标签：#t

I spent last month migrating 23 client sites from JPEG/PNG to WebP and AVIF. Here's what I wish someone told me before I started. AVIF vs WebP: the real numbers AVIF is about 30% smaller than WebP at the same quality level. But Safari support is still patchy — if your traffic is 40%+ iOS, you need <picture> tags with WebP fallback. No way around it. The biggest win wasn't the format The single biggest reduction came from capping max image width at 1200px and setting quality to 80. One site went from 9.4MB to 318KB per page — a 97% reduction — just from those two settings plus lazy loading. The format switch was the cherry on top, not the cake. Tools I used daily SmartImgKit — quick batch conversions in the browser. No uploads, no signup, drag and drop. Handles the 80% case where you don't need a CLI pipeline. Supports JPG, PNG, WebP, AVIF, GIF, BMP, TIFF. ImageMagick — server-side batch jobs for when you need automation. Squoosh — one-off fine-tuning with visual comparison. Sharp (Node.js) — build pipeline integration. The HEIC surprise Every iPhone user's photos are HEIC. Most web tools crash on them. You need a converter that handles them before the pipeline — SmartImgKit's HEIC converter works locally in-browser, no uploads. The 80/20 rule Format + max width + lazy loading = 80% of the gain. Everything else is diminishing returns. Don't over-engineer it.

I Deployed 6 AI Systems Live — Here's What Actually Broke A few weeks ago I wrote about the 5 bugs that cost me 60+ hours building 49 AI systems. Every one of those bugs lived inside the code itself wrong array layout, a renamed model class, a serialization mismatch. This article is the second half of that story, and it taught me something more uncomfortable: code that runs perfectly on your machine can fail completely the moment it leaves your machine for reasons that have nothing to do with your code. I took 6 of my pinned GitHub projects and deployed every one of them live on Streamlit Cloud. Locally, all 6 worked without a single error. Deploying them surfaced 5 failures I had never seen before, none of which were bugs in my logic. Here they are, in the order I hit them. Failure 1 — A Module That Existed Yesterday, Gone Today My RAG chatbot used this import, unchanged for weeks: from langchain.chains import ConversationalRetrievalChain Locally: works. Deployed: instant crash. ModuleNotFoundError: No module named 'langchain.chains' The cause had nothing to do with my code. My local environment had an old, cached version of LangChain installed months ago. The deploy environment did a clean install and pulled whatever the latest version was at that moment and recent LangChain releases moved legacy chain classes like this one out of the core package entirely. The fix that actually worked pin the exact version that still contains the class, rather than chasing the newest API pattern under deployment pressure: langchain = =0.3.7 langchain-community = =0.3.7 The lesson: "it works on my machine" is frequently true specifically because your machine never reinstalled anything recently. A clean deploy environment has no such luxury it gets whatever is newest the moment it builds. Pin your versions before you ever need to debug this at 1 AM. Failure 2 — A File That Exists, Until It Doesn't My construction RAG project loads a prebuilt FAISS vector index from disk: vectorstor

The Kotlin Multiplatform (KMP) ecosystem moves fast. To stay at the cutting edge, ImagePickerKMP has recently undergone a major architectural upgrade in version 1.0.42 , adopting the latest stable versions of Kotlin and Compose Multiplatform. For the latest requirements and installation guides, always refer to https://imagepickerkmp.dev/ . Major Version Upgrades The v1.0.42 release brings significant updates to the core dependencies of the library: Dependency New Version Previous Version Kotlin 2.3.20 2.1.x Compose Multiplatform 1.10.3 1.9.x Ktor 3.4.1 3.0.x Android Gradle Plugin 8.13.2 8.x Warning: Kotlin 2.3.x brings breaking ABI changes. Projects using Kotlin < 2.3.x will fail to compile with an "ABI version incompatible" error when using ImagePickerKMP 1.0.42. Why the Upgrade Matters Performance: Kotlin 2.3.20 includes numerous compiler optimizations that result in smaller and faster binaries for both Android and iOS. Stability: Compose Multiplatform 1.10.3 resolves several rendering issues on iOS and Desktop, providing a smoother user experience. Future-Proofing: By moving to these versions, ImagePickerKMP is ready for the upcoming features in the Kotlin roadmap. How to Upgrade Your Project To use the latest version of ImagePickerKMP, you must update your build.gradle.kts file: plugins { kotlin ( "multiplatform" ) version "2.3.20" id ( "org.jetbrains.compose" ) version "1.10.3" } dependencies { implementation ( "io.github.ismoy:imagepickerkmp:1.0.42" ) } If your project is not yet ready for Kotlin 2.3.x, you can continue using version 1.0.41 of the library, which maintains compatibility with older Kotlin versions. Conclusion Staying updated is crucial for security, performance, and developer happiness. ImagePickerKMP makes it easy to leverage the power of the latest Kotlin features while maintaining a simple, unified API for media picking. Explore the full API reference and new features at https://imagepickerkmp.dev/ . References [1]: ImagePickerKMP Documentati

A few weeks back I shared some early Zen code examples. Since then, a lot has changed. We're now at v1.1.1 and the language actually has real tooling. What's new: Full CLI with package management zen publish - publish packages directly from CLI zen install - install packages from the registry zen list - browse all published packages with pagination Language improvements Struct support with literals and returns Regex with POSIX ERE ( matchRegex ) File I/O with binary support FFI bindings to C functions 162 stdlib functions across math, strings, fs, os, http, crypto, path utilities Package Registry (v1.0.0) JWT-based authentication GitHub-hosted packages Support for both runnable apps and libraries Semantic versioning The reactive variables concept from the first post is still there (that was my favorite feature), and now you can actually write real programs and share them with the community. Full docs: https://jishith-dev.github.io/zen-doc/site/ Install: curl -fsSL https://raw.githubusercontent.com/jishith-dev/Zen/main/install.sh | bash Next up: HTTP server APIs, better imports, and whatever the community asks for. Open to feedback and collaborators 💻 ✨ zen #programming #compiler #llvm #packagemanager #opensource #programminglanguage

Santa Clara, 2029. A speculative fiction about hegemony, sanctions, and the playbook nobody followed.

Original post: tachiosports.com What we're building By the end of this guide, you'll have a Discord bot that posts live CS2 match scores to a channel, updates every 60 seconds, and shows team names, current map, and odds. No database required — everything comes straight from the API. Prerequisites You'll need Node.js installed (v18 or newer), a Discord bot token from the Discord Developer Portal, and a free Tachio Sports API key. Sign up on the homepage with GitHub to get yours. Step 1 — Create the Discord bot Go to discord.com/developers/applications and create a new application. Under the Bot tab, click Add Bot and copy the token. Invite the bot to your server with the 'bot' and 'Send Messages' permissions. Keep your token secret — it's like a password for your bot. Step 2 — Set up the project mkdir cs2-discord-bot cd cs2-discord-bot npm init -y npm install discord.js Step 3 — The complete bot code const { Client , GatewayIntentBits , EmbedBuilder } = require ( " discord.js " ); const DISCORD_TOKEN = process . env . DISCORD_TOKEN ; const API_KEY = process . env . TACHIO_API_KEY ; const CHANNEL_ID = process . env . CHANNEL_ID ; const client = new Client ({ intents : [ GatewayIntentBits . Guilds , GatewayIntentBits . GuildMessages , ], }); async function fetchLiveMatches () { const res = await fetch ( " https://api.tachiosports.com/esports/live/cs2 " , { headers : { " x-api-key " : API_KEY } }, ); if ( ! res . ok ) return []; const data = await res . json (); return data . matches ?? []; } function buildEmbed ( match ) { const home = match . teams . home . name ?? " TBD " ; const away = match . teams . away . name ?? " TBD " ; const score = match . score ?. display ?? " vs " ; const map = match . current_map ?? "" ; const format = match . match_format ?? "" ; const league = match . league . name ?? "" ; const oddsHome = match . odds . match_winner . home ?? " – " ; const oddsAway = match . odds . match_winner . away ?? " – " ; return new EmbedBuilder () . setColor (

What We Know: The Basics of the Breach Polymarket, one of the largest prediction market platforms in the crypto space, confirmed on X that hackers stole funds from users after attackers compromised a third-party vendor. The breach allowed the attackers to inject malicious code directly into Polymarket's website, though the company specified the code ran "for some users" — a detail that raises immediate questions about whether the attack was deliberately targeted or only partially executed before detection. Polymarket spokesperson Connor Brandi confirmed to TechCrunch that the vendor compromise resulted in direct theft of user funds. Beyond that confirmation, the company declined to answer specific questions about the incident, leaving the scale of the financial damage, the identity of the compromised vendor, and the exact mechanism of the malicious code injection all officially unaddressed. The platform says it has contained the breach and is reaching out directly to affected users, committing to full refunds. No figure for total stolen funds has been disclosed. Blockchain monitoring firm PeckShield flagged suspicious activity around the same time Polymarket made its public announcement, adding an independent layer of confirmation that something significant moved on-chain during the incident window. What stands out immediately in the crypto security community is where the failure originated. The Polymarket platform itself was not the direct point of entry — a third-party vendor was. That distinction matters enormously. Users who trusted Polymarket's smart contract security and on-chain transparency had no visibility into the web infrastructure dependencies sitting between them and the prediction market interface. The malicious code injection attack, a technique that exploits trusted website supply chains, bypassed the decentralized architecture that crypto platforms often promote as a security feature. The incident joins a growing list of Web3 platform breaches wher

Ship Faster with TanStack, Cost Less with Cloudflare Discussion | Link

Most multi-tenant email-agent setups give every customer the same caps. Your free-tier user who signed up an hour ago and your enterprise account doing thousands of sends a day hit the exact same daily send limit, the exact same storage ceiling, the exact same retention window. That's fine right up until a free trial account starts hammering your infrastructure, or an enterprise customer files a ticket because their agent stopped sending at noon UTC and nobody can explain why. Free-tier and enterprise tenants shouldn't share the same caps. They have different risk profiles, different contractual obligations, and different billing. The trick is to make the quota a property of the tier, not a property of each individual account — so when you provision a new tenant you don't compute limits, you just drop them into the right bucket and the limits come along for free. With Nylas Agent Accounts that bucket is a workspace , and the caps live on a policy you attach to it. Set up one policy per tier, attach each to its tier's workspace, and every Agent Account in that workspace inherits the policy's send, storage, and retention limits automatically. No per-account configuration, no drift. I work on the Nylas CLI, so the terminal commands below are the exact ones I reach for when I'm wiring this up. As always, I'll show both the raw HTTP call and the CLI equivalent for every step, because half of you live in scripts and the other half live in your app code. What you actually get An Agent Account is just a Nylas grant with a grant_id — a managed mailbox that can send and receive on a domain you've registered. Everything grant-scoped works against it: Messages, Drafts, Threads, Folders, the lot. There's nothing new to learn on the data plane. A policy is a reusable bundle of limits and spam settings. One policy can govern many accounts. The limits we care about for tiering are: limit_count_daily_email_sent — how many messages an account can send per day. limit_storage_total — t

Unit converters are everywhere online, but they all seem to either require an account, run ads that cover half the screen, or send your input to a server for no reason. I built one that runs entirely in your browser, with no dependencies, no tracking, and no round-trips. 👉 https://unit-converter-dev.pages.dev What It Does Seven conversion categories, 70+ units, real-time bidirectional conversion: Category Example units Length mm, cm, m, km, in, ft, yd, mi, nmi, light-year Weight mg, g, kg, t, oz, lb, st, short ton Temperature °C, °F, K, °R Volume ml, l, m³, fl oz, cup, pint, quart, gallon, tbsp, tsp Area mm², cm², m², km², ha, acre, ft², in², mi², yd² Speed m/s, km/h, mph, ft/s, knot, Mach Data bit, byte, KB/KiB, MB/MiB, GB/GiB, TB — both SI and binary Features: Bidirectional — type in either field, the other updates instantly Swap button — flip from/to with one click All-units panel — see your input converted to every unit in the category simultaneously Formula display — shows the conversion factor (e.g. "1 Mile = 1.609344 Kilometer") Zero dependencies — single HTML file, no build step, no npm Implementation Notes Linear vs. non-linear conversions Most unit conversions are linear: multiply by a factor to get to the base unit, divide by another factor to get to the target. The approach: function convert ( catKey , fromUnit , toUnit , value ) { const base = toBase ( catKey , fromUnit , value ); // → base unit return fromBase ( catKey , toUnit , base ); // base unit → target } function toBase ( catKey , unit , value ) { const u = CATEGORIES [ catKey ]. units [ unit ]; if ( u . toBase ) return u . toBase ( value ); // non-linear (temperature) return value * u . factor ; } Temperature is the classic non-linear case. You can't just multiply to convert between Celsius, Fahrenheit, and Kelvin — you need offset arithmetic: temperature : { units : { C : { toBase : v => v + 273.15 , // °C → K fromBase : v => v - 273.15 , // K → °C }, F : { toBase : v => ( v - 32 ) * 5 / 9 + 2

Most job search tools focus on submitting more applications. I wanted to solve a different problem: reaching the people actually making hiring decisions. So I built PitchHired , an AI-powered platform that helps job seekers find hiring managers, generate personalized outreach emails, review them with AI, and send them from their own Gmail account on a business-hours schedule. The goal isn't to replace the job search, it's to remove repetitive work while keeping the candidate in control. I also chose a one-time credit model instead of monthly subscriptions because job seekers shouldn't have to keep paying while they're between opportunities. PitchHired is still evolving, and I'd genuinely appreciate feedback from fellow developers. What features would you want in a tool like this, and what would make you trust (or not trust) AI-assisted job search?

I started a gaming guide blog six weeks ago. Twenty-one articles later, it's getting traffic from Google, I have four affiliate programs set up, and I have never written a single line of code. This is not a "how to make money blogging" post. This is a practical breakdown of the tools, the workflow, and the mistakes I made so you can skip them. The blog is yxgonglue.com. It covers PC and console game guides — GTA VI pre-order comparisons, VPN setups for gaming, cloud gaming platform rankings, extraction shooter loot guides. Niche stuff. The kind of content people search for when they have a specific problem. Here is the stack that runs it. THE STACK WordPress + Kadence Theme Hosted on a standard shared hosting plan. Kadence is a free WordPress theme that loads fast and does not fight you. No page builder. No Elementor. Just the block editor and Kadence blocks for tables and formatting. The biggest lesson here: your theme does not matter as much as your content structure. Pick something lightweight. Stop theme-shopping. Start writing. Yoast SEO The free version. It gives you a red/yellow/green score for each post based on keyphrase density, subheading distribution, link count, and meta length. Is it perfect? No. Is it a useful checklist for someone who does not do SEO for a living? Absolutely. One thing Yoast taught me the hard way: Custom HTML blocks are invisible to the plugin. If you paste your article into a Custom HTML block, Yoast reads zero words, zero links, zero headings. Everything turns red. Use the regular editor. If you need a table, use a table block. Keep it simple. Google Search Console This is where you see what people actually searched before they clicked your article. The gap between what you think people search for and what they actually search for is enormous. Search Console closes that gap. Submit every new post URL manually. It takes ten seconds. Do not wait for Google to discover your site on its own. THE CONTENT WORKFLOW One Article Per Day Tw

Playwright and Selenium are great until you hit bot detection. Google OAuth, Cloudflare, and Vercel checkpoints all flag headless browsers. Here's how to control a real Chromium instance via CDP using Python and websockets. Why Not Playwright? Playwright launches a headless browser with automation flags. Even in headed mode with Xvfb, Google detects it. The CDP Approach Launch Chromium with remote debugging: chromium-browser --user-data-dir = /path/to/profile --remote-debugging-port = 9222 --no-first-run Connect via WebSocket in Python: import asyncio , json , websockets , urllib . request async def get_page_ws (): resp = urllib . request . urlopen ( ' http://localhost:9222/json ' ) targets = json . loads ( resp . read ()) for t in targets : if t [ ' type ' ] == ' page ' : return t [ ' webSocketDebuggerUrl ' ] async def cdp_call ( ws , method , params = None ): msg_id = cdp_call . id = getattr ( cdp_call , ' id ' , 0 ) + 1 msg = { ' id ' : msg_id , ' method ' : method } if params : msg [ ' params ' ] = params await ws . send ( json . dumps ( msg )) while True : resp = json . loads ( await ws . recv ()) if resp . get ( ' id ' ) == msg_id : return resp Key Advantages Real browser fingerprint, no automation flags Persistent sessions, cookies survive across runs Google OAuth works, existing sessions carry over No bot detection, it IS a real browser Follow for more tutorials on browser automation and AI agent architecture.

5 Ferramentas de IA Gratuitas que Todo Desenvolvedor Deveria Usar em 2026 A inteligência artificial não é mais o futuro — é o presente. E o melhor: muitas ferramentas poderosas são gratuitas . Neste artigo, vou compartilhar 5 ferramentas de IA que transformaram minha produtividade como desenvolvedor. 1. 🤖 GitHub Copilot (Gratuito para opensource) O Copilot se tornou indispensável para qualquer desenvolvedor. A versão gratuita oferece: Autocomplete de código em tempo real Sugestões contextuais inteligentes Suporte a +30 linguagens Como usar: Instale a extensão no VS Code e comece a digitar. O Copilot sugere código automaticamente. # Exemplo: Digite um comentário e o Copilot gera a função # Função para ler JSON de um arquivo def read_json_file ( filepath ): import json with open ( filepath , ' r ' ) as f : return json . load ( f ) 2. 🔍 Perplexity AI (100% Gratuito) Pesquisa com IA que cite fontes. Perfeito para: Pesquisar documentação Entender conceitos complexos Encontrar soluções para bugs Dica: Use o modo "Pro Search" para respostas mais detalhadas. 3. 🎨 v0.dev (Vercel) — Frontend com IA Gere componentes React/Next.js com descrições em linguagem natural. # Exemplo de prompt: "Um card de produto responsivo com imagem, preço e botão de compra" O v0 gera o código completo, estilizado com Tailwind CSS. 4. 📝 Notion AI (IA gratuita integrada) O Notion permite usar IA para: Resumir documentos longos Gerar templates de código Traduzir conteúdo automaticamente Criar documentação técnica Atalho: Pressione Ctrl/Cmd + J para ativar a IA em qualquer bloco. 5. 🔧 Cursor (Editor com IA) O Cursor é um fork do VS Code com IA integrada nativamente: Chat com IA sobre seu código Edição por comando ("adicione tratamento de erros") Compreensão automática do contexto do projeto Diferencial: Ele lê todo seu projeto e entende o contexto, não apenas o arquivo atual. 💡 Dica Extra: Combinando as Ferramentas O segredo não é usar uma ferramenta isolada, mas combiná-las : Perplexity para pesquisa

Headline: The wins come from context, not cleverness. An AI with your codebase, your design system, and your deploy logs in scope writes code that ships. Without that scope, it writes plausible code that doesn't. Two years ago, AI coding tools were autocomplete with attitude. In 2026 they are a credible second engineer — provided you build the workflow around them. This is the workflow I run today at Devya Solutions and on personal projects like eng-ahmed.com . The Stack Claude Code in the terminal — long-horizon, multi-file edits with skills and subagents. MCP (Model Context Protocol) servers for live access to docs, deployments, browser, and design tools. Cursor or VS Code for inline edits when I want to stay in the IDE. Why Context Is Everything The single highest-leverage move in AI-assisted dev is feeding the model the right context. MCP servers do this without prompt stuffing. Docs MCP — pulls current library docs at call time, so the model doesn't hallucinate the Tailwind v3 API in a v4 codebase. Browser MCP (Claude-in-Chrome) — lets the agent open the running dev server, screenshot the page, and verify the change actually rendered. Vercel MCP — fetches deploy logs and runtime errors directly. No more pasting logs. Context-mode MCP — keeps file scans, search results, and command output in a sandbox, only surfacing what's relevant to your conversation. A Real Workflow The blog page redesign I just shipped was built in a single 45-minute session. Rough flow: State the goal — two sentences, not a spec doc. Let the agent scout — Claude Code greps, reads a few files, proposes a plan. Iterate visually — screenshot the result, feed it back. The agent fixes the sticky-filter scroll bug in one turn. Commit and push — a single cm shortcut runs build, commits, and pushes. Vercel deploys on push. What the Agent Is Still Bad At Holistic taste — it copies the closest example in your codebase. If that's mediocre, the new feature is mediocre. Domain knowledge — it doesn't kn

Your quiet space for reading articles later. Discussion | Link

What This Guide Covers When you start building LLM-powered applications, one pattern becomes unavoidable: RAG (Retrieval-Augmented Generation) . LLMs only know what they were trained on. Your company's internal documents, the latest spec sheets, project-specific information — none of that exists in the model. To handle data the model doesn't know, you need a system that retrieves relevant knowledge in real time and injects it into the context. That's RAG. In this guide, we'll implement a RAG system from scratch using pgvector and Gemini, then extend it step by step through Tool Use, AI Agents, MCP, and cloud deployment. Step 1: Embedding · Vector DB · RAG — core implementation Step 2: AI Architect perspective — design decisions explained Step 3: Tool Use — LLM autonomously searches the DB Step 4: AI Agents — combining multiple tools Step 5: MCP — exposing tools as a server Step 6: Cloud deployment — Render × Supabase Three Concepts to Understand First Embedding Computers can't measure "semantic similarity" from raw text. Embedding converts text into a list of numbers (a vector), and semantically similar words produce numerically similar patterns. "dog" → [ 0.82 , 0.75 , 0.10 , ... ] 768 numbers "cat" → [ 0.78 , 0.72 , 0.12 , ... ] ← similar pattern to "dog" "bank" → [ 0.08 , 0.10 , 0.85 , ... ] ← completely different Gemini's embedding model handles this conversion. Vector DB A regular DB searches by keyword matching. A vector DB searches by numeric distance — meaning it finds semantically related documents even when the exact words don't match. -- Regular search (misses if keywords don't match) SELECT * FROM docs WHERE body LIKE '%F1 score%' ; -- Vector search (finds semantically related docs) SELECT * FROM docs ORDER BY embedding <=> query_vector LIMIT 3 ; Search for "how to measure model performance" and it finds "F1 score calculation" — even without matching words. We use pgvector , a PostgreSQL extension, for this. RAG LLMs are limited to their training data. R

The Agent Skills open standard today, and the 2026 research on agents that write their own skills. TL;DR: In late 2025, "Agent Skills" became a thing — a dead-simple way to teach an AI agent a task: a folder with a SKILL.md file (some instructions in Markdown). It's already an open standard. The wild part is what's coming next: agents that write their own skills. I built a demo where an agent solves a task the hard way once, saves a real SKILL.md , and then reuses it — cutting its total effort almost in half. ~130 lines, no API key. First, what's a "skill"? If you've used Claude Code or similar tools lately, you've probably seen SKILL.md files. The idea is refreshingly low-tech. A "skill" is just a folder with a Markdown file that says how to do something : --- name : csv-to-markdown description : Turn comma-separated text into a Markdown table. Use when the input looks like CSV and the user wants a table. --- # CSV to Markdown ## Instructions Split the text into rows on newlines and columns on commas. Make the first row the header, add a `---` divider row, then format every row as `| a | b | c |`. That's it. No SDK, no config. Anthropic introduced this in October 2025 and then published it as an open standard ( agentskills.io ) in December 2025, so the same skill folder now works across ~30+ different agent tools (Claude Code, Cursor, Copilot, and more). The full rules are short ( agentskills.io/specification ): the only required fields are name (1–64 chars, lowercase-with-hyphens, and it must match the folder name) and description (≤1024 chars, saying what it does and when to use it ). Everything else — license , metadata , compatibility , allowed-tools — is optional. That's the whole spec. The SKILL.md files my demo writes follow it to the letter, so they'd load unmodified in any compatible CLI. The clever trick: progressive disclosure Here's the smart part. If you just dumped 50 skills' worth of instructions into the agent's context, you'd fill it up and leave n

Have you noticed how every demo of "an AI agent" looks impressive in the video and falls apart the moment you ask a sharper question? The agent confidently does the wrong thing. It forgets what it just decided. It tries to call a tool that doesn't exist. It loops forever rewriting the same file. It calmly tells you the deployment succeeded when it didn't. These aren't failures of the model. They're failures of the workflow around the model. Because that's all an agent really is: a software workflow where a language model can pick the next step and call tools. The "intelligence" sits in the prompt and the orchestration around it, not in some secret agent-flavoured fairy dust. Strip the word "agent" away and you've got five pieces of plumbing: planning, tool use, memory, constraints, verification. Every production-grade agent stands or falls on those five. This is a long walk through each one. Not the marketing version. The kind of detail you actually need before you ship something that talks to your database. The Loop You're Actually Building Before we touch any pillar individually, hold the whole loop in your head. A useful agent does roughly this on every turn: Read the goal (and whatever memory is relevant to it). Decide the next action: answer directly, call a tool, ask a clarifying question, or stop. If it called a tool, observe the tool's result and feed it back in. Update memory if anything is worth remembering. Check constraints: are we over budget, out of iterations, touching something off-limits? Verify the output before declaring success. Loop until done or stopped. That's it. Every framework (LangGraph, OpenAI Agents SDK, Claude Agent SDK, smolagents, whatever ships next month) is a different shape of the same loop with different defaults. agent-loop.ts async function runAgent ( goal : string , ctx : AgentContext ) { const state = ctx . startState ( goal ); for ( let step = 0 ; step < ctx . maxSteps ; step ++ ) { const decision = await ctx . model . decid

I'm building TTT -- a terminal text editor and IDE written in Go. Single binary, zero config, runs anywhere. Think VS Code but in your terminal. It has syntax highlighting, LSP integration, a plugin system, an integrated terminal, git integration, etc... The source is on GitHub and I develop it with Claude Code as my pair programmer. This is the story of how a frustrating limitation turned into something genuinely useful: a built-in scripted interaction system that lets AI agents (or anyone) drive the editor like Playwright drives a browser. The problem I was deep in revamping the widget system and building out a Lua plugin API. Phases of work stacking up -- widget rendering, panel support, tree views, input fields, command registration, keybinding hooks. The kind of work where you need to see what's happening. Click a tree node, check if it expands. Open a panel, verify focus moves correctly. Run a plugin, confirm the dialog appears. Here's the thing: Claude Code can run shell commands and read files. It cannot interact with a live TUI session. The editor launches, takes over the terminal, and that's it -- Claude is blind. Step 1: tui-use (what we had) The project already had functional tests using tui-use , a JavaScript library that drives a real terminal binary. It can type, press keys, wait for text to appear, and take snapshots: const tui = await start ( " bin/ttt " , [ " test-file.go " ]); await tui . waitFor ( " test-file.go " ); await tui . exec ( " editor.joinLines " ); const screen = await tui . snapshot (); expect ( screen ). toContain ( " joined line " ); This works. But it's slow -- each test spawns the binary, waits for screen renders, polls with timeouts, and parses terminal escape codes. And critically, it can't click . Mouse events aren't supported. For a widget system with tree views, buttons, and split panels, that's a dealbreaker. Step 2: Debug commands (the workaround) So we added a Debug: Simulate Click command to the editor itself. Open the co