Recently updated open-source projects

Use PRISM for hard-to-reverse choices in architecture, security, or multi-agent design. Prepare one technical brief, send it independently to three or more models without sharing their answers, validate every claim against the actual code, and let a human choose the outcome. The framework then records the decision in an ADR and tracks recurring hallucinations as institutional memory. Reversible, low-impact decisions are intentionally left to a lighter workflow.

Launchpad coordinates wave-planner, plan-review, Linear issue creation, and hive-mind-execution as a four-stage Claude Code workflow with explicit gates. It can resume from the last completed stage, start from a chosen phase, skip review, or route infrastructure work differently, and verifies that Linear issues exist before execution. The companion skills and integrations are required.

Use this lightweight skill when you already know the gameplay-video workflow and want a concise output. It delivers six image prompts, video prompts, and a minimal UI dictionary while keeping four-view character references, HUD anchors, opening-frame guidance, and aspect-ratio constraints. It omits process explanations, checklists, generation steps, and post-production notes to save context.

Use this Claude Code skill to run Screaming Frog headless crawls, query existing Derby databases, diff crawl results, and render client-facing HTML audit reports. The bundled checks cover broken links, redirect chains, titles, H1s, thin content, page size, noindex pages with inlinks, and orphan candidates. Fresh headless crawls require a paid Screaming Frog license; queries and reports can use existing GUI-produced crawls.

Use this skill to build a keyword hierarchy and search-focused content plan for Douyin or TikTok videos. It helps creators look beyond recommendation feeds by reviewing how titles, phrasing, and topic coverage align with likely search queries. The workflow is intended for short-video creators, social media operators, and local businesses seeking more discoverable content.

nudgr keeps reminders, tasks, and scheduled Claude Code runs in a local SQLite database, then mirrors them to Calendar.app or Google Calendar through several sync options. Platform timers can launch prompts in detached tmux sessions and surface or recover missed runs after a sleeping laptop wakes. It targets macOS, Linux, and WSL2, remains pre-1.0, and does not support native Windows.

This standalone agent skill scans local files, lockfiles, processes, persistence points, containers, and CI surfaces for indicators from documented npm and PyPI worm campaigns. Its key safety rule is to stop persistence and quarantine evidence before revoking credentials, then release a dependency-ordered rotation plan only after a clean second scan. It supports macOS, Linux, Windows, WSL, containers, and CI, but remains an incident-response runbook rather than a guarantee of safety.

This experimental Claude Code skill helps solo builders surface the production concerns that an MVP can hide. Before coding, it asks four classification questions, assigns a risk tier, and can guide assumptions, security-tool checks, critical-path walkthroughs, handoff documents, and a cool-down before launch. It does not modify code, enforce deployment controls, or provide legal, compliance, or security certification; the skill content is currently German.

compbio-pi gives Claude Code an opinionated decision framework for routine computational-biology work, answering with a decision, reason, and next step instead of a menu of options. It favors simple baselines, reproducibility, interpretability, targeted ablations, and reviewer-style stress tests. The skill still asks the researcher before changing a core scientific claim or journal tier, purchasing new data or compute, seeking collaborators, or handling ethics and data agreements.

Use this Japanese-first Claude Code skill to design, implement, or review presentation dashboards with HTML and ECharts. It organizes requirements through 5W1H, uses a six-part grid, recommends four basic chart types, and checks contrast and accessibility with a 26-item review flow. The scope is presentation dashboards rather than exploratory analysis, and the included guidance and templates are primarily Japanese.

Use mdrise when a README, report, or article needs a polished web version without changing its wording. The Claude skill chooses a visual treatment based on the document type, adds a floating table of contents, supports many writing systems, and embeds a source hash so unchanged Markdown is not rendered again unnecessarily.

ndemo converts a YAML playbook into a narrated demo of a web application. Claude Code inspects the page's accessibility tree, authors and tests browser actions segment by segment, and then renders an MP4 with text-to-speech narration; it requires Node.js 20+, ffmpeg, Claude Code, and an OpenAI API key for voice generation.

Process clinical dental photos into white-balanced images, AACD-view crops, before/after composites, and a 12-photo accreditation board. The Claude Code version runs locally with RAW and JPEG support, while the ChatGPT version accepts JPEG uploads only. Follow applicable patient-privacy rules; the toolkit performs photo post-processing rather than diagnosis or treatment advice.

Use this Claude Code skill when a human-facing result benefits from sections, comparisons, or visual structure. It creates one self-contained HTML file with inline CSS, JavaScript, and SVG, using a warm editorial style and responsive layout; the file works offline and is opened in a browser. Short answers, linear explanations, and code-only output remain Markdown.

Resume Craft guides Claude Code through a seven-stage workflow from job-description analysis and information collection to layout, length checks, and export. It can work from pasted details, local files, or GitHub repository information, then produce HTML, PDF, DOCX, and Markdown resumes using discipline templates and visual themes. Profiles and generated files stay in local folders, while rendering requires the listed Python and Node.js dependencies.

Measure tempo from MP3, WAV, FLAC, M4A, OGG, and similar audio files with a two-stage analysis. A coarse pass checks the metric level, then a fine pass refines the BPM within a constrained range; the JSON output includes the primary BPM, metric-level BPM, result status, alternative half/double or 3:2 candidates, and the first 20 beat timestamps. It uses librosa and soundfile.

Point Claude at a function or failing request, and the plugin reads nearby code, selects strategic breakpoints, attaches through the Chrome DevTools Protocol, and waits while you trigger the code path. It then inspects locals, expressions, and the call stack before reporting findings in plain language. The debugger is Node.js-only and requires Node.js 22, a service started with --inspect, and a usable attach configuration in .vscode/launch.json.

ADVPL Dev gives Claude domain guidance for building on TOTVS Protheus, including naming and source conventions, database locks and transactions, MVC entry points, REST APIs, error handling, performance, essential functions, SX tables, and common failures. It is distributed as an installable Claude skill with reference material compiled from TOTVS documentation.