someone built a VISUAL cheatcode book for Claude Code that takes you from “/hello” to full agent orchestra in a weekend
10 modules, 11–13 hours, all copy‑paste:
1) slash cmds
2) memory
3) skills
4) subagents
5) MCP
6) hooks
7) plugins
8) checkpoints
9) advanced
10) CLI
5,900+ devs already using it like a Claude OS layer for their projects
perfect visual: screenshot of the learning roadmap table + one Mermaid diagram showing slash commands → hooks → subagents pipeline
2
1
68
someone built an OPEN-SOURCE voice stack that eats hour-long audio + talks back in real time
it’s called VibeVoice by microsoft:
1) ASR: 60 min in ONE pass → who spoke, when, and what, in 50+ languages
2) TTS: 90‑min multi-speaker convos
3) Realtime: ~300ms latency, 0.5B params
attach: demo waveform + spectrogram GIF, screenshot of ASR “who/when/what” transcript, and a short clip of realtime TTS responding while text streams in.
1
1
55
someone basically built a fully autonomous PHD STUDENT that writes its own workshop paper and gets it PEER REVIEWED 🤯
AI-Scientist v2:
1) takes your topic .md
2) generates hypotheses + experiments via agentic tree search
3) runs the code on GPU
4) analyzes results
5) spits out a full PDF paper
no human templates, no hand-holding — it scouts ideas, debugs failing paths, checks novelty via Semantic Scholar, then drafts the manuscript end-to-end
cost is “a few bucks” for ideation + ~$20–25 for a full run with Claude/GPT, but you get a timestamped folder with logs, HTML tree viz of the whole search, and a final workshop-level paper
this used to be a whole lab of grad students
now it’s a Python repo, a CUDA GPU, and your API keys
1
25
someone built a *live* deepfake studio that runs on your laptop in 3 clicks 😵💫
Deep-Live-Cam:
1) real-time face swap from a **single** image
2) works on webcam, videos, even movies
3) multi-face mapping + “mouth mask” so your own lip sync stays perfect
press is calling it “truly terrifying” – it’s open source
(attach: demo GIF from README, plus a short screen recording of webcam face-swap + multi-face meme scene side-by-side)
1
159
someone built a “drop a prompt, spawn an AI dev team” layer on top of Claude Code and it’s kinda cracked
it’s called oh-my-claudecode — teams-first multi-agent orchestration:
1) `autopilot: build a REST API…`
2) `/team 3:executor "fix all TypeScript errors"`
3) 32+ specialized agents, tmux workers for Claude/Codex/Gemini
no YAML, no config, no prompt temples. just natural language → team-plan → exec → verify → auto-fix loops
visuals: screenshot of the HUD/statusline + a tmux session with multiple AI panes running in parallel
3
1
43
someone built a research agent that reads the entire internet’s *last 30 days* for you and turns it into one sane brief 🔥
it’s called `/last30days`
you type:
> /last30days "cursor vs windsurf"
it:
1) scrapes Reddit, X, Bluesky, YouTube, TikTok, IG, HN, Polymarket + web
2) ranks by what’s actually upvoted, shared, bet on, and said on camera
3) runs 3 parallel passes and gives you a side‑by‑side: strengths, weaknesses, verdict
4) auto-saves a markdown brief into `~/Documents/Last30Days/` so you’re quietly building your own intel library
this used to be “open 12 tabs, doomscroll 2 hours, still miss half the alpha”
now it’s a single command in Claude Code or Gemini CLI and a 2–8 min wait while your personal OSINT intern works in the background
visual: screenshot of a `/last30days` comparison run + the generated markdown brief in a Documents/Last30Days folder
1
60
someone built a FUSION REACTOR for floorplans in the browser
it’s called Pascal Editor – a 3D building editor using React Three Fiber + WebGPU, with a game-engine style core:
1) node-based scene graph (walls, slabs, roofs, scans, guides)
2) dirty-node systems that regen geometry on the fly
3) separate stores for scene, viewer, editor (all Zustand + undo/redo via Zundo)
walls get mitered + boolean-cut for doors/windows, slabs/ceilings/roofs are generated by systems, and a spatial grid tells you where items can *actually* go
dev-wise it’s a Turborepo monorepo:
- @pascal-app/core → schemas, state, systems, spatial grid, event bus
- @pascal-app/viewer → R3F/WebGPU viewer with level modes (stacked / exploded / solo)
- apps/editor → tools like WallTool, ZoneTool, ItemTool on top
you run `bun dev` once at the root and get hot-reload across everything
this is basically “indie BIM engine” vibes—except it’s just TypeScript + React and lives in your browser
attach:
- short GIF of drawing walls → dropping doors/windows → toggling floor levels
- screenshot of exploded-view levels
- code snippet shot of the dirtyNodes + system update loop
1
55
someone basically shipped an OPEN‑SOURCE AI STAFF that researches, codes & ships products while you sleep
it’s called DeerFlow 2.0 — a “super agent harness” that:
1) spins up sub‑agents
2) runs them in sandboxes
3) keeps long‑term memory
4) connects to Slack / TG / Feishu so you just DM it work
runs via Docker or local `make dev`, plugs into GPT‑4/5, Claude, DeepSeek, Kimi, OpenRouter, even custom MCP servers
this used to be bespoke infra at BigCo scale — now it’s a Python repo and `make up` away
attach: GIF of the web UI orchestrating multiple sub‑agents + screenshot of config.yaml showing multi‑model + IM channels setup
2
42
someone open-sourced a FULL AI TRADING FLOOR in one Python framework
it’s called TradingAgents: a multi-agent LLM setup where:
1) analysts read fundamentals, news, sentiment, charts
2) bulls vs bears debate
3) trader + risk + portfolio agents decide the move
runs via CLI or `TradingAgentsGraph().propagate()` with GPT, Gemini, Claude, Grok, OpenRouter or even local Ollama
this is “quant firm but make it open-source”
attach:
- screenshot of the CLI ticker/date selection screen
- diagram of the multi-agent pipeline (analyst → researcher debate → trader → risk → portfolio)
- a code snippet image showing the NVDA example from the README
3
1
50
someone built an open source MONEY PRINTER that runs your whole content hustle for you 😂
it’s called MoneyPrinter V2:
1) auto-posts to Twitter on a schedule
2) cranks out YouTube Shorts
3) plugs into Amazon affiliate + Twitter
4) scrapes local businesses + cold outreach
clone, drop your keys in config.json, `python src/main.py` and it just…goes
visuals:
- screenshot of the CLI dashboard running
- short GIF of a YouTube Short being auto-uploaded
- config.json snippet (keys blurred) to show how simple setup is
1
29
someone basically open-sourced Stripe/Ramp/Coinbase’s internal AI dev bots and called it… Open SWE 🔥
it’s an async coding agent you run *inside* your org:
1) triggers from Slack / Linear / GitHub (`@openswe`)
2) spins an isolated cloud sandbox per task
3) edits code, runs tests, then opens a draft PR automatically
4) reads `AGENTS.md` so it actually follows your conventions
5) can spawn subagents for parallel subtasks
this is “AI teammate for your repo” architecture that used to be locked inside elite eng orgs — now it’s a Python framework you customize, built on LangGraph + Deep Agents
visuals:
- 1 screenshot of a Slack thread where `@openswe` reacts with 👀 and drops a PR link
- 1 diagram of the sandbox + tools + middleware flow
- optional GIF of a Linear issue → comment `@openswe` → PR created end-to-end
1
42
someone built a PDF parser that actually understands your docs instead of blenderizing them for your RAG stack
it’s called opendataloader-pdf and it’s benchmarking #1 overall (0.90) with 0.93 table accuracy
1) PDF → Markdown/JSON/HTML with bounding boxes
2) hybrid AI mode for nasty tables, formulas, scans (80+ OCR langs)
3) Python / Node / Java SDKs, local mode is 0.05s/page
and the wild part: same engine is becoming the first open-source end‑to‑end auto‑tagger to Tagged PDF for accessibility (PDF Association + veraPDF crew involved) — manual $50–200/doc remediation turns into “run script, ship compliance”
visuals: share the annotated PDF screenshot (bounding boxes), plus the benchmark chart comparing engines for max flex
2
1
41












