Recently updated open-source projects

Socratic Mirror creates one interactive HTML lesson for a concept. Two unrelated parables introduce the shared mechanism, followed by staged multiple-choice questions on comprehension, misconceptions, analogy, recall, and transfer; each first answer locks and receives immediate feedback. Pages work offline without dependencies, but the method intentionally uses recognition questions rather than free-text explanations.

Import existing CLI and VS Code sessions into Claude Desktop by creating pointer records for transcripts the app does not list yet. The skill performs a dry run, flags live sessions, archives copies, skips duplicates, and supports undo. It is macOS-specific, requires Desktop Code mode to have opened once, and needs an app restart before imported sessions appear.

fumeng combines gpt-image-2 mockups, MCP-fetched components and assets, and a structured frontend design method. Its three playbooks help users explore a visual direction, translate an existing mockup into code, or audit a live interface against the intended design. It works with Claude Code, Cursor, and Windsurf; image-led exploration requires gpt-image-2 access, while the fuller workflows depend on relevant MCP tools.

LinkedIn Outreach Skill combines a profile screenshot with one business configuration file to create personalized outreach drafts. It scores prospects from role, activity, and buying signals, then suggests a non-pitching first message, follow-up cadence, value comments, and reply handling while filtering spammy or overly sales-focused wording. It is designed mainly for existing first-degree connections and responsible low-volume outreach, not mass automation.

This skill collection gives Claude Code package-oriented guidance for more than 250 R packages across data wrangling, visualization, machine learning, statistics, bioinformatics, spatial work, web applications, and package development. After installation, users can request R analysis or code in natural language; R 4.0 or newer is required.

Build a structured profile for a feeling from lexicons, first-person language, affective frameworks, and visual references, then generate creative that separates it from nearby emotions and avoids named clichés. Five profiles and blind comparison records are included. Current validation uses independent model judges rather than market or human studies; the era/cohort layer is fully grounded only for coziness and remains stubbed for the other four feelings.

Pyramid Principle is a Claude Code writing skill that turns a report, proposal, email, or presentation idea into a conclusion-first structure before polishing the wording. It uses SCQA, MECE grouping, question-and-answer logic, and problem trees, and can diagnose vague claims, buried conclusions, or overlapping sections. It is useful when the main difficulty is organizing thought rather than generating prose.

Give Codex or Claude a file, diff, pull request, or repository, and the skill produces a severity-ranked security review with an attack-surface map, CWE and OWASP labels, file locations, assumptions, validation questions, and patches in the original language. Results can be formatted as Markdown, SARIF, GitHub issues, JIRA tickets, or executive summaries. It is defensive only and does not replace a professional penetration test.

Use this Claude Code skill to inspect available MCP capabilities and run narrow, time-bounded, read-only SRE queries. It keeps observed evidence separate from inference and refuses writes, deploys, restarts, migrations, deletes, scaling changes, or failovers without explicit confirmation. Missing data is reported as unobserved, not treated as proof.

Provide a screenshot, sketch, or text prompt to generate a browser-ready, single-file HTML wireframe built from predefined tokens, components, and icons. A bundled linter rejects hard-coded colors, emoji icons, and improvised component classes, keeping drafts structurally consistent. It is intended for layout and information hierarchy, not polished visual design.

Use it before design or implementation to resolve conflicting inputs, missing goals, priorities, page states, and no-touch boundaries. After a short diagnostic interview, it creates a read-only design brief plus agent-specific execution constraints for Claude Code, Cursor, or Codex. Checkpoints can require human approval before work continues, helping prevent scope drift and repeated rework.

Give it a product recommendation, store visit, or personal experience and receive a Xiaohongshu-style post with a stronger title, mobile-friendly short paragraphs, emoji pacing, and layered hashtags. It is aimed at creators, brands, e-commerce teams, and casual reviewers who want to reshape stiff source material into copy that feels native to the platform.

Start with one or two X accounts you already trust, then rank the people they repeatedly amplify through reposts, quotes, replies, or mentions. Cross-seed overlap raises a candidate, but the skill still walks you through checking past claims instead of declaring anyone trustworthy. It requires a free X API key and never follows accounts for you.

Git & GitHub Workflow Skill gives Claude Code stack-agnostic rules for pre-flight checks, atomic commits, branches, pull requests, branch protection, CI, tags, and releases. Its default CI model requires lint, typecheck, test, and build on PRs and pushes to main, with copyable templates included. The rigor may be unnecessary for tiny throwaway repositories.

Query the NHL's public web and stats endpoints for player and goalie numbers, standings, scores, schedules, rosters, playoffs, and filters such as nationality or team. The skill needs no API key, but the endpoints are unofficial and may change, so field names and season behavior require care.

Before planning a change, inspect the codebase, identify the forks that affect implementation, and ask structured multiple-choice questions with a recommendation and explicit tradeoffs. Related questions are batched, then the resolved choices become a sequenced plan. It is intended for open decisions, not work that is already fully specified.

InvestEye walks a founder through 20 investor-readiness checkpoints one question at a time, covering the problem, market, team, distribution, business model, execution, and networking. It steelmans the skeptical investor before awarding a green result, then creates a stage-aware dossier with a score, strengths, gaps, and weekly actions.

Linode-API Skill combines a stdlib-only Python CLI with a matching Claude skill for operating the Linode API. It keeps tokens in OS-native storage, classifies requests into read, mutating, destructive, billable, financial, and privilege tiers, and requires explicit flags before mutations. Every mutation is recorded locally. The project is an early alpha, so production use calls for dry runs and a code audit first.