Recently updated open-source projects

decide researches current industry practice for an explicit implementation question, then acts autonomously when evidence converges on a reversible choice. When sources disagree, it asks for a factual piece of project context instead of presenting an arbitrary preference poll. This repository is now a legacy mirror whose active development moved to Decision Kernel, and its workflow depends on external research tools.

This Claude Skill prototype creates a 20-question multiple-choice English reading session from two or three authentic passages. Questions cover tone, main idea, vocabulary, references, restatement, analogy, and other UTBK/TOBK literacy skills, with a 20-minute timer, one-question navigation, automatic scoring, category breakdowns, and answer review. It relies on Claude Skills and Visualizer, works best in a desktop browser, and may not render well on mobile devices.

Ingest a resume, GitHub repository, or other documents into a reusable experience bank, then tailor an HTML resume to a job description with weighted keywords, STAR rewrites, and a match report. A mock interview probes the same claims from basics to technical and behavioral depth. The skill refuses invented metrics or credentials; browser-only use cannot persist the bank on your computer.

The skill asks which Apple platform, framework, and interface you are building before loading only the relevant HIG references. It applies platform-correct navigation, semantic colors, Dynamic Type, SF Symbols, touch targets, VoiceOver, and motion guidance across SwiftUI, UIKit, AppKit, and Apple-style web or Figma mockups.

Set up a three-document project record: a stable charter, a rolling work plan, and an append-only journal that records intent before risky operations. A lightweight heartbeat flags stale build markers, journal debt, and open questions, while session rules keep handoffs and code changes aligned. Existing files are offered for merging rather than overwritten.

Fetch a YouTube transcript from a supported URL and save it as Markdown and plain text. Optional actions hand the saved transcript to the agent for a summary, bullet list, claim critique, or follow-up questions; the script itself does not run an LLM or require an API key. Python, the transcript package, and network access are required.

Route web pages, PDFs, Office files, video, audio, and images through media-specific accessibility tools, then produce JSON and HTML reports with WCAG criteria, severity, source tool, and review status. The skill marks missing tools and out-of-scope checks instead of treating them as passes. Native mobile apps, formal conformance claims, and several interactive behaviors still need separate testing.

Nomira guides a naming session from strategy to shortlist: it clarifies the idea a name must carry, generates many candidates, removes awkward or derivative options, and compares survivors through sound, language, and trademark pre-checks. Each finalist includes a rationale and risk flag; the pre-check is not legal clearance, so the final choice remains human.

Run the skill against a website or code repository to receive a scored audit, concrete findings, and prioritized actions for AI search visibility and agent readability. Recommendations are labeled by source tier and confidence, and every report includes practices the evidence does not support. Its knowledge base is dated May 2026, so fast-changing platform behavior should be rechecked.

Use the skill to scaffold Fabric Apps with TypeScript models, relationships, row-level authorization, GraphQL client calls, and Rayfin CLI configuration. It focuses on the preview workload and tells the agent to check current Microsoft documentation because decorators and signatures may change. Deploying a real app still requires a Fabric workspace, capacity, and an administrator-enabled preview workload.

Use the dove commands to import approved DLT templates, send individual or bulk messages, generate or verify OTPs, schedule campaigns, and decode delivery reports. The engine distinguishes gateway acceptance from actual delivery and explains DLT failures such as DND or template mismatch. Indian DLT registration is required, and bulk messages cost money and require consent.

Scaffold a runnable MapLibre web map from a description with standard coordinate systems, open basemaps, colorblind-safe palettes, attribution, and documented data provenance. The included Kenya demonstration map is illustrative: its town centroids and intensity labels are editorial, not exact event measurements. The skill also provides a checklist for auditing the result.

Run First Principles, JTBD, 5 Whys, OKR, SWOT, Porter, BCG, Blue Ocean ERRC, and Inversion independently, then compare which conclusions recur across lenses. The output focuses on root diagnoses, blocked decisions, death scenarios, and questions only the team can answer, with anti-flattery checks throughout. The workshop takes about two hours and is heavier than a single-framework review.

Web Design Engineer Pro guides non-technical users from a business description to a finished site or landing page. It generates several standalone browser previews, lets the user choose a direction, then collects a short brief, sets up a Next.js project, builds the sections and motion, and runs build, accessibility, SEO, and performance checks. It requires Node.js 20.9 or newer and can also extract content from an existing site for redesign work.

This repository provides two Claude Code skills with a shared warm-paper visual language. One renders Markdown into a local PDF, while the other turns a sentence, outline, or document into an editable PPTX assembled from reusable slide blocks. Image generation through Gemini or Unsplash is optional; without keys, the deck still builds with placeholders. Python 3.9 or later is required.

For work likely to touch many files or require repeated exploration, record a timestamp and short Git commit at the start, delegate the heavy work, and collect only a structured summary of up to 200 words. The workflow helps limit context and editor memory growth during large refactors, but adds overhead to small edits or tightly coupled reasoning.

Ask the plugin to review a command-line tool in the current project, and it will exercise real commands rather than judging documentation alone. Parallel evaluators test discovery, help, errors, output, accessibility, safety, and other criteria before producing a scored report, remediation plan, metrics file, and regression script. Its scope is UX and developer experience, not internal code architecture, and the evaluation does execute the target CLI.

auto-project-builder takes a generated or user-supplied idea through research, scoring, GitHub deduplication, implementation, QA checks, README generation, and optional pushing. It targets Next.js, shadcn/ui, TypeScript, and SQLite projects, with an automatic loop for type, lint, and build failures. The autonomous mode is intended for users who accept broad code-writing and external-research actions.