words: it takes time to rebuild your workflows around it, but there's a payoff waiting. 🗝️ What to actually do - Run /doctor in Claude Code to trim your CLAUDE.md and skills. - If Opus 5 feels stubborn, disable your skills for one session and try the same task clean. - Try it on LOW thinking before you decide it's dumb.

📜 Anthropic just published the new rules for context engineering — and the headline is that they deleted over 80% of Claude Code's own system prompt for Opus 5 and Fable 5 with no measurable loss on their coding evals. If their prompt was too long, yours probably is too. ⚓ The six reversals - Rules → judgement. Hard bans like "never write comments" are wrong for some tasks. Their new line: "write code that reads like the surrounding code." - Examples → interface design. Examples now box the model in. Spend the effort on expressive tool parameters instead. - Everything upfront → progressive disclosure. Stop treating CLAUDE.md as the warehouse. Use a tree of files and skills that load when needed. - Repeat yourself → say it once. Instructions belong in the tool or skill description, not restated in the main prompt. - Memory in CLAUDE.md → auto-memory. Claude saves relevant memories itself now. The # hotkey habit is largely obsolete. - Simple specs → rich references. A test suite, a real function, or an HTML mockup beats a paragraph describing it. ⚓ The one command to run They shipped the whole thing as a command: run /doctor in Claude Code and it will rightsize your skills and CLAUDE.md files for you. ☠️ Watch out - Conflicting instructions are the real tax. Their own transcripts had "leave documentation as appropriate" fighting "DO NOT add comments" in the same request. Claude burns reasoning resolving your contradiction before it does your work. - This is guidance for Claude 5 models. If you are running something older, the old guardrails still earn their keep. 🗝️ Where to read it I've added a new section, "What Changed with Claude 5 Models," to the lesson 🏴‍☠️ What Actually Belongs in CLAUDE.md in 📜 CLAUDE.md & Context Mastery. Then/now list, what it means for your file, and the /doctor step. Credit where it's due: this comes from Thariq Shihipar at Anthropic (@thariq). https://claude.com/blog/the-new-rules-of-context-engineering-for-claude-5-generation-models

📜 Anthropic shipped Claude Opus 5 today (July 24, 2026). Frontier-level intelligence at roughly half the cost of Fable 5, at the same price as Opus 4.8, and it is already live in Claude Code. ⚓ The numbers (Anthropic's own benchmark table) - Agentic terminal coding (Frontier-Bench v0.1): 43.3% — Opus 4.8 was 21.1%, GPT-5.6 34.4%, Fable 5 33.7%. Roughly double 4.8. - Knowledge work (GDPval-AA v2): 1861 vs Fable 5's 1747 and Opus 4.8's 1593. - Novel problem-solving (ARC-AGI-3): 30.2% — about 3x the next best score. - Computer use (OSWorld 2.0): 70.6%, top of the table. - Business workflows (AutomationBench): 26.0% vs 17-18% for everything else. - Agentic search (BrowseComp): 90.8%. Not a clean sweep: GPT-5.6 still edges it on DeepSWE v1.1 (72.7% vs 68.8%), and Mythos 5 leads on health. ⚓ Price and where to get it - $5 per million input tokens, $25 per million output — identical to Opus 4.8. The "half the cost" line is versus Fable 5, not a price drop. - API model ID: claude-opus-5 - New default model on Claude Max; strongest model available on Pro - Live in Claude Code and Claude Cowork - Fast mode: about 2.5x faster output for 2x the price ⚓ What actually matters for us - The gains land exactly where we live: long-running agents, terminal coding, computer use, multi-step automation. This is an agent model, not a chat upgrade. - It self-verifies. Anthropic's example: on one task it could not see a drawing, so it built its own computer vision pipeline, then rebuilt the 3D model until it was right. That is the verification loop we keep teaching, baked in. - Same price as 4.8 means there is no reason to stay on 4.8. Switch with /model in Claude Code. ☠️ Watch out - Fast mode costs double. Handy for a long grind, expensive if you leave it on. - Deliberately capped on cyber and bio. It sits behind Mythos 5 on offensive-security and biology-research tasks on purpose. - These are the vendor's benchmarks. Run it on your own repo before you rebuild your workflow around a percentage.

I started the newsletter build on 5 June, and as of today I have 762 sub. I made my first ad revenue - $80 - two days ago. I think people are hungry for local information hubs that pull 50 facebook groups together without the facebook bs that goes with them, all into one place. I'll talk more about this build and what I did, but it started with a tweet that I copy and pasted into claude code. __________________________ Can you build this project out in a folder? I'd use substack. Call it the Barrhaven Bugle. Confirm spelling. Task second brain, researching the idea to flesh it out and save findings. If I had to start one AI side hustle in 2026, this would be it: a local newsletter that pulls in 2k a month and runs entirely by a team of four AI agents. The founder (Lucas Walter) has never written a single edition. This is a copy-pasteable system that could be applied to any city in the world. Here's the agent setup: 1. A growth engineer agent scrapes every local news site, subReddit, and event calendar at 6:00 a.m. daily. 2. A content director reads the database and writes the weekly newsletter in his voice every week. 3. A sales director handles every inbound advertiser email back-and-forth, pitching AI-generated ad creative. 4. A CEO agent orchestrates all three and reports back to him. Results from Lucas's newsletter: Spokane Pulse, 7,400 subscribers, 47.5% open rate (industry average 20% to 30%), 11% click-through rate, 2k a month in ad revenue, and growing under 4 hours a week of human in the loop. Now imagine running the same system in your city. I link to the Spokane Pulse in the comments so you can check it out for yourself. ___________________________ From that tweet I just started building it out. Now I ultimately ended up with a different system than this. That looks much like a news aggregator. I do have seven Hermes agents doing it. I am pushing out five-question interviews to people by email and building features. I've added a cartoon daily, a crossword, and a swipe deck pulse.