Recently updated open-source projects

perplexity-plugin teaches Claude Desktop and Cowork how to route fresh-information tasks across Perplexity's search, answer, reasoning, research, computation, history, and diagnostic tools. It also handles plan-gated features by falling back to free-tier search patterns when needed. The package does not install an MCP server: users must connect one themselves, and plan limits should be checked as they change.

Equip an AI coding agent with task-specific workflows for startup naming, brand audits, Shopify development and store reviews, venture validation, website cloning, and agency marketing research. VeyraSkills is organized into installable packs and individual skills, so teams can add only the workflows they need. The Markdown-based skills work across several agent environments, including Claude Code, Cursor, Windsurf, Gemini CLI, and GitHub Copilot.

search-science-literature queries 13 open scholarly sources, including OpenAlex, Crossref, PubMed, Europe PMC, Semantic Scholar, arXiv, and Unpaywall. It normalizes abstracts, removes duplicates, checks for retractions, and exports BibTeX, RIS, or CSV records. Optional APIs add coverage when their keys and quotas are available; the skill explicitly skips Google Scholar scraping and paywalled databases.

Turn past Claude Code sessions into a searchable local index, then reuse a previous method, print a resume command, or reopen the original session in a new terminal. Session-end hooks create compact recap cards, with an optional handoff loop for continuity after compaction. It is local-first and has no package dependencies, but the full open/spawn workflow is built for macOS and requires Python 3.9+ plus a logged-in Claude CLI.

Claude LazyCodex Skill lets Claude delegate coding work to Codex and the LazyCodex harness in a live tmux pane. It selects planning, looping, or team commands, watches the session, answers questions, and independently runs builds or tests before reporting back. The setup requires Claude Code inside tmux plus a signed-in Codex CLI and LazyCodex.

Dynamic Soul Sliders lets Claude Code or Hermes adjust seriousness, energy, and warmth while keeping the underlying content and identity unchanged. It is a Markdown-only skill with no fine-tuning or state file. The author notes that silent always-on calibration is unreliable on current models, so deliberate invocation or a visible persona-file anchor works better.

Chameleon replaces hardcoded colors with a small semantic CSS contract that can be reused across HTML artifacts. Its Claude Code skill can create themed pages or propose a dry-run conversion for existing files, while a Chrome extension applies the reader's saved theme across compatible pages. For durable or offline artifacts, the theme can be baked into the HTML and pinned to a contract version.

unifusion runs an OpenCode orchestrator that sends one difficult brief to several panelists in parallel and keeps the synthesizer separate from the panel. It captures the final analysis, panel reports, session context, task events, and saved provenance, making the harness useful for comparing independent model views before a recommendation.

Provide hierarchical data and drilldown-table renders the same corner-header report layout to XLSX, DOCX, or PPTX from one shared engine. It supports row, column, and two-dimensional drill-downs, uneven two-to-four-level branches, automatic totals, and three visual themes. Users can run it as a Claude or Codex skill, a Windows desktop app, a browser tool, or a Python library.

notebooklm-research wraps NotebookLM with a four-stage handoff: choose a research variant and scale, create a search outline, apply citation and fact/opinion/inference rules, then verify the finished report. It checks decision-grade numbers, time markers, named entities, and recurring anomalies, while preserving a human-in-the-loop exchange between the agent and NotebookLM. The repository includes installers for native skill hosts and pointer-based setup for rules-driven agents.

Artic helps a team set a homepage's design direction before implementation, grounding it in reusable rules drawn from professional and open-source references. It turns the brief into project-specific strategy and DESIGN.md documents, validates the result, and explicitly avoids cloning a reference site or brand.

Give an agent one command-line workflow for model discovery, spend controls, image generation and editing, video and audio creation, and image-to-3D jobs. The public CLI connects to a hosted service that returns durable asset URLs, recoverable job state, and cost receipts without requiring separate provider API keys. It depends on that online service, runs on Node.js 20 or newer, and guides a fresh user through anonymous signup and local credential storage.

ClawSuite is a self-hosted command center for OpenClaw agents, delivered primarily as a web application and installable PWA. It combines multi-agent mission control, live output streaming, chat, cost analytics, task boards, cron management, memory and file browsers, terminal access, and gateway diagnostics. It suits users who need a central operational view of agent activity, while the native desktop application and managed cloud edition are still described as works in progress.

project-guides documents a finished app, feature, or toolkit as a matched pair: a self-contained GUIDE.html for operators and an exhaustive DEVELOPER-INDEX.md for people who need to modify the code. Separate authoring roles read the real source, verify both outputs against it, and render the guide for visual inspection; an optional AI-CONTEXT.md gives a shorter orientation. The Codex edition uses native tools and local validation without requiring the Claude CLI.

tech-audit routes a codebase review into quick, security, release, deep, or full modes across a 13-dimension framework. Stack-aware extensions cover PHP/Laravel, Python/FastAPI, TypeScript/Node, shell, and Docker, while calibrated severity keeps real blockers separate from minor findings. It suits PR checks, release gates, and deeper technical reviews.

Chatpad is a free web and desktop interface for chatting with ChatGPT. It keeps application data in local storage and states that it uses no tracking or cookies, making it suitable for people who prefer a lightweight client rather than another hosted workspace. You can use the published web or desktop app, or run the project yourself with a Docker container.

This benchmarking toolkit runs ARC-AGI tasks through configurable adapters for providers such as OpenAI, Anthropic, and Gemini, with rate limiting, retries, saved submissions, and scoring built in. It supports both single-task debugging and asynchronous batch runs. Full evaluations require separately cloned ARC-AGI data, and hosted models require their respective API credentials.

This skill uses Pencil MCP to translate design files into UI code for iOS, Android, Flutter, or web projects. Its staged workflow saves progress checkpoints, can run with parallel agents or a slower linear mode, and uses differential audits to update only the parts affected when a design changes.