Recently updated open-source projects

opus-mind audits CLAUDE.md-style system prompts against 11 structural checks, while BOOST identifies missing context, constraints, verification, and decomposition in everyday requests. It runs as a Claude Code skill or deterministic CLI without model calls or API keys. The score reflects prompt structure rather than semantic correctness, security, or multi-turn behavior.

lights-hooks merges Claude Code lifecycle hooks into `~/.claude/settings.json` so the macOS Lights app can show red while work runs, yellow for permission states, and green when idle. It saves a timestamped backup before changing existing hooks and removes only its own port-9876 entries on uninstall. The setup requires macOS, Lights.app running locally, and Claude Code.

Send one or more common audio or video files to Claude Code and receive structured Markdown with detected language, estimated duration, topics, 4–15-second timestamped segments, speaker labels, and uncertainty notes. Processing can run in parallel and supports combined or JSON output, but it requires a Gemini API key and uses generic speaker numbers.

Use this Claude Code skill to audit Metabase SQL, correct cards, simplify wide tables, and wire dashboard filters through all required mapping layers. It emphasizes semantic checks such as duplicate counting, snapshot-versus-history confusion, and unsafe joins, while preserving visualization settings during updates. Warehouse table names must be adapted to the target schema.

Add persistent goal tracking to Claude Code: define a goal, break it into subtasks, save state, and resume work across conversations until the goal is complete or paused. The skill stores state in a local JSON file and supports status, pause, resume, and cancel commands; it is a user-level workflow rather than native runtime state.

Generate US Letter worksheet PDFs for kindergarten math, alphabet tracing, sight words, CVC phonics, and cursive practice. The LITE skill also includes listing-thumbnail and listing-copy helpers, but not the Pro edition’s larger catalog, KDP cover generator, or auto-bundling. Python, ReportLab, and Pillow are required.

Every coding task is classified as XS, M, or L after a branch and backlog preflight. Small fixes get a smoke test, while larger changes require a user-visible result description, an RFC, stronger review, and progressively deeper security checks; parallel batches can be dispatched to isolated worktrees. This is an opinionated policy skill rather than a code framework, and its main instructions are written in Russian.

This universal skill routes plain-language requests to LaunchMyStore surfaces: the lms CLI, App Bridge, WASM Functions, Aqua themes, or the merchant MCP server. It covers app scaffolding, admin UI, checkout logic, theme authoring, and store operations, so users can work through one intent-driven entry point.

Elements is a focused desktop application for displaying the chemical periodic table in its own window. Prebuilt downloads are provided for macOS, Windows, and Linux, making it useful when a simple standalone reference is preferable to a browser page. Running or building the source requires Node.js, Git, and Bower.

Use this collection for short-form educational video production across TikTok, YouTube Shorts, Reels, and Xiaohongshu. Separate skills cover spoken scripts, pronunciation drills, vertical slides, cover thumbnails, and Descript export polishing, while the included SOP connects them from a locked script through publishing.

Give the skill a trusted PPTX, PDF, HTML deck, older presentation, or screenshots, and it extracts a compact style profile for Codex or Claude to follow when creating or revising HTML slides. Supporting checks look for layout overflow, overlap, and visible drift from the reference style. It is aimed especially at academic presentations, but it does not produce an editable native PPTX or promise a high-fidelity PowerPoint reconstruction.

Render a Markdown report, specification, or page draft as HTML, then highlight text, draw region boxes, add comments, or test CSS changes in the browser. The tool exports only new annotations as one structured prompt for the next Claude iteration and tracks their status locally across review rounds. Annotation works from a static file with no browser extension, MCP connection, or server; converting Markdown to HTML requires Python with the markdown package.

This skill creates self-contained HTML explainers in which inline SVG diagrams carry the meaning and prose stays caption-like. It plans 6–13 sections from reusable patterns such as flows, timelines, decision trees, and risk matrices, supports five baked-in themes, and is meant for durable reference documents rather than marketing pages, essays, decks, or photo-heavy work.

job-hunter combines multi-source job discovery, match scoring, a local Kanban tracker, salary analysis, ATS form filling, and per-job resume tailoring in a CLI and localhost web app backed by SQLite. Sensitive profile values and session cookies stay outside model context. Applications can pause in shadow mode, while automatic submission requires reliability checks and explicit consent.

NorthData Connectors wraps the NorthData API with a Python CLI, an MCP server, and a Claude skill that share one client and credit guard. It exposes German company profiles, owners, representatives, financials, and publications to shell scripts, CI jobs, and MCP-capable agents without duplicating API logic. A NorthData API key is required, usage remains subject to NorthData's terms, and the wrapper is an unofficial community project.

This experimental skill turns a plain-language Revit request into a pyRevit handler and MCP manifest, validates it, installs it in a personal library, and tests it against the open project. It targets Windows with Revit 2024 and pyRevit 4.8+, has only been verified on Revit 2024, and should be tried on test models first.

m_AI_l coordinates ongoing projects with a roster, per-thread mailboxes, and archives stored as ordinary Markdown files. Each Claude conversation can leave messages, act on its inbox, and hand work to a successor when its context fills, without a database or custom server. It is aimed mainly at non-developers seeking minimal setup and requires Claude Desktop with filesystem MCP access; the web version is not supported.

Turn a research corpus into atomic claims with A–G evidence grades, source locations, and cross-references. The skill surfaces contradictions, inflated claims, gaps, and downstream content seeds, with larger corpora handled in different extraction modes. It produces a queryable inventory rather than prose and works only from documents you provide.