Cursor died on me in a year. I don't miss it. Two computers under my TV run Claude Code 24/7. Sometimes 10 instances, sometimes 20. Sometimes #Claude Code manages the other Claude Codes.

People asked for water after seeing desert. So I built water. Turns out that's the whole content strategy. Post. Watch what lands. Build the next thing they want. (Scheduled via Publora — so the agent posts while I build.) Marketing teams write 40-slide decks about this process. Indie builders just do it.

Coding agent memory built from session history already on disk: 97.2% R@5 on LongMemEval-S, indexes 19,195 sessions in 17.6s, no LLM, single Go binary. Works with @claudeai Code, @cursor_ai, @OpenAI Codex via MCP. github.com/vshulcz/deja-vu

MCP server from Apify lets AI agents call thousands of ready-made scrapers for sites, maps, and search — works with @claudeai, @ChatGPTapp, @cursor_ai via HTTPS or local stdio. 11k stars. github.com/apify/apify-mcp-s…

Agent skill that turns complex answers into a single offline HTML page: ~8× fewer tokens (612 vs 4893) and 2.6× faster (12s vs 31s) than asking an LLM for HTML directly. Works with @claudeai Code, @OpenAI, @cursor_ai, @opencode. github.com/QingYunA/answer-m…

28 reusable agent skills for the full dev loop—spec, implement, review, auto-research—compatible with @claudeai Code, @OpenAI Codex, @cursor_ai and 70+ skill-aware agents. Install via npx skills add. github.com/dzhng/skills

Personal AI agent that deploys to your own Cloudflare account via `npx create-talorys@latest` — chat, memory, tasks and reminders stored in a single SQLite Durable Object, no telemetry, free-tier friendly. 🛠️ @Cloudflare github.com/rociiu/talorys

95% of intercontinental internet traffic runs through undersea cables. Cables. Not satellites. Not magic. Not the cloud. Actual physical cables on the ocean floor that sharks occasionally chew on. We built $5 trillion worth of AI mythology on top of infrastructure a plumber would recognize. (Meanwhile I'm automating posts across 10 platforms via Publora. Very cyberpunk. Very physical cables.)

Curated list of 64 AI tools for UI/UX work: 25+ design skills (Impeccable, Material Design 3, Swiss Design System), MCP servers that fetch real components instead of generating from scratch, plus @cursor_ai / @claudeai Code / @OpenAI configs. github.com/maxbogo/awesome-a…

Autoharness gives @claudeai Code a self-learning skill layer: reflection every 50 tool calls, consolidation every 250, skills capped at 1024 chars, pruned if unused. Pure Python 3.11+, no deps, no daemon. github.com/tigerless-labs/au…

Prompt-master is a @claudeai skill that builds optimized prompts for 30+ AI tools via an 8-stage pipeline, 13 auto-selected templates (RTF, CO-STAR, ReAct…), and detects 37 credit-wasting patterns. 13.6k★ github.com/nidhinjs/prompt-m…

Menu-bar switcher to point any coding agent at any model: @claudeai Code on @Kimi_Moonshot, @OpenAI on @deepseek_ai, 14+ agents, 20+ providers, edits configs surgically without touching comments. <15MB Go binary. github.com/yetone/magpie

Hippo-memory: biologically-inspired memory layer for AI agents with decay, retrieval strengthening, and consolidation. TypeScript, SQLite, zero runtime deps, MCP support for @claudeai Code and @cursor_ai. Hits R@5 = 74% on LongMemEval with just BM25. github.com/kitfunso/hippo-me…

Self-organizing markdown knowledge base for @claudeai Code + Obsidian: 18 agent skills, 72 slash commands, 6 subagents, 4 zero-dep Python scripts. New sources update existing pages instead of piling up; contradictions logged with dates, not overwritten. github.com/undefined-ui/seco…

Rockstar spent $1B–$2B building GTA VI over 12 years. Someone on Reddit made GTA India. For free. I scheduled this post across 10 platforms via Publora while a single developer out-created a billion-dollar studio. The gap between production cost and creative output keeps getting harder to explain with a straight face.

Claude Code plugin that captures your corrections via regex + semantic analysis, then syncs them to CLAUDE.md and AGENTS.md with confidence scores from 0.60–0.95. 160 tests, runs on macOS/Linux/Windows, no WSL. @claudeai github.com/BayramAnnakov/cla…

Tencent's BrowserSkill lets AI agents drive your already-logged-in Chrome or Edge via a CLI + extension, with human-in-loop for CAPTCHAs and logins. Works with @cursor_ai, @claudeai Code, @OpenAI, @deepseek_ai Harness. 4.2k stars. github.com/Tencent/BrowserSk…

Persistent memory for coding agents: 95.2% retrieval accuracy (R@5), 92% fewer tokens, 54 MCP tools, works with @claudeai Code, @cursor_ai, @GitHubCopilot and 17 others. No external DB, state lives in ~/.agentmemory. github.com/rohitg00/agentmem…

Reticle gives coding agents runtime perception: it runs your app, inspects network, storage & console, then returns pass/fail with file:line. Works with @cursor_ai, @claudeai Code, Copilot via MCP. @reactjs, Vue, @sveltejs, @electronjs, @TauriApps. github.com/reticlehq/reticle

MCP server serving 2,320 pages from 832 real sites so coding agents can learn design — extracts palette, fonts, CSS vars into DESIGN.md, plus 68 copy-paste JSX components. Install via `npx -y inspo-mcp install`. github.com/Nutlope/inspo

Graft builds a local code graph via tree-sitter (~3ms rebuilds) for @claudeai Code, @cursor_ai, @OpenAI & @GeminiApp. On SWE-bench Verified: 54%→66% correctness, -23% tokens, -32% time. github.com/trailhq/Graft