Recently updated open-source projects

loop-creator helps decide whether recurring work deserves a bounded agent loop, adapts one of 50 patterns, or designs a new loop through a plain-language interview. It can diagnose missing brakes, reward hacking, self-approval, and stale state, with a separate checking instance and explicit stop or human-question conditions.

Bazi Master uses deterministic Python calculations for the Four Pillars, then applies rules drawn from nine classical texts to produce a 19-chapter HTML report covering the chart, career, relationships, luck cycles, and annual trends. It also includes a historical calibration chapter and compatibility analysis. The project presents Bazi as traditional cultural reference, not medical, investment, or marriage advice.

IPFS Companion is a browser extension for Chromium-based browsers and Firefox that connects everyday browsing to a locally running Kubo node. It can handle ipfs:// addresses, redirect content-addressed sites and file paths to the local gateway, and provide quick actions for importing, pinning, unpinning, and copying CIDs or public gateway links. A local node must be running through IPFS Desktop or the command-line daemon.

CNKI Search Skill gives Claude Code a structured workflow for Chinese academic literature retrieval. It saves searches as reusable workspaces, then supports result inspection, detail fetching, facets, citation export, and PDF or CAJ downloads with JSON output. It reports login, captcha, and permission states rather than bypassing CNKI access controls.

This skill lets Claude Code hand off expensive or specialized work to Google Antigravity CLI, including web research, image generation, codebase analysis, browser debugging, and scientific database queries. Claude receives the result without carrying the entire delegated workflow in its context. Users must install and authenticate the agy binary and set its correct path in the skill before use.

smooth-scroll-storytelling provides a no-build recipe for quiet editorial pages with elastic scrolling and staged reveals. It combines a proxy-driven GSAP scroll layer, word-level blur reveals, sticky scenes, a drawn SVG guide line, and a scroll-responsive canvas visual, with reduced-motion and no-JavaScript fallbacks. The included skeletons and demo are meant to be reused in native HTML, CSS, and GSAP projects.

Cloud Operating Model helps an organization describe, compare, score, and recommend centralized, decentralized, federated, or platform-engineering models. Its main deliverable layers an executive assessment over technical rationale and ends with a phased transition roadmap, including ownership shifts, exit criteria, and workforce sizing; an adaptable draw.io diagram is optional. The same workflow is available as a Claude skill and GitHub Copilot prompt file, with current vendor terminology checked against live sources when needed.

Run the skill in an existing codebase to create or audit root and per-directory CLAUDE.md files from the repository's actual architecture, dependencies, and conventions. It also generates four specialized agents for analysis, implementation, testing, and documentation, plus a guide for handing tasks through them in order. Because initialization writes files autonomously and the project is still early, users should preserve a rollback point and review the resulting diff.

taiwanizer helps authors adapt Chinese prose to Taiwanese Mandarin. Its pure-Python checker distinguishes hard cases, gray areas, false friends, and whitelist terms, works with Claude Code, Codex, Cursor, and other agents, and offers three detection modes plus editable feedback for tuning the dictionary.

AgentManager inventories AI development CLIs on macOS, Linux, and Windows, checks their versions in parallel, and installs or updates them through package managers such as npm, PyPI, Homebrew, and native installers. It offers both CLI and interactive terminal interfaces, ships an offline-first catalog, and can expose management through a tray helper, REST API, or gRPC.

dev-forge stores bug descriptions, root causes, impact, solutions, and lessons as plain Markdown and YAML on your disk. Log a problem with Git context, search past causes before repeating the work, resolve it with the fix commit and time spent, then forge lessons, reports, or resume bullets from the same records. Optional auto-capture can run silently, but it is off by default and does not require a hosted account.

FoFo SDLC separates the context that writes tests from the context that writes code, then uses a referee to grade both the implementation and the tests. Scripts enforce assertion intent, traceability, separation, and secret checks with real non-zero exit codes, so quality cannot be reduced to instructions the agent claims to follow. The plugin is stack-agnostic and sits beneath an existing specification or TDD workflow. It targets Claude Code and requires Python plus Node.

fable-setup-optimizer reads Claude Code settings, plugins, agents, and external CLI availability, then presents a done/partial/not-done assessment. With approval before each edit, it can route exploration to cheaper models, load plugins per project, reduce context overhead, tier effort, and delegate typing. Changes to model, plugins, or CLAUDE.md take effect in a later session.

Formant turns a conversational brief into a form schema and a single self-contained HTML file with one-question-at-a-time navigation, keyboard controls, and light or dark mode. Forms can run offline or deploy to Cloudflare or Vercel. Add a backend only when you need stored responses, analytics, exports, or shared administration.

short-form-video-script-writer generates a topic and angle, calibrates the presenter’s voice, selects one of six frameworks, and builds hooks, a 5–7 second re-hook, open-loop body copy, and a CTA. Its default is one 45–60 second script adapted across TikTok, Instagram Reels, and YouTube Shorts, followed by a ten-point revision checklist and an optional performance-feedback loop.

ClickUp Operator organizes ideas, to-dos, and project work in ClickUp through a shared set of routing and setup rules. The package keeps its portable logic, templates, examples, and references in one core, then adds separate installation wrappers for OpenClaw or ClawHub and Claude Code. Maintainer notes and validation materials are kept apart from the runtime package, so the same operator behavior can be reused across both agent environments.

pharma-review-skills packages four independent reviews for pharmaceutical work: evidence provenance and usability, depth and interpretation of business analysis, whether key messages lead to a doctor action without compliance traps, and whether an academic event or forum serves strategy and customer experience. In a multi-agent environment, the skills can add blinded panel review, rebuttal, and synthesis.

Enter a visual idea, choose an aesthetic model, and set controls such as dialogue, music, output type, aspect ratio, duration, and negative prompts. Real combines those choices with a curated example library to assemble a structured Veo 3 director brief, with an emphasis on fashion editorials, runway, beauty, and luxury campaigns. It is available as a web app, MCP server, Gemini CLI extension, Chrome extension, and Claude skill; generation uses Gemini, and the controls are currently prompt-guided rather than strictly validated.