Recently updated open-source projects

The skill records a founder or small team’s skills, geography, constraints, network, and goals, then structures ideas, compares candidate verticals, brings in external research, and pressure-tests a beachhead recommendation. It supports both a guided sequence and an open advisor mode, but the current MVP stops at intake, exploration, vertical comparison, and beachhead selection; positioning, customer evidence, GTM, and ongoing review are deferred.

The skill interviews founders to create a slide-by-slide outline, timed speaker script, and rehearsal checklist when no deck exists. For an existing deck, script, or pitch transcript, it scores the material against the DePitch framework and returns a small set of prioritized fixes instead of rewriting everything.

This Claude Code skill requires suspected bugs to be checked against tests, callers, and concrete execution traces before they are reported. Its strongest coverage is a 20-section C and C++ checklist with four severity buckets, while reference guides also cover Rust, Go, Python, Java, frontend frameworks, architecture, performance, and security.

This skill turns hard-won Starknet game-shipping lessons into 13 rules that Claude applies when a session touches contract code. It focuses on failure-prone areas such as transaction handling, account state, confirmation, proofs, and deployment, with wrong and correct patterns plus the underlying failure mode. It targets projects using the Dojo, Cartridge, Garaga, Noir, and starknet.js stack.

The skill moves from scope to measurable rubrics, bias-aware judge prompts, and a runnable Python harness. Generated tooling supports content-hash caching, watch mode, per-step trajectory attribution, and calibration against human labels. The standalone example uses an Anthropic API key, while Claude Code guides the design and scaffolding.

Follow bilingual step-by-step guides from a Claude Design handoff through React Native and Expo setup, Firebase integration, device builds, and a TestFlight upload. Claude Code can run local shell work while the user completes Apple and Firebase account and credential steps. The workflow targets iOS on macOS, excludes Android and EAS, and is centered on TestFlight distribution.

Slacksmith first fixes the topic, scope, and audience, then rewrites the same facts for a boss or technical teammate in several tones. It is a Claude Code/Codex drafting skill focused on reviewable message text rather than autonomous posting, so the final send remains with the user.

Holocron adds role-specific agents and engineering commands to Claude Code for planning, implementation, testing, security review, debugging, and documentation. Its SQLite and FTS5 “crystal” stores corrections and recalls relevant lessons in later sessions, so repeated guidance can become durable project knowledge. The plugin requires Node 18 or newer and an npm install for its native database dependency.

claude-skills-rj0217 packages repeatable Claude Code audits around SOPs, references, and deterministic scripts. Its pre-launch skill reviews common web attack vectors, dependency risks, and client-bundle exposure, with active testing gated on authorization. A second skill examines how AI crawlers reach a site through robots.txt, sitemaps, llms.txt, entry pages, SPA behavior, and framework-specific pitfalls, producing a structured report.

Create text or template messages from natural-language commands or the included Python CLI, with support for multiple recipients and AIM or SMS fallback. The skill is useful for agent workflows or scripts integrating a 5G messaging interface. It requires valid API credentials and remains subject to the interface's rate limits.

Install two focused skills: one for sending email, running campaigns, managing contact lists, verifying domains, and registering webhooks; the other for writing, building, and deploying the Astro Starlight documentation site. They target 3AVA Mail, a self-hosted transactional and marketing email platform with a Resend-compatible wire format. API tasks require a configured key, while documentation work uses a separate skill.

Paste a YouTube, Bilibili, or other yt-dlp-supported URL and Videoink downloads the audio, transcribes it, and creates a local Markdown article bundle. In Claude Code mode, faster-whisper handles transcription locally and the current Claude session writes the article without a separate LLM API. The standalone CLI can instead use OpenAI, Anthropic, OpenRouter, or Ollama, and the project is currently alpha.

AHP Decision Helper guides you from defining alternatives and criteria through pairwise comparisons, then calculates weights, overall scores, and a consistency check with its Node.js calculator. It works for choices such as homes, job offers, or devices where several tradeoffs matter. The recommendation formalizes the preferences you enter, so it supports a transparent decision rather than supplying an objectively unique answer.

Run the skill near the end of a long coding session to capture the current task, concrete progress, stopping point, next actions, decisions, gotchas, references, and open questions. It emits one fenced handoff block capped at roughly 300 tokens, omitting empty sections and linking to project instructions by path instead of copying them. The core prompt can be reused across several coding agents.

The skill detects the stack, classifies frontend, backend, data, jobs, integrations, and infrastructure layers, explores routes and services, then writes a portfolio poster, a detailed multi-page diagram, and an optional user journey. A state file enables cheaper incremental updates after later commits, while the first run is more context-intensive and diagrams.net or draw.io is needed to view and edit the results.

Use this skill to decide where project documentation belongs when files are duplicated, outdated, or packed into an oversized AGENTS.md or CLAUDE.md. It assesses project size, recommends a matching directory structure, and classifies documents by audience and status. Templates cover small, medium, and large projects and work across common AI coding agents.

Start with the learner's target technology, experience, goal, environment, and available study time, then create a learning plan and progress tracker. Lessons advance through small hands-on tasks, staged hints, and code reviews that distinguish bugs, design issues, and optional improvements. The skill supports five teaching languages and several dedicated stacks, but it is intentionally educational rather than a way to rush production code.

Use four modules to clarify purpose, test a decision against personal values, address recurring discipline failures, or conduct a periodic life audit. The process emphasizes written evidence and concrete changes without imposing a spiritual or moral framework. It is not therapy: when signs of acute or prolonged distress appear, the skill stops the exercise and redirects the user toward qualified human or emergency support.