Videos / Categories / Developer tools

Developer Tool Launches

Explore real developer tool launches, product demos and announcements from founders building coding tools, infrastructure, APIs, automation and developer productivity software.

These launches are also written up as startup ideas for 2026.

Looking for something to build?

Explore developer tool startup opportunities inspired by these real launches.

Explore 114 developer tool startup ideas

115 films

Game Rebuild Kit launch videoGame Rebuild KitPlaysGary's post says one click analyzes any game and rebuilds the whole implementation. The repository is GF007/game-rebuild-kit. The skill text is MIT. The readme says one-click automation and a complete rebuild are still goals. This release is an agent skill and one worked example, distilled from a single Android and Unity look. Fitted numbers, not a check that the result matches the original game. The kit does not include extracted assets, a format decoder, a viewer, or an automated rebuild. Game footage in the film keeps its own rights. The post says any game, one click. The folder is one example and a skill. This page did not rebuild a game.
Game Rebuild Kit · @Fun63880905 · Nov 20, 2026Developer tools · 0:36
Safari browser agent launch videoSafari browser agentPlaysNart's post says WebKit opposes WebMCP, so he built an open-source Safari agent that supports it. The repository is nartmadi/safari-browser-agent. The license is AGPL-3.0. You add the folder as a temporary unsigned extension. A sidebar can call tools a page exposes through the WebMCP polyfill. The model can be local Ollama, or a key for OpenAI, Anthropic, Gemini, or Bedrock. The readme calls it an experimental preview. A tool call can change the site. If a request times out, look at the page before you try again. A consequential tool asks Allow or Cancel. A read-only hint is respected. The tools button shows up only when that page exposes tools. A normal site has no button. This page did not click one.
Safari browser agent · @nartmadi · Nov 19, 2026Developer tools · 0:46
openRocode launch videoopenRocodePlaysNoa's post says he built openRocode so a sentence becomes code, and the harness runs that code as guarded motion on a robot. The repository is wrencanfly/openRocode. The license is Apache 2.0. Claude, GPT, or any OpenAI-compatible model can write the step. The body checks the workspace and the force, a person can step in, and an emergency stop cuts a program that is already running. A motion that stops early is not retried, and the rest of that turn is cancelled. The robot in the readme is one MuJoCo tabletop. The gripper floats. There is no arm yet, so reach and joint limits are not modelled. The code runner has a time budget. It is not a security sandbox. One run is one episode. The post says a model can drive a robot. The gripper in this film is not attached to an arm. This page did not move a robot.
openRocode · @noahatwhere · Nov 19, 2026Developer tools · 0:18
Pulse launch videoPulsePlaysVikash's post says a dozen agents, a few bots, and some cron jobs left him unsure which one was waiting. He built Pulse as mission control for Kiro Crew. The repository is ivikash/KiroCrew-Pulse. The license is Apache 2.0. One page lists who is running, what is waiting on you, what runs next, and what today cost, then lets you answer and approve a tool call without opening the session. It needs Kiro Crew 0.12 or later. The readme says there is no storage, no cron, no network, and no backend of its own. The waiting row is whatever the crew already reported. The post's word is stuck. The page does not add a detector. This page did not approve a tool call.
KiroCrew Pulse · @know_vikash · Nov 18, 2026Developer tools · 0:24
debugealo launch videodebugealoPlaysCarlos's post says he built one keyboard-first terminal for Docker and for debugging Python, Go, and Node. The install line is debugealo. The repository is carlosrivas2021/debuguealo. The license is MIT. Go debugging needs Delve, and a missing Delve is only a warning. Debug ports listen on localhost. After install, the tool works offline. The readme says Windows has not been rechecked since the Python attach and the F5 and F9 changes. The post shows one terminal. The folder you clone is not the word you type to install it. This page did not attach a debugger.
debugealo · @adevlikeyou · Nov 18, 2026Developer tools · 0:52
Connect Anything launch videoConnect AnythingPlaysTerp's post says he built Connect Anything so an agent on one harness can start an agent on another, on a subscription the person already pays. Local or cloud. The repository is OnlyTerp/connect-anything. The license is Apache 2.0. The readme lists local CLIs including Claude Code, Codex, Cursor, and Grok, and a shorter list of cloud agents. A spawn starts at most 10 cloud jobs, and 20 can run at once. Local jobs share the machine, so six is the cap, and one stops after 30 minutes. Read-only is the default. The line that matters: an agent called through this server cannot call further agents. The server refuses at one hop. The post says any agent can work with any other. The called one stops there. This page did not start an agent.
Connect Anything · @OnlyTerp · Nov 18, 2026Developer tools · 0:23
EDwaRd launch videoEDwaRdPlaysThomas's post says he built his own macOS security monitor, EDwaRd, and then asks what enterprise antivirus has that his agent and OSINT do not. His answer is OS privilege from Apple, and telemetry from millions of their users. They can see a threat once and block it everywhere. The repository is LlamzonAmazon/edward. The license is Apache 2.0. It is a small Go program. It polls programs, network connections, apps, login items, and new files, and writes a local JSON file. A page on localhost shows it. An agent can read that store. The readme says it has no kernel sensor, no malware signatures, and no blocking. It cannot tell you whether a file is malicious, and it never blocks or deletes anything. It shows what a normal user on this Mac can already see. It polls snapshots, so a program that runs for less than a few seconds may never appear. The post says the enterprise product can block a threat everywhere. This one cannot say the file is bad. This page did not run the monitor.
EDwaRd · @thomasllamzon · Nov 17, 2026Developer tools · 0:16
AnyMachine launch videoAnyMachinePlaysApek's post says he built a 100-foot walking robot in Blender, entirely from code, using AnyMachine. It starts from one reference image, makes more views, and Fable builds the parts. His stilt walker was 160 unique parts, with hydraulics, a rig, and a walk cycle. He says the method generalizes to other machines. The repository is apekshik/anymachine. The license is MIT. The readme's examples are that stilt, a laptop, a quadruped, a spider, and backpacks. Every control is a 0 to 1 parameter, and a check sweeps for overlaps, clamps a value that is out of range, and measures foot slip. The roll-top backpack passes. The rucksack and the daypack fail where the controls interact. The generator gave the spider six legs. The eight come from the brief. The post says it generalizes really well. The readme's own backpacks do not all pass. This page did not build a rig.
AnyMachine · @apekshik · Nov 16, 2026Developer tools · 1:27
agemu launch videoagemuPlaysPiotr's post says agents write mobile code faster than it can be checked, so the emulator became the bottleneck. He built agemu. The same 18-step checkout, on his numbers: adb 74 seconds, Argent 30, mobile-mcp 29, agent-device 24, agemu 10.6. Open source. His reply says one command adds it to Claude, and it works with Codex and Cursor. MIT. The repository is piotr-graczyk-dev/agentic-emulator. The readme says it replaces the control layer and leaves the devices stock. A screenshot that took about 150 milliseconds is 4 to 5. A UI tree that took about 2 seconds is 2 to 23 milliseconds. A reset that took 14 seconds is 2.1. It drives the Android emulator and the iOS Simulator. It does not drive a real phone. Boots, app launches, and the app's own animations take as long as before. That time belongs to the OS and the app. iOS needs an Apple silicon Mac and private Xcode APIs. Android on Windows is a preview, and the taps are not verified there. The post's 10.6 seconds is the checkout. It is not the boot. This page did not run the checkout.
agemu · @piotrgraczykk · Nov 16, 2026Developer tools · 0:38
V-NOC launch videoV-NOCPlaysYared's post says he released V-NOC, an open-source extension for VS Code and Cursor. It lets you focus on one function, explore its call tree, and follow the code without jumping between files. The Python analyzer is built on ty, with his own layer for the call tree. Markdown docs can be attached to a function, a class, a file, or a folder, and they stay in the repo. Next, he says, is a CLI so agents can organize docs around the code. The repository is yaredtsy/vnoc. The license is Apache 2.0. The readme says Python is the language it reads now. JavaScript, TypeScript, and Rust are planned. The agent CLI, the MCP server, and editing on the canvas are not in this release. The docs live in a .vnoc folder and have to be committed if anyone else should see them. The post says the CLI is next. It does not say an agent can ask the tree today. This page did not open a call tree.
V-NOC · @tsyared · Nov 16, 2026Developer tools · 0:18
TheoLDR launch videoTheoLDRPlaysAlex's post says he built a tool that shortens Theo's videos by cutting only. Nothing generated, nothing moved around. The first film is the point of a 45 minute video in under 2 minutes. The repository is alexandreeberhardt/TheoLDR. The license on the code is MIT. The readme says it is for Theo's channel, t3.gg. Claude, through the Claude Code CLI, decides which sentences to keep. ffmpeg cuts in the silences between words. The agent never picks a timestamp. If YouTube has no captions yet, Whisper transcribes the video, and Whisper does not mark who is speaking. Ask the agent for more coverage and the readme says it glues in tiny fragments until the video turns into confetti. The post says nothing was generated. The picture is only cuts. The list of what to keep is Claude's. This page did not run a cut.
TheoLDR · @Alex62580010 · Nov 15, 2026Developer tools · 1:44
Adari launch videoAdariPlaysRahul's post says he built a Mac app on top of Herdr. The agent that is waiting on you goes to the top of the sidebar, instead of sitting in pane 7 until you notice. Free and open source. The site is adari.ai. The repository is adari-app/adari. The license is Apache 2.0. Adari 0.1.0 is for Apple silicon, signed and notarized, on macOS 13 or later. Chat is first class for Claude Code and Codex. A longer list is marked early: the app reads those sessions and can miss things. The code for herdr-core is in 0.1.0, and it has not been tested against a live herdr-core. The post says the app is on top of Herdr. It does not say herdr-core is untested. This page did not open the sidebar.
Adari · @RahulNK · Nov 14, 2026Developer tools · 0:16
anchormem launch videoanchormemPlaysChirag's post says coding agents remember the code, then the code changes and the memory goes stale. He built anchormem. Every memory is anchored to the function it describes, so it lights up the moment that code changes. Open source. The repository is ChiragPatankar/anchormem. The license is MIT. A memory points at a file, a symbol, or a line range. A check fingerprints the code with the whitespace removed. CHANGED, MOVED, and MISSING come from that fingerprint. FLAGGED means an agent called the memory outdated even though the code did not change. The dashboard with --serve updates on localhost as the code changes. Hooks that hand the memory over are written for Claude Code. Codex and Cursor are told, in AGENTS.md, to run the check themselves. Python symbols use Python's parser. JavaScript, TypeScript, and the other listed languages use a brace matcher. Line-range anchors are fragile. The check flags the memory. It does not rewrite it. The post says it lights up the moment the code changes. It does not say Codex has to run the check. This page did not save a memory.
anchormem · @Chiragp1605 · Nov 13, 2026Developer tools · 0:30
watch-my-ram launch videowatch-my-ramPlaysSi's post says parallel Claude Code sessions filled a 16 GB Mac, so he built watch-my-ram. It shows RAM and memory pressure above the prompt, lists sessions, leftover dev processes, and open ports, and warns when one session starts tests while another is already testing. It only reads and warns. It does not kill anything. The repository is nguyenphusi/watch-my-ram. The license is MIT. It is a Claude Code mod for macOS, and it needs Claude Code 2.1.287 or later. It reads vm_stat, sysctl, footprint, and lsof. The pull-request column needs the GitHub CLI, signed in. The readme says it never blocks or changes a command. A reply asked whether the warning stops the second test. It does not. This page did not close a process.
watch-my-ram · @nguyenphusi13 · Nov 13, 2026Developer tools · 0:08
usage-watch launch videousage-watchPlaysFadhil's post says he kept typing /usage and still hit the Claude limit mid-task, so he built usage-watch for Hermes Agent. Every few turns it posts one line with how much of the session and the week is left, and when it resets. No LLM calls. The change is pull request 154167 on NousResearch/hermes-agent. The pull request is still open. The default is one line every five completed turns, not every turn. It warns and does not switch the model. Nous credits are not covered. The API server has no place for a plugin to post, so those posts are refused. An idle timer can only refresh a line that already exists. The post says every few turns. It does not say the pull request is still open. This page did not install the plugin.
usage-watch · @Fadiru_ · Nov 12, 2026Developer tools · 0:33
Locust launch videoLocustPlaysColin's post says he built Locust so Claude Code, Codex, Cursor and Copilot work as one team on your own computer. Each teammate runs on the agent and model you pick. You approve what matters. Free, MIT, no account. The site is locust.lol. The repository is StopBanning/locust. The license is MIT. It also runs OpenCode, Muse Code, Antigravity and Kilo. Voice is on Windows for now. The docs say Locust can stop a run only where the agent asks first. Cursor's command line has no way to stop and ask. Kilo and Muse run without asking. Discord, Slack and Telegram, if you turn them on, can read what goes through them. There is no Locust account and no telemetry. You still pay for the plans those agents use. A native Rust window started this weekend is early, beside the Electron app. The post says you approve what matters. It does not say Cursor can be stopped. This page did not run a teammate.
Locust · @Scamtwt · Nov 9, 2026Developer tools · 0:36
T3 Sidebar launch videoT3 SidebarPlaysV1rgul's post says he rebuilt the T3 Code sidebar for Nous Research Hermes Desktop: Working, Input, and Done at a glance, pin, snooze, visible unsent drafts, token and API-equivalent cost, and Codex and Claude quotas pooled across accounts with a run-out forecast. The repository is V1rgul/hermes-plugin-t3sidebar. The license is MIT. It needs Hermes 0.23 or newer. Install is from Git, as an agent plugin and a desktop UI. The Limits page pools accounts in units weighted by plan. The behavior notes say the sidebar works in the selected gateway or profile, and does not silently add other accounts or servers. A snooze never stops or resumes the agent. It is a session list, not a replacement for Hermes or for T3 Code. The readme says it stores no credentials and sends no telemetry. The post says the quotas are pooled. It does not say a snooze pauses the run. This page did not install the plugin.
T3 Sidebar · @V1rgul · Nov 8, 2026Developer tools · 1:14
Learn with Claude launch videoLearn with ClaudePlaysBaba's post says Claude writes the code in minutes, and a week later he cannot explain it. So he built a free plugin that quizzes you on the changes while Claude codes. npm i -g learn-with-claude, then learn-with-claude install. The repository is babakhalid/learn-with-claude. The license is MIT. It is a plugin for Claude Code. The agent critiques the request in the chat. The readme says the session is never blocked waiting for a quiz answer. The post says it quizzes you while Claude codes. It does not say Claude keeps going if you never answer. This page did not take a quiz.
Learn with Claude · @BabaKhaliid · Nov 8, 2026Developer tools · 1:03
Coolant launch videoCoolantPlaysAndrew's post says macOS fan policy does not care if the CPU and GPU are at 100% and the battery overheats, so he built a fan control that should keep the Mac healthy long-term. Free and open source. The repository is arterialist/coolant. The license is MIT. It is for macOS 14 or newer on Apple Silicon. Reading the SMC needs no privileges. Writing a fan needs root. Releases are not notarized, because there is no Apple Developer ID, so macOS refuses to open a downloaded copy until the quarantine flag is cleared. Apple does not document the temperature keys, so no tool can prove its absolute numbers. The post says it should keep the machine healthy. The readme says the degrees cannot be proven. This page did not change a fan.
Coolant · @Arterialist · Nov 8, 2026Developer tools · 0:15
Lucidly launch videoLucidlyPlaysValerio's post says he built Lucidly to show when Claude Code is working, stuck, or needs you, at a glance. Free, open source. No installation needed, just npx @vale-dinap/lucidly, and it requires Node 22.18 or newer. The repository is vale-dinap/lucidly, on the master branch. The license is Apache 2.0, copyright 2026 Valerio Di Napoli. It opens in the browser on that folder's Claude Code session. The readme says it is a viewer, not a controller. It does not send commands to the agent, approve or block actions, or change how the agent behaves. The current target for development and testing is Claude Code. The post says no installation. npx still fetches the package, and the thing that needs you still has to be answered in Claude. This page did not open a session.
Lucidly · @valedinapoli · Nov 7, 2026Developer tools · 0:58
Notchi launch videoNotchiPlaysThe post says a MacBook notch was wasted space, so Notchi puts what's playing in any app into the notch, moves volume and brightness there, and takes notchi notify from the terminal. Free and open source. The repository is hardikdhingra150/Notchi. The app is GPL-3.0. The plugin SDK is MIT. Built-in themes and assets are CC BY 4.0. The readme says it is not affiliated with the other open-source project named notchi. Since macOS 15.4, only Apple-signed processes can read what is playing, so Notchi loads a tiny helper into /usr/bin/perl. The post does not name that helper, or the license split. This page did not hover a notch.
Notchi · @Snifferz_haadd · Nov 7, 2026Developer tools · 0:24
Lightpath launch videoLightpathPlaysAayush's post says Lightpath is an MCP server so an agent stops asking you to click and report back. It clicks the button itself and sees every line that ran, in the browser and on the server, plus the lines that did not. The repository is Unknown-333/Lightpath. The license is MIT. Node 22.12 or newer. Chromium only, Chrome or Edge, because coverage uses the DevTools Protocol. Node backends. A Python agent is a preview and is not joined into click traces yet. run_action does click, Enter, and Space, with fields filled first. Not hover, drag, file upload, dialogs, popups, or iframes. An answer is at most 16,000 characters. When a trace is bigger, the least-used files are left out. The post says every line. The readme leaves some out. This page did not click a button.
Lightpath · @Aayush__Paudel · Nov 7, 2026Developer tools · 0:14
Forge Desktop launch videoForge DesktopPlaysOmar's post says Forge Desktop is a Mac app that runs AI coding agents off a task board. Write a card in Linear, Basecamp, or GitHub Issues, press Start. Agents plan, build on a branch, review, test, and hand you a pull request. Free and open source. The repository is Donnadieu/forge-desktop. The license is Apache 2.0. You need a Mac with Apple Silicon and macOS 14 or newer. The dmg is signed and notarized. Keys live in the Mac Keychain and nowhere else. Commits go out under your name. Status is early, in daily use. The post says free. It does not say Apple Silicon, or that the agents still spend your keys. This page did not start a card.
Forge Desktop · @OverviewEffect6 · Nov 5, 2026Developer tools · 1:02
Bastion launch videoBastionPlaysSoumil's post says his AI firewall told him three things that were not true. One of them let delete_thing on prod-db walk past a deny rule. Bastion is the open-source gateway he built between agents and their MCP tools. An adversarial review of his own code turned those up. The repository is SoumilBhandari/Bastion. The license is Apache 2.0. It is local-first: one config, one command, no cloud, no database. The readme says a denied call is blocked before it reaches the upstream and is written to the audit log. The post says a delete walked past a deny. The readme does not describe that miss. This page did not send a tool call.
Bastion · @SoumilBhandari · Nov 5, 2026Developer tools · 0:29
Botburn launch videoBotburnPlaysThomas's post says his Grok bot hit 90% of the weekly limit by Tuesday, so he built Botburn, a free skill that shows which bot, routine, and day burned the usage. He says OnlyBots was replying to fans 700 times a week. The repository is apwn/botburn. The license is MIT, 2026. Version 0.1.0 is an early public kit: a skill, docs, and a synthetic sample. It is not a GitHub release until review, and it is not an official product. The pictured audit is labeled fiction, not a real account. The sample write-up uses about 700 runs a week as example language. Figures are shares of all Grok Bot usage in the window unless the report says it renormalized. The post reads like a measurement. The repo's 700 is the sample. This page did not open an account.
Botburn · @ThomasKanze · Nov 5, 2026Developer tools · 0:28
Network X-ray launch videoNetwork X-rayPlaysSam's post says the same URL works on his machine and fails on yours, so he built Network X-ray for Hermes Agent. Capture one URL on two machines, see where DNS, TCP, TLS, or HTTP first differ, then export the evidence. Built for a Nous Research plugin hackathon. The repository is rusty4444/hermes-network-xray. The license is MIT. It is a local-first plugin. No service account, API key, hosted backend, or model call is required. It collects bounded observations and offers possible next tests. It does not invent a root cause. Two different healthy CDN addresses are not a network failure. The post says it finds the first difference. It does not say a healthy split is not a bug. This page did not capture a URL.
Network X-ray · @mrru5s3ll · Nov 5, 2026Developer tools · 1:00
Shadowfenced launch videoShadowfencedPlaysWelly's post says people paste anything into AI, even API keys, so he built something to catch it first. The repository is wellytg/meta-hackathon-shadowfenced. The license is Apache 2.0. It is a hackathon build, and the security note says use it at your own risk. Gemini, Claude, and meta.ai are covered. Grok is added but does not yet stop the send. If the local app cannot check, nothing is sent. 27.5% of messages still need a person. Harmless messages with sensitive words are escalated. The local judge is Muse Glimmer 30B, and the readme says it does not fit for training on a 12 GB GPU. The post says it catches the key. It does not say Grok still lets the send through. This page did not paste a key.
Shadowfenced · @wellytg · Nov 5, 2026Developer tools · 1:57
Checkride launch videoCheckridePlaysPreetham's post asks whether an AI agent can delete files without asking, then says he built Checkride to scan MCP servers and agent tools, trace issues to the code, and suggest fixes. pip install checkride. The repository is PreethamNoelP/checkride. The license is Apache 2.0. Python 3.11 or later. No dependencies, no account, no network, and it never runs the code it reads. A check placed after the action does not count. A tool argument named confirm does not count, because the model sets it. MCP elicitation does. Every finding carries a concrete fix. The post says it scans and suggests. It does not say the scanner never executes the agent. This page did not run it.
Checkride · @preetham_noel_p · Nov 2, 2026Developer tools · 0:58
Colorizer launch videoColorizerPlaysviol3's post says a $100 Unity asset for recoloring low-poly models, so he built his own with Claude in 15 minutes and made it open source, free forever. The repository is viol3/colorizer. The license is MIT. It recolors models that use a color atlas. Unity 2021.3 or newer. The package is editor-only and adds nothing to builds. Status is early, v0.0.1, and the readme says you may hit bugs. The post does not say it only works on atlas models, or that it never ships in the build. This page did not open Unity.
Colorizer · @viol3dev · Nov 1, 2026Developer tools · 0:14
India Trading Skills launch videoIndia Trading SkillsPlaysShubham's post says 23 free, open-source Claude Code skills for NSE, BSE, NIFTY F&O, Indian taxation, and trading costs. Ten calculators. 200 tests. The repository is Shubhamkumartrichy/claude-trading-skills-india. The license is MIT. The readme says a delivery round trip costs about 0.23% to break even, not the 0.1% most guides quote. After the April 2026 STT hike, a NIFTY futures lot needs about 14 points just to cover charges. The disclaimer says these are educational tools, not investment, trading, or tax advice, and not from a SEBI-registered adviser. It cites SEBI's finding that about 91% of individual F&O traders lost money in FY25. Live broker data is optional, through an MCP login. The post does not print the 91% line, or the not-advice line. This page did not run a calculator.
India Trading Skills · @TheTortAndHare · Nov 1, 2026Developer tools · 0:26
NestJS Benchmark launch videoNestJS BenchmarkPlaysSajan's post says a live benchmark of NestJS HTTP adapters: Express, Fastify, and Node, one route at a time. On his machine Nest plus Node led on all three routes. Yours may differ. The repository is sajanv88/nestjs-benchmark-workspace. The license is MIT. The readme says the numbers are unofficial measurements from one machine. They are not comparable to another computer, or to the PR's M1 Pro results. The order held in every run here. The size of the lead did not: with the same 100 connections split across two load generators, Node's lead over Fastify on GET / fell from +54% to +6%. A single generator may become the limit before the faster adapters do. The post says the order on one machine. It does not say the ratios move that much. This page did not run the chart.
NestJS Benchmark · @sajanvtech · Oct 31, 2026Developer tools · 2:42
App Store Localization launch videoApp Store LocalizationPlaysGiacomo's post says he built a Mac app so he could stop living in App Store Connect. One window for every app: translate listings into every App Store language, price 175 territories for purchasing power, reply to reviews, preflight, and submit. A reply the same minute says it is free to look around and $79 once to apply changes, and that the API key stays in the Keychain. Calls go from the Mac to Apple. The site says the same split. The free download reads and plans. A license applies the plans. macOS 26 or later. Translation is from en-US by your own AI. Prices come from World Bank purchasing-power data, snap to Apple's ladder, and never go below 25% of Apple's price by default. A territory at 0.60 or richer keeps Apple's price. The preflight picture on the site is an illustration of a made-up weather app, not a screenshot. The post does not say macOS 26, or that the translation is your model. This page did not submit a version.
App Store Localization · @BigBalli · Oct 30, 2026Developer tools · 2:44
Agent Retry Rodeo launch videoAgent Retry RodeoPlaysRP's post says an agent reported done and posted the same comment twice, so he built a test rig for retries that go wrong. Bring your own agent, or add a case. The repository is RishabhPande/agent-retry-rodeo. The license is MIT. The readme calls it a local test range. The quick start is six scripted runs, no login, no API keys, synthetic data only, Python 3.10 or later. One case is a lost acknowledgement that duplicates a write. Another finishes halfway. It is not a sandbox. It does not block network access or files, and it cannot stop an infinite loop. No model connector is included. The post's duplicate comment is a scripted fault, not a live agent on a real ticket. This page did not run the suite.
Agent Retry Rodeo · @Rishpande · Oct 30, 2026Developer tools · 0:31
Revit Model Compare launch videoRevit Model ComparePlaysMonica's post says a pyRevit tool compares two MEP models by geometry: coverage, missing, extra, moved, and wrong-size parts, plus screenshots. Open source. The reply links kidge-io/revit-model-compare. The readme says MIT, free, built by KIDGE Ltd. You need Revit, developed on 2024, and pyRevit. Both models must share the same internal origin. The tool does not align coordinates. Tolerances are fixed for now, 5 mm, 50 mm, and 1 m. Issue Shots should run on copies. It adds views and does not edit elements. Paid work sits beside the tool: KIDGE will run the comparison and deliver a report, or help a team adopt it. The post does not say the models have to share an origin, or that the report is a separate service. This page did not open Revit.
Revit Model Compare · @ru_lin59968 · Oct 29, 2026Developer tools · 0:19
WebView Native UI Engine launch videoWebView Native UI EnginePlaysMasahiko's post says native ads scroll with web content, and web modals sit over native UI. He built an engine for Capacitor and Cordova. The repository is webview-native-ui/engine. The license is MIT. The readme says it is a library for plugin developers, on iOS and Android. It projects native appearance into the web layer. It does not reorder the actual native view inside the DOM. A native view above a WebView does not automatically join DOM layout, clicks, or live video. If a suitable image is missing, the integration uses its web fallback. Platform limits still apply, including Liquid Glass. The 23-second demo in the repo was recorded on the iOS Simulator. The post does not say the scroll is a projection, or that the film is a simulator. This page did not install a plugin.
WebView Native UI · @rdlabo · Oct 29, 2026Developer tools · 0:23
MotionSites launch videoMotionSitesPlaysViktor's post says he built the landing pages with Claude and Kimi, using his MotionSites skill. The skill is free and open source. The repository is Motionsites-AI/claude-skill. The license file splits in two. Instructions, references, and scripts are MIT, except the examples and media folders. Those prompts and the videos they point at stay copyright MotionSites. You may use them to build sites, including for clients. You may not resell or redistribute them as a prompt or media collection. The readme says Claude reads two or three of 40 example prompts, and picks from 50-plus hosted hero videos. Installing it in Claude needs code execution turned on. The post calls the skill open source. It does not say the media shelf is outside the MIT grant. This page did not build a page.
MotionSites · @viktoroddy · Oct 28, 2026Developer tools · 0:12
Capturra launch videoCapturraPlaysJerry's post says good recorders want a subscription, and free ones leave a shaky cursor and no zoom. He built Capturra in Rust and Tauri: auto-zoom, a smooth cursor, free and open source. The readme says MIT, and macOS 13 or later. It records the screen with the cursor off, logs mouse and keyboard timing, then redraws a smoothed cursor and zooms in the editor. Export is MP4 or GIF. macOS asks for Screen Recording, Accessibility, and Input Monitoring. The status line says it is early. Some features are checked by unit tests and short runs. The post does not say the cursor in the video is a redraw, or name the three permissions. This page did not record a screen.
Capturra · @iamjerrybarry · Oct 28, 2026Developer tools · 1:02
desk launch videodeskPlaysZac's post says he asked Claude to tidy a desktop, and it took the mouse. desk is an open-source CLI that gives Claude Code eyes and hands on a Mac, and the film is at 20 times speed. The repository is zacfink/screen-intelligence. The license is MIT. macOS only. Claude plans. A small model called Jev, from TypeSafe, does the looking and the clicking. Jev needs TYPESAFE_API_KEY, its own Python 3.12 venv, and the readme says Images and Box are TypeSafe beta features. After five idle minutes the launcher waits three seconds, then takes the mouse. A find under 0.6 confidence is refused. The setup note says the paths are for his machine. The 20 times line is in the readme, for the drag demo. The post does not mention the key, or the beta. This page did not move a window.
desk · @zacfinkelstein · Oct 27, 2026Developer tools · 0:17
Quillwork launch videoQuillworkPlaysMarc's post says Quillwork lets any AI agent edit a real Word file without breaking it. You get the same file back. Only what you asked for changes. Everything else stays byte for byte. The repository is gardan4/quillwork. The license is Apache 2.0. Status is alpha, 0.0.x, and the API can still change. Edits are written as Word tracked changes. A person accepts or rejects them in Word. Untouched parts stay byte for byte, which matches the post. The limitations say docx and docm only. Password-protected files are refused. There is no doc, odt, xlsx, or pptx. Footnotes, hyperlinks, field insertion, and image edits are not implemented. Images can be listed. The post does not say alpha, or that a hyperlink is still out. This page did not edit a contract.
Quillwork · @MarcMeijers01 · Oct 27, 2026Developer tools · 0:14
Ebb launch videoEbbPlaysLucas's post says his T3 Code setup has a provider called Pool, and CLIProxyAPI spreads each request across accounts by what each one has left. He built Ebb, a menu bar app, because the missing piece was seeing that remainder. A reply says it is native SwiftUI and open source, keeps the 5-hour and weekly numbers separate, retunes the proxy weights, and covers Claude, Codex, Devin, Kimi, and more. The readme says MIT, and macOS 26 or later. The 5-hour and weekly figures are never added across providers. xAI, Vertex, a Gemini API key, AI Studio, and OpenAI-compatible logins are health only. They are never counted in a pool. Meta Muse can show "Limits not available" when the proxy cannot send the device-code token. The post does not name the macOS floor, or the accounts that stay out of the pool. This page did not open the menu bar.
Ebb · @nicecache · Oct 27, 2026Developer tools · 0:16
CrychicUI launch videoCrychicUIPlaysThe post says CrychicUI is fully open source, with 1,500 MyGo components. The website is not ready for a phone yet. Custom settings and variants are later. The site is crychicui.zacharyzhang.com. The readme says it is a Go library in 38 packages, drawn on the GPU by MyGo, with no HTML, JavaScript, or WebView. Install is go get, and it asks for Go 1.27.1 and MyGo v0.5.5. Every component animates. Interactive ones play a synthesized sound, on by default. The license file is MIT. The readme does not mention the license. The post does not mention the MyGo version, or the sound. This page did not compile a component.
CrychicUI · @ZacharyZhang · Oct 27, 2026Developer tools · 4:00
Orplex launch videoOrplexPlaysThe post says Orplex is an open-source coding agent on your own machine. Bring your own key, across 12 providers. Parallel agents in separate git worktrees. Built to run on an 8 GB laptop. The site is orplex-code.vercel.app. The readme says MIT. It is an Electron app. Worktree runs wait for an explicit merge or discard. Bring-your-own-key is any OpenAI-compatible endpoint plus your key. Conversation history is stored in Supabase Postgres. Sign-in is Google or GitHub through NextAuth. The readme does not list 12 providers, or an 8 GB floor. The post does not say the chat history leaves the laptop. This page did not start a run.
Orplex · @WhiteD92151 · Oct 27, 2026Developer tools · 1:12
Dry Run Lab launch videoDry Run LabPlaysPaul's post says Dry Run Lab is a free, open-source network simulator in public beta. Load your real router and switch configs, try the change, and see exactly what breaks before the window. The readme says Apache 2.0. It is a beta, for a small vendor-style network. You draw the devices, paste a config, and confirm the cables it inferred. Show commands are computed on this machine by Batfish. Nothing is uploaded, and it never opens a session to a real device. Results are computed, not proof that the live network, or another software release, will match. It is not affiliated with Cisco. The post says exactly what breaks. The readme says the result is not that. This page did not load a config.
Dry Run Lab · @pawel7 · Oct 26, 2026Developer tools · 1:16
OrgExplorer launch videoOrgExplorerPlaysRahul's post says OrgExplorer v1.0.0 is an open-source look at a GitHub organization's activity, built during GSoC 2026 with AOSSIE. The readme says GPLv3. It runs in the browser on GitHub's REST API. There is no backend. Signed out, the diagram in the readme caps the hour at 60 requests. A personal access token raises that. Deleted repositories and complete historical totals are not in the numbers. The site is orgexplorer.aossie.org. The post does not mention the hourly cap, or the history that is missing. This page did not open an organization.
OrgExplorer · @_Rahul_vyas · Oct 25, 2026Developer tools · 1:01
PandaCode launch videoPandaCodePlaysThe post says PandaCode Connect links machines and runs a task on another one. Live output every 100 ms, nothing polls. End-to-end encrypted. Fully free, with a 10-device limit. Open source, no telemetry. A mobile app is later. The readme says the desktop app is free, with no account, no license key, and no telemetry, until you turn remote access on. Connect is optional. It wants a GitHub sign-in, then a relay that forwards the devices' encrypted messages. Nothing is sent before that setting. The license is MIT plus one line: a public or commercial fork must say it is a fork of PandaCode and link the repo. The readme does not print a 10-device cap. The post does not say Connect starts with a GitHub login. This page did not pair a machine.
PandaCode · @aicodeking · Oct 25, 2026Developer tools · 0:15
whyfull launch videowhyfullPlaysBehnam's post says a Mac disk fills up with node_modules, Xcode simulators, Docker, npm cache, Ollama models, and agent caches. whyfull finds them, explains them, and clears them safely. The readme says MIT. You need macOS 15 or later. It changes nothing until you review and confirm. Caches that apps rebuild are deleted. Your own files go to the Trash. A folder it does not know stays "Not recognized," and it never sends a report on its own. The post does not say caches are gone for good, or that the floor is macOS 15. This page did not delete a cache.
whyfull · @baz7c8 · Oct 24, 2026Developer tools · 0:44
claude-grok-bridge launch videoclaude-grok-bridgePlaysCosimo's post says he was copying between Claude Code and Grok Bot, so he built a bridge. Type "ask my agent if we ship today" in the terminal, and the answer comes back. Open source. The readme says MIT. It is for agents you build in Grok Bot, SpaceXAI's desktop app. Not the Grok chat, and no xAI API key. No servers and no keys to paste. In the usual setup the agent installs both sides, writes the grok-bridge skill on your Mac through Grok Bot's local exec, sends itself a test, and answers PONG. The readme also says the agent can steer a session at 2 a.m. with the laptop closed. The post does not say the chat product is out, or that the skill is written on a Mac. This page did not send a message.
claude-grok-bridge · @CosimoCosimo95 · Oct 22, 2026Developer tools · 0:20
Ratatoskr launch videoRatatoskrPlaysOana's post says keeping Codex and Claude Code skills in sync is the hard part, across machines and operating systems. She built Ratatoskr. Free and open source. The readme says MIT, and it is pre-1.0. One Git source renders a different copy for each machine, OS, and agent. Identical copies are not the point. A new machine writes to a staging folder, so the real agent folders stay untouched until you switch to live. A sync refuses a dirty, diverged, or wrong-branch checkout. The downloads are macOS and Windows. Apple notarization and Microsoft publisher signing are not configured, so the first open can warn. The post does not say the live folders stay untouched, or that Linux is not a download. This page did not sync a skill.
Ratatoskr · @OanaGoge · Oct 22, 2026Developer tools · 1:11
termkit launch videotermkitPlaysTiru's post says a Sessions pane colors every Claude Code window. Red needs him, yellow is working, green is done. Click to jump to that tab. Old chats are searchable by what you remember. The readme says termkit is a set of Claude Code plugins, MIT. The hub is one plugin. The colored pane is a second plugin, window-board, installed with its own command. A click on an open window switches to it. A closed session is searchable, and a click resumes it in a new tab. The same repo also ships a Chrome-dino runner that plays while Claude works. The post does not say the pane is a separate install, or that a closed chat opens somewhere new. This page did not install the plugin.
termkit · @tirupathipodugu · Oct 21, 2026Developer tools · 0:45
QAmcp launch videoQAmcpPlaysAbhi's post says one command tests an MCP server in 12 seconds. Tools, bad inputs, prompt injections, and leaked secrets. Free and open source. The readme says MIT, Node.js 20 or newer. It connects like an AI app and sends nothing except to the server you name. The demo starts a server they broke on purpose and takes about 30 seconds. It flags a hidden instruction that tells the AI to read an SSH key, and an error that shows a database password. Tools that change data are not run, ever. A read-only tool with no safety label is skipped in safe mode and runs with --owner. The post's 12 seconds is the length of the film. The readme's demo is about 30. This page did not test a server.
QAmcp · @AbhiRam2371650 · Oct 21, 2026Developer tools · 0:12
Full-Duplex Code launch videoFull-Duplex CodePlaysSebastian's post says he can boss Claude Code, Codex, or Pi around out loud, and the voice part is open source. The readme says a browser companion, powered by GPT Live 1, follows the agent's messages, edits, and tool results, and can hand a request over mid-turn. The tested setup is macOS, Node.js 22.18 or newer, Chrome, a microphone, and the agent already signed in. It also wants an OpenAI key with access to GPT Live 1. The license is PolyForm Noncommercial 1.0.0. Free for noncommercial use. Commercial use is not granted. The post says open source. The license does not grant a company the right to ship it. This page did not talk to an agent.
Full-Duplex Code · @Sebasti54919704 · Oct 21, 2026Developer tools · 0:22
GitSense launch videoGitSensePlaysNitin's post says GitSense answers "how does auth work?" on a pasted GitHub repo, with code citations. Java, Spring Boot, pgvector, and RAG. He links lookwhatibuilt.live and the repository. The readme titles the same repo DevPilot. It indexes repositories and chats with them. You need Docker, Java 21, Node.js 20 or newer, an OpenAI key, and a GitHub OAuth app. Chat is GPT-4o-mini. Embeddings are text-embedding-3-small. Sign-in is GitHub OAuth, then a repository sync. The post says paste a repo. The readme starts with a login and two keys. The readme does not print a license. This page did not index a repository.
GitSense · @nitinmehra_09 · Oct 21, 2026Developer tools · 0:30
UPIlerify launch videoUPIlerifyPlaysHarsha's post says UPIlerify lets you get paid in UPI with zero gateway fees. No Razorpay, no 2% cut, no waiting, and you know the moment someone pays. He says he built it in August. The readme says MIT, self-hosted, for Next.js and Node. The customer pays a UPI QR. The bank emails a credit alert to Gmail. The app reads that mail over IMAP, matches the amount and a 12-digit UTR, and sends a signed webhook. It recognizes 16 banks and apps. You need Node.js 20.9 or newer, a Gmail app password, a bank that emails on credits, and a host with a disk. The readme says a large or regulated business should use a licensed gateway, and important payments should still be checked on the statement. The post does not mention Gmail, or the 16 banks. This page did not run a checkout.
UPIlerify · @harsha_varma02 · Oct 20, 2026Developer tools · 0:09
unwreck launch videounwreckPlaysAyush's post says an agent can delete files git does not track, so he built unwreck. Undo, and the files are back. Open source, one Rust binary. The readme says hooks snapshot the working tree before Claude, Codex, or Gemini run, including untracked files. Snapshots sit outside the repository. It does not write inside .git. It does not restore chat history. A file over 16 MiB is recorded as too large and comes back as MISSING. It does not replace version control. Garbage collection drops snapshots older than 30 days. Prebuilt binaries cover Linux musl, macOS, and Windows. The post says everything is back. The readme does not. This page did not run a restore.
unwreck · @ayushdoesdev · Oct 20, 2026Developer tools · 0:29
AI Limit Tracker launch videoAI Limit TrackerPlaysBahadır's post puts Claude, Codex, and Cursor limits in the Mac menu bar. A later reply says it is free and Mac only, pings at 75% and 90%, holds more than one account, and a card shows the last 13 weeks. The readme says MIT, macOS 14 or later, Apple silicon and Intel. You sign in once in your own browser. Cursor is read from the Cursor app. Tokens stay on the Mac. There is no server of his, but the app talks to Anthropic, OpenAI, Cursor, and GitHub for updates. It reads undocumented endpoints, so it can break when they change. The app is not notarized, and not affiliated with those companies. The post does not say undocumented. This page did not install it.
AI Limit Tracker · @nuroglubahadir · Oct 19, 2026Developer tools · 0:40
Readback launch videoReadbackPlaysNikhil's post says Readback shows what Claude Code planned against what it changed, as a picture. Open source. The site says it is in development. It is a Claude Code plugin. It reads the git diff and a map at .readback/map.json. Init drafts that map from folders, and you rename the tiles. The readme says MIT, Python 3.8 or newer, and git. A maybe tile is asked once and does not raise a flag by itself. Unplanned and not-touched flags need a real plan, or a Readback-Plan line on the pull request. The log stores tile ids and counts, not the code, and it is gitignored. The post says every angle: user, product, team. Those angles are the map, and the map starts as folder names. This page did not install the plugin.
Readback · @nikhil_kadalge · Oct 19, 2026Developer tools · 0:44
Wispstage launch videoWispstagePlaysYaleson's post says UI automation steals the mouse and pulls the test window forward. Wispstage is a background end-to-end plugin for Claude. It runs a SwiftUI app behind the desktop, using macOS Accessibility, so the person keeps the machine. The readme says MIT, macOS 14 or later, Swift 6, and both Accessibility and Screen Recording. Background mode cannot test focus rings, a first click that only activates the window, real hover, or a global hotkey. The post says it checks layouts and motion. The readme says motion recording assumes a flat background, no shadows, and only spring curves. The post does not name Screen Recording. This page did not run a suite.
Wispstage · @yalesony · Oct 18, 2026Developer tools · 0:48
LinPad OS launch videoLinPad OSPlaysVali's post says LinPad OS is Linux for iPad. VS Code, a terminal, git, Python, Node.js, and Claude Code listed in the Store. Free, open source, and pre-release. The site says Alpine, GPLv3, a fork of iSH, on an iPad with M1 or newer and iPadOS 17 or later. It is sideloaded with your own Apple ID. The FAQ says it will not go on the App Store, because it generates native code and downloads executables. A free Apple ID signature expires after 7 days. Without StikDebug the programs run on an interpreter and are several times slower. The app has no analytics. Docker, Steam, and x86 games are not supported. The post does not say sideload, or the 7-day expiry. This page did not install it.
LinPad · @AmbsdOP · Oct 18, 2026Developer tools · 0:22
ChatVault launch videoChatVaultPlaysSunny's post says ChatVault keeps Claude Code, Codex, and Cursor chats in one place, across accounts. Search, local backups, free, open source, Windows, and portable. The readme says the same three tools, MIT, Windows 10 or 11, 64-bit, and a portable exe beside the installer. It reads the chat files, does not upload them, and needs no account. The only network request is to GitHub, to check for a new version. The screenshots are labeled sample data, not real chats. The post does not mention that version check. This page did not open a chat archive.
ChatVault · @sanaullahrais · Oct 17, 2026Developer tools · 1:08
Hermes UI+ launch videoHermes UI+PlaysRobby's post says he built an app for Hermes Agent so a checklist, a choice, a comparison, or a command approval is a tap, and the tap goes back to Hermes. Open source. It runs on your own server. The page calls it Hermes Interactive UI+, unofficial, alpha 0.1, MIT. It needs Hermes Agent v0.21 or newer. You run a bridge next to it. The app talks to that bridge, not to a service of theirs. No accounts and no telemetry. Android, web, macOS, and Linux. Android in 0.1 has no notification while the app is closed. The web app can push. The post does not say unofficial or alpha. This page did not pair a bridge.
Hermes UI · @CamiAi__ · Oct 15, 2026Developer tools · 0:53
Corply launch videoCorplyPlaysCorply's post says one prompt to an agent can incorporate a Delaware C-Corp, open a bank account, and accept credit card payments. A reply says less than 30 minutes. The page lists initial formation at $399, a Delaware filing plus formation service at $329, a managed 83(b) at $70 per founder, a free EIN, and a free corporate banking setup. A $69 monthly plan is marked required with formation. A registered agent is $7 and a virtual mailbox is $39 on that ongoing list. Delaware government fees are separate. A human reviews every state filing. The page says Corply is software from 0Lumen Labs Corp, not a law firm and not a bank. It names invoices and payments, not the words credit card. It does not say 30 minutes. Claude Code, Codex, and Cursor are the agents it names. This page did not file a company.
0Lumen Labs · @CorplyDev · Oct 15, 2026Developer tools · 0:22
OwlLayer launch videoOwlLayerPlaysn14o's post introduces OwlLayer, an open-source SDK that lets a model call the buttons, links and forms already on the page. A reply says the agent sees what the user sees, and that there is no limitation. The docs say the model never takes control of the DOM. Tools register for the current screen and unregister when it goes away. High-risk steps, such as a payment or clearing the cart, wait for the person. The readme shows the same wait before an order is confirmed. React, Vue, Svelte, Angular, vanilla, Shopify and WooCommerce. MIT. The package is @owllayer/core. The docs nav says changelog v0.5.0. No price.
OwlLayer · @nightwalkeryao · Oct 13, 2026Developer tools · 2:40
Helpdock launch videoHelpdockPlaysOday's post says Helpdock is an open-source help desk you run yourself. Email, WhatsApp, Instagram, Telegram and live chat in one inbox. SLAs, rules, and AI that answers from your docs or hands off. English and Arabic. No per-agent seats. AGPL-3.0, v0.5.0. The v0.5.0 release exists. The readme says pre-1.0, and that email, the chat widget, SLAs, rules and the help center, in English and Arabic, have shipped. The same status block says Telegram, AI and the public API are still being built. WhatsApp, Facebook and Instagram are listed as in development, milestone M10. Install is Docker Compose on a Linux server with 2 vCPU, 4 GB of memory, and two DNS names. The readme prints no seat price.
Helpdock · @Oday_Hunter · Oct 13, 2026Developer tools · 1:27
Inkpad launch videoInkpadPlaysRayen's post introduces React Native Inkpad, a native ink canvas. It draws with Android's Ink API and Apple's PencilKit, without a WebView. Pen, marker, highlighter and eraser. Coming from react-native-signature-canvas, change one import. The readme says androidx.ink and PencilKit, MIT, React Native 0.80 or later with the New Architecture. iOS 15.1 or later. Android 7.0 or later. Expo Go is not supported. The import change is to react-native-inkpad/signature-canvas. The same section says some props are ignored, the footer looks different, the eraser removes whole strokes rather than pixels, and background images wait for 0.3. Web is still on the roadmap. No price.
Inkpad · @WeskerTN · Oct 11, 2026Developer tools · 0:58
Context Siege launch videoContext SiegePlaysroberottt's post says he built Context Siege while waiting on Claude: a low-poly shooter you upgrade only with the tokens Claude Code burns. Open source. The readme says you start with 20K TK, and the rest comes from transcripts under ~/.claude since you started playing. Input and output count once, cache writes at 0.25, cache reads at 0.02. In each 10 minutes the first 100K TK count in full. Above that, growth is the square root. If Claude produced tokens in the last 90 seconds, the ship overclocks. Node 18+, no npm dependencies, localhost:4321. A badge says it runs 100% local. The requirements also say Three.js and the font load from a CDN, so it needs a network. The GitHub repository has no license file. Windows session detection is untested. This page did not play it.
Context Siege · @roberottt · Oct 10, 2026Developer tools · 0:15
RoadRaven launch videoRoadRavenPlaysMateusz's post says he built RoadRaven, a local roadmap tree Claude Code updates while it works, with an MCP server, skills, and a Claude Code mod. The readme says alpha v0.8.9. The plan is a roadmap.json in the repo. The MCP server exposes 21 tools. It binds to 127.0.0.1. Windows and Linux install today. macOS is next. Two skills, roadraven:orchestrate and roadraven:work-node, come with the plugin, and the plugin needs Node.js 24 or later. The wizard and npx paths give the tools only. The license is FSL-1.1-MIT: free to use, you cannot sell or host it as a competing product, and each release becomes MIT two years later. The readme says that is not OSI open source. No price. The Windows build is unsigned.
RoadRaven · @Shuffzord · Oct 10, 2026Developer tools · 0:15
rdyrct launch videordyrctPlaysAndrea Bruno's post introduces rdyrct 1.0.0, a link shortener and QR generator that runs on your own Cloudflare account. Custom domains, teams, click analytics with no IP stored. MIT. The site is also a hosted product. Free is $0: 30 links, 3 members, 7 days of analytics, a QR on every link, and an API and MCP. No credit card. A link made with no account lasts 24 hours. Hobby is $4 a month: 500 links, 5 members, 1 custom domain, 30 days, a QR with a logo. Pro is $9 a month: 3,000 links, 3 organizations, 25 members, 5 domains, a year of analytics. Only the owner pays. The readme says QR codes are a paid feature, and that custom domains are Pro. The pricing page gives the free plan a plain QR, and Hobby one custom domain. The one-click deploy is not a complete install.
rdyrct · @baronunread · Oct 10, 2026Developer tools · 0:16
norma launch videonormaPlaysJuanca's post says he built norma: one way of working for AI-built projects, the same whichever agent is driving, enforced by git. Open source. The thread names the loop as orient, spec, plan, implement, verify, review, document, deliver, and three stops: the spec, the task list, and delivery. The readme's loop is orient, enrich, branch, plan, implement, verify, review, document, close, deliver, and the stops are after enriching, after planning, and before delivery. Review is a fresh session. The pre-commit hook refuses a commit without a fresh stamp. A skipped test, an eslint-disable, or fewer assertions is flagged for a person. POSIX sh, no dependencies, MIT, version 0.21.0. Install is brew install jcarlosrodicio/tap/norma. No price.
norma · @JuancaRodicio · Oct 9, 2026Developer tools · 0:32
Clawd View launch videoClawd ViewPlaysAhnaf Iqbal's post introduces Clawd View as a free, open-source plugin for Claude Code. The post says it is a live checklist in plain words, it pauses for your OK, it does not hide when it is stuck, and it tracks limits, context, and cost. The post says it works in the terminal and the Claude desktop app. The readme says Clawd View's own code is MIT, and that a stuck step says what happened instead of pretending it worked. The post's install lines are /plugin marketplace add thedepegger/clawd-view and /plugin install clawd-view@clawd-view. The readme also gives /plugin install clawd-view --marketplace thedepegger/clawd-view. No price on the repo.
Clawd View · @thedepegger · Oct 9, 2026Developer tools · 2:37
Mission Control launch videoMission ControlPlaysBurak Vural's post says he built mission-control for the Windows Claude Code desktop app. The post says the panel shows what Claude is doing and what it is costing, session and weekly limits, without leaving the prompt. The readme calls it a status panel above the prompt, MIT. It was built and tried in the desktop app on Windows. In a terminal the readme says the panel falls back to plain text, without icons, animation, or ticking clocks. No price.
Mission Control · @b0rv0r · Oct 9, 2026Developer tools · 1:02
Command Code Cloud launch videoCommand Code CloudPlaysThe post says you run Command Code in the cloud on a GitHub repo, close the computer, and it can push changes and open a draft pull request. The changelog for October 7 calls cloud a beta in v1.77.0. The docs say the agent works on its own branch and never pushes to the base branch. A question, or a turn with no changes, ends without a pull request. A stopped session costs nothing. During the beta, the docs bill cloud time at $0.10 an hour, on top of model usage. The pricing table says cloud access is on every plan except Go, at $0.10 an hour. The docs say the GOAT plan and above. The pricing page says GOAT is $10 a month.
Command Code · @CommandCodeAI · Oct 7, 2026Developer tools · 0:34