Prompt & Done
AI tools & productivity, daily.
Used in real work
before it is listed.
From the videos
65Prompt packThe PROMPT: Claude built this videoThe prompt to start your own 60-second 3D video built entirely in code with Claude and Three.js.ToolThe RELINK packKytyPS5 and AnyPS5 (free, open source), the GTA V and Astro Bot sources from the reel, and our count of AnyPS5 commits that credit Claude.Prompt packThe FLAPPY packAll three games from the reel (Claude Opus 5.5, GPT-6.1 Sol and free Qwen 3.5 on a laptop), the exact one-line prompt, and all three rounds.Prompt packThe ARCADE packBoth games from the reel (Haiku 5.5's 3¢ one and Opus 5.5's 39¢ one), the exact one-line prompt, and all three race results.Prompt packThe Long-Chat ResetThe handoff-note prompt from the reel: pack a long chat into one note, open a fresh chat, and keep going. Plus Anthropic's own tips for making your usage last longer.Tool10 AI Tools for Job SeekersThe full toolkit from the reel — ChatGPT, Claude, Teal, Jobscan, LinkedIn, Canva, Grammarly, Huntr, Levels.fyi, and Final Round AI, each linking to its official site.ToolOpenAI DotsOpenAI's always-on AI agents — give yours a goal and it works across 4,000+ apps with its own computer and browser, asking your approval before anything big.Prompt pack2 ChatGPT Prompts: 0 to 30M ViewsThe exact 2 prompts from the reel — one turns ChatGPT into a viral strategist for your niche, the other writes 30 days of scripts in your voice.GuideAI Engineering from ScratchA complete AI engineering degree, free on GitHub — 523 lessons across 20 phases, from math to LLMs to agents, and every lesson ships a real artifact: a prompt, a skill, an agent, or an MCP server. ~58K stars, MIT.ToolAI Job SearchFree, open-source framework that turns Claude Code into a full job-search machine — it scores postings, tailors your CV, writes cover letters, and preps you for interviews. ~44K stars, MIT.ToolGod's Eye ViewFree 3D "spy satellite" in your browser — live aircraft, ships, satellites, earthquakes and public cameras on a photorealistic globe, with voice control. No API keys, no signup.ToolOmniRouteFree, open-source AI gateway — one endpoint routes you through 1,200+ models with quota-aware auto-failover and token compression that stretches every token up to 10x further.ToolVoiceStudioFree, open-source local ElevenLabs alternative — clone any voice from a 3-second clip, dub videos, transcribe, make audiobooks. No account, no API key.ToolClaude for Financial ServicesAnthropic's free, open-source agent pack for finance — pitch decks, DCF models, earnings reviews, and more.ToolImpeccableFree, open-source design guidance for AI coding agents — 1 skill, 24 commands, 61 rules that kill ugly AI website tells.Toolvideo-useFree, open-source video editing by AI coding agent — drop in raw footage, it cuts, grades, captions, and self-checks the render.ToolAXGoogle's open-source Kubernetes for AI agents — declare a task in YAML, run it in an isolated sandbox.Toolagent-skills25 free engineering skills that turn your AI coding agent into a senior engineer — testing, reviews, specs.ToolUnbagrndFree, open-source — removes photo backgrounds on your own device. No account, no uploads, no limits.ToolJevThe AI model that never writes a word — typed decisions with probabilities, in milliseconds, for fractions of a cent.ToolOpenCodeReviewAlibaba's internal AI code reviewer, now open source — line-level precision, 1/9 the tokens.ToolOrcaFree, open-source — fan one prompt across a fleet of coding agents in isolated worktrees, merge the winner.ToolOpenCutFree, open-source CapCut clone in your browser — 4K export, auto captions, no watermark.ToolDuckDB SkillsAsk your data files questions in plain English — no Python scripts.ToolaffiliateWispr FlowAI voice dictation that cleans up filler words. Free to start.ToolAgent-ReachGives your AI agent eyes on the whole internet.ToolBrowserSkillTencent's tool: your AI agent borrows a browser tab, already logged in.ToolbragTurn the project you just shipped into a launch video with one command.Toolsecurity-audit-skillCloudflare's skill: tell your AI to audit your code — and it actually does it.ToolHyperFramesWrite HTML. Render video. Built for agents.ToolMuse Gadgets SDKMeta's open SDK for building your own Muse gadgets — ESP32 and Linux firmware, Apache 2.0. Grab an API token, point your coding agent at it, build screens, buttons, sensors.ToolClaude Code ModsTiny TypeScript add-ons that plug straight into Claude Code — rewrite what it says, block what it does, even redraw the screen. Four ship built in: diff view, security guard, usage stats, project instructions.ToolProcedural Isometric WorldsA skill that generates entire miniature 3D worlds from code — a rocket launch with exhaust shaders and pooled smoke, a waving robot, even a garden center. No Blender, no modeling skills: describe what you want, get a world.ToolBeamReflection AI's 501B open-weight model — only 23B parameters fire per token, with a 1M-token context. Join the waitlist for early access.ToolREAMCP server that turns your coding agent into a reverse engineer — point it at any app, no source code needed.ToolAnatomyA free, open-source Claude Code skill: ask how something works and it builds an interactive machine you can drag, press and take apart. Install line and the exact prompt from the reel inside.ToolChief of StaffA free Claude Code skill (beta): give Claude one goal and it hands the work to background agents while you keep chatting. Install line, how to run it and the exact goal from the reel inside.ToolOpenMontageA free, open-source toolkit: one prompt to Claude Code made a 30-second Apollo 11 documentary from real footage (4 approvals, 29 min, no API keys). The repo, the exact prompt and our run notes inside.ToolDIORAMA promptOne prompt makes Claude write a complete animated 3D world as a single HTML file (Three.js). Change one line, get a new world. The prompt, the two lines from the reel and our run notes (2 of 3 first tries worked) inside.Toollast30daysA free (MIT) Claude Code skill: ask a question and it reads the last 30 days of Reddit, Hacker News, GitHub and Polymarket, then writes a brief with links and a "choose X if" verdict. Install lines, the exact command from the reel and five prompts to try inside.Tool/code-review bug huntClaude Code's built-in /code-review reviews your branch before you merge. The command with effort levels and --fix, the demo shop repo with five planted bugs, and what our run found (all five plus five we didn't plant) inside.ToolHeadroom token testHeadroom (free, Apache-2.0) squeezes logs before Claude reads them. Our test: a big log sent straight to Claude went from 149,481 to 8,754 tokens with the same answer; inside Claude Code it saved about 2%. The Python snippet for your own agent, the exact numbers and the scripts to rerun it inside.ToolPonytail: 59 lines or 6Ponytail (free, MIT) makes Claude Code check what already exists before it writes code. Our test, same prompts with and without it: 1,046 lines vs 243 across four tasks, all 24 versions built. The two install lines, the four prompts and the real 6-line date picker inside.Tool/goal: 5 finish lines/goal is built into Claude Code: you type the finish line and a second model checks it after every turn. Our test: a broken demo site went from Lighthouse 56/47/62/73 to 100 x4, the judge said "not yet" when Claude rewrote three links, and the goal closed on turn 2. Five finish lines, the recipe and the audit script inside.ToolThe INKLIGHT questionWhatsApp's help pages had hidden notes for its support AI, including "Do not recommend contacting any ... regulator" (Redact, WIRED; WhatsApp says a bug, now removed). We gave Claude the same page and note: wait 120 days, 3 of 3. One question changed that. The question, our test and how to check the archived pages inside.ToolMOTION: one prompt, one launch videoHyperFrames (free, Apache-2.0) lets Claude Code write a video as a web page and render it frame by frame on your computer. Our runs: a 12-second launch video in 7 min 18 s (it caught and fixed a missing logo itself), a channel intro in 2 min 35 s, an animated chart in 1 min 51 s. The two setup lines and all three prompts inside.ToolCRYWOLF: an AI security audit that argues with itselfCloudflare's open-source security-audit skill: hunter agents find suspects, then a fresh agent tries to disprove each one. On our own small notes app with 5 planted holes and 5 decoys it confirmed all 5 holes and no decoy, in about 10 minutes. Setup and our run inside.ToolREFCHECK: catch fake and retracted AI citationsWe asked Claude for sources 6 times (60 citations): 2 were a retracted study, 2 had fake DOIs, 2 pointed at another paper. A free MIT citation skill fixed each list in about a minute. The prompt and setup inside.ToolBLENDER: one prompt, a whole 3D sceneOne prompt to Claude Code drove free Blender through the open-source Blender MCP: it built, lit and rendered a snowy cabin in 11 min 45 s and fixed two of its own mistakes by looking. Setup and prompt inside.ToolLOWEFFORT: same prompt, five Claude effort levelsOne kanban prompt on Claude Opus 5.5 at low, medium, high, xhigh and max: all five passed the same ten checks. Low took 32 s ($0.17), max 40 min ($7.01). The prompt, the checks and every number inside.ToolFANOUT: eight games from one promptClaude Code with one subagent per game built 8 browser games in 45 s, versus 126 s one after another (about 2x the tokens). Both prompts and all eight playable games inside.ToolCHATTERBOX: a free voice clone on your own computerChatterbox (Resemble AI, MIT) clones a voice from a short clip and runs offline on your own computer; on our laptop a 5-second line took about 3.4 s. Setup and our test inside.ToolDEPTHPOP: one photo to a 3D sceneFree Depth Anything V2 Small script: photo to depth map and a looping 3D move, offline on a laptop CPU (0.28 s per photo in our test).ToolLLAMACODE: Claude Code on a free local modelollama launch claude runs Claude Code on a free local model. Our test: Qwen 3.5 9B built a small website from one sentence in about 3 minutes for $0 (small jobs; games were hit and miss). The commands inside.ToolNIGHTSHIFT: 8 reels from one promptThe exact prompt that had Claude Code make 8 finished reels one at a time while I slept, plus the setup, trimmed skills, templates and the usage check.ToolPLAYWRIGHT: Claude fills any web formClaude Code plus the Playwright MCP filled our 35-field sample rental form from one pasted email and stopped on the review page. The setup, the prompt and our run numbers.ToolCLONE: the self-checking page clone loopClaude Code renders its copy, diffs it against the original and fixes the red. On our Wikipedia test it went from 17% of pixels wrong to 2.5%. The prompt and the small compare script.ToolPUSHY: one sentence that makes skills fireAnthropic's skill-creator says to make skill descriptions a little pushy. One added sentence took Haiku from 10 of 64 to 39 of 64 right skill loads in our test. The sentence and the test.ToolGRAPHIFY: map a whole codebase for freeFree, Apache-2.0 Graphify mapped Django's 3,140 files in about 43 seconds with zero tokens, and answers how two parts connect. Install and commands for Claude Code and Cursor.ToolCHORUS: free local songs with ACE-StepACE-Step 1.5 (MIT) sings Claude's lyrics on a Mac: three 60-second songs in 87 seconds. Install steps, the exact requests, our numbers and three full songs to listen to.ToolHAIKU: Haiku vs Opus vs Fable, same game promptOne tower-defense prompt, three Claude models: $0.05, $0.87 and $2.05. The exact prompt and all three games to play in your browser.ToolKLEIN: free photo edits on a MacFLUX.2 klein 4B (Apache 2.0) makes a photo from one sentence, then edits it with typed instructions, all on a Mac for $0. The install command, prompts and our test numbers.ToolFENCE: the Claude Code sandbox settingsWith permissions off, a poisoned dependency sent our fake key out 6 of 6 times; with /sandbox on, 0 of 6, and the tests still passed. The settings and our test rig.ToolMINIAGENT: an AI agent in 30 linesA 30-line, four-tool loop turned a free local model (Qwen via Ollama) into an agent that fixed two buggy projects in about 27 seconds each. The code, a sample project and setup.ToolSLOP: make AI websites stop looking the sameSame Claude, same prompt: the plain page failed Hallmark's linter 29 times, the /hallmark page passed clean. Install, the prompt, both pages and our numbers.
No matches — try a different search.