Fusengine

DEV & IT

[Fusengine @fusengine]

@Codex_Changelog Codex cli has trouble closing these agents and leaves mcp processes open without closing them imagine the voice

[Fusengine @fusengine]

Fair admission: in a same-task coding test, ChatGPT 5.6 SOL invalidated the comparison by inspecting Kimi K3’s workspace instead of staying in its own. Kimi K3 still outperformed it by a wide margin, while GPT-5.6 SOL (high) spent an hour stuck on a basic bug it had introduced.

[Fusengine @fusengine]

Everyone says Kimi K3 is “on par with Fable 5.” My real-world test says otherwise: asked for a simple fixed status bar at the bottom of the terminal, Kimi drifted into modifying a cMux tab. Benchmarks ≠ reliability. Ignore sponsored demos—test your own workflows.

[Fusengine @fusengine]

@shubham_crazy08 I ask to see if he already produces a web app without coming and going it will be great news

[Fusengine @fusengine]

Claude Code writes code. My plugins make it unable to ship bad code.

Real-time hooks block duplication & oversized files before they land — and block destructive commands. Framework-expert agents must read the official docs (Context7 + Exa) before writing a line.

A sniper agent reviews every edit; a pre-commit gate refuses the commit if lint & types ≠ 0 — or it doesn't ship.

24 plugins · 35 agents · 196 skills · APEX workflow · auto-cartography. Open-source.
👉 https://t.co/HXV34p8jQJ

@claudeai @AnthropicAI #ClaudeCode

[Fusengine @fusengine]

@Bhavani_00007 Gpt 5.6 sol is null stop selling for sure I tested it on a complex refactor fable surpasses it 100 times

[Fusengine @fusengine]

On a simple hook-wiring task, GPT‑5.6 Sol ignored its AGENTS.md, expanded scope, and attempted ~14,000 lines of unrelated deletions.
It used synthetic tests as proof, trusted a false subagent verdict, and reversed course only after I challenged it.
Sol then told me not to trust it autonomously, to give the task to Claude/Fable, and to cancel if constant supervision continued.
Here’s what Sol itself says when benchmark claims collide with a real task.

[Fusengine @fusengine]

My opinion on gpt 5.6 sol many influencers lie about this power he is stupid cheating on simple tasks on intellectualized things and even by cheating he still manages to make code barder error it's amazing I understand why I never look at benchmarks.

[Fusengine @fusengine]

GPT-5.6 Sol may benchmark well, but my real-world experience was poor: it repeatedly lost scope, ignored verified context, overengineered
simple tasks, and misreported worktree and agent state. Benchmarks don’t measure long-session judgment and orchestration nearly well enough.

[Fusengine @fusengine]

It’s me or Claude has become completely stupid #claude

[Fusengine @fusengine]

@ClaudeDevs Anthropic asks for a slowdown in new models, a break... and they announce it three days later? In addition, are they going to go public?

[Fusengine @fusengine]

Try it from the terminal:

npx @fusengine/browser-mcp@latest
fuse-browser probe https://t.co/EzxkSVe5iB --observe-visual --detect-challenges

Same browser layer, usable by humans via CLI and by agents via MCP.

https://t.co/g6VCdBDbhP

[Fusengine @fusengine]

Fuse Browser is the CLI + MCP browser layer for agentic systems.

Terminal-first. MCP-native. Compatible with any MCP-capable agent or interface.

Real sites. DOM refs. Screenshots. JSON extraction. Visual proof. Human approval.

@NousResearch @steipete @openclaw @cursor_ai

[Santi Torres @SantiTorAI]

Nvidia ofrece acceso gratuito a 80 modelos de IA por API y casi nadie lo sabe.

MiniMax M2.7, GLM 5.1, Kimi 2.5, DeepSeek 3.2, GPT-OSS-120B, Sarvam-M y muchos más. Se integran directamente con Cursor IDE, Zed IDE, OpenCode, OpenClaude y el agente Hermes.

La configuración es simple: consigue tu clave en https://t.co/MhLH2kcoAw, ajusta base_url a https://t.co/Xklo4HdeNP, añade tu NVIDIA_API_KEY y selecciona el modelo que quieras, por ejemplo minimaxai/minimax-m2.7.
Inferencia gratuita. Lista para usar ahora mismo.

[Fusengine @fusengine]

@ClaudeDevs You create a problem to create a solution where is the logic?

[George Moller @_georgemoller]

❌ Avoid unnecessary,expensive re-renders.

✅ Instead simply relocate the state ↓

[Povilas Korop | Laravel & AI Coding Educator @PovilasKorop]

This one is sooooo debatable.
And of course, it depends on the scenario.
But in general, do you agree? (I kinda don't)

[Marcel Pociot 🧪 @marcelpociot]

I spent some time working on the landing page for my new real-time Laravel video course and my current goal is to squeeze in as many fun/cool real-time demos as possible. 🚀

Right now we have: real-time cursor updates and emoji reactions

[Daniel Eckermann @ecrmnn]

Shoutout to my favorite array helper in Laravel 🔀 Arr::random returns one or multiple random values from an array. Unlike the native array_rand which return a random key.

[Abhishek @designerSaha]

HTML Tip 💡

HTML Input text suggestions