Start with the job, then find an open-source answer

skill-validator is a CLI for checking Agent Skill packages beyond basic specification compliance. It validates structure and frontmatter, detects broken or orphaned references, flags unnecessary files, reports token usage, and analyzes instruction density and cross-language contamination. Optional model-based evaluation can score dimensions such as clarity, actionability, and novelty. The tool can be installed through Homebrew or Go, used as a pre-commit check, integrated into CI, or imported as a Go library for custom validation workflows.

AgenticX combines a Python SDK, the agx CLI, a Studio server, and the Machi desktop app into one platform for building and operating multi-agent systems. It supports focused automation assistants, graph-based workflows, and collaborating agent teams, with multiple model providers, MCP tools, hierarchical memory, managed skill lifecycles, sandboxed execution, and session management. Developers can embed the core ReAct agent primitive or manage runs through the studio and server interfaces.

ToolGen represents tools as dedicated vocabulary tokens so a model can select tools, call them, and continue generating language within one process. The repository includes models, data for its three-stage training procedure, and evaluation code for research reproduction. Its local inference example also requires a ToolBench key and a StableToolBench deployment.

Fokkyp/claude-skills packages two reusable workflows for product managers: a PRD generator and a competitive-analysis skill for individual products or multi-product comparisons. Users copy the selected folder into a personal or project-level Claude Code skills directory, then invoke it through natural language or slash commands. Deep comparisons can consume substantial context, so the repository recommends a model with roughly a one-million-token context window or a narrower, multi-stage research scope.

grok-skill lets Codex or Claude send research queries to an OpenAI-compatible Grok endpoint and receive structured JSON containing both an answer and its sources. It is aimed at time-sensitive checks such as software versions, release dates, documentation, installation details, and troubleshooting. Configuration can be supplied through local overrides or environment variables. The included one-click installer targets Windows PowerShell, and the skill still requires you to provide a working 2api or other compatible Grok endpoint.

EdgeMark keeps a Markdown notes panel one gesture or global shortcut away from any macOS app. You can pin the panel while copying between windows, then organize plain `.md` files with folders, tags, search, previews, and batch actions without locking them into a proprietary database. The app requires macOS 15.7 or later.

pitchcraft structures project communication around Hook, Context, Proposal, Evidence, and Ask. Before drafting, it prompts for missing material, the people in the room, and what those people care about, then adapts the result for kickoffs, status updates, wrap-ups, investor pitches, or solution selling. The skill is designed to secure understanding and support rather than merely summarize completed work, and can be used through several coding agents or pasted into web chats.

comine brings video, audio, file, torrent, and gallery downloads into one cross-platform application. It is available through GitHub releases for Windows, Linux, macOS, and Android, with the stable release recommended for most users. The project is still in early development, so users should expect missing features and occasional bugs.

Use this Codex skill after a single-slide reference image has established the visual direction. It crops complex illustrations, icons, or maps that are impractical to redraw, while recreating titles, body text, cards, arrows, labels, and conclusion bars with editable PowerPoint objects. The workflow works page by page from individual PNG references and explicitly avoids placing the whole screenshot as a flat slide background, preserving editability where it matters.

sql_to_ER is a browser-based, front-end-only tool that turns SQL CREATE TABLE statements or DBML into Chen-model entity-relationship diagrams. Users can drag nodes, edit them by double-clicking, zoom or rotate the canvas, and run an automatic layout pass after generation. The hosted version requires no sign-in or installation, and the project can also be installed as an Agent Skill. It targets Chen diagrams rather than logical models, and local source use must run through Vite instead of opening the HTML file directly.

MDFlux is a Windows desktop app for turning PDFs, Office files, ebooks, web formats, images, and audio into structured Markdown without using a terminal. Built-in OCR recovers text from scanned or image-only PDFs, and folder batching provides progress and per-file diagnostics. The first launch needs internet access to prepare a local Python environment; later conversions run offline, while macOS support is still only planned.

This reference architecture shows how to build a Strands Agents chatbot on Amazon Bedrock AgentCore with persistent memory, MCP-connected tools, code execution, browser automation, voice interaction, and A2A delegation to specialist agents. Terraform modules and a working frontend/backend structure make it useful for teams evaluating or extending advanced agent workflows on AWS.

These skills help compatible agents write, edit, and debug Typst by consulting bundled documentation and examples instead of guessing newer syntax. The general `typst-author` skill covers ordinary documents, while `touying-author` focuses on presentation decks, animations, and themes. Installation requires cloning the repository and copying the desired skill directories into an assistant that supports the Agent Skills standard.

Polymarket Agent Skills gives developers and coding assistants structured guidance for authentication, orders, market data, WebSocket streams, positions, token split or redemption flows, bridging, and gasless transactions. The main SKILL.md keeps common endpoints and patterns compact, while detailed references load only when needed. Any workflow that executes trades still requires correctly secured signing keys and platform credentials.

UI Skill Lab is a reference workflow that turns a PRD into a visible UI target before an agent writes code. After review, it extracts visual specifications, tokens, layout rules, a component tree, evidence, and prohibited assumptions for Codex or Claude Code to implement. Screenshot acceptance and focused diff tasks guide repairs. It works best for HTML/CSS-heavy dashboards, SaaS products, admin interfaces, mobile screens, and ordinary landing pages. It is not a complete UI generator and does not automatically slice images or export Figma, PSD, or Sketch assets.

distil-cli-skill teaches Claude to run a Distil training workflow for task-specific small language models. It covers datasets for classification, question answering, tool calling, and multi-turn tool use, then creating a model, uploading data, evaluating with a teacher model, and starting training. Afterward, it can guide downloads and deployment through llama.cpp, vLLM, or Distil-managed infrastructure. It is an instruction package for the Distil CLI and platform, so training still requires the corresponding account and service access.

This library packages repeatable agent workflows for research, intelligence gathering, design, and content creation. Its catalog includes Hacker News and Product Hunt research, app-store intelligence, trademark checks, motion patterns, design tokens, creative direction, and visual QA. Users can install the entire collection or a selected skill through npx, with support for Claude Code, OpenClaw, Codex, Cursor, and similar tools; some integrations are keyless, while others require service-specific API credentials.

Tensorlake Skill is a lightweight instruction package that tells Claude Code, Codex, and similar agents to verify Tensorlake APIs against current documentation before suggesting code. It fetches the live documentation index, selects relevant pages for sandboxes, orchestration, or SDK tasks, and falls back to a bundled snapshot only when fetching fails. This is guidance rather than a runtime tool server: using Tensorlake still requires the appropriate Python, TypeScript, or CLI installation and the user's own API configuration.