Young Tries AI

Claude Code projects enter beta, and ChatGPT gets 17 video plugins

Claude Code projects are in beta for select Pro and Max subscribers. A project takes a goal and a repo, then hands the work to parallel cloud threads. A Higgsfield guide covers 17 video plugins for ChatGPT, and Claude Code v2.1.325 ships fixes. Check cloud-session access first.

6 usable today, 2 partly, 0 not yet

  1. Claude CodeProjects

    CodingFor developers, office workers

    What happened.Redesigned Claude projects are in beta in Claude Code for select Pro and Max subscribers. You set a goal plus a repo or context, and Claude scopes the request, delegates to parallel threads, reviews the outputs and assembles the result. Each thread is a cloud session on its own branch and copy of the repo.

    Why it matters.Overlapping edits show up as a normal merge conflict. Each thread is a full session, so projects can hit usage limits faster. Access widens over the coming week, with Team and Enterprise later.

    Can you use it today?Partly

    With a Pro or Max account that has cloud sessions in Claude Code and no existing projects, create a project and pick a goal and repo. Otherwise, join the waitlist.

    Source: Hacker News한국어로 읽기

  2. OpenAIChatGPT WalletChatGPT

    BusinessFor business and marketing, beginners

    What happened.ChatGPT Wallet lets agents pay without seeing the user's actual card details. The user reviews items, total and card before each purchase and approves with a security check such as Face ID or a fingerprint scan. It is built with American Express, Mastercard and Visa.

    Why it matters.Availability is limited: it starts through dots for ChatGPT Pro subscribers in the US. OpenAI also published a Commerce Partner Playbook for businesses that want agents to reach their products.

    Can you use it today?Partly

    Business owners can claim their business in ChatGPT and update the product feed now. Paying through Wallet requires a US ChatGPT Pro account.

    Source: OpenAI News, producthunt.com한국어로 읽기

  3. ChatGPTHiggsfieldRunwayOpenArtHeyGeninvideo

    Video and imageFor creators, beginners

    What happened.A Higgsfield guide says ChatGPT does not generate video itself. You connect a video app from the ChatGPT Plugins Directory, and ChatGPT helps write the brief and request changes. The guide covers 17 video plugins as of November 2026, including Higgsfield, Runway, OpenArt, Magnific, HeyGen and invideo.

    Why it matters.A ChatGPT plan does not pay for video. Higgsfield, Runway, OpenArt, HeyGen and Descript charge generations to your credits with that service, and Higgsfield in ChatGPT needs an active subscription.

    Can you use it today?Yes

    Install a video plugin from the Plugins Directory in ChatGPT and sign in to that service. Describe the shot with subject, action, setting, camera, light and mood, and ask for the credit cost before generation.

    Source: Higgsfield Blog한국어로 읽기

  4. Claude Codev2.1.325

    CodingFor developers

    What happened.Claude Code v2.1.325 fixes a security issue in how zstd-compressed responses are decoded. It also fixes subagents with their own model being moved to the plan-mode model when the model setting is opusplan or haiku, and sessions on many third-party gateways and Foundry stopping with a context length error on every turn.

    Why it matters.Other fixes cover recent turns missing from a resumed session after an unclean stop, and a saved sonnet[1m] setting rewritten to a model id the API rejects on a new install. A new error line points to /rewind or /clear when organization policy blocks a request.

    Can you use it today?Yes

    Update Claude Code to v2.1.325 to pick up the fixes.

    Source: Claude Code releases한국어로 읽기

  5. ClaudeOpus 5.5

    AutomationFor developers

    What happened.Anthropic published a guide to always-on agents, which mostly sleep and wake periodically for a run: receive an event, finish the task, save what happened. The five core primitives are event inbox, timers, memory, permissions and outbox.

    Why it matters.The inbox is a durable list of events, usually a database table, with an HTTP handler that writes events and a worker that reads them. Events can be deduplicated by source identifiers, and one inbox per user is suggested to start.

    Can you use it today?Yes

    Read the guide and build the inbox-plus-worker pattern for a scheduled or webhook-triggered Claude agent.

    Source: Claude Blog한국어로 읽기

  6. HiggsfieldSeedance 2.5Kling 4.0RunwayOpenArt

    Video and imageFor creators, business and marketing

    What happened.Higgsfield's own comparison covers 8 micro-drama tools. Higgsfield saves a series cast as Elements and generates dialogue with synchronized lip movement in Seedance 2.5 clips of up to 30 seconds. Seedance 2.5 starts on the $49 plan with 1,000 credits. Kling 4.0 makes 3- to 30-second 9:16 clips with stereo audio and lip-synced dialogue.

    Why it matters.The comparison comes from a vendor that sells one of the tools. Kling AI saves up to 50 Elements on the $10 Standard plan, and OpenArt Director builds multi-scene videos of up to five minutes from the $14 Starter plan.

    Can you use it today?Yes

    Pick Higgsfield, Kling AI or Runway to direct shots with a saved cast. Pick OpenArt, Topview, invideo or Picsart to have an agent turn a script into episodes.

    Source: Higgsfield Blog한국어로 읽기

  7. llama.cppggmlv1.0.0

    CodingFor developers

    What happened.The llama.cpp and ggml libraries (libllama and libggml) reach v1.0.0. API and ABI incompatibilities will now be signalled with major-version bumps. The release supports more than 150 architecture variants and all mainstream speculative decoding methods.

    Why it matters.MAJOR covers breaking changes to the public C API, MINOR covers compatible features, backend updates or model support, and PATCH covers bug fixes. Builders shipping on the libraries get a versioning rule to pin against.

    Can you use it today?Yes

    Install llama.cpp from a listed channel such as Homebrew or Winget, and pin downstream apps to the v1 API.

    Source: Hugging Face Blog한국어로 읽기

  8. HiggsfieldTalent Contest

    Video and imageFor creators

    What happened.Higgsfield's Talent Contest offers $1,000,000 in cash prizes across 20 awards: three main prizes, two Audience Choice prizes and 15 category awards. Entries are original AI films of three to seven minutes, solo or in teams, made in a dedicated Higgsfield Talent Contest project between November 12 and November 26, 2026, 7:00 PM UTC.

    Why it matters.Entry is free with no purchase or subscription required, though generation credits are used as usual. Every submitted project, including Private entries, becomes permanently public at the deadline with its prompts, generation history and assets.

    Can you use it today?Yes

    Start a Talent Contest project in Higgsfield and submit a three-to-seven-minute film before November 26, 2026, 7:00 PM UTC. Potential winners are announced December 3.

    Source: Higgsfield Blog한국어로 읽기

Today’s finds

New skills, plugins, MCP servers, repos and small products, described only from their own README or landing page.

  • Skill32 stars on GitHub

    lowdevco/Clip-Forge

    Claude skill that takes a YouTube link or video file, finds 8 to 12 moments and scores each out of 100. It waits for your pick, then renders vertical clips for YouTube Shorts, Instagram Reels and TikTok with a title, caption and hashtags per clip.

    Video and imageFor creators

    Usable todayIn Claude Code, paste an install request pointing to the GitHub repo, start a new session and run /clip-video with a link.

    Source: GitHub new repos한국어

  • Skill74 stars on GitHub

    mileson/chaojifeng-video-content-creator

    Chinese-language video skill that runs topic, voice-over script, storyboard, filler-word cleanup, effects overlay, subtitles and a publish pack in one pipeline with gate files per stage. It includes 148 HTML/GSAP effects timed to subtitle timestamps.

    Video and imageFor creators, developers

    Usable todayRun npx skills add mileson/chaojifeng-video-content-creator, then tell the agent to use it on a topic or a raw video file.

    Source: GitHub new repos한국어

  • MCP server66 stars on GitHub

    hridaya423/phonedriver

    MCP server that lets Claude Code, Codex, Cursor or any MCP client operate an Android phone or an iPhone by reading the screen as text. It asks the user before it taps Pay, and it claims 6.9x fewer tokens than mobile-mcp across nine phone jobs.

    Skills and MCPFor developers

    Usable todayRun claude mcp add phonedriver -- npx -y phonedriver mcp. It needs an Android emulator, a USB-debugging Android phone or the iOS Simulator.

    Source: GitHub new repos한국어

  • MCP server22 stars on GitHub

    tammo1007/alyn

    Remote MCP server that needs no account or API key and lets an agent ask which skill fits a task, load it and follow it. It says it knows 6,134 open source skills by 445 creators, refreshed nightly.

    Skills and MCPFor developers

    Usable todayAdd it with claude mcp add using the http transport and the name alyn-mcp, then work as usual. Claude web uses Settings > Connectors > Add custom connector.

    Source: GitHub new repos한국어

  • MCP server26 stars on GitHub

    OnlyTerp/turnpack

    Small local MCP server with three tools that fold dependent agent round trips into fewer calls. Its ship tool applies edits across files and runs your test command; either everything passes and stays, or every file is restored byte-for-byte.

    CodingFor developers

    Partly usableFollow the Set it up section of the README. The install command is not given here, so check the README for it.

    Source: GitHub new repos한국어

All finds

Worth trying this week