正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

A solo developer building an ocean ship game with Claude Opus 5.5 has shared his workflow and costs, alongside a free playable demo. → He runs a separate Claude session for each game system and talks only to a 'manager' session that messages the rest → A decisions.md file... show more

0xMarioNawfal

266,441 subscribers

49,413 次观看 • 5 天前 •via X (Twitter)

艺术游戏科学技术

Anya Rossi• Live Now

Private livecam show

7 条评论

VkDream 的头像

VkDream5 天前

一个人用多个 Claude 会话造远洋船游戏只跟一个 manager 会话说话省的是上下文贵的是那份 decisions.md 得自己盯多会话编排的门槛从来不在工具在记录

Vipul Kumar Kewat 的头像

Vipul Kumar Kewat5 天前

The manager-session architecture is the standout idea.

고조 체인 的头像

고조 체인5 天前

that decisions.md file is a genius way to avoid feature creep in long projects

why 的头像

why5 天前

A decisions.md file alongside separate system chats seems like a practical way to keep the project’s logic from drifting.

Alex Vainshtein 的头像

Alex Vainshtein5 天前

This is very close to how I’ve ended up working too. The part I’d add is aggressive verification - game agents can pass every automated check and still produce something visibly wrong. Separate workers + clear ownership work well, but for 3D/game work I’ve learned the rendered pixels still have to be the final judge.

Found Play 的头像

Found Play5 天前

a four-funnel liner in a fishing game is a bold boat choice. that free demo should be on

noel 的头像

noel5 天前

The [USER]/[AI] tag on each decision is the useful part. When something breaks later, the AI-tagged choices are the ones to reopen first, and the user-tagged ones can stay put.

相关视频

Higgsfield MCP is now FREE. Every new user gets 100 free credits to generate content with Fable 5 driving Higgsfield MCP on Claude. Grab the free trial plan and get full MCP access for 3 days. This is a limited offer.

Higgsfield MCP is now FREE. Every new user gets 100 free credits to generate content with Fable 5 driving Higgsfield MCP on Claude. Grab the free trial plan and get full MCP access for 3 days. This is a limited offer.

Higgsfield AI 🧩

106,135 次观看 • 3 个月前

A developer used Claude to build Scroll Studio, a free open-source engine that creates cinematic scroll-driven landing pages locally, the kind of sites that usually depend on paid tools. → Uses LTX-2.3 for video, Z-Image for images, Blender for 3D and Depth Anything for photo depth → Runs locally with no API keys, from a local app, the command line or by asking Claude → MIT licensed, with a live showcase of sites already built with it Reddit:

A developer used Claude to build Scroll Studio, a free open-source engine that creates cinematic scroll-driven landing pages locally, the kind of sites that usually depend on paid tools. → Uses LTX-2.3 for video, Z-Image for images, Blender for 3D and Depth Anything for photo depth → Runs locally with no API keys, from a local app, the command line or by asking Claude → MIT licensed, with a live showcase of sites already built with it Reddit:

0xMarioNawfal

42,273 次观看 • 5 天前

Claude Opus 5.5 with this prompt is a f*cking cheat code paste it in and Opus 5.5 designs the whole system for you: > the agents and what each one owns > the prompt each agent runs on > the workflow, handoffs and stop conditions between them Claude Code tip: run it in plan mode first, cut every agent without a clear output, then let it build grab it before it becomes everyone's default playbook:

Claude Opus 5.5 with this prompt is a f*cking cheat code paste it in and Opus 5.5 designs the whole system for you: > the agents and what each one owns > the prompt each agent runs on > the workflow, handoffs and stop conditions between them Claude Code tip: run it in plan mode first, cut every agent without a clear output, then let it build grab it before it becomes everyone's default playbook:

Mr. Buzzoni

33,817 次观看 • 4 天前

Haiku 5.5 is now available in the Claude Platform and Claude Code. On average, it costs around 75% less to run than Haiku 4.5. It pairs well with Opus 5.5 or Sonnet 5.5 as a subagent. Use it for high-volume, cost-sensitive tasks like summaries, compactions, or database queries.

Haiku 5.5 is now available in the Claude Platform and Claude Code. On average, it costs around 75% less to run than Haiku 4.5. It pairs well with Opus 5.5 or Sonnet 5.5 as a subagent. Use it for high-volume, cost-sensitive tasks like summaries, compactions, or database queries.

ClaudeDevs

1,218,614 次观看 • 4 天前

THIS GUY HIT HIS DAILY CLAUDE LIMIT BUILDING A KNOWLEDGE GRAPH. THE SYSTEM KEPT RUNNING WITHOUT HIM he opened obsidian at 6am and claude had already read everything new connections, a brief, one question worth thinking about today his coworkers are still copy-pasting context into every session he stopped doing that 3 months ago one markdown file changed how he works with claude forever Bookmark & like it if you’re tired of starting from zero every session

THIS GUY HIT HIS DAILY CLAUDE LIMIT BUILDING A KNOWLEDGE GRAPH. THE SYSTEM KEPT RUNNING WITHOUT HIM he opened obsidian at 6am and claude had already read everything new connections, a brief, one question worth thinking about today his coworkers are still copy-pasting context into every session he stopped doing that 3 months ago one markdown file changed how he works with claude forever Bookmark & like it if you’re tired of starting from zero every session

leopardracer

87,564 次观看 • 5 个月前

Found my lost childhood game — and rebuilt it in 3D with Fable 5 + Godot. AirXonix — one of the first games I ever played. My sister and I wore it out on a single PC. Nostalgia hit hard. So I gave Claude Fable 5 the full toolbox: Godot MCP + Blender MCP + APIs for textures, 3D models, SFX, even the music. The whole pipeline in one chat. Honestly? More fun than I've had in months. And the game came out genuinely good. Full build:

Found my lost childhood game — and rebuilt it in 3D with Fable 5 + Godot. AirXonix — one of the first games I ever played. My sister and I wore it out on a single PC. Nostalgia hit hard. So I gave Claude Fable 5 the full toolbox: Godot MCP + Blender MCP + APIs for textures, 3D models, SFX, even the music. The whole pipeline in one chat. Honestly? More fun than I've had in months. And the game came out genuinely good. Full build:

Stefan 3D AI

14,443 次观看 • 2 个月前

here's how to build a game like this with Claude Opus 5.5 + Blender + Unreal Engine + Meshy... the trick is giving every tool one job, but there's a smart way to set it up: - step 1: install Claude Code, Unreal Engine 5 and Blender 5.1+, all free except the Claude plan - step 2: connect Blender through its MCP add-on, and Meshy through Meshy MCP with an API key - step 3: brainstorm 5 game ideas with Opus first, pick one, have it expand the idea before anything gets built - step 4: tell it your GPU, so it pushes the graphics as far as your machine can run - step 5: Meshy generates the props and characters, Opus imports them into Blender and cleans them up - step 6: Opus brings everything into Unreal: the level, lighting, HUD, minimap - step 7: play it, then add one feature per prompt: combat, enemies, quests who does what: > Meshy: the 3D models > Blender: cleanup, placement, materials > Unreal: lighting, physics, the actual game > Opus: runs all three from one chat and the real cost... one full first build took about 4 hours and ~19% of a weekly Max plan build the first level well, then the next ones go much faster because the assets already exist

here's how to build a game like this with Claude Opus 5.5 + Blender + Unreal Engine + Meshy... the trick is giving every tool one job, but there's a smart way to set it up: - step 1: install Claude Code, Unreal Engine 5 and Blender 5.1+, all free except the Claude plan - step 2: connect Blender through its MCP add-on, and Meshy through Meshy MCP with an API key - step 3: brainstorm 5 game ideas with Opus first, pick one, have it expand the idea before anything gets built - step 4: tell it your GPU, so it pushes the graphics as far as your machine can run - step 5: Meshy generates the props and characters, Opus imports them into Blender and cleans them up - step 6: Opus brings everything into Unreal: the level, lighting, HUD, minimap - step 7: play it, then add one feature per prompt: combat, enemies, quests who does what: > Meshy: the 3D models > Blender: cleanup, placement, materials > Unreal: lighting, physics, the actual game > Opus: runs all three from one chat and the real cost... one full first build took about 4 hours and ~19% of a weekly Max plan build the first level well, then the next ones go much faster because the assets already exist

rewind

156,881 次观看 • 3 天前

5 weeks ago, this fishing game didn’t exist. Now it has a full 3D world, dynamic water, day/night cycles, fishing mechanics, rarity systems, custom UI and even its own in-game editors. Almost everything is being built with AI. Claude handles most of the game, Claude + Blender MCP handles 3D work, and ChatGPT helps with image generation. The interesting part is that the workflow is changing too. Instead of asking AI 20 times to move a lighthouse or UI element into the right position, he had Claude build custom editors so he can tweak and save everything manually. Week 5 of seeing how far one person can push AI game development.

5 weeks ago, this fishing game didn’t exist. Now it has a full 3D world, dynamic water, day/night cycles, fishing mechanics, rarity systems, custom UI and even its own in-game editors. Almost everything is being built with AI. Claude handles most of the game, Claude + Blender MCP handles 3D work, and ChatGPT helps with image generation. The interesting part is that the workflow is changing too. Instead of asking AI 20 times to move a lighthouse or UI element into the right position, he had Claude build custom editors so he can tweak and save everything manually. Week 5 of seeing how far one person can push AI game development.

Hakimi Eiqbal

43,159 次观看 • 1 个月前

THIS GUY HAD OPUS 4.8 BUILD A FULL LEAGUE OF LEGENDS CLONE IN UNDER A DAY AND CALLED IT "LMAO" temu league of legends, fully playable, multiplayer, and all built by claude ALL the art is claude too he had claude generate every character, animation, and background as SVG code then converted them to procedural canvas for smooth animation. used dedicated sub-agents for each champion the tech stack: typescript, react, canvas, and partykit for multiplayer the workflow: > used /goal to kick off the entire project > ultracode workflows for big tasks like optimization > /goal again to queue up lists of bug fixes > treated claude like a dev team the original champion names claude came up with: the original champion names claude came up with: Teehee" instead of Teemo. "CtrlAltDefeat" as a champion name. he had to run through and clean them all up so nothing infringed on riot's IP claude even built the bots on its own with "chill," "normal," and "sweaty" difficulty levels the token bill was 2.7 billion total tokens consumed. mostly cache reads. 15.5 million actual output tokens. would have cost roughly $6,600 at list price but he just used his pro max subscriptions people are already benchmarking claude by how good of a game clone it can build game developers are watching someone build a multiplayer MOBA in a day with zero art skills and zero game dev experience 2026 is wild (and its going to get even wilder)

THIS GUY HAD OPUS 4.8 BUILD A FULL LEAGUE OF LEGENDS CLONE IN UNDER A DAY AND CALLED IT "LMAO" temu league of legends, fully playable, multiplayer, and all built by claude ALL the art is claude too he had claude generate every character, animation, and background as SVG code then converted them to procedural canvas for smooth animation. used dedicated sub-agents for each champion the tech stack: typescript, react, canvas, and partykit for multiplayer the workflow: > used /goal to kick off the entire project > ultracode workflows for big tasks like optimization > /goal again to queue up lists of bug fixes > treated claude like a dev team the original champion names claude came up with: the original champion names claude came up with: Teehee" instead of Teemo. "CtrlAltDefeat" as a champion name. he had to run through and clean them all up so nothing infringed on riot's IP claude even built the bots on its own with "chill," "normal," and "sweaty" difficulty levels the token bill was 2.7 billion total tokens consumed. mostly cache reads. 15.5 million actual output tokens. would have cost roughly $6,600 at list price but he just used his pro max subscriptions people are already benchmarking claude by how good of a game clone it can build game developers are watching someone build a multiplayer MOBA in a day with zero art skills and zero game dev experience 2026 is wild (and its going to get even wilder)

Om Patel

336,682 次观看 • 4 个月前

SOMEONE WITH ZERO CODING EXPERIENCE BUILT A FULL REALISTIC FIGHTER JET GAME WITH CLAUDE IN 4 DAYS the guy has never written a line of code before. he just described what he wanted to claude and kept iterating, and 4 days later he has a real flight combat game > you fly actual fighter jets, lock onto targets and fire missiles > it launches off a carrier, dogfights, the whole thing > claude code and opus 4.8 wrote and debugged all the game logic in godot > the planes and missiles are free 3d models, claude handled 100% of the code > hes even adding missions to it now imagine showing this 10, 20, 40 years ago. a person talking to a mic, going back and forth, and a fighter jet game comes out the other end. the barrier to building was never the idea, it was the years of learning to code. that barrier is just gone now

SOMEONE WITH ZERO CODING EXPERIENCE BUILT A FULL REALISTIC FIGHTER JET GAME WITH CLAUDE IN 4 DAYS the guy has never written a line of code before. he just described what he wanted to claude and kept iterating, and 4 days later he has a real flight combat game > you fly actual fighter jets, lock onto targets and fire missiles > it launches off a carrier, dogfights, the whole thing > claude code and opus 4.8 wrote and debugged all the game logic in godot > the planes and missiles are free 3d models, claude handled 100% of the code > hes even adding missions to it now imagine showing this 10, 20, 40 years ago. a person talking to a mic, going back and forth, and a fighter jet game comes out the other end. the barrier to building was never the idea, it was the years of learning to code. that barrier is just gone now

Om Patel

72,285 次观看 • 2 个月前

Pokee Isaac now runs inside Claude Code. And you can use it a few different ways: have Claude call Isaac for complete one-shot generations, switch directly to Isaac as your model mid-session, or run the entire Claude Code session on Isaac. For the workflow shown below, Claude plans and verifies. Isaac generates the complete artifact in a single pass. Integration built and open-sourced by ⭕ AI & Design (Marco). Here's a quick how-to video in order to get started! You can access the repo at:

Pokee Isaac now runs inside Claude Code. And you can use it a few different ways: have Claude call Isaac for complete one-shot generations, switch directly to Isaac as your model mid-session, or run the entire Claude Code session on Isaac. For the workflow shown below, Claude plans and verifies. Isaac generates the complete artifact in a single pass. Integration built and open-sourced by ⭕ AI & Design (Marco). Here's a quick how-to video in order to get started! You can access the repo at:

Pokee AI

41,078 次观看 • 1 个月前

New open source project: Easel, a macOS workspace for AI-assisted product design and frontend iteration powered by the Codex SDK. Think of it as "Claude Design for Codex." It brings together a local project library, design-system setup, Codex chat, and a live web preview with a point-and-click inspector. Set up your design system however you like. Import one from Figma , bring your own, or have the agent generate it as a design.md. Then use it to build prototypes and slide decks. Codex-first, with Claude and other AI providers and local models planned for future releases. OpenAI Developers

New open source project: Easel, a macOS workspace for AI-assisted product design and frontend iteration powered by the Codex SDK. Think of it as "Claude Design for Codex." It brings together a local project library, design-system setup, Codex chat, and a live web preview with a point-and-click inspector. Set up your design system however you like. Import one from Figma , bring your own, or have the agent generate it as a design.md. Then use it to build prototypes and slide decks. Codex-first, with Claude and other AI providers and local models planned for future releases. OpenAI Developers

James Rochabrun

61,608 次观看 • 4 个月前

THIS GUY TURNED ONE OBSIDIAN VAULT INTO A SHARED BRAIN FOR 5 CLAUDE AGENTS BUILT TO HANDLE 80% OF A SOLO DEVELOPER’S WORK. 00:03 he zooms out and hundreds of connected notes appear on screen. it is the perfect visual for Claude working from one shared project memory instead of starting from zero every session. most developers give Claude 1 file and 1 prompt. this setup gives it the architecture, coding rules, tests, client briefs, and previous decisions before it writes anything. obsidian becomes the memory layer, storing everything in markdown files Claude can read alongside the codebase. every agent works from the same context instead of repeating the same questions. then 5 specialists split the job. one researches, one plans, one codes, one reviews, and one deploys while the developer only controls the important decisions. one vault, 5 agents, 2 model tiers, and a $50 to $150 monthly stack can turn Claude from a chatbot into a small development team.

THIS GUY TURNED ONE OBSIDIAN VAULT INTO A SHARED BRAIN FOR 5 CLAUDE AGENTS BUILT TO HANDLE 80% OF A SOLO DEVELOPER’S WORK. 00:03 he zooms out and hundreds of connected notes appear on screen. it is the perfect visual for Claude working from one shared project memory instead of starting from zero every session. most developers give Claude 1 file and 1 prompt. this setup gives it the architecture, coding rules, tests, client briefs, and previous decisions before it writes anything. obsidian becomes the memory layer, storing everything in markdown files Claude can read alongside the codebase. every agent works from the same context instead of repeating the same questions. then 5 specialists split the job. one researches, one plans, one codes, one reviews, and one deploys while the developer only controls the important decisions. one vault, 5 agents, 2 model tiers, and a $50 to $150 monthly stack can turn Claude from a chatbot into a small development team.

Gipp 🦅

10,799 次观看 • 3 个月前

This is the new bar for PMs: "I'm able to do now in a day what used to take me a week." In this episode of How I AI, Daniel Blum demos how he uses Claude Cowork as an AI system that compounds over time, instead of starting from scratch with every chat. He shows how he - uses Notion as his persona brain - contextualizes Claude with voice memos, links, and updates - how his system "learns" company jargon - built a "workstation" plugin to bootstrap everyone else onto the same system Finally - we cover the 20% Claude still cant do (for now) As always, a huge ty to our sponsors: Optimizely - Your AI agent orchestration platform for marketing and digital teams: Atlassian Jira AI SDLC - Get your tokens’ worth with Jira: Full ep on YT:

This is the new bar for PMs: "I'm able to do now in a day what used to take me a week." In this episode of How I AI, Daniel Blum demos how he uses Claude Cowork as an AI system that compounds over time, instead of starting from scratch with every chat. He shows how he - uses Notion as his persona brain - contextualizes Claude with voice memos, links, and updates - how his system "learns" company jargon - built a "workstation" plugin to bootstrap everyone else onto the same system Finally - we cover the 20% Claude still cant do (for now) As always, a huge ty to our sponsors: Optimizely - Your AI agent orchestration platform for marketing and digital teams: Atlassian Jira AI SDLC - Get your tokens’ worth with Jira: Full ep on YT:

claire vo 🖤

19,004 次观看 • 1 个月前

5 GitHub repos with ⭐50K+ combined stars that show how far people are pushing Claude Code beyond a single coding agent 1) instructkr/claude-code 2.3K⭐ -> a research snapshot of Claude Code’s source with the internals exposed for inspection. [ ] 2) codeaashu/claude-code 3.2K⭐ -> another reconstructed Claude Code codebase built for studying how the agent works under the hood, including tools, prompts, MCP. [ ] 3) Donchitos/Claude-Code-Game-Studios 20K+⭐ -> turns one Claude Code session into a game development studio with 49 specialized agents, 73 skills, 12 hooks, 11 rules and 41 templates. [ ] 4) VoltAgent/awesome-claude-code-subagents 24K+⭐ -> 154 specialized Claude Code subagents split across 10 categories. [ ] 5) tanbiralam/claude-code 2.2K⭐ -> a research-oriented Claude Code source reconstruction for inspecting the terminal agent architecture, plugins, tools and internal implementation. [ ] Save these if you want to understand what people are actually building around Claude Code instead of just using the default CLI.

5 GitHub repos with ⭐50K+ combined stars that show how far people are pushing Claude Code beyond a single coding agent 1) instructkr/claude-code 2.3K⭐ -> a research snapshot of Claude Code’s source with the internals exposed for inspection. [ ] 2) codeaashu/claude-code 3.2K⭐ -> another reconstructed Claude Code codebase built for studying how the agent works under the hood, including tools, prompts, MCP. [ ] 3) Donchitos/Claude-Code-Game-Studios 20K+⭐ -> turns one Claude Code session into a game development studio with 49 specialized agents, 73 skills, 12 hooks, 11 rules and 41 templates. [ ] 4) VoltAgent/awesome-claude-code-subagents 24K+⭐ -> 154 specialized Claude Code subagents split across 10 categories. [ ] 5) tanbiralam/claude-code 2.2K⭐ -> a research-oriented Claude Code source reconstruction for inspecting the terminal agent architecture, plugins, tools and internal implementation. [ ] Save these if you want to understand what people are actually building around Claude Code instead of just using the default CLI.

Lummox

12,775 次观看 • 1 个月前

Someone asked Claude Opus 5.5 to make a game inspired by the Antikythera mechanism, often called the world's first computer, and it delivered a playable 3D dive adventure packed into a single 3 MB HTML file. → Every visual and sound is procedurally generated in real time from code Opus 5.5 wrote → Set in 1901 Greece, where sponge divers stumble on ancient bronze fragments and you help find the rest → Full controls to look, walk, turn, jump, scan with sonar and clean finds with a brush → Playable right now in the browser as a Claude artifact Credits: edwin

Someone asked Claude Opus 5.5 to make a game inspired by the Antikythera mechanism, often called the world's first computer, and it delivered a playable 3D dive adventure packed into a single 3 MB HTML file. → Every visual and sound is procedurally generated in real time from code Opus 5.5 wrote → Set in 1901 Greece, where sponge divers stumble on ancient bronze fragments and you help find the rest → Full controls to look, walk, turn, jump, scan with sonar and clean finds with a brush → Playable right now in the browser as a Claude artifact Credits: edwin

0xMarioNawfal

49,950 次观看 • 18 天前

A guy on Reddit got so fed up with scrolling LinkedIn for internships that he built his own job search engine with Claude Code. The tool is called Pinloop. It's an open-source CLI, and he built it with Claude Code too. Here's how it works: → It pulls millions of postings a month, worldwide, straight from hiring systems like Greenhouse and Workday (50+ of them) plus job boards like LinkedIn, refreshed hourly → It stores your resume and preferences → Every day, Claude Code pulls the new postings, reads each one against your profile, and writes a verdict with its reasoning → The good fits land in a shortlist you check in the morning His reasoning: every application already gets screened by AI on the employer's side, so he built a screener for his own side. You don't even have to set it up yourself. Paste this into Claude Code, Codex, or Cursor: "Run npm install -g pinloop, then run pinloop welcome and follow the instructions"

Vaibhav Sisinty

19,240 次观看 • 18 天前

This workflow will save you thousands with Claude. Run Opus 5.5, Sonnet 5.5, Haiku 5.5, and Fable 5.1 together, and stop spending Opus tokens on work that doesn't need Opus. The entire idea in one line: Opus plans, Sonnet edits, Haiku reads, Fable reviews at decision points. Roles, broken down: - Opus 5.5, high effort, owns the plan and reviews the final code - Sonnet 5.5, medium effort, is the worker: edits files, runs tests - Haiku 5.5, low effort, splits into explorer (searches and reads the codebase) and researcher (pulls docs). it's the first Haiku with an effort setting - Fable 5.1, set with /advisor fable. Opus calls it at decision points, and it gets the full transcript each time Three moments Opus tends to call it: → before committing to a plan: is this the right approach? → the same error shows up again: is this going nowhere? → before marking the task done: did something get skipped? Why Haiku only reads: Anthropic's launch post says Sonnet 5.5 and Opus 5.5 are still the better choice for complex agentic coding (Terminal-Bench 4.0: 39.2% for Haiku 5.5, 70.6% for Sonnet 5.5) and that Haiku 5.5 fits narrowly scoped subagent work. so it gets the lookups, not the edits. Anyone still running one model for everything is paying $4 per million input tokens to check whether a file exists. Haiku 5.5 does it for $0.10. Drop this into Claude Code 👇 "Rebuild my Claude Code setup around this structure: Confirm Claude Code is v2.1.293 or later, so the haiku alias resolves to Haiku 5.5. If it isn't, stop and tell me to run claude update. Look through ~/.claude/agents and .claude/agents for subagents already covering explorer, worker, and researcher. Only create new ones for roles that are missing. Set model: haiku, effort: low on explorer and researcher, with no Edit or Write tools. Set model: sonnet, effort: medium on worker. If an existing subagent is locked to a different model, leave it as is and just list it. Name the explorer subagent Explore so it overrides the built-in one, which otherwise runs on my main model. In ~/.claude/settings.json, set advisorModel to fable, and set effortLevel to high for claude-opus-5-5 under modelSettings. A top-level effortLevel in user settings doesn't apply to Opus 5.5. Check for anything disabling the advisor: CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, or anything blocking feature-flag fetches. Also check CLAUDE_CODE_EFFORT_LEVEL, which overrides subagent effort settings, and CLAUDE_CODE_SUBAGENT_MODEL_FORCE, which makes Claude Code ignore subagent model fields. Report what you find. Don't change any of it yet. Add one line to ~/.claude/CLAUDE.md: consult the advisor before a big plan, when the same error shows up twice, and before marking a long task done. Show every change as a diff first. Wait for my go-ahead before touching anything."

This workflow will save you thousands with Claude. Run Opus 5.5, Sonnet 5.5, Haiku 5.5, and Fable 5.1 together, and stop spending Opus tokens on work that doesn't need Opus. The entire idea in one line: Opus plans, Sonnet edits, Haiku reads, Fable reviews at decision points. Roles, broken down: - Opus 5.5, high effort, owns the plan and reviews the final code - Sonnet 5.5, medium effort, is the worker: edits files, runs tests - Haiku 5.5, low effort, splits into explorer (searches and reads the codebase) and researcher (pulls docs). it's the first Haiku with an effort setting - Fable 5.1, set with /advisor fable. Opus calls it at decision points, and it gets the full transcript each time Three moments Opus tends to call it: → before committing to a plan: is this the right approach? → the same error shows up again: is this going nowhere? → before marking the task done: did something get skipped? Why Haiku only reads: Anthropic's launch post says Sonnet 5.5 and Opus 5.5 are still the better choice for complex agentic coding (Terminal-Bench 4.0: 39.2% for Haiku 5.5, 70.6% for Sonnet 5.5) and that Haiku 5.5 fits narrowly scoped subagent work. so it gets the lookups, not the edits. Anyone still running one model for everything is paying $4 per million input tokens to check whether a file exists. Haiku 5.5 does it for $0.10. Drop this into Claude Code 👇 "Rebuild my Claude Code setup around this structure: Confirm Claude Code is v2.1.293 or later, so the haiku alias resolves to Haiku 5.5. If it isn't, stop and tell me to run claude update. Look through ~/.claude/agents and .claude/agents for subagents already covering explorer, worker, and researcher. Only create new ones for roles that are missing. Set model: haiku, effort: low on explorer and researcher, with no Edit or Write tools. Set model: sonnet, effort: medium on worker. If an existing subagent is locked to a different model, leave it as is and just list it. Name the explorer subagent Explore so it overrides the built-in one, which otherwise runs on my main model. In ~/.claude/settings.json, set advisorModel to fable, and set effortLevel to high for claude-opus-5-5 under modelSettings. A top-level effortLevel in user settings doesn't apply to Opus 5.5. Check for anything disabling the advisor: CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, or anything blocking feature-flag fetches. Also check CLAUDE_CODE_EFFORT_LEVEL, which overrides subagent effort settings, and CLAUDE_CODE_SUBAGENT_MODEL_FORCE, which makes Claude Code ignore subagent model fields. Report what you find. Don't change any of it yet. Add one line to ~/.claude/CLAUDE.md: consult the advisor before a big plan, when the same error shows up twice, and before marking a long task done. Show every change as a diff first. Wait for my go-ahead before touching anything."

Alvaro Cintas

55,249 次观看 • 3 天前

Claude Code tip: once Opus 5.5 runs your main session, stop leaving Fable 5.1 on the bench put it on call with /advisor run /advisor fable Opus 5.5 keeps writing the code Fable 5.1 reads the whole session, every tool call included, and only speaks up at three moments: → before a plan: is this the right approach? → when the same error comes back: am I digging in the wrong place? → before "done": what did I miss? Fable 5.1 reviews. Opus 5.5 ships jev engineering is the same move one layer down: the forks that need no thinker (which file, which tool, retry or stop) go to jev in under half a second, and the big model only sees the ones that split - the full tree > Opus 5.5 on high runs the main session > explorer reads the code > worker edits and runs tests > researcher pulls the docs > all three on medium > Fable 5.1 on call as the advisor paste the tree and this prompt into Claude Code ↓ "Rebuild my Claude Code setup around this tree: 1. Check ~/.claude/agents and .claude/agents for subagents that already fit explorer, worker and researcher. > Draft new ones only for missing roles > Give each model: opus, effort: medium > Skip any that pin a different model and list them 2. Set the main session to high via effortLevel in ~/.claude/settings.json, and set advisorModel to fable 3. Find anything that keeps the advisor off (CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, any variable that stops feature-flag fetching) plus CLAUDE_CODE_EFFORT_LEVEL, which overrides subagent effort. Report them, change nothing 4. Add one rule to ~/.claude/CLAUDE.md: consult the advisor before a large plan, when an error repeats, and before calling a long task done Show me every change as a diff first. No edits until I say go." ↳

Claude Code tip: once Opus 5.5 runs your main session, stop leaving Fable 5.1 on the bench put it on call with /advisor run /advisor fable Opus 5.5 keeps writing the code Fable 5.1 reads the whole session, every tool call included, and only speaks up at three moments: → before a plan: is this the right approach? → when the same error comes back: am I digging in the wrong place? → before "done": what did I miss? Fable 5.1 reviews. Opus 5.5 ships jev engineering is the same move one layer down: the forks that need no thinker (which file, which tool, retry or stop) go to jev in under half a second, and the big model only sees the ones that split - the full tree > Opus 5.5 on high runs the main session > explorer reads the code > worker edits and runs tests > researcher pulls the docs > all three on medium > Fable 5.1 on call as the advisor paste the tree and this prompt into Claude Code ↓ "Rebuild my Claude Code setup around this tree: 1. Check ~/.claude/agents and .claude/agents for subagents that already fit explorer, worker and researcher. > Draft new ones only for missing roles > Give each model: opus, effort: medium > Skip any that pin a different model and list them 2. Set the main session to high via effortLevel in ~/.claude/settings.json, and set advisorModel to fable 3. Find anything that keeps the advisor off (CLAUDE_CODE_DISABLE_ADVISOR_TOOL, DISABLE_TELEMETRY, any variable that stops feature-flag fetching) plus CLAUDE_CODE_EFFORT_LEVEL, which overrides subagent effort. Report them, change nothing 4. Add one rule to ~/.claude/CLAUDE.md: consult the advisor before a large plan, when an error repeats, and before calling a long task done Show me every change as a diff first. No edits until I say go." ↳

Hanako

52,065 次观看 • 10 天前

THIS GUY VIBE CODED A FULL CAPYBARA FOOD DELIVERY GAME IN 2 WEEKS WITH CLAUDE CODE you play as a capybara delivering food on a bike. orders stack on the back, you have a phone with apps in-game and the whole delivery system is realistic 2 weeks, zero game dev experience, and ENTIRELY AI generated the full stack: > claude code for all the code > three.js for the 3D engine > suno for original music > elevenlabs for sound effects and voice > GPT images-2 and grok for textures and illustrations > tripo3d for generating all the 3D assets the cinematics are all in-game too. he asked claude to build a cinematic editor with timeline controls, camera animation, and transitions. then he just placed the cameras himself his workflow was more planning than coding (obviously): > come up with the core mechanic > plan every feature using claude /plan mode > generate assets with AI tools > spend most of his time on the final polish, prop placement, and making the design feel right he said the human part is what most vibe coded games are missing. AI can generate everything but having taste for what looks good and what feels right is still on you the game is playable right now in the browser this is what vibe coding is actually capable of in the game dev space right now a year ago this would have taken a small team of developers, a sound designer, and an artist working together for months now one person with no experience can ship a polished playable game with story, music, and mechanics in 14 days the tools keep getting better and the barrier to making real games keeps getting lower

THIS GUY VIBE CODED A FULL CAPYBARA FOOD DELIVERY GAME IN 2 WEEKS WITH CLAUDE CODE you play as a capybara delivering food on a bike. orders stack on the back, you have a phone with apps in-game and the whole delivery system is realistic 2 weeks, zero game dev experience, and ENTIRELY AI generated the full stack: > claude code for all the code > three.js for the 3D engine > suno for original music > elevenlabs for sound effects and voice > GPT images-2 and grok for textures and illustrations > tripo3d for generating all the 3D assets the cinematics are all in-game too. he asked claude to build a cinematic editor with timeline controls, camera animation, and transitions. then he just placed the cameras himself his workflow was more planning than coding (obviously): > come up with the core mechanic > plan every feature using claude /plan mode > generate assets with AI tools > spend most of his time on the final polish, prop placement, and making the design feel right he said the human part is what most vibe coded games are missing. AI can generate everything but having taste for what looks good and what feels right is still on you the game is playable right now in the browser this is what vibe coding is actually capable of in the game dev space right now a year ago this would have taken a small team of developers, a sound designer, and an artist working together for months now one person with no experience can ship a polished playable game with story, music, and mechanics in 14 days the tools keep getting better and the barrier to making real games keeps getting lower

Om Patel

243,227 次观看 • 5 个月前