Recently updated open-source projects

This Claude skill helps Higher community members and onchain creators draft posts, manifestos, and onboarding copy in a short, direct voice. It can also evaluate an idea against four Higher principles and return a structured verdict. Install the SKILL.md in Claude Code or paste it into custom instructions.

code-agent-finops helps diagnose coding sessions that drifted, spent too many credits, passed tests while violating design intent, or repeatedly reloaded excessive context. It produces a calibrated prompt contract for the next session with explicit scope, semantic requirements, validation boundaries, and a stop condition. The method works well for bounded implementation tasks rather than open-ended architecture or brainstorming.

Changeflow lets an agent create website monitoring sources, list their status, search detected changes by source, date, or keyword, trigger immediate checks, and pause or resume monitoring. It works with Claude Code, Cursor, and other Agent Skills hosts, but requires a Changeflow account and API token rather than running as a self-contained crawler.

With Claude in Chrome connected to a running development server, the skill captures the viewport, console errors, failed network requests, and page structure, then traces the problem back to source files. It can edit the code and recheck the browser, or instead produce an HTML report, file a GitHub issue, or watch for new errors.

Search a curated catalog of more than 3,900 MCP servers from Claude Code and receive Chinese descriptions, editorial notes, star counts, and copyable installation commands. The same skill also returns a daily AI briefing and weekly trending tools, making it useful for Chinese-speaking developers who want discovery and setup in one workflow.

BRD Specialist turns an existing PDF or DOCX, a raw meeting transcript, or a guided questionnaire into a complete business requirements document for financial-system projects. It is designed for Claude with code execution and handles document tooling through dependencies such as docx, Poppler utilities, and Pandoc, making it suitable for payments, compliance, funds, and similar requirements work.

Naze combines UI structure, state, and interactions in a declarative language that compiles to WebAssembly and renders through Canvas2D rather than the DOM. Its toolchain includes compilation, type checking, testing, examples, and project-context export for AI agents, but the project is still experimental and explicitly not ready for production use.

Enter a text prompt in this OpenClaw skill to generate an image through the Neta API and receive a single URL for embedding or download. The command supports several size presets and can take a picture UUID as a style reference. It requires Node.js 18 or later and a valid NETA_TOKEN.

Google Ads Pilot helps media buyers review exported search-term data, CSV files, or a written account description. It can suggest negative keywords, budget shifts, account fixes, weekly analysis, and RSA copy, while its core method checks search intent, the triggering keyword, and ad-group fit so each recommendation has an explicit reason.

Start with the lightest launch option for a local HTML folder or development server: create a temporary Cloudflare Tunnel URL, move to a persistent Pages deployment when needed, and optionally connect a domain the user purchases separately. Tunnel links stop working with the process. Pages requires a Cloudflare account and API token, while domain purchase remains manual and automated DNS setup needs a separate scoped token.

Use it as a peer-level advisor for platform choices, governance exceptions, capability maturity, transition architectures, and roadmaps. Responses lead with a recommendation, connect the reasoning to ADM phases, the Content Framework, reference models, or governance mechanisms, and close with the matrices and diagrams worth producing. It is designed for practitioners who already understand TOGAF terminology.

Classify Claude Code actions such as reads, file writes, network requests, credential handling, and destructive commands into proportional risk tiers. Routine work can continue, while sensitive actions trigger notices, approval pauses, or stops and are logged for review. This repository is the Markdown-only skill, so enforcement depends on model compliance; the companion plugin adds deterministic hooks.

Guide Claude Code through architecture review, code audit, API attack testing, and SOC 2 CC6, CC7, and CC8 checks. The materials cover injection, JWT flaws, IDOR, SSRF, cryptography, uploads, and Playwright security probes across several backend stacks. The workflow treats critical findings, missing authentication or validation, and hardcoded secrets as blocking issues.

Set up a durable Claude Code Remote Control session using the native service manager on each platform: tmux and systemd on Linux, tmux and launchd on macOS, or PowerShell and Task Scheduler on Windows. The skill inspects the machine and shows its plan before changing anything. Claude Code must already be installed and authenticated, with platform-specific dependencies available.

Use Claude Code to query Bags tokens, pools, swap quotes, claimable fees, and wallets linked to social handles. The kit can scaffold fee-sharing apps, trading bots, or token launchers, and provides guided workflows for token launches, fee allocation, and performance analysis. Running the bundled tools requires Node.js 18 or later and a Bags API key.

Describe an app or provide its App Store link to create marketing previews focused on the hook, core value, and a key feature. The skill produces 1242×2688 assets with generated headlines, gradients, and campaign-style compositions. The repository positions the free workflow as a preview, while the complete five-image set is offered through a separate paid service.

Collect OpenClaw health, deep status, and all-agent session output into a self-contained HTML dashboard. It visualizes channel state, agent activity, token use, model distribution, and aborted or stale sessions, then highlights issues that deserve attention. The workflow requires Python 3.8+, a running OpenClaw gateway, and terminal or MCP access to its CLI.

Hand it a multi-step plan and Claude Code will sequence tasks, parallelize independent work, track progress, and stop repeated failure loops with a circuit breaker or external model review. It also supports resumable tmux sessions, service monitoring, and optional ntfy phone notifications. Fully unattended execution still depends on your Claude permission settings.