Recently updated open-source projects

DataEcho packages one skill for publishing a file, static site, or Dockerfile-based app to a live URL and another for persistent agent memory. Publishing supports anonymous uploads with a later claim, incremental deploys, and private Drives with scoped share tokens. The memory skill keeps a small index and one fact per file, with atomic writes, history, undo, and handoff between agents. The scripts use Bash and Python’s standard library, while credentials and stored memories remain account-scoped.

publish-to-github-skill creates a new GitHub repository, uploads files through the browser, and checks the result through raw GitHub content rather than trusting the upload screen. It also preflights README, license, and ignore-file hygiene and calls out the executable-bit loss that can affect uploaded scripts. A signed-in Chrome session with Claude browser tools is required; no API key or gh CLI is needed.

Neel helps Nim developers create desktop tools with a web frontend while retaining access to Nim on the backend. It handles routing, JSON conversion, and two-way calls between JavaScript and Nim, and release builds can embed the frontend assets. The library has been tested on macOS, Windows, and Linux, but it currently opens Chrome in app mode by default and is described as too early for large commercial applications.

design-psychology-skill audits or builds websites and user interfaces against roughly 60 research-backed principles, from Hick’s Law and Fitts’s Law to trust, motivation, and cognitive load. It detects a live URL or local codebase, derives the product’s own visual tokens and voice, and returns prioritized findings or rationale-backed design changes. An honesty-focused module constrains tactics that would become dark patterns.

Run /roast in Claude Code to review a repository across security, architecture, customer flows, and growth readiness, with findings cited to files and lines and the top three priorities summarized at the end. Optional Semgrep, Gitleaks, and dependency checks add deterministic evidence; their own setup may fetch rules, but audit findings are not uploaded by default. Live-URL checks contact the requested site and Google PageSpeed only when explicitly enabled.

mask rewrites bland copy in the voice of a chosen character by drawing on linguistic fingerprints such as self-reference, catchphrases, sentence rhythm, and diction. You can use it for a pasted line or draft and switch among example voices such as Sherlock Holmes, Yoda, or Shakespeare. The goal is a consistent, recognizable character voice rather than a shallow word swap.

This repository documents how Claude Code features work and packages the reverse-engineering process as a reusable skill. It extracts the embedded cli.js from the installed Bun binary, reformats the minified source, locates features through surviving string literals, annotates relevant code, and produces evidence-backed reports on commands, skills, agents, caching, and remote control; the workflow requires a local Claude Code installation and substantial source processing.

Godplans turns a software idea into a single PLAN.mdx before implementation begins. It asks a small batch of high-leverage questions, records scope and irreversible decisions, maps applicable domains, and emits requirements, acceptance criteria, architecture, risks, and phased tasks. A policy screen can refuse prohibited uses or add mitigations, and the skill works across several agent clients.

Use godaudits before a release, during a handoff, or for technical due diligence. One command identifies the project type, audits the applicable domains, verifies each finding at a specific file and line, and writes scores plus checkbox remediation tasks to AUDIT.mdx. The audit is read-only: it does not run the application or tests, connect to live systems, or apply the fixes itself.

seele-s-html injects a presenter view into an HTML deck that can be paged in a browser. Pressing S opens current and next-slide previews, speaker notes, and a timer, with navigation synchronized to the audience window. The zero-dependency Python postprocessor leaves the original generator alone and produces a portable HTML file; PNG and PPTX inputs are outside its scope.

urimal-for-socialworker edits Korean social-work plans and reports for natural wording while preserving facts, numbers, dates, names, and quotations. It applies a 36-point Korean usage guide alongside social-work-specific checks, offers fast and strict modes, and includes kordoc support for HWP, HWPX, PDF, and DOCX inputs. The project is tested for Claude Code; other Claude interfaces may not discover it reliably.

fullstack-pro-max is a Claude Code skill that selects a stack from project needs and organizes UI, backend, data, authentication, security, testing, SEO, and launch work across 13 modules. It includes standard-library-only Python helpers and can optionally use connected services such as Supabase or Vercel for live operations. The project reports only two author-run end-to-end field tests so far, so routing and generated results still require human verification.

WebWhiz crawls a product website, turns its content into searchable context, and gives you a customizable chatbot that can be embedded for customer questions without writing integration code. It supports recurring content updates, offline-message handling, and a visual builder. Self-hosting is a multi-service deployment with an API server, JavaScript and Python workers, MongoDB, and Redis, and an OpenAI API key is required.

Give it a topic to research official sources, choose an appropriate manual, chapter, example, or quiz structure, and build a themed static handbook in the language you are using. In a terminal agent it can run its review checks and publish the site to GitHub Pages automatically. End-to-end deployment requires shell access, git, an authenticated GitHub CLI, and network access; cloud chat environments generally require manual publishing.

TAR Engine's audit layer discovers and checks SKILL.md files, Codex skill definitions, Claude commands, OpenCode configs, and adjacent helper scripts from the CLI, CI, or an MCP-compatible agent. Static rules run by default; semantic and adversarial analysis requires an explicitly supplied model key. The hosted mode sends submitted skill text to tarai.dev, so sensitive material can use the self-hosted backend instead.

LPIM uses a site's sitemap and client materials to build a product-line taxonomy, classify pages, and produce a Product Interest Map plus a findings brief. It is aimed at B2B manufacturers, suppliers, services, and SaaS sites, and deliberately pauses for human review; only the mapper is currently shipped, not visitor qualification or CRM write-back.

seo-writer asks for the audience and publishing method, researches competitors, real questions, and authoritative data, then pauses for outline approval before drafting a Traditional Chinese article. It adds direct-answer sections, FAQ, structured data, and SEO metadata, followed by a second quality checkpoint. The result can go through WordPress MCP or be saved as Markdown, and the optional humanizer skill changes the copy style.

rsi-loop selects the highest-priority unchecked medium task from BACKLOG.md, implements the smallest correct change, runs the repository’s tests, typecheck, and build, then commits it. When the backlog is empty it can regenerate more work, and .rsi-loop.json controls the commit target, perpetual mode, and blast-radius limit. Because it has no question or checkpoint step, use it only where the task boundaries and checks are already trustworthy.