Recently updated open-source projects

This Claude Code skill structures substantial engineering work around understanding, specifying contracts, pre-mortems, small changes, and behavior-focused verification. It adds 30 red-flag checks, software and security thinking layers, and situation-specific playbooks grounded in ten engineering books. The project states that real-world use remains unproven at scale.

Run static checks across common web frameworks and dynamic browser probes for language, headings, ARIA landmarks, form labels, contrast, focus visibility, and related issues. The plugin maps findings to all 106 RGAA 4.1.2 criteria and can generate an accessibility statement, a three-year accessibility plan, an annual action plan, and a traceable CSV grid. Its automated scan intentionally favors false positives, so expert verification remains necessary before official publication.

Use one CLI workflow for Comfy Cloud image, video, and multi-step generation tasks. It resolves S/A/B/C quality tiers, inspects model schemas to translate dimensions and flags, and includes recipes for characters, storyboards, products, thumbnails, and moodboards. Premium is the default, while budget mode downshifts quality; a Comfy API key is required.

This Codex and Claude skill maps thesis roles—title, abstract, headings, body, captions, and references—to Word styles using a generated structure map. It supports default or supplied school, journal, template, or PDF formatting rules, backs up the input, and writes a new DOCX while preserving text, table data, and image assets. It does not edit prose.

Prompt Driven Development treats prompts, persistent project context, and review criteria as maintained project artifacts rather than disposable chat input. It provides workflows for setup, research, planning, prompt generation, revision, review, and evaluation. Simple features can follow a short Context-to-Prompts-to-Review path, while complex work adds research and planning. The shared core is packaged for Claude Code, GitHub Copilot, Codex, and Google Antigravity.

Detect the project’s framework and mobile setup, then apply structural fixes for safe areas, notches, keyboards, input zoom, viewport height, splash screens, status bars, and hybrid storage or authentication. The skill preserves the existing visual design, verifies lint/build/cap sync, and reports manual device and store checks for PWA or Capacitor targets.

Use this skill for aerospace, pharmaceutical, medical-device, automotive, nuclear, or legal documents where terminology and modal force matter. It pins glossary terms per chunk, critiques safety-critical passages, checks numbers, sections, references, and warning order, and can back-translate flagged content. Outputs are working drafts, not certified translations, and scanned PDFs need OCR first.

KISS gives Claude Fast and Deep operating modes, rules for separating durable project memory from temporary ideas, and a token-discipline framework that limits repetition. It suits multi-session development and ongoing product work, using CLAUDE.md and memory files as the integration surface. The skill shapes workflow and response depth rather than project behavior.

mailoutofspam guides a ten-step deliverability repair: audit SPF, DKIM, DMARC, and MX, add suppression handling, implement RFC 8058 unsubscribe, parse bounces, throttle by provider, and verify raw headers through IMAP before resuming volume. It focuses on compliant recovery rather than spam evasion and includes standalone Python helpers. DNS, SMTP, and IMAP access are needed for a real diagnosis, and sending remains an operator-controlled action.

This Claude Code skill inspects a task’s type, language, and constraints, recommends a model available through an OpenAI-compatible gateway, and asks for confirmation before invoking it through Bash. Setup stores the API key in a local permission-restricted file rather than chat history, and the workflow can reuse your own prompt files. It requires access to a compatible provider or internal gateway.

Use the scream CLI to speak arbitrary text, replay the last assistant message, choose among 30 Gemini Flash TTS voices, or trigger audio automatically when a Claude Code run stops. It can fall back to the offline macOS say command, while Gemini voices require an API key. The current release is macOS-only, caps spoken text at about 1,200 characters, and does not provide transcription or streaming synthesis.

3D Website Builder turns a ready-made roughly seven-second MP4 into a vanilla landing-page hero that advances frame by frame as the user scrolls. It asks for basic brand details, adds baseline SEO metadata, and uses a static poster on small screens or when reduced motion is preferred. It needs an input MP4 but no Node.js or FFmpeg.

3D Animation Creator turns a video into a single-file Apple-style scroll-animation site whose canvas playback follows scroll position. It adds responsive mobile treatment and annotation cards, can use a URL or pasted text for content, and runs with FFmpeg on macOS or Linux.

Ask Claude for a 3D wireframe HTML presentation and the skill produces a single file that opens directly in a modern browser. Slides use Three.js backgrounds, horizontal snap scrolling, parallax depth, responsive layouts, and automatic theme-color transitions. The presentation depends on an internet connection because its external libraries are loaded from CDNs.

launchspecs turns a product brief or landing-page context into paste-ready launch kits for Product Hunt and Uneed. Each kit follows the selected platform's field limits, categories, image sizes, scheduling guidance, and operations checklist, with cross-aligned copy when both platforms are in scope. The project is unofficial and community-maintained, currently covers only those two platforms, does not submit forms automatically, and refuses vote-gaming tactics or turnkey Product Hunt comments that violate platform rules.

Provide a resume and job description to generate gap-focused hands-on labs, platform-specific commands, a two-minute introduction, JD-to-experience mapping, STAR stories, a theory cheatsheet, and quick Q&A. It works across languages and major Claude surfaces, but it does not score jobs, rewrite a CV, or submit applications.

Repair an oversized Codex JSONL rollout without deleting the conversation. The scripts first remove superseded checkpoints, replace embedded screenshots with text markers, and trim oversized tool output, preserving the original backup and verifying that the result is safe to load. Only as a last resort does the workflow drop older turns.

This Claude Code skill sends mechanical text work such as list classification, filtering, and formatting to a local Ollama Mistral model through curl. It avoids an MCP server, logs each call locally, and offers an audit mode for success rates and estimated token savings. Ollama must be running with Mistral available.