Recently updated open-source projects

Beauty Diagram teaches an agent to use the bd CLI to polish existing Mermaid or PlantUML, render generated diagram source, and return share or direct-embed links for README files, Notion, and blogs. Node.js is required. Anonymous use is limited and watermarked, while unwatermarked exports, higher quotas, and prompt-to-diagram generation require an API key; AI generation is restricted to Pro or Premium plans.

printable-docs helps create Traditional Chinese A4 reports, quotes, contracts, minutes, forms, resumes, slides, and study sheets from self-contained HTML. It includes 39 templates, a style contract, reusable grid CSS, offline SVG infographic modules, and a page-assembly tool. Its explicit embeddable font stack addresses the macOS Chrome problem where Chinese text can disappear in printed PDFs. Before delivery, the project recommends printing a real PDF and checking selectable text, page count, and table breaks.

Embed Claude Code, Codex, Gemini, or OpenCode in a Rust application without writing separate subprocess and event parsers for each tool. gate4agent normalizes Pipe, PTY, and ACP sessions into transport-neutral events, supports session resume and multi-turn ACP callbacks, and tracks history and context usage. It is a transport layer, not an agent harness or security sandbox.

Describe the genre, mood, theme, or artist references for a track and receive the fields needed for Suno's Advanced Mode: styles, lyrics or instrumental guidance, exclusions, slider settings, a title, and production tips. The skill translates artist references into descriptive musical language and adapts recommendations to the user's plan and model version, with an optional consistency profile for an EP. It prepares prompts; it does not generate the audio itself.

This portable skill combines interface design systems, motion guidance, design-token conventions, and pre-ship checks into reusable instructions for Claude and compatible coding tools. After loading it, commands such as polish, animate, and audit can refine hierarchy, motion, and accessibility without repeating a long design prompt in every session.

Claude Prospector reads local Claude Code session files to break token use down by model, agent, skill, and project across the supported billing windows, with an optional interactive HTML dashboard. It also inventories the effective agent and skill configuration to flag collisions, overlap, and avoidable bloat. A one-time setup step creates a plugin-owned Python environment before these features become active.

Ethical Response Reviewer checks a draft AI response before delivery and chooses approve, revise, refuse, clarify, or escalate. It explains the decision, supplies a safer rewrite or escalation note when required, and emits machine-readable audit tags for later logging. The skill is aimed at high-risk contexts such as law, medicine, finance, employment, public services, minors, and sensitive data. It is a structured risk layer, not a guarantee or substitute for professional advice, human responsibility, access controls, or platform safety.

Research Clinician skill helps Chinese medical students and early-career clinicians turn bedside observations into research questions, evidence notes, tool evaluations, reproducible mini-runs, bias audits, and study plans. At each gate, the AI exposes its work, sources, and uncertainties while the clinician decides what to accept, authorize, and sign. Real cases must be de-identified and cleared through ethics and data-governance processes. Without code execution, it drafts an executable plan instead, and it does not provide individual diagnosis or treatment decisions.

content-wand can atomize an article, transcript, notes, topic, or URL into posts for nine platforms, or convert between formats such as podcasts, blogs, newsletters, email courses, and scripts. It can match a personal or client writing style and applies an AI-pattern cleanup pass before delivery. Web and URL processing also includes defenses against indirect prompt injection.

disk-cleaner-skill scans a macOS or Linux machine, classifies space usage by safety, and presents the findings in a local browser dashboard. It never deletes during discovery and asks for explicit approval before acting: pure caches may be removed, while review items move to the recoverable Trash and protected data is excluded. A dry-run, per-item approval, self-contained report, and append-only action log support inspection. The project is still v0.1 alpha, so the dry-run and backups remain important.

Claude Competitive Intel runs a nine-step research workflow around a business idea or existing codebase, covering competitors, feature matrices, market gaps, customer segments, pricing, complaints, branding, and technology recommendations. It generates DOCX reports, a multi-sheet Excel workbook, and self-contained interactive HTML dashboards. The workflow requires web search access, Node.js with the docx package, and Python with openpyxl.

Code Architecture Auditor examines a language-agnostic codebase for structure, data flow, dependencies, dead code, complexity, error handling, tests, and documentation, then produces a report and cleanup plan. Report-only mode is read-only by default. Refactoring requires a captured baseline, an impact-based gate, approval for each change, and validation that output still matches; deletion candidates are checked for reflection, dependency injection, entry points, and public APIs first. The project uses PolyForm Noncommercial licensing, with commercial use requiring separate permission.

Polish Korean LLM output by targeting stiff formality, empty intensifiers, filler connectives, template openings, hidden agents, excessive hedging, and other catalogued patterns while preserving the text’s meaning. The repository includes a prompt and skill for Claude Code and Codex, plus a shorter prompt for other LLMs.

Portfolio README Writer documents a finished project for either a personal portfolio audience or a hackathon judge. It reads the repository, asks only for details it cannot infer, and writes a structured README with a pitch, quick start, stack, architecture, caveats, and mode-specific sections. It avoids emoji and inflated language, shows a diff before replacing an existing README, and does not modify code, media, or deployment.

Super Translator uses a learner’s mother language as the explanation anchor, then compares several languages to expose nuance and offers register, rewrite, vocabulary, and grammar lenses. It reads targets and preferences from config.md and adapts a version-controlled profile.md after each interaction. If the skill itself needs a structural change, it proposes an auditable edit for approval rather than rewriting silently. The result is a local Claude Code workflow for non-native speakers who want deeper understanding and more professional writing.

Governance Encoded turns AI governance material from 12 jurisdictions and eight sectors into developer-facing directive cards, code patterns, and checklists. Its three modes map requirements to a system, compare agent frameworks such as LangGraph and CrewAI, or create handoff protocols for operations spanning jurisdictions. References carry source and currency information, but the project warns that summaries may be incomplete or outdated and are not legal advice.

Inspect a code project read-only, identify its stack and heuristic learning gaps, and match them to Turing books or public resources with reasons. The skill can produce a trackable weekly plan and aggregate gaps across repositories, while keeping book content and paid access boundaries intact. Prices, stock, and course status should be checked on the official site.

piped-clover turns a pasted research article or transcript into three linked deliverables: a Marp slide deck, an editable PowerPoint, and a zoomable interactive mind map. A deterministic parser extracts headings, bullets, numbers, frameworks, and timestamps; Claude shapes the visual story and adds a source citation plus a personal “So What” takeaway. URL fetching is not yet automated, and first-time renderers may need npm or PyPI access.