Recently updated open-source projects

xuyangsheng-perspective distills public talks and educational ideas associated with Yangsheng Xu into a reflection framework for questions such as career transitions and education. It is a lens for self-reflection rather than a direct-answer engine. The material is based on public sources and is not Xu's own speech; it is not intended for technical fixes, short-term financial choices, or crisis response.

Six-Layer Product Research guides Claude Code through a structured problem-definition process before solution work begins. It examines customer language, ecosystem perspectives, behavioral frictions, failed status-quo attempts, new technology enablers, and the changes a team can realistically influence. Separate findings are then synthesized into a multi-perspective definition that helps product teams test assumptions and understand why earlier approaches failed.

This skill routes image briefs between OpenAI GPT-Image-2 and Google Gemini Nano Banana, with presets for concepts, high-resolution output, web and social assets, references, edits, sessions, image analysis, and saved style fingerprints. It works from Claude or a standalone Python script, but requires both provider credentials and careful protection of generated bundles or key-bearing files.

Parallax starts from a topic, proposes an editable roster of analysts with distinct agendas, and sends each perspective to research before combining the results into an executive synthesis, disagreements, open questions, and optional comparison tables. Runs can be sequential for mid-course control or parallel where subagents are supported. It needs at least one research source such as web search or MCP, and can export the finished report as Markdown, Word, or PDF when the required converters are available.

Use 396 named bash scripts that map directly to Twenty CRM REST endpoints for CRUD, filtering, batch operations, duplicate merging, attachments, and more than 30 object types. Agents can discover the right script and inspect its expected JSON before execution, reducing malformed requests. The toolkit works as a Claude or OpenClaw skill, through MCP, or in shell automation, and requires bash, curl, jq, and a Twenty API key.

This skill enriches frontend conversations with relevant MDN documentation for CSS, JavaScript, DOM, HTML, HTTP, SVG, accessibility, and related web-platform features. It can find, render, search, browse, and refresh a local slug index, but it does not provide live compatibility tables or interactive examples, and it does not cover Node, Deno, or Bun runtime APIs.

Convert Feishu Dou or Feishu Minutes transcripts into an eight-section note for meetings, interviews, courses, podcasts, and other dense recordings. The skill separates recap, business implications, reusable content, methods, and actions, attaches stable IDs with source, value, and timestamp labels, and can render a Mermaid relationship map or an HTML card version.

Cert-Lifecycle-Harness guides certificate renewal, migration, and audit through inventory, risk assessment, planning, layered review, execution, rollback, and verification. It produces L3 decision, L2 strategy, and L1 execution documents so reviewers can work in bounded time, and it may perform Import or Modify writes only after six gates are satisfied. Delete operations stay script-only, while cloud APIs, kubectl, and production changes still require explicit authorization and human approval.

Zilliz Launchpad guides an agent through collecting data, configuring a use case, planning an index, ingesting records, evaluating search, and deploying a runnable application. It supports text and multimodal inputs through an agent skill or MCP server, targeting local Docker-based Milvus or Zilliz Cloud, with Python, Node.js, an embedding API key, and the chosen deployment environment required.

Paste an existing Etsy listing or describe a new product and the skill researches current autocomplete, competitor results, and Etsy policies before returning an optimized title, tags, attributes, description, and Pinterest copy. It checks tag limits and trademark terms and can remember listings in the Claude path. It needs web access for live research and does not promise rankings or solve image and conversion problems.

Turn image prompts stored in a Markdown file into ready-to-use URLs for articles or documentation. The skill generates each image through Gemini, uploads it to Cloudinary, and returns automatically optimized CDN links; it requires both Gemini and Cloudinary credentials.

architecture-solver turns workload requirements into infrastructure designs through seven dimensions: TIME, SPACE, WORK, STATE, FAILURE, COST, and TRUST. It applies formulas such as Little's Law, Erlang-C, USL, availability calculus, and cost optimization, then matches technologies and proven patterns. The skill can produce sized and costed recommendations, compile them to Terraform, CDK, or Pulumi, and define load, resilience, security, and cost checks.

The skill first asks domain questions and produces an ADR covering aggregates, commands, events, projectors, and reactors for review. After approval, it generates tests first, then the domain code and migrations, and runs the suite. It targets greenfield Laravel domains using Spatie Event Sourcing v7 rather than converting existing CRUD code.

Claude Dependency Auditor runs native audit tools for Node.js, Python, Rust, Go, Ruby, JVM, and PHP projects, then separates actionable runtime issues from development-only, unreachable transitive, or disputed findings. It can also check license compatibility, generate CycloneDX or SPDX SBOMs, report stale dependencies, and propose remediations that are applied only after confirmation and verified with tests.

Agency Toolkit captures a brand’s colors, fonts, logo, industry, and voice, then applies them to HTML proposal templates through a Node-based CLI or Claude Code skill. It includes English/Spanish copy variants and a curated vanilla web stack, with landing pages, diagnostics, design systems, and dashboards represented in its templates or roadmap; Node 18+ is required.

Instrument synchronous or asynchronous Python code with context-manager spans or decorators, capturing timing, attributes, events, and exception status. Trace data can stay in memory or persist as local JSON files. A terminal dashboard and MCP tools expose individual traces and aggregate statistics, while thread locks and ContextVar tracking preserve isolation across concurrent work.

Decision Memory watches product conversations for moments such as accepting, rejecting, deferring, merging, or shipping an idea. It asks for the reasoning and any revisit or notification details, then formats the result as a searchable decision record and closure-message draft. The skill works on its own with Markdown output, while an optional IdeaLift MCP connection can persist records and link them to ideas.

This skill transcribes spoken video with Whisper word timestamps, groups words into 2–5-word chunks around punctuation and pauses, and burns styled captions with ffmpeg. You can choose fonts, colors, brand plates, aspect-ratio presets, languages, and a glossary; it requires a Whisper package and ffmpeg.