The best gifts, researched and ranked by AI.

Officially licensed KPop Demon Hunters toys, dolls and merch for kids, teens and grown-up fans, picked from hands-on reviews by Brickset, The Toy Box Philosopher and others. Ordered by how each did in hands-on reviews, then by how many independent editors recommend it, then by fit for kids and price. Check the ages: Mattel rates its Rumi dolls and figure 4 and up, the light stick, Nerf sword and Furblet are 6 and up, Little People Collector is 8 and up despite the toddler brand, and the LEGO set is 9 and up. Which Rumi doll? The singing Golden doll plays three clips of the song but has five joints. The How It's Done doll lists for about $6 less, has 11 joints and her sword, and The Toy Box Philosopher liked it better. For posing, the 6-inch figure has 21 joints and swappable faces. For a 6-year-old fan, the light stick and the Derpy Furblet come to about $30. For game night, UNO is the easier deck (7 and up) and Monopoly Deal the trickier one (8 and up), about $17 for both. Teens and adult fans get the most from the LEGO set, the art book or, for crafters, the Woobles kit. Prices checked October 5, 2026.

The best Figma MCP servers for turning designs into code with Claude Code, Cursor, Codex or VS Code, and for letting an agent edit the canvas, ranked by independent tests and recommendations. Start with Figma's own server if your AI client is on Figma's approved list and you have a Dev or Full seat, which costs $12 a month billed annually for Dev. On the free Starter plan, or a View or Collab seat, Figma allows only 20 calls a month, and Framelink, which goes through Figma's API, runs into a similar limit of 20 file reads. The plugin servers, Figma Console MCP, TalkToFigma and figwright, work through a plugin in Figma's desktop app instead, so those limits don't apply. Read or write: Framelink only reads. Figma's server writes to the canvas from its remote version, free during the beta and a paid, usage-based feature after that. The plugin servers read and write, so keep your client's tool approvals on: a write changes the live file. Ordered by how many independent tests and recommendations each server has, in blog posts, videos and Hacker News threads. Stars and downloads only break ties. Prices checked October 17, 2026.

The best extensions for the Pi coding agent (some people call them plugins), picked from the setups developers publish and the write-ups that tested them, not from download counts alone. Pi is deliberately bare: no sub-agents, no plan mode, no permission prompts. You add what you miss with one command, pi install npm:<package>. If you are starting out, three cover most of the gap with Claude Code: pi-subagents for delegation, pi-web-access for search, and rpiv-ask-user-question so the model asks before it guesses. Add Plannotator if you want plan mode, and pi-permission-system if you don't want the agent free to run anything. Two things have changed that older guides miss. Pi has had MCP support built in since version 0.99, so pi-mcp-adapter is now an upgrade and no longer a requirement. And several popular names have lookalikes: npm has at least four other packages named pi-mcp-adapter under other people's scopes, and more than one /btw extension, so install by the exact name shown here. Extensions run with the same access to your machine that Pi has. Pi's catalog lists more than 5,000 packages, so read the source of anything that touches your shell or credentials, and pin versions you depend on. Ordered by how many independent users and reviewers recommend each one (Kay Rhodes's setup guide, Composio's tested top ten, developers on Hacker News and Pi's own GitHub discussions), then by npm downloads and GitHub stars. Downloads count automated installs too, so treat them as a rough signal. Numbers checked October 18, 2026.

The best OpenCode plugins, and which ones work on OpenCode 2, ranked by how many developers outside each project say they use them, in blog posts, hands-on reviews and Hacker News threads. If your sessions run out of context, start with Dynamic Context Pruning. If you want the agent to plan before it codes, get Superpowers, or Plannotator to review plans in a browser. For a team of subagents, oh-my-opencode-slim is the one that runs on the current version. For memory that stays on your machine, opencode-mem. The thing no plugin page says loudly: OpenCode 2.0 came out in September 2026 with a new plugin system, and plugins written for OpenCode 1 do not load on it, even when you rename the config entry. Check the Works with line on each pick before you upgrade. OpenCode 1 still gets releases. Two names to keep straight. The OpenCode here is the one at opencode.ai, now in the anomalyco/opencode repository on GitHub. An earlier Go project with the same name, opencode-ai/opencode, is archived and continues as Crush. And oh-my-opencode was renamed oh-my-openagent. Plugins that sign you in with a Claude or Gemini subscription are left out. Google and Anthropic treat that as against their terms, the popular Antigravity one is archived, and accounts have been flagged. A plugin runs with your full permissions, so read what you install. Ordered by how many independent recommendations each plugin has, then by downloads and stars. A plugin that does not load on the current OpenCode ranks below the ones that do. Numbers checked October 17, 2026.

The best FPV simulator is the one that teaches Acro flying before you crash a real drone. These picks rest on round-ups by Oscar Liang, Unmanned Tech, Coptrz and Droneblog, plus pilots on IntoFPV. For a first sim on a gaming PC, start with Liftoff. To race, VelociDrone has the physics MultiGP races on. For freestyle with friends, Uncrashed has online play and big maps. Tiny whoop pilots get their own game in Liftoff: Micro Drones. For cinematic lines through detailed scenery, try TRYP FPV. The best free FPV simulator is FPV.SkyDive, which also runs on phones. On an old laptop, FPV Freerider runs on 2 GB of RAM. What the store pages don't say: a drone simulator teaches most when you fly it with the radio you will use on the real drone, not a gamepad, and Oscar Liang suggests 5 to 10 hours before buying a drone. The DRL Simulator is still sold on Steam, but it hasn't been updated since 2023 and Oscar Liang and Unmanned Tech both warn its servers fail, so it isn't here. VelociDrone isn't on Steam: its own shop sells the licence in pounds only, Team BlackSheep sells it in dollars, and the freestyle pack costs extra. FPV.SkyDive is free, but extra maps and lessons are paid add-ons. Liftoff needs a dedicated graphics card. Steam refunds a game within 14 days if it was played for under 2 hours, so a sim that feels wrong can go back. Ordered by how many named reviewers pick each sim as the best for some kind of flying, then by price, lowest first. Prices checked October 17, 2026.

The best FPV drone for most US buyers is a bind-and-fly quad that matches the goggles and radio they already own, or one starter kit with both. These picks rest on hands-on tests by Oscar Liang, FirstQuadcopter, DroneXL and TechRadar. New to FPV and own no gear? The BetaFPV Cetus X kit puts drone, radio and goggles in one box, and the Aquila20 kit adds position hold for outdoor practice. More goggle-included boxes are on our Best FPV Drone Kits list. For a first bind-and-fly whoop to fly with your own ExpressLRS radio and analog goggles, the Air75 II and Meteor75 Pro II are the cheapest here. For DJI footage from a quad under 250 g, the Pavo20 Pro II is the strongest, and the Pavo Pico II and Mobula8 are smaller. For 5-inch freestyle, the GEPRC Vapor X5. To carry a full-size GoPro, the Cinelog35 V3. What the boxes don't say: BNF means no radio or goggles, and PNP usually means no receiver either, though a DJI O4 drone can fly with DJI's own remote. Digital DJI picks need DJI goggles. Real flight times run far under the claims: Oscar Liang got about 5 minutes from a Meteor75 Pro II rated for 10:30. The Vapor X5 and Cinelog35 weigh over 250 g, so US flyers must register them with the FAA. Most analog video transmitters legally need a Technician ham licence. DJI's US store doesn't sell the Avata 2, and the FCC's December 2025 Covered List blocks new foreign drone models, though models authorised before then can still be sold. Links go to the brand's own Amazon listing where one exists at a fair price, otherwise to the maker's store or a US FPV shop. Ordered by how many named reviewers tested each drone hands-on, then by price, lowest first. Prices checked October 17, 2026.

FPV goggles show video only from the system they were built for, so the first question is which video your drone sends. These picks rest on hands-on reviews by Oscar Liang, DroneXL, First Quadcopter and Unmanned Tech. For a DJI Avata 2, Neo or O4 build, the DJI Goggles N3 is the cheapest way in. For O3 gear, buy the DJI Goggles 3. Walksnail pilots on a budget start with the Avatar Goggles L, and the Goggles X adds analog and HDZero inputs. For analog on a budget, the Skyzone Cobra SD is the value box, and the SKY04X Pro is the slim pair that takes an HDZero or Walksnail receiver over HDMI. HDZero racers want the HDZero Goggle 2. The best FPV goggles for beginners match the kit they fly: our FPV drone kits list has kits with goggles in the box, and the DIY drone kits and builds list names the system each build uses. FPV goggles, FPV headsets and drone goggles are the same thing. What the boxes don't say: no digital system talks to another. The DJI N3 won't connect to an O3 air unit. Walksnail Ascent is a separate, cheaper system that doesn't work with Avatar gear. The BetaFPV VR04 HD sees only ArtLynk. DJI's US store now lists the Goggles Integra and Goggles 2 for support only. Analog box goggles cost about twice what they did: the $74.99 BetaFPV VR04 is sold out and the Eachine EV800D sells for $159.99. Since December 22, 2025, the FCC Covered List blocks new foreign-made drone gear from approval, but goggles already approved can still be sold. Ordered by how many named reviewers tested each pair hands-on, then by price, lowest first. Links go to Amazon when the brand or an authorised dealer sells there at list price, otherwise to the maker's store. Prices checked October 17, 2026.

An FPV drone kit puts drone, goggles and radio in one box. These picks rest on hands-on tests by Oscar Liang, TechRadar, Space.com and Droneblog, and each weighs under 250 g, so US recreational flyers skip FAA registration. For a first kit that flies outdoors and can hover on its own, start with the BetaFPV Aquila20, or the Aquila20 HD for a sharp digital picture. On the smallest budget, the brushed Cetus is built to fly indoors around people. For flying around the house, the Aquila16 lasts longest per pack. To learn Acro on a drone that runs Betaflight, get the Cetus X. To learn freestyle outdoors, the Tinyhawk III Plus Freestyle flies on one battery or two. Every kit here includes goggles, which FPV needs. Expect 3 to 10 minutes per battery. What the boxes don't say: most of these kits take batteries only their maker sells, and Oscar Liang warns that beginners outgrow kit goggles and radios fast. Spend 5 to 10 hours in a simulator first. Analog video is cheaper and reacts faster, while digital is sharper but ties you to one goggle system. US recreational flyers must pass the free TRUST test and, with goggles on, need a visual observer beside them. Most analog video transmitters legally need a Technician ham licence. DJI's US store doesn't sell the Avata 2 or the Neo with goggles, and the Neo 2 never launched in the US. The FCC's December 2025 Covered List blocks new foreign drone models, but models authorised before then can still be sold. Links go to the brand's own Amazon listing or the maker's store. Ordered by how many named reviewers tested each kit hands-on, then by price, lowest first. Prices checked October 17, 2026.

These DIY drone kits and builds come from published parts lists and build guides by Oscar Liang, Joshua Bardwell, PX4 and others, priced part by part. Building rarely costs less than buying the same drone built. To build your own drone with no soldering, start with Oscar Liang's 1S toothpick, which plugs together in about 30 minutes. For a drone that holds position and flies home on GPS, the Hawk's Work F450 kit comes in one box with its radio. To learn coding, the Crazyflie 2.1+ or the $49 LiteWing fly from a phone. To learn to solder and repair, Joshua Bardwell's sub-250 g 3-inch build has a video for every step. A DIY drone kit, a drone building kit and a build-your-own drone kit all mean you put it together. A DIY FPV drone adds a camera and goggles, and an FPV drone build usually means soldering. What the parts lists leave out: the radio, goggles and charger cost about as much as a small quad. An ELRS radio starts at $32.99 (BetaFPV LiteRadio 3) and analog goggles at $159.99 (Eachine EV800D, since the $74.99 BetaFPV VR04 is sold out). Walksnail Goggles L cost $199.99, and a 4S to 6S charger with its power supply $76.98. Anything 250 g or more needs FAA registration and Remote ID (the NewBeeDrone BeeID is $36.99), and every flyer takes the free TRUST test. Most analog video transmitters need a US Technician ham licence. Since December 22, 2025, new foreign-made drone parts, DJI included, can't get FCC approval, so spares may get scarce. Ordered by how much soldering a build needs, none first, then by its parts total, lowest first. Totals count the batteries each build's spec names, not shared gear or shipping. Prices checked October 17, 2026.

The skills Codex users and independent testers keep recommending, each with the exact command that installs it in Codex, and what it costs you in tokens or setup. Where Codex skills go: a skill is a folder with a SKILL.md file. Codex reads them from .agents/skills in your repo, or ~/.agents/skills for every project, and $skill-installer adds one by name. Plugins bundle skills with MCP servers and hooks; install them with /plugins or codex plugin marketplace add, in the CLI or the ChatGPT app but not the IDE extension. Do Claude skills work in Codex? Plain SKILL.md skills do once they're in .agents/skills, and Codex can add a Claude plugin marketplace directly. Claude slash commands and subagents don't carry over, and hooks have to be approved again in /hooks. What older guides get wrong: OpenAI's openai/skills repo is deprecated in favor of its plugin marketplace, and ~/.codex/skills is a legacy folder. Measured savings also run smaller than advertised: caveman cut Codex's tokens by about 4% in one independent test. Ordered by how many independent testers and developers recommend each skill for Codex, then by GitHub stars. Numbers checked October 17, 2026.

These Grok Bot templates are the shared bots with outside evidence they work, from hands-on reports by users like Moritz of Prompt Warrior to picks by template reviewers Avid and Steal What Works. Most directory listings have none. A Grok Bot template is a bot someone shared as an x.ai/bot link: its instructions, skills and routines, without the maker's logins, files or chat history. What can Grok Bot do? Each bot works on a cloud computer with a browser and your connected apps, keeps going after your laptop closes, and waits for your yes before it sends, buys or deletes. The use cases that keep recurring are a chief of staff that routes work to other bots, inbox and calendar triage, sourced research, and an outer loop that hands code to Cursor agents. If you plan to build your own bots, add dr eggbot first. Projects Manager suits anyone juggling several projects at once. Parents get the most from Tradbot, developers from loops. Vet every template before adding it. Read its share page, then every skill and routine once it's in your app: it acts with whatever you connect, and SpaceXAI does not verify third-party bots. All your bots share one cloud computer, so a new template can use logins your other bots opened. Check who made it: several makers here work on Grok Bot, and X now pays invited creators when people use their templates, so a loud launch post is not proof. Bouncer, below, gives a verdict on any share link. Directory numbers mean little yet. ShareGrokBots counts single-digit adds and gtemplate.net counts views of a launch post on X, not installs. Every template needs Premium ($30 a month) or higher on The X Subscription and draws on the same weekly usage pool. Ordered by independent hands-on reports first, then by how many independent reviewers recommend each template, then by launch-post views. Prices checked October 16, 2026.

Useful gifts for dad under $25: things he'll use every week, each a winner in a hands-on test by Serious Eats, Wirecutter, America's Test Kitchen, Project Farm or ToolGuyd. No novelty socks or joke books. Match it to the dad: for the cook, the Lodge skillet or the Javelin thermometer. For the griller, the Weber chimney and OXO tongs (together about $46, a good pair from two kids). For the fix-it dad, the Klein screwdriver, the Milwaukee utility knife, or the $9 tire gauge as a stocking stuffer. For the commuter, the Contigo mug. And for the dad who loses things, the AirTag. Ordered by how often a dad is likely to use each one, everyday things first. Prices checked October 4, 2026.

Officially licensed KPop Demon Hunters costumes for kids and adults: Rumi, Mira, Zoey and the Saja Boys, each with who it suits and what to watch for. We flag the add-ons that aren't licensed. Spirit Halloween or Disguise? Spirit's sets, about $50 for kids and $70 for adults, are separate pieces: a real zip-up jacket, top, shorts and boot covers. Disguise's, about $25, are a printed one-piece suit with a jacket on top. They're cheaper and quicker to get into, but they look printed up close. Sizing: Spirit's kids' sizes go by height. XS fits 36 to 41 inches, S 41 to 47, M 52 to 55, L 57 to 61 and XL 61 to 64. Spirit says to size up when in doubt, and parents say the tops run small. Disguise comes in 4–6X, 7–8 and 10–12. On a budget: Disguise's Rumi ($25) plus the Getgal braid ($10) is the whole look for about $35, or about $45 with the glow tattoos. A Disguise HUNTR/X trio (Rumi, Mira and Zoey) is about $76. Ordered with the HUNTR/X costumes first, the closest to the movie on top, then the Saja Boys, then the add-ons. Prices checked October 4, 2026.

The best portable jump starters, picked from hands-on tests by Car and Driver, TechGearLab and Project Farm, which jumped dead batteries in V6s, trucks and beaters. A portable jump starter is a battery pack with clamps: it starts a dead car without jumper cables or a second car. Quick picks: the WOLFBOX MegaVolt24 is the best overall, the GOOLOO GP2000 the best value, the Strauto is Car and Driver’s budget pick, and the $45 AstroAI S8 is the cheapest one that passed a real test. How to choose: match the pack to your engine rather than the peak-amp number on the box, which isn't comparable between brands. The NOCO GB40 is rated for gas engines up to 6 liters, the GOOLOO GP2000 up to 8 and the WOLFBOX up to 10. For a V8 truck or a diesel, get the GOOLOO GT6000, or the NOCO GB70 if you want the name brand. The small lithium packs fit in a glovebox; the Stanley is a heavier lead-acid unit meant for the garage. Keep it charged: a pack that's flat when you need it is no help. Top it up every few months. The GP2000 is rated to hold a charge for up to two years. Ordered by how each pack did in those tests on an ordinary car, then the ones for big engines and the garage. Prices checked October 4, 2026.

Small models for text classification you can run on a laptop: spam, intent, routing and moderation with no API bill, picked from independent tests by Red Hat's AI safety team, Oliver Borchers and the BTZSC benchmark. It started with a build on Made with Jev: a developer swapped Jev, TypeSafe's hosted decision model, for a DeBERTa large classifier on a base M1 Mac to decide, inside a Claude Code hook, whether a message was just a question, and found it just as fast and just as good for that job. Jev is the yardstick here. In Borchers' field test on 400 fresh arXiv papers it scored 86.3% with no training, for about $0.04 per million input tokens. How to choose comes down to labels. With none, try Clef-flash on a 16 GB Mac with a Pro or Max chip, or GLiClass on anything else. With a few dozen examples per label, embed the text and train a logistic regression on top. With a few hundred, fine-tune DeBERTa-v3 or ModernBERT. No labels yet? Have Jev or a large LLM label a few hundred of your own examples, then train the small model on those. Skip facebook/bart-large-mnli, still the default in Hugging Face's zero-shot pipeline: it dates from 2019, runs a full pass for every label, and trails newer models in both the 28-task zero-shot benchmark and BTZSC. The list is ordered by how well each model did in those tests for the work it takes to set up, with the specialists for PII, prompt injection and secrets after the general-purpose models. Numbers checked October 5, 2026.

These Grok Bot use cases and workflows are the jobs paying users report running, drawn from walkthroughs by Claire Vo, Matthew Berman, Peter Yang and Alex Finn. The ones that hold up end in a draft you approve. Grok Bot is SpaceXAI's always-on agent. Each bot works on a cloud computer that keeps going after your laptop closes. Show it a workflow once and it saves the steps as a skill, then a routine reruns that skill on a schedule. A template is a shareable link to a bot with its skills and routines. If you will run more than two bots, start with a chief of staff, the one bot nearly every heavy user keeps. For a single first job, inbox triage that leaves drafts in Gmail has the most reported wins. Developers get most from the outer loop that hands code to Cursor cloud agents. At home, a school-email digest or a fare watch is the easy start. What can Grok Bot do on its own? Read, research and draft around the clock. Sending, buying and deleting wait for your yes by design. Free and Plus plans don't include it. What SpaceXAI's pages don't stress: all your bots share one cloud computer and its logins, so a second bot is not a security boundary. The weekly usage pool runs out fast, and a Cursor forum user on the $200 Ultra plan used 99% of a week in three days. Amazon reportedly blocks Grok Bot from placing orders. Guides quoting $200 or $300 a month describe launch-week access, since replaced by The X Subscription, where Premium at $30 a month is the cheapest plan with Grok Bot. Ordered by how many independent users we found reporting each job running, then by whether they reported a measurable result. SpaceXAI staff guides count only to show a template exists. Prices checked October 16, 2026.

The best Apple Vision Pro apps are the visionOS games, players and tools that reviewers keep recommending. These picks rest on hands-on coverage from Tom's Guide, PCMag, Road to VR, UploadVR, MacStories, AppleInsider and independent writers. To show someone what the headset can do, start with Encounter Dinosaurs and Explore POV. For games, get Synth Riders, or Retrocade if you own a Bluetooth gamepad. For work, Fantastical and Things 3 for Vision are the usual picks. For watching, Disney+ has 3D films in themed theaters, YouTube finally has its own app, and Portal streams a PS5 or Xbox to a giant screen. Several games here are Apple Arcade titles: free to download, but they need the $6.99-a-month Arcade subscription. Things 3 for Vision is a separate $29.99 app and now needs visionOS 27. Juno, the YouTube app many older guides recommend, was pulled from the App Store in 2024, and Google's own app arrived in February 2026. Netflix still has no Vision Pro app, so it plays in Safari. visionOS 26 added widgets you pin to walls and support for Sony's PS VR2 Sense controllers, though few games use the controllers yet. visionOS 27, out since September 14, 2026, adds Siri AI and lets you turn panoramas into Environments. The list is ordered by how many independent reviewers recommend each app, then by price. Prices checked October 14, 2026.

The best GEO tools track how often ChatGPT, Gemini, Perplexity and Google's AI answers mention your brand. This list rests on hands-on tests by Zapier, agency teams and named SEO practitioners, not on vendors ranking themselves. Answer engine optimization (AEO) tools, AI search optimization tools and GEO tools are three names for the same job: checking whether ChatGPT, Gemini, Perplexity and Google's AI answers mention you, and which pages they cite. For a solo site, start free: Bing Webmaster Tools shows how often Copilot actually cites your pages, and HubSpot's AI Search Grader gives a one-off snapshot. The cheapest trackers after that are OtterlyAI at $29 a month and HubSpot AEO at $50. A small team or agency should try Peec AI first: every plan has unlimited seats, and it won the head-to-head tests. If you already pay for Ahrefs or Semrush, use their add-on before buying anything new. Enterprises with a budget and an analyst go to Profound, which now sells only custom plans. Read the plan, not the headline price. Cheap tiers often track three or four engines and charge extra for Gemini, Claude or Google AI Mode. Prompts, credits and checks are different units, so compare what a month of your own questions would cost. This market also changes hands fast: Adobe bought Semrush, Sitecore bought Scrunch, and Hall closed to new sign-ups after Tracksuit bought it. Treat every score as an estimate. Apart from Bing's own report, no AI engine gives these tools a data feed; they sample answers, and a SparkToro study found less than a 1 in 100 chance that the same prompt returns the same list of brands twice. Watch your trend over weeks, spot-check a few prompts by hand, and expect different tools to disagree. None of them writes the fix for you. The list is ordered by how many independent testers recommend each tool, then by price, lowest first. Prices checked October 14, 2026.

The best AI note taking device is a card, pin or ring that records a conversation and hands back a transcript and summary. These picks rest on hands-on tests by WIRED, TechCrunch, Wirecutter, PCWorld and other named reviewers. If you want the safest all-rounder, get the Plaud Note Pro. If you never want a monthly bill, the Comulytic Note Pro and Pocket transcribe with no minute cap, and the HiDock P1 does the same for calls you take through earbuds. If you want something to wear, look at the Soundcore Work or the Plaud NotePin S. The hardware price is only part of the cost. Plaud, Soundcore and Vocci include 300 free transcription minutes a month, which is five hours. Past that, Plaud and Soundcore charge $99.99 a year for 1,200 minutes a month or $239.99 a year for unlimited. TicNote and viaim include 600 minutes. Pocket, Comulytic and HiDock transcribe free without a cap and sell extras such as speaker labels and longer history: Pocket Pro is $239.88 a year, Comulytic Premium is $10 a month billed annually and HiDock Pro is $199 a year. Older guides still recommend devices you can no longer order. The Limitless Pendant was pulled from sale after Meta bought the company, and Amazon's Bee and the open-source Omi were both sold out on their makers' stores when checked. Sale prices in this category change often, so the regular price is given alongside. Recording law depends on where you are: US federal law and most states let you record a conversation you take part in, but some states, including California and Florida, require everyone's consent, so tell people before you press record. The list is ordered by how many independent reviewers recommend each device, then by what it costs to own. Prices checked October 11, 2026.

The best OpenClaw skills, chosen from what developers and reviewers who run the agent say they keep installed, then checked against each skill's source and its ClawHub security audit. For email and calendar, start with Gog. For code, GitHub. For long articles, videos and podcasts, Summarize. For notes, Obsidian or Notion. To stop the agent repeating a mistake, self-improving agent. For cleaner prose, Humanizer. Safety first. ClawHub, the registry, lets anyone with a GitHub account publish, and a skill acts with whatever access your agent has. Security firms have counted more than a thousand malicious skills there since January 2026, and in June Unit 42 and Trail of Bits each showed skills getting past its scanners. OpenClaw's own June data labelled 25,504 of 67,453 skill versions suspicious and 206 malicious. Every skill page now has a security audit with an outcome such as Pass or Review. Read it, and the SKILL.md, before you install. Older guides are out of date in three ways. Most of the skills they tell you to install now ship with OpenClaw and switch on when their command-line tool is present. Web search, self-learning and memory search are built in, and Tavily has a first-party plugin, so the Tavily and QMD skills are no longer needed. And names are copied: at least five publishers use the slug gog, so install with the owner, as in @steipete/gog. Skill Vetter and Capability Evolver are on most lists and left out here. Skill Vetter's publisher links to a GitHub profile that no longer exists. ClawHub's audit marks Capability Evolver for review because it can run jobs handed to it by a remote server. Skills are ordered by how many independent reviewers and users recommend them, then by downloads or stars. All are free. Numbers checked October 11, 2026.

The best OpenRouter alternatives, hosted routers and gateways you run yourself, picked from developers who switched and reviewers who measured them. Each one still gives you one API for many AI models. Match the pick to your reason for leaving. If it is the 5.5% fee on credits, Vercel AI Gateway charges no markup, and Cloudflare AI Gateway is free when you bring your own provider keys. If you want no middleman at all, self-host LiteLLM, or Bifrost if LiteLLM's latency bothers you. If your data must stay in Europe, look at Cortecs or Requesty. If you are tired of one model behaving differently from host to host, buy from a single host such as Fireworks AI, Together AI or Groq, or use OpenCode Zen's short tested list. This market changed fast in 2026, so older guides are wrong in places. Stripe announced in August that it is buying OpenRouter. Palo Alto Networks bought Portkey in May. Helicone has been in maintenance mode since Mintlify bought it in March, and TensorZero's repository is archived, which is why neither is here. LiteLLM had two poisoned releases on PyPI in March (1.82.7 and 1.82.8), so pin your version. Self-hosting takes the gateway company out of the path, but your prompts still go to whichever model provider you call. And OpenRouter's fee is charged when you buy credits, with an $0.80 minimum, so small top-ups cost more than 5.5%. The list is ordered by how many independent developers and reviewers recommend each option, then by adoption, then by price. Prices checked October 11, 2026.

The best local LLM for coding depends on how much memory your machine has. This list ranks open-weight models you can run yourself, by independent benchmark runs and by developers who use them in coding agents. With 24 to 32 GB, start with Qwen3.8-27B. Its 17 GB 4-bit file matched the full model on Terminal-Bench 2.1 in Quesma's test. With 128 GB of unified memory, Qwen3.8-Flash-Next posted the top score in WonderRico's benchmark. With 16 GB, gpt-oss-20b fits whole on the card, and one developer reports useful work from a 2-bit Qwen3.8-27B. With 8 GB, try Qwen3.5-9B and keep the jobs small. At 64 GB, run the 27B models at 8-bit, a 29 GB file, with a longer context. The step above that needs about 100 GB. A coding agent only works if the model emits tool calls your server can parse. Gemma 4 shipped with broken tool calling that was fixed later, and Qwen chat templates have needed community fixes, so update Ollama, LM Studio or llama.cpp before blaming the model. Agents also need room: LM Studio loads models with an 8,192-token context unless you raise it. Scores on a model card are the maker's own. A mixture-of-experts model computes with a few billion parameters per token but still has to hold all of them in memory, and context takes memory on top of the file. Qwen3.8 thinks at its highest effort by default, which is slow on home hardware. Qwen3.8-Flash-Next is not Apache licensed. Older guides still recommend Qwen2.5-Coder, Qwen3-Coder and Devstral, which general-purpose models beat in the hands-on tests cited here. The list is ordered by how many independent testers and developers recommend each model for coding, then by downloads. Numbers checked October 11, 2026.

The best Claude skills are the SKILL.md packages that Hacker News users and independent reviewers keep recommending. This list ranks collections and single skills you can install today, with what each adds and its catch. If you want Claude to plan and test before it codes, get Superpowers. If you want it to question you before it starts, get grill-me from Matt Pocock's skills. For better-looking pages, start with frontend-design and add Impeccable to review the result. To build your own skills, or to make real Word, Excel, PowerPoint and PDF files, use Anthropic's repository. For marketing work, use Corey Haines's pack. Skills install in three ways. In Claude Code, run /plugin install from a marketplace, or copy a folder into ~/.claude/skills. For most coding agents, npx skills add owner/repo does the copying. In the Claude apps, upload a ZIP under Customize > Skills. Skills follow the open Agent Skills format, so most of these also run in Codex, Cursor and Gemini CLI. Older guides are now wrong in places. Claude Code has merged custom commands into skills. Superpowers and Matt Pocock's skills are in Anthropic's official marketplace, so adding a separate marketplace is no longer needed. The Karpathy repository moved to a new owner. Star counts are huge and say little, and a link list such as awesome-claude-skills installs nothing. A skill is instructions plus scripts that run with your agent's permissions. Snyk scanned 3,984 skills from ClawHub and skills.sh and found a critical security issue in 13.4% of them. Read a skill before you install it, and install a few, not all. Items are ordered by how many independent people recommend each one, then by GitHub stars. Everything here is free. Numbers checked October 11, 2026.

The best Claude Code plugins, picked from the setups developers publish after pruning their own installs and from Hacker News threads where people say what they kept, not from install counts alone. A plugin is a bundle. One install can carry skills, slash commands, subagents, hooks and MCP servers. A skill is a single instruction file Claude loads when it judges it relevant, and an MCP server is a connection to an outside tool, so a plugin can contain both. Type /plugin inside Claude Code to browse Anthropic's official marketplace, which is connected from the start, or run /plugin install name@claude-plugins-official. For anything else, add the maker's marketplace first with /plugin marketplace add owner/repo. If you install one thing, make it Superpowers, plus the LSP plugin for your language. Add Security Guidance if you ship code you don't read line by line, Plannotator if you want to mark up plans in a browser, and Claude HUD to see how full the context is. Every enabled plugin adds to the context on every turn, and several of the writers quoted here deleted more than they kept. Plugins run code on your machine with your permissions, and community marketplaces are not reviewed, so read what a plugin will install before you say yes. Two names trip people up. Claude Code has a built-in review command alongside the Code Review plugin. And Anthropic keeps a second, smaller marketplace in the anthropics/claude-code repo, where the loop plugin still goes by ralph-wiggum and not ralph-loop. The list is ordered by how many independent people recommend each plugin, then by installs and GitHub stars. Prices checked October 11, 2026.

The best OpenClaw alternatives are other personal AI agents you message from a chat app, self-hosted or hosted. These picks rest on write-ups by people who switched and on what developers on Hacker News say they run now. People leave OpenClaw for four reasons: updates that break things, token bills, the work of running it, and security. If it kept breaking, most switchers land on Hermes Agent, which imports your OpenClaw memory and skills. If you want code you can read and agents locked in containers, pick NanoClaw. For a small VPS, a Raspberry Pi or a $10 board, look at ZeroClaw or PicoClaw. If you never want to run a server again, Claude Cowork, Grok Bot and Manus do the hosting for a monthly fee. A few things the makers' pages won't tell you. None of the open-source agents is free to run: you still pay a model provider, and switchers report $10 to $20 a month on cheap models and far more on top ones. People who moved to Claude's own tools say Anthropic stopped Claude subscriptions from covering OpenClaw in April 2026. NemoClaw is a sandbox that runs OpenClaw or Hermes inside it, not a replacement. QwenPaw was called CoPaw until April 2026. Perplexity's Computer and Personal Computer are different products. Hosted agents bill in credits or weekly allowances that are hard to predict. And a smaller codebase is easier to audit, but any agent holding your logins can still be tricked by a hostile web page or message. Ordered by how many independent users and reviewers recommend each one, then by GitHub stars, then by price. Prices checked October 11, 2026.