Recently updated open-source projects

Fireteam uses a .fireteam directory and Markdown files to preserve mission context, task state, checkpoints, handoffs, and daily notes between AI coding sessions. It can resume after context limits or crashes and supports two to six agents coordinating through shared files, without a server or database. A Claude skill is provided for setup, while the file-based protocol can also be used by other coding agents that read project files.

sitedex-audit-fix reads a site's live SiteDex report, maps failed checks to repair recipes, generates the expected discovery and metadata files, and verifies the deployed result. It can run as a Claude Code skill or through standalone shell scripts, and is specifically designed for SiteDex AI-readiness scoring rather than a general-purpose website audit.

Use this skill to generate email HTML with table layouts, Outlook conditionals, responsive stacking, dark-mode handling, accessible presentation tables, and durable CTA buttons. It is designed around real client differences rather than web-page CSS patterns, and it references tested header, hero, body, CTA, and footer components. The output is markup guidance and templates; sending, delivery monitoring, and inbox placement are outside its scope.

Use this Claude skill to architect the four email flows that commonly drive e-commerce retention: welcome, abandoned cart, post-purchase, and win-back. It emphasizes timing, list hygiene, conflict resolution between overlapping flows, transactional-versus-marketing separation, and the limits of open-rate tracking. The skill references a free template API for tested HTML components; it teaches flow design rather than sending campaigns itself.

This Claude skill grounds analysis of Chinese classical texts in provenance, authorship, editions, biography, historical context, and commentarial traditions before moving to interpretation. It preserves original passages, avoids unsupported attribution and fabricated citations, and postpones modern analogies unless requested. The scope covers classics, histories, philosophy, Buddhist and Daoist scriptures, and Yi studies; plain translation and simple lookup requests are outside its trigger.

This internal Claude Code marketplace packages one outreach-agents plugin for the dave.io business-development team. Its multi-agent workflows cover revenue, research, meeting preparation, deals, personalization, and omnichannel work, with connectors for Apollo, Notion, Gmail, Slack, and WebSearch. Every install needs a filled product-context.md with positioning, ICP, value propositions, and trigger events; without it, outputs are intentionally generic.

Use this Agent Skill to inspect and manage Easypanel instances through the tRPC API: deploy, restart, stop, monitor, back up configuration, restore it, or plan migrations between servers. Its key safety rule is read-modify-write because many update calls reset omitted fields; restores are dry-run by default and mutations require explicit confirmation. You need an Easypanel URL and API token, and volume-data restoration is outside the API.

Find What You Live By uses guided conversation to help someone derive personal principles from meaningful life experiences rather than adopting a generic list. Each principle is translated into observable behavior, and the skill is designed to preserve the user’s own language and agency. A quiet 30–60 minute session suits reflective work, but the project explicitly excludes acute crises, severe trauma, active psychiatric symptoms, and uncontrolled addiction.

Use this skill for a first-pass review of Luxembourg SCSp or ASLP limited partnership agreements against AIFMD II, CSSF Circular 25/901, ILPA Principles 3.0, and Luxembourg fund-formation rules. It produces an executive summary, cited findings with severity, and draft redlines, but it is not a complete legal opinion and does not cover US, non-EU, tax, or AML operations. Regulatory currency must be checked at run time when possible.

Configure topics, RSS feeds, brand voice, and target channels, then run the skill to filter relevant items and check keyword or trend signals using the best available source. It writes a daily Markdown file with drafts tailored for LinkedIn, X, blogs, or newsletters. It never publishes automatically, and basic operation requires an RSS or web-fetching tool such as Firecrawl or web_fetch.

Design MD converts a website, brand guide, screenshots, or a mood-board description into two paired artifacts. DESIGN.md captures semantic tokens, component rules, visual directives, and named anti-patterns for AI agents, while design-showcase.html presents the same system in a self-contained page that humans can open without a server. The shared source tokens help keep implementation guidance and visual review aligned.

This skill monitors an owned X account, reads engagement and KOL signals, and queues posts or replies as local files. Official X API writes use scoped OAuth 2.0 PKCE and require an explicit approval keypress; KOL reads use the bundled bird-search path, while automatic likes, follows, and quote-posts are excluded.

Use this framework to audit and improve a public website one readiness level at a time. It guides the addition of discovery files, crawler policies, Markdown negotiation, MCP/A2A and skill catalogs, OAuth metadata, WebMCP, and related public surfaces, then reruns an external audit to show progress. Platform notes cover common Next.js, Astro, SvelteKit, Cloudflare, Vercel, Netlify, and static-site setups; commerce protocols are opt-in and should appear only when the site has a real commerce or payment flow.

exiftool-skill translates natural-language requests into ExifTool commands for image, video, and audio metadata. It covers GPS handling, capture-time correction, tag copying, date-based organization, structured JSON or CSV export, video metadata, and stripping private fields before publication. ExifTool must be installed on PATH before use.

This paired workflow turns a one-line request into a Codex-ready brief with goals, scope, constraints, acceptance criteria, deliverables, validation, and output rules. A second skill on the Codex side consumes that brief, executes the task, and reports a summary, changed files, validation results, and follow-ups. It is useful when planning and implementation should be separated; the two skills are installed in different environments.

This skill installs a Hybrid Context Management Protocol with universal instructions, agent-specific routing, maintained architecture notes, a generated component registry, and context files for decisions, domain rules, and terminology. A Python generator refreshes the structural map, while setup detects the repository, asks focused questions, validates links, and avoids blindly overwriting existing guidance.

Relay lets one orchestrator break cross-repository work into a dependency-aware ticket graph and route tasks to live Claude or Codex agents in the relevant repos. Channels retain the board, decisions, execution feed, and pull-request state, while the CLI, TUI, and desktop app share local data under `~/.relay/`. It is still pre-1.0 beta software, so interfaces and file layouts may change.

This Claude skill rewrites a LinkedIn profile for a target job description while preserving the user's voice. It covers headline, About, experience, skills, and Featured sections; can compare pasted competitor profiles; and iterates against clarity, specificity, anti-buzzword, story arc, and keyword-match scores. It also flags sensitive resume details and refuses to invent achievements.