Find the open-source solution that fits.

SkillWard triages Agent Skills in three stages: rule-based static analysis, LLM-assisted intent evaluation, and isolated Docker execution for unresolved cases. The runtime guard records evidence of suspicious network access, sensitive writes, credential risks, and other behavior. It supports single or batch CLI scans, with an optional web interface for reviewing results.

jspaint.exe packages JS Paint as a desktop application for a nostalgic, classic Microsoft Paint-style drawing workflow with added features. Releases are available for macOS, Windows, and Linux, although the documented testing coverage differs by platform and the Linux build has not been tested recently. The app uses Chrome to display its web interface and may download and install Chrome when it is missing.

Collect single-choice, multi-choice, and free-form questions from parallel AI agents in a separate terminal inbox, then route your answers back to the correct sessions. AUQ integrates through MCP, an OpenCode plugin, or Agent Skills and works over SSH with native notifications; Bun is recommended for its default terminal renderer.

This repository packages 20 modular Agent Skills for idiomatic Go development, distilled from Google, Effective Go, Uber, and community guidance. Coding agents can load procedures for code review, concurrency, context handling, errors, interfaces, naming, performance, testing, and related decisions, supported by 48 references and eight check scripts. The collection works across multiple coding tools through the common Agent Skills installation flow. Version-sensitive recommendations are tracked separately so users can check minimum Go releases and fallbacks.

clideck puts Claude Code, Codex, Gemini CLI, OpenCode, and other real terminal sessions into one chat-style local dashboard. It adds live status, previews, projects, notifications, session resume, and agent-to-agent requests without rewriting prompts or outputs. An end-to-end encrypted mobile relay enables remote control while agents stay on the host machine. macOS and Windows are supported; Linux remains untested.

video-editing-skill gives OpenClaw and Codex a local workflow for talking-head, interview, and vlog footage. It can transcribe speech, split sentences, choose and merge clips, burn subtitles, and prepare plans for short-form batches, hook variants, jump-cut fades, picture-in-picture overlays, and vertical reframing. Optional YOLO analysis can identify scenes or subjects, while a lighter mode produces sampled frames for review. It suits creators who want a script-driven process with JSON and Markdown artifacts, not a fully graphical one-click editor.

Math Modeling Skills provides one skill for solving and another for paper writing across CUMCM and MCM/ICM problem types. It helps teams decompose a prompt, search literature, select models, use algorithm cookbooks and code templates, then move into abstracts, validation, figures, English polishing, and formatting checks. It is a structured competition workflow rather than a substitute for verifying the mathematics, code outputs, and final argument.

Wreckit takes a text backlog and moves each item through codebase research, planning, story-by-story implementation, and pull request creation for you to review. State and artifacts stay in Git-trackable JSON and Markdown files, making runs inspectable and resumable. It requires an AI agent or API access and can consume substantial tokens; an optional Sprite-based sandbox isolates riskier execution.

This skill turns a working prototype into a structured Figma review artifact. It creates a frame for each meaningful interaction state, maps elements to available design-system components, and adds flow arrows, conditions, edge cases, annotations, and optional Code Connect links. It requires Figma MCP tooling, and clients with limited tool support may produce a specification document instead of building the file directly.

Agentlas OS and its Hephaestus engine compile plain-language requests into portable agent or team packages with roles, tools, memory boundaries, permissions, routing, and verification contracts. Users can borrow public Hub specialists or store owned packages in a private Agent Cloud, while the chosen host executes them. Credentials, local files, and machine permissions stay local.

Moonshine-Image is a local Windows desktop tool for removing objects, text, and watermarks from images and video. Users can draw masks for Lama or MAT inpainting, use SLBR for visible semi-transparent watermarks, process folders in batches, and edit video masks on a timeline with resumable segments and FFmpeg fallback export. Releases are split into CPU and CUDA packages with bundled or external models; CPU processing is much slower, while MAT requires CUDA and its weights are licensed for non-commercial use only.

Octagon Skills provides agent workflows for financial statements, growth metrics, valuation, stock performance, market data, and broader company research. The skills are installed through npx, a Claude plugin, or manual copying, while live data access is supplied by the separate Octagon MCP service. Users therefore need to configure that MCP connection and an Octagon API key before the data-driven workflows can run as intended.

EasyClaw is a desktop installer that reduces OpenClaw setup to downloading the app, running the installer, and entering an AI provider key. It can detect and install required components such as WSL, Node.js, and OpenClaw on Windows, and it also supports connecting the resulting agent to a Telegram bot. Builds are available for Intel and Apple Silicon Macs as well as Windows, although Windows may still display a security warning while code signing is pending.

Open Record/Replay captures a demonstrated macOS workflow as structured mouse, keyboard, window, selection, and accessibility events. The resulting session can be validated and packaged as input for an agent to create a reusable skill, which is useful for cross-application tasks without a clean API. The project is alpha software, requires macOS permissions, and does not include screenshots in its current core recording path.

MapLibre Agent Skills gives coding assistants curated context for decisions that commonly go wrong in mapping projects. Current skills cover choosing tile sources, building serverless PMTiles workflows, and migrating Mapbox GL JS applications to MapLibre. They are plain Markdown files that can be installed together or individually with `npx skills`, or copied into an agent's project instructions; the repository supplies guidance rather than a map runtime.

Loom combines a XeLaTeX document class with a Claude skill for creating notes that students both read and complete. It supports blanks, proof skeletons, do-it-yourself exercises, recall prompts, and a distinctive woven visual system; the companion skill can organize a chapter, draft the notebook, compile it, and check the result. Local use requires XeLaTeX with TeX Live 2023 or newer, and non-macOS users may need to swap the bundled system fonts or use one of the Overleaf-ready editions.

Reframed combines screen capture and editing in one macOS application. It records a display, window, region, or connected iOS device with system audio, microphone, and an optional webcam overlay, then opens the result directly in a timeline editor. Users can trim tracks, reduce microphone noise, generate captions, and add cursor-following zooms before exporting MP4, MOV, or GIF files with configurable resolution, frame rate, and platform presets.

Product Manager Skills equips Claude Code, Codex, Cursor, and Windsurf with reusable product-management workflows and reference material. It can turn vague requests into measurable problem framing, analyze SaaS health, critique PRDs and roadmaps, pressure-test priorities, support discovery, and coach career transitions. The core is inspectable Markdown for local or repository-based use, not a collaborative web application with approvals, comments, or sharing.