GitHub Trending

Privacy first, AI meeting assistant with 4x faster Parakeet/Whisper live transcription, speaker diarization, and Ollama summarization built on Rust. 100% local processing. no cloud required. Meetily (Meetly Ai - https://meetily.ai) is the #1 Self-hosted, Open-source Ai meeting note taker for macOS & Windows. Understand How to write meeting minutes

This is MCP server for Claude that gives it terminal control, file system search and diff file editing capabilities

Codex plugin for Claude Code Use Codex from inside Claude Code for code reviews or to delegate tasks to Codex. This plugin is for Claude Code users who want an easy way to start using Codex from the workflow they already have. What You Get /codex:review for a normal read-only Codex review /codex:adversarial-review for a steerable challenge review /codex:rescue, /codex:transfer, /codex:status, /codex:result, and /codex:cancel to delegate work, hand off sessions, and manage background jobs Requirements ChatGPT subscription (incl. Free) or OpenAI API key. Usage will contribute to your Codex usage limits. Learn more. Node.js 18.18 or later Install Add the marketplace in Claude Code: /plugin marketplace add openai/codex-plugin-cc Install the plugin: /plugin install codex@openai-codex Reload plugins: /reload-plugins Then run: /codex:setup /codex:setup will tell you whether Codex is ready. If Codex is missing and npm is available, it can offer to install Codex for you. If you prefer to install Codex yourself, use: npm install -g @openai/codex If Codex is installed but not logged in yet, run: !codex login After install, you should see: the slash commands listed below the codex:codex-rescue subagent in /agents One simple first run is: /codex:review --background /codex:status /codex:result Usage /codex:review Runs a normal Codex review on your current work. It gives you the same quality of code review as running /review inside Codex directly. !NOTE] Code review especially for multi-file changes might take a while. It's generally recommended to run it in the background. Use it when you want: a review of your current uncommitted changes a review of your branch compared to a base branch like main Use --base for branch review. It also supports --wait and --background. It is not steerable and does not take custom focus text. Use [/codex:adversarial-review when you want to challenge a specific decision or risk area. Examples: /codex:review /codex:review --base main /codex:review --background This command is read-only and will not perform any changes. When run in the background you can use /codex:status to check on the progress and /codex:cancel to cancel the ongoing task. /codex:adversarial-review Runs a steerable review that questions the chosen implementation and design. It can be used to pressure-test assumptions, tradeoffs, failure modes, and whether a different approach would have been safer or simpler. It uses the same review target selection as /codex:review, including --base for branch review. It also supports --wait and --background. Unlike /codex:review, it can take extra focus text after the flags. Use it when you want: a review before shipping that challenges the direction, not just the code details review focused on design choices, tradeoffs, hidden assumptions, and alternative approaches pressure-testing around specific risk areas like auth, data loss, rollback, race conditions, or reliability Examples: /codex:adversarial-review /codex:adversarial-review --base main challenge whether this was the right caching and retry design /codex:adversarial-review --background look for race conditions and question the chosen approach This command is read-only. It does not fix code. /codex:rescue Hands a task to Codex through the codex:codex-rescue subagent. Use it when you want Codex to: investigate a bug try a fix continue a previous Codex task take a faster or cheaper pass with a smaller model !NOTE] Depending on the task and the model you choose these tasks might take a long time and it's generally recommended to force the task to be in the background or move the agent to the background. It supports --background, --wait, --resume, and --fresh. If you omit --resume and --fresh, the plugin can offer to continue the latest rescue thread for this repo. Examples: /codex:rescue investigate why the tests started failing /codex:rescue fix the failing test with the smallest safe patch /codex:rescue --resume apply the top fix from the last run /codex:rescue --model gpt-5.4-mini --effort medium investigate the flaky integration test /codex:rescue --model spark fix the issue quickly /codex:rescue --background investigate the regression You can also just ask for a task to be delegated to Codex: Ask Codex to redesign the database connection to be more resilient. Notes: if you do not pass --model or --effort, Codex chooses its own defaults. if you say spark, the plugin maps that to gpt-5.3-codex-spark follow-up rescue requests can continue the latest Codex task in the repo /codex:transfer Creates a persistent Codex thread from the current Claude Code session and prints a codex resume command. Use it when you started a debugging or implementation conversation in Claude Code and want to continue that same context directly in Codex. Examples: /codex:transfer /codex:transfer --source ~/.claude/projects/-Users-me-repo/.jsonl The plugin's existing SessionStart hook supplies the current transcript path automatically; --source is available as a manual override. The transfer uses Codex's external-agent session importer, so it follows the same conversion rules as importing Claude history in the Codex App and creates visible turns that can be continued in the App or TUI. The source must be under ~/.claude/projects, and older Codex versions that do not expose session import must be upgraded before using this command. /codex:status Shows running and recent Codex jobs for the current repository. Examples: /codex:status /codex:status task-abc123 Use it to: check progress on background work see the latest completed job confirm whether a task is still running /codex:result Shows the final stored Codex output for a finished job. When available, it also includes the Codex session ID so you can reopen that run directly in Codex with codex resume . Examples: /codex:result /codex:result task-abc123 /codex:cancel Cancels an active background Codex job. Examples: /codex:cancel /codex:cancel task-abc123 /codex:setup Checks whether Codex is installed and authenticated. If Codex is missing and npm is available, it can offer to install Codex for you. You can also use /codex:setup to manage the optional review gate. Enabling review gate /codex:setup --enable-review-gate /codex:setup --disable-review-gate When the review gate is enabled, the plugin uses a Stop hook to run a targeted Codex review based on Claude's response. If that review finds issues, the stop is blocked so Claude can address them first. [!WARNING] The review gate can create a long-running Claude/Codex loop and may drain usage limits quickly. Only enable it when you plan to actively monitor the session. Typical Flows Review Before Shipping /codex:review Hand A Problem To Codex /codex:rescue investigate why the build is failing in CI Start Something Long-Running /codex:adversarial-review --background /codex:rescue --background investigate the flaky test Then check in with: /codex:status /codex:result Codex Integration The Codex plugin wraps the [Codex app server. It uses the global codex binary installed in your environment and applies the same configuration. Common Configurations If you want to change the default reasoning effort or the default model that gets used by the plugin, you can define that inside your user-level or project-level config.toml. For example to always use gpt-5.4-mini on high for a specific project you can add the following to a .codex/config.toml file at the root of the directory you started Claude in: model = "gpt-5.4-mini" model_reasoning_effort = "high" Your configuration will be picked up based on: user-level config in ~/.codex/config.toml project-level overrides in .codex/config.toml project-level overrides only load when the project is trusted Check out the Codex docs for more configuration options. Moving The Work Over To Codex Delegated tasks and any stop gate run can also be directly resumed inside Codex by running codex resume either with the specific session ID you received from running /codex:result or /codex:status or by selecting it from the list. This way you can review the Codex work or continue the work there. FAQ Do I need a separate Codex account for this plugin? If you are already signed into Codex on this machine, that account should work immediately here too. This plugin uses your local Codex CLI authentication. If you only use Claude Code today and have not used Codex yet, you will also need to sign in to Codex with either a ChatGPT account or an API key. Codex is available with your ChatGPT subscription, and codex login supports both ChatGPT and API key sign-in. Run /codex:setup to check whether Codex is ready, and use !codex login if it is not. Does the plugin use a separate Codex runtime? No. This plugin delegates through your local Codex CLI and Codex app server on the same machine. That means: it uses the same Codex install you would use directly it uses the same local authentication state it uses the same repository checkout and machine-local environment Will it use the same Codex config I already have? Yes. If you already use Codex, the plugin picks up the same configuration. Can I keep using my current API key or base URL setup? Yes. Because the plugin uses your local Codex CLI, your existing sign-in method and config still apply. If you need to point the built-in OpenAI provider at a different endpoint, set openai_base_url in your Codex config.

CubeSandbox Instant, Concurrent, Secure & Lightweight Sandbox Service for AI Agents 中文文档 · Quick Start · Documentation · Changelog · X(Twitter) · End User Program Cube Sandbox is a high-performance, out-of-the-box secure sandbox service built on RustVMM and KVM. It supports both single-node deployment and easy scaling to multi-node clusters. It is compatible with the E2B SDK and can create a hardware-isolated, fully serviceable sandbox in under 60ms with less than 5MB of memory overhead. 📰 News v0.5: AutoPause, Terraform deployer, ARM64 & network policy hardening AutoPause/AutoResume — idle sandboxes auto-suspend and wake on the next request. Terraform one-click cluster deploy ARM64 native full-stack support network policy hardening — per-sandbox traffic tokens, policy-routing egress. Changelog → · Terraform deploy → v0.4: Safer egress, easier ops Credential vault — Agents call external APIs as usual; keys never enter the sandbox. Dashboard — version matrix and template health checks; see at a glance whether templates need rebuilding after upgrades. Changelog → · Security proxy guide → · WebUI guide → Snapshot, Clone & Rollback at hundred-millisecond granularity CubeSandbox 0.3.0 introduces the CubeCoW Copy-on-Write snapshot engine, enabling event-level snapshots, instant cloning, and rollback to any saved state. Changelog → 🎉 Initial open-source release Cube Sandbox is now open source! Millisecond boot, hardware-level isolation, E2B-compatible sandbox for AI Agents. Changelog → Product Highlights ⚡ Sub-60ms boot · High density · Auto pause/resume Average <60ms cold start, <5MB overhead per instance — run thousands of Agents on one node. Supports automatic sandbox pause and resume for cost optimization Quick start → 🔒 Hardware-level isolation Each sandbox gets its own Guest OS kernel — no Docker shared-kernel escapes; run untrusted LLM-generated code safely Architecture → 🔌 Seamless E2B migration Native E2B SDK compatibility — swap one URL env var, zero business code changes Examples → 🖥️ Web console Manage sandboxes, templates, nodes, and version matrix in the browser — open :12088 right after install WebUI guide → 🔐 Credential vault Agents call LLMs and external APIs as usual — keys never enter the sandbox, model context, or logs Security proxy guide → 🛡️ Egress control Domain allowlists, instant block on unauthorized egress, full audit logs for compliance Security proxy guide → 📸 Snapshot · Clone · Rollback Hundred-millisecond checkpoints on running sandboxes — roll back or fork from any saved state v0.3 changelog → 📦 Template system Turn OCI images into templates in one step, install official presets from the Template Store, auto-distribute across nodes Templates guide → 🤖 AgentHub digital assistants Spin up OpenClaw assistants in one click — snapshots, rollback, and assistant template publishing Digital assistant → Demos Installation & Demo Performance Test RL (SWE-Bench) Snapshot · Clone · Rollback Benchmarks In the context of AI Agent code execution, CubeSandbox achieves the perfect balance of security and performance: | Metric | Docker Container | Traditional VM | CubeSandbox | |---|---|---|---| | Isolation Level | Low (Shared Kernel Namespaces) | High (Dedicated Kernel) | Extreme (Dedicated Kernel + eBPF) | | Boot Speed Full-OS boot duration | 200ms | Seconds | *Sub-millisecond ( Sub-150ms sandbox delivery under both single and high-concurrency workloads CubeSandbox base memory footprint across various instance sizes (*Blue: Sandbox specifications; Orange: Base memory overhead). Note that memory consumption increases only marginally as instance sizes scale up. Quick Start ⚡ Millisecond-level startup — watch the fast-start flow above. Cube Sandbox requires an x86_64 Linux environment with KVM support. The guide walks you through everything in four steps — provisioning a server, installing Cube Sandbox, creating a sandbox template, and running your first agent code. No source build needed, up and running in minutes. Choose your deployment path: 🖥 PVM · Cloud VM → 🏆 Recommended 🏗 Bare Metal → 💻 Dev-Env → ⚠️ Not recommended — poor performance First thing after install: open the Web console 🖥️ Visual management — from overview to creating a sandbox and streaming logs, all in your browser. After one-click deployment, open in your browser: http://:12088 Recommended three steps: Check overview — Open Overview, confirm nodes are Ready and capacity looks healthy Prepare a template — Install an official preset from Template Store; skip if you already have a READY template under Templates Create a sandbox — Sandboxes → + New sandbox, pick a READY template, and view live logs on the detail page within seconds See the full WebUI console guide. Deep Dive Documentation Home — complete guide navigation ☁️ PVM Deployment — deploy on ordinary cloud VMs without bare metal or nested virtualization Template Concepts — image-to-template concepts and workflows Example Projects — hands-on examples (code execution, browser automation, OpenClaw integration, RL training, and more) 🖥️ WebUI Console — visual management right after install (:12088) 🔐 Security Proxy & Credential Vault — CubeEgress domain filtering, injection, and auditing 🤖 Digital Assistant AgentHub — create and manage OpenClaw assistants (Preview) 💻 Development Environment (QEMU VM) — no KVM access? Try Cube Sandbox inside a disposable OpenCloudOS 9 VM Architecture | Component | Responsibility | |---|---| | CubeAPI | High-concurrency REST API Gateway (Rust), compatible with E2B. Swap the URL for seamless migration. | | CubeMaster | Cluster orchestrator. Receives API requests and dispatches them to corresponding Cubelets. Manages resource scheduling and cluster state. | | CubeProxy | Reverse proxy, compatible with the E2B protocol, routing requests to the appropriate sandbox instances. | | Cubelet | Compute node local scheduling component. Manages the complete lifecycle of all sandbox instances on the node. | | CubeVS | eBPF-based virtual switch, providing kernel-level network isolation and security policy enforcement. | | CubeEgress | OpenResty-based egress security gateway: L7 domain filtering, credential injection, and access auditing; works with CubeVS kernel policies so sandbox traffic cannot bypass inspection. | | CubeHypervisor & CubeShim | Virtualization layer — CubeHypervisor manages KVM MicroVMs, CubeShim implements the containerd Shim v2 API to integrate sandboxes into the container runtime. | 👉 For more details, please read the Architecture Design Document and CubeVS Network Model. Community & Contributing We welcome contributions of all kinds—whether it's a bug report, feature suggestion, documentation improvement, or code submission! 🐞 Found a Bug or have questions? Submit an issue on GitHub Issues. 💡 Have an Idea? Join the conversation in GitHub Discussions. 🛠️ Want to Code? Check out our CONTRIBUTING.md to learn how to submit a Pull Request. 📝 Want to contribute docs? Submit bilingual PRs to our community doc channels: Troubleshooting, Use Cases, and Integrations. 💬 Want to Chat? Join our Discord. Roadmap Coming soon — see the full roadmap for details. | Feature | Description | |---|---| | Kubernetes-Native Deployment | Deploy and operate CubeSandbox entirely within a K8s cluster using CRDs, operators, and native scheduling — no out-of-band orchestration | | Volume Support | Persistent and shared volume support compatible with the E2B volume protocol | | Cross-Node Pause & Resume | Suspend a sandbox on one node and resume it on another with full memory and filesystem state preserved | | E2B API Compatibility | Close remaining gaps with the E2B specification for full drop-in compatibility | | Control Plane / Data Plane Separation | Decouple the control plane from the data plane so control plane upgrades or failures never affect sandboxes already in flight | | Sandbox Fault Recovery | Automatic detection and recovery of crashed VMs, stuck shim processes, and network partitions with configurable recovery policies | | Scheduling & Operations Enhancements | Resource-aware placement, affinity rules, live rebalancing, and node drain with sandbox migration | License CubeSandbox is released under the Apache License 2.0. The birth of CubeSandbox stands on the shoulders of open-source giants. Special thanks to Cloud Hypervisor, Kata Containers, virtiofsd, containerd-shim-rs, ttrpc-rust, and others. We have made tailored modifications to some components to fit the CubeSandbox execution model, and the original in-file copyright notices are preserved. Cube Sandbox is listed in the CNCF Landscape.

Abseil - C++ Common Libraries The repository contains the Abseil C++ library code. Abseil is an open-source collection of C++ code (compliant to C++17) designed to augment the C++ standard library. Table of Contents About Abseil Quickstart Building Abseil Support Codemap Releases License Links About Abseil Abseil is an open-source collection of C++ library code designed to augment the C++ standard library. The Abseil library code is collected from Google's own C++ code base, has been extensively tested and used in production, and is the same code we depend on in our daily coding lives. In some cases, Abseil provides pieces missing from the C++ standard; in others, Abseil provides alternatives to the standard for special needs we've found through usage in the Google code base. We denote those cases clearly within the library code we provide you. Abseil is not meant to be a competitor to the standard library; we've just found that many of these utilities serve a purpose within our code base, and we now want to provide those resources to the C++ community as a whole. Quickstart If you want to just get started, make sure you at least run through the Abseil Quickstart. The Quickstart contains information about setting up your development environment, downloading the Abseil code, running tests, and getting a simple binary working. Building Abseil Bazel and CMake are the official build systems for Abseil. See the quickstart for more information on building Abseil using the Bazel build system. If you require CMake support, please check the CMake build instructions and CMake Quickstart. Support Abseil follows Google's Foundational C++ Support Policy. See this table for a list of currently supported versions compilers, platforms, and build tools. Codemap Abseil contains the following C++ library components: base The base library contains initialization code and other code which all other Abseil code depends on. Code within base may not depend on any other code (other than the C++ standard library). algorithm The algorithm library contains additions to the C++ `` library and container-based versions of such algorithms. cleanup The cleanup library contains the control-flow-construct-like type absl::Cleanup which is used for executing a callback on scope exit. container The container library contains additional STL-style containers, including Abseil's unordered "Swiss table" containers. crc The crc library contains code for computing error-detecting cyclic redundancy checks on data. debugging The debugging library contains code useful for enabling leak checks, and stacktrace and symbolization utilities. flags The flags library contains code for handling command line flags for libraries and binaries built with Abseil. hash The hash library contains the hashing framework and default hash functor implementations for hashable types in Abseil. log The log library contains LOG and CHECK macros and facilities for writing logged messages out to disk, stderr, or user-extensible destinations. memory The memory library contains memory management facilities that augment C++'s `` library. meta The meta library contains type checks similar to those available in the C++ `` library. numeric The numeric library contains 128-bit integer types as well as implementations of C++20's bitwise math functions. profiling The profiling library contains utility code for profiling C++ entities. It is currently a private dependency of other Abseil libraries. random The random library contains functions for generating pseudorandom values. status The status library contains abstractions for error handling, specifically absl::Status and absl::StatusOr. strings The strings library contains a variety of strings routines and utilities. synchronization The synchronization library contains concurrency primitives (Abseil's absl::Mutex class, an alternative to std::mutex) and a variety of synchronization abstractions. time The time library contains abstractions for computing with absolute points in time, durations of time, and formatting and parsing time within time zones. types The types library contains non-container utility types. utility The utility library contains utility and helper code. Releases Abseil recommends users "live-at-head" (update to the latest commit from the master branch as often as possible). However, we realize this philosophy doesn't work for every project, so we also provide Long Term Support Releases to which we backport fixes for severe bugs. See our release management document for more details. License The Abseil C++ library is licensed under the terms of the Apache license. See LICENSE for more information. Links For more information about Abseil: Consult our Abseil Introduction Read Why Adopt Abseil to understand our design philosophy. Peruse our Abseil Compatibility Guarantees to understand both what we promise to you, and what we expect of you in return.

agent multiplexer that lives in your terminal.

Extracted system prompts from Anthropic - Claude Fable 5, Opus 4.8, Claude Code, Claude Design. OpenAI - ChatGPT GPT-5.6, Codex GPT-5.6, GPT-5.5. Google - Gemini 3.5 Flash, 3.1 Pro, Antigravity. xAI - Grok, Cursor, Copilot, VS Code, Perplexity, and more. Updated regularly.

Never stop coding. Free AI gateway: one endpoint, 231+ providers (50+ free), connect Claude Code, Codex, Cursor, Cline & Copilot to FREE Claude/GPT/Gemini. RTK+Caveman stacked compression saves 15-95% tokens, smart auto-fallback, MCP/A2A, multimodal APIs, Desktop/PWA.

Orca Español · Português · 中文 · 日本語 · 한국어 The AI Orchestrator for 100x builders. Run Codex, ClaudeCode, OpenCode or Pi side-by-side — each in its own worktree, tracked in one place. Download Orca Features Mobile Companion Monitor and steer your agents from your phone — get notified when an agent finishes and send follow-ups from anywhere. iOS App Store · TestFlight · Android APK 0.0.27 · Docs → Parallel Worktrees Fan one prompt across five agents, each in its own isolated git worktree — compare the results and merge the winner. Docs → Terminal Splits Ghostty-class terminals with WebGL rendering, infinite splits, and scrollback that survives restarts. Docs → Design Mode Click any UI element in a real Chromium window to send its HTML, CSS, and a cropped screenshot straight into your agent's prompt. Docs → GitHub & Linear, Native Browse PRs, issues, and project boards in-app — open a worktree from any task and review without a context switch. Docs → SSH Worktrees Run agents on a beefy remote box with full file editing, git, and terminals — auto-reconnect and port forwarding included. Docs → Annotate AI Diffs Drop comments on any diff line and ship them back to the agent — review, edit, and commit without leaving Orca. Docs → Drag Files to Agents VS Code's editor with autosave everywhere — drag files or images straight into an agent prompt. Docs → Orca CLI Agents drive Orca too — script every workflow with orca worktree create, snapshot, click, and fill. Docs → Also in the box: Quick open — Search across worktrees, files, agents, commands, and repo context without leaving your flow. Account switcher & usage tracking — See Claude and Codex usage and rate-limit resets, and hot-swap accounts without re-logging in. Rich repo previews — Preview Markdown, images, PDFs, and repo docs in the workspace. Computer Use — Let agents operate desktop apps and visible UI when a workflow needs real interaction. Notifications and unread state — Know when an agent finishes or needs attention, then mark threads unread to come back later. And many, many more — we ship daily, so this list is perpetually behind. The changelog is the real feature list. Supported Agents Works with any CLI agent — if it runs in a terminal, it runs in Orca. Claude Code Codex Grok Cursor GitHub Copilot OpenCode MiMo Code Amp OpenClaude Antigravity Pi oh-my-pi Hermes Agent Devin Goose Auggie Autohand Code Charm Cline Codebuff Command Code Continue Droid Kilocode Kimi Kiro Mistral Vibe Qwen Code Rovo Dev any CLI agent Install Desktop — macOS, Windows, Linux Download from onOrca.dev Or grab a build directly: macOS Apple Silicon · macOS Intel · Windows (.exe) · Linux AppImage · All builds Running orca serve on a headless Linux server? See the headless Linux server guide. Or via a package manager: macOS (Homebrew) brew install --cask stablyai/orca/orca Arch Linux (AUR) — or stably-orca-git to build from source yay -S stably-orca-bin Mobile Companion — iOS, Android Pair with your desktop app to monitor and steer your agents from your phone. iOS: Download on the App Store or join TestFlight Android: Download APK 0.0.27 Community & Support Discord: Join the community on Discord. Twitter / X: Follow @orca_build for updates and announcements. WeChat: Scan the QR code to join the community. If the first group is full, use the backup group. Feedback & Ideas: We ship fast. Missing something? Request a new feature. Privacy: See the privacy & telemetry docs for what anonymous usage data Orca collects and how to opt out. Show Support: Star this repo to follow along with our daily ships. Developing Want to contribute or run locally? See our CONTRIBUTING.md guide. License Orca is free and open source under the MIT License.

/watch Give Claude the ability to watch any video. Claude Code (recommended — auto-updates via marketplace): /plugin marketplace add bradautomates/claude-video /plugin install watch@claude-video Codex, Cursor, Copilot, Gemini CLI, or any of 50+ Agent Skills hosts: npx skills add bradautomates/claude-video -g (-g installs globally for your user, available across all projects. Drop it to scope per-project.) More install options (claude.ai web, manual) in the Install section below. Zero config to start — yt-dlp and ffmpeg install on first run via brew on macOS (Linux/Windows print exact commands). Captions cover most public videos for free. Whisper API key is only needed when a video has no captions. Claude can read a webpage, run a script, browse a repo. What it can't do, out of the box, is watch a video. You paste a YouTube link and it has to either guess from the title or pull a transcript that's missing 90% of what's on screen. With Claude Video /watch you can paste a URL or a local path, ask a question, and Claude fetches captions first, downloads only what it needs, extracts frames (scene-aware, or fast keyframes at efficient detail), pulls a timestamped transcript (free captions when available, Whisper API as fallback), and Reads every frame as an image. By the time it answers, it has seen the video and heard the audio. /watch https://youtu.be/dQw4w9WgXcQ what happens at the 30 second mark? What people actually use it for Analyze someone else's content. /watch https://youtu.be/ what hook did they open with? Claude looks at the first frames, reads the opening transcript, breaks down the structure. Same for ad creative, competitor launches, podcast intros, anything where the how matters as much as the what. Diagnose a bug from a video. Someone sends you a screen recording of something broken. /watch bug-repro.mov what's going wrong? Claude watches the recording, finds the frame where the issue appears, describes what's on screen, often catches the cause without you ever opening the file. Summarize a video. /watch https://youtu.be/ summarize this does the obvious thing — pulls the structure, the key moments, what was actually said and shown. Faster than watching at 2x. Cut the hype out of an update video. /watch https://youtu.be/ what's actually new — skip the hype Strip a "game-changer" feature drop down to the few things that matter, so you get the substance without ten minutes of intro and overselling. Turn a playlist into notes. /watch https://youtu.be/ summarize this to a note Run it across a series and file a per-video summary, so a channel or course becomes a searchable set of notes instead of hours you have to sit through. How it works You paste a video and a question. URL (anything yt-dlp supports — YouTube, Loom, TikTok, X, Instagram, plus a few hundred more) or a local path (.mp4, .mov, .mkv, .webm). yt-dlp checks captions first. At transcript detail, captioned URLs return without downloading video. Otherwise, or when Whisper needs audio, it downloads only what the run needs. ffmpeg extracts frames at the chosen detail. efficient decodes keyframes only (near-instant); balanced/token-burner prefer scene-change frames and fall back to the duration-aware uniform sampler when they under-produce. JPEGs are 512px wide by default and clamped to 1998px tall for Claude Read compatibility. The transcript comes from one of two places. First try: yt-dlp pulls native captions (manual or auto-generated) from the source. Free, instant, accurate-ish. Fallback: extract a mono 16 kHz 64 kbps mp3 audio clip (~480 kB/min) and ship it to Whisper — Groq's whisper-large-v3 (preferred — cheaper and faster) or OpenAI's whisper-1. Frames + transcript are handed to Claude. The script prints frame paths with t=MM:SS markers and the transcript with timestamps. Claude Reads each frame in parallel — JPEGs render directly as images in its context. Claude answers grounded in what's actually on screen and in the audio. Not "based on the description" or "according to the title." It saw the frames. It heard the transcript. It answers the way someone who watched the video would. Cleanup. The script prints a working directory at the end. If you're not asking follow-ups, Claude removes it. Frame budget — why it matters Token cost is dominated by frames. Every frame is an image; image tokens add up fast. The script's auto-fps logic exists so you don't blow your context budget on a sparse scan of a 30-minute video that would have been better answered by a focused 30-second window. | Duration | Default frame budget | What you get | |----------|---------------------|--------------| | ≤30 s | ~30 frames | Dense — basically every key moment | | 30 s - 1 min | ~40 frames | Still dense | | 1 - 3 min | ~60 frames | Comfortable | | 3 - 10 min | ~80 frames | Sparse but workable | | > 10 min | 100 frames (capped modes) | "Sparse scan" warning — re-run focused, or --detail token-burner for full uncapped coverage | When the user names a moment ("around 2:30", "the last 30 seconds", "from 0:45 to 1:00"), pass --start / --end. Focused mode gets denser per-second budgets, capped at 2 fps. Far more useful than a sparse pass over the whole thing. Frame deduplication Frame selection — keyframes (efficient), scene-change detection (balanced/token-burner), or the uniform sampler it falls back to — can still surface near-identical frames: a screen recording that holds one slide for 90 seconds produces a dozen, each billed as a separate image. A dedup pass drops them before frames reach Claude. It runs by default on every frame mode (--no-dedup turns it off): One ffmpeg call scales each extracted JPEG to a 16×16 grayscale thumbnail. Everything after is pure-stdlib Python — no image libraries. For each frame, compute the mean absolute difference against the last frame that was kept (average per-pixel brightness change, 0–255 scale). If that difference is at or below the threshold (2.0), the frame is a near-duplicate and is dropped. Otherwise it's kept and becomes the new reference. The frame-budget cap applies after dedup, so the budget is spent on distinct frames. Comparing against the last kept frame (not the previous one) catches slow fades that never trip a frame-to-frame threshold. The threshold is deliberately low and measures absolute brightness rather than structure, so a one-line code diff, a terminal scrolling a row, or two differently-colored flat slides all survive. The Frames line reports what was collapsed, e.g. 6 selected from 14 candidates (… 8 near-duplicates dropped …). On always-moving footage nothing is dropped and you pay what you would have anyway. Detail modes — measured The --detail dial trades speed and token cost for visual fidelity. Numbers below are from a real run against a 49:08 YouTube video (1280×720, English auto-captions) — a long, mostly-static screen recording, the case that stresses the caps hardest. Extraction times are local CPU against a pre-downloaded copy; the one-time download was ~37 s / 76 MB, shared by the three frame modes. | Mode | Engine | Frames | Cap | Extraction time | Temporal coverage | Est. image tokens | |------|--------|--------|-----|-----------------|-------------------|-------------------| | transcript | none (captions) | 0 | — | ~4.5 s (one yt-dlp call, no download) | full (text) | 0 (≈26.6k text tokens) | | efficient | keyframe (-skip_frame nokey) | 50 | 50 | 0.5 s | 0:00 → 49:04 (full) | 9.8k | | balanced | scene-change | 100 | 100 | 20.9 s | 0:00 → 48:38 (full) | 19.7k | | token-burner | scene-change | 116 | uncapped | 21.0 s | 0:00 → 48:38 (full) | 22.8k | Image tokens use Anthropic's (width × height) / 750 — at the default 512px width these 720p frames are 512×288, ≈197 tokens/frame; --resolution 1024 roughly 4×s that. The transcript is surfaced in every captioned mode and on long videos is often the larger cost. One sampling rule across frame modes. Each detects all candidates across the full range, then even-samples (first + last always kept) down to its cap. The modes differ only in candidate source (keyframes vs. scene cuts) and cap, never in how coverage is spread — so the last frame always lands at the end, not partway through. efficient is the speed tier (0.5 s) — it only reconstructs keyframes, so it's 40× faster than the scene modes, which decode every frame to find cuts. It can also return more frames than balanced on low-motion footage (keyframes outnumber scene cuts); "efficient" means fast extraction, not fewer frames. token-burner only diverges from balanced past the cap. This clip had 116 cuts, so balanced sampled 100 and token-burner kept all 116. On high-motion video with hundreds of cuts, token-burner keeps everything (and trips the >250-frame token warning) while balanced thins to 100. End-to-end from a cold URL, transcript is the cheapest mode by far; the frame modes add the shared ~37 s download on top of the extraction times above. Install | Surface | Install | |---------|---------| | Claude Code | /plugin marketplace add bradautomates/claude-video then /plugin install watch@claude-video | | Codex, Cursor, Copilot, Gemini CLI, +50 more | npx skills add bradautomates/claude-video -g | | claude.ai (web) | Download watch.skill → Settings → Capabilities → Skills → + | | Manual / dev | git clone then symlink skills/watch into your host's skills dir (see below) | Claude Code /plugin marketplace add bradautomates/claude-video /plugin install watch@claude-video Update later with /plugin update watch@claude-video. Codex, Cursor, Copilot, Gemini CLI, and 50+ other hosts The Agent Skills CLI installs the skill into whatever agents it detects: npx skills add bradautomates/claude-video -g -g installs globally for your user (/.codex/skills, /.cursor/skills, etc.); drop it to install into the current project instead. Useful flags: -a, --agent — target specific hosts, e.g. -a codex -a cursor -l, --list — list the skills in this repo without installing --copy — copy files instead of symlinking (for filesystems without symlink support) The CLI discovers the skill from skills/watch/SKILL.md and copies the whole folder — SKILL.md plus its scripts/ runtime — as a self-contained unit. SKILL.md resolves its own scripts relative to wherever it was installed, so it works the same on every host. Update later with npx skills update watch -g. claude.ai (web) Download watch.skill from the latest release. Go to Settings → Capabilities → Skills. Click + and drop the file in. Enable "Code execution and file creation" under Capabilities first — the skill shells out to ffmpeg and yt-dlp, so it won't run without it. Manual (developer) Clone the repo and symlink the self-contained skill folder into your host's skills directory — the symlink keeps the install in sync with your working tree as you edit: git clone https://github.com/bradautomates/claude-video.git ln -s "$(pwd)/claude-video/skills/watch" /.claude/skills/watch # or /.codex/skills/watch For claude.ai, build the .skill bundle from source: bash skills/watch/scripts/build-skill.sh produces dist/watch.skill. First run On the first /watch call, the skill runs scripts/setup.py --check. If ffmpeg / yt-dlp aren't on your PATH, or no Whisper API key is set, it walks you through fixing it: macOS — auto-runs brew install ffmpeg yt-dlp. Linux — prints the exact apt / dnf / pipx commands. Windows — prints the winget / pip commands. API key — scaffolds ~/.config/watch/.env (mode 0600) with commented placeholders for GROQ_API_KEY (preferred) and OPENAI_API_KEY. After setup, preflight is silent and /watch just works. The check is a sub-100ms lookup, so it doesn't slow you down on subsequent runs. Bring your own keys Captions cover the majority of public videos for free. The Whisper fallback only kicks in when a video genuinely has no caption track — typically local files, TikToks, some Vimeos, and the occasional caption-less YouTube upload. | Capability | What you need | Cost | |------------|---------------|------| | Download + native captions | yt-dlp + ffmpeg | Free | | Whisper fallback (preferred) | Groq API key — whisper-large-v3 | Cheap, fast | | Whisper fallback (alt) | OpenAI API key — whisper-1 | Standard pricing | | Disable Whisper entirely | --no-whisper | Free, frames-only when no captions | Usage /watch https://youtu.be/dQw4w9WgXcQ what happens at the 30 second mark? /watch https://www.tiktok.com/@user/video/123 summarize this /watch ~/Movies/screen-recording.mp4 when does the UI break? /watch https://vimeo.com/123 what tools does she mention? Focused on a specific section — denser frame budget, lower token cost: /watch https://youtu.be/abc --start 2:15 --end 2:45 /watch video.mp4 --start 50 --end 60 /watch "$URL" --start 1:12:00 # from 1h12m to end Other knobs (passed to scripts/watch.py): --detail transcript|efficient|balanced|token-burner — fidelity/speed dial. transcript skips frames (transcript only); efficient uses fast keyframes (cap 50); balanced uses scene-aware frames (cap 100); token-burner is scene-aware and uncapped. --timestamps T1,T2,… — grab a frame at each absolute timestamp (SS/MM:SS/HH:MM:SS). Claude reads the transcript first, then targets the moments the presenter flags ("look here", "as you can see"). Added on top of the detail frames (reserved against the cap); out-of-window cues are dropped in focus mode; with --detail transcript these become the only frames. --max-frames N — lower the frame cap for a tighter token budget. --resolution W — bump frame width to 1024 px when Claude needs to read on-screen text (slides, terminals, code). --fps F — override the auto-fps calculation (still capped at 2 fps). --whisper groq|openai — force a specific Whisper backend. --no-whisper — disable transcription entirely; frames only. --no-dedup — keep near-duplicate frames. By default a frame-delta pass drops frames that are visually near-identical to the one before them (held slides, static screen recordings, paused video), so the frame budget is spent on distinct content; this flag turns that off. --out-dir DIR — keep working files somewhere specific (default: auto-generated tmp dir). Limits Long-video accuracy depends on the detail mode. On the capped modes (efficient, default balanced) coverage thins out past ~10 minutes — the frame cap spreads across the whole clip, so the script prints a "sparse scan" warning and you're better off re-running focused with --start/--end. token-burner lifts the cap and keeps every scene-change frame across the full video, so it stays complete on longer clips at the cost of more image tokens. The 10-minute mark is guidance for the capped modes, not a hard ceiling. Detail is one dial. Defaults are balanced: scene-aware frames, 2 fps max, 100-frame cap. Use --detail efficient for a fast 50-frame keyframe pass, or --detail token-burner for uncapped scene candidates. Set WATCH_DETAIL in ~/.config/watch/.env to change the default. Structure . ├── skills/watch/ # self-contained skill — copied as a unit by every installer │ ├── SKILL.md # skill contract — the source of truth across all surfaces │ └── scripts/ │ ├── watch.py # entry point — orchestrates download → frames → transcript │ ├── download.py # yt-dlp wrapper │ ├── frames.py # ffmpeg frame extraction + auto-fps logic │ ├── transcribe.py # VTT parsing + dedupe + Whisper orchestration │ ├── whisper.py # Groq / OpenAI clients (pure stdlib) │ ├── config.py # shared config (~/.config/watch/.env) │ ├── setup.py # preflight + installer │ └── build-skill.sh # build dist/watch.skill for claude.ai upload (dev-only) ├── hooks/ # SessionStart status hook (Claude Code only) ├── .claude-plugin/ # plugin.json + marketplace.json (Claude Code) ├── .codex-plugin/ # plugin.json — Codex/agents manifest ("skills": "./skills/") ├── .agents/plugins/ # marketplace.json — Agent Skills marketplace listing ├── AGENTS.md → CLAUDE.md # generic-agent entry point ├── tests/ # pytest suite (ffmpeg-synthesized clips, no network) └── .github/workflows/ # release.yml — auto-builds watch.skill on tag push Develop Run the test suite (stdlib + pytest; ffmpeg required for frame tests): python3 -m pytest -q Build the claude.ai upload bundle: bash skills/watch/scripts/build-skill.sh # → dist/watch.skill Releasing: tag vX.Y.Z, push the tag. The workflow builds dist/watch.skill and attaches it to the GitHub release. Keep the version in sync across skills/watch/SKILL.md, .claude-plugin/plugin.json, and .codex-plugin/plugin.json. See CHANGELOG.md for version history. Open source MIT license. Built on yt-dlp, ffmpeg, and Claude's multimodal Read tool. Whisper transcription via Groq or OpenAI. Built by Brad Bonanno — I make content about building with AI on YouTube (@bradbonanno), and build AI operating systems for businesses at Solaris Automation. If /watch saves you from scrubbing through a video, come say hi on the channel. Star History github.com/bradautomates/claude-video · @bradbonanno · Solaris Automation · LICENSE