Start with the job, then find an open-source answer

Give it a research question and the Claude Code skill decomposes it, searches the open web with Tavily, and, when relevant, adds GitHub, academic indexes, or current library documentation. It grades and corroborates sources before producing a research plan, cited report, source file, and claim-to-evidence map. Claude Code, Tavily MCP, and Python 3.10 or later are required.

webcpu-agent-skill gives coding agents verified references, examples, and templates for building browser-based Canvas2D scenes with EASEL.js 0.5.0. It packages the same guidance for Claude, Codex, OpenCode, Copilot, and other agent environments, with a rule to report unknown APIs rather than invent them. Rendering still requires a browser DOM canvas; Deno can type-check, bundle, and serve the app but does not provide headless Canvas2D by itself.

Clawproof Checks evaluates an agent system across ten operational areas, including least privilege, audit trails, prompt injection, human escalation, rollback, secrets, and regression testing. The packaged Claude skill can inspect a codebase, runbook, or configuration and return a scored report with prioritized remediation. The same checks are also available as YAML and consolidated JSON for vendor-neutral programmatic use.

routine-writer teaches agents to express inbox triage, pull-request reviews, alert checks, briefings, trackers, and fixers as terse routines rather than long operational playbooks. It defines a title, tagline, and body pattern with verb-first instructions, explicit iteration, optional actions, empty-state handling, and a short report. The skill follows the Agent Skills specification and can be installed in compatible harnesses.

Fix My Resume interviews you about your experience, target role, and real achievements before building or rewriting an ATS-friendly single-column DOCX. It diagnoses vague bullets, passive voice, repeated verbs, formatting problems, and missing job-description keywords, then provides a heuristic score and tuning workflow. It refuses to invent metrics and asks for evidence instead; language checks are English-focused, and neither the score nor the resume can guarantee interviews.

html-to-pdf-agent-skill uses Playwright to render remote HTML pages as local PDFs. It can handle one page or wait for a single-page application to load, extract real documentation links from the rendered sidebar, convert each page, and merge the results. Page size, margins, styles, and authentication can be customized; Playwright with Chromium is required, and full-site merging additionally depends on PyPDF2.

Personal Intelligence Skill guides you through combining direct RSS feeds and Google News queries into a focused daily feed, then filtering, scoring, deduplicating, and archiving the results. The reference architecture runs on GitHub Actions and publishes a static site to Cloudflare Pages, with optional OpenAI TTS audio briefs. It suits people who want direct control over sources and relevance without maintaining a backend.

Feed it chats, social posts, photos, and your own context to map the relationship timeline, estimate its direction, and suggest practical next moves. It is designed for reflection around crushes, mixed signals, or cold spells, not as professional counseling, and it declines serious abuse or self-harm scenarios.

Claude Think Twice combines a reflection skill with a Claude Code pre-push hook. When a push is attempted, it inspects the pending file types, presents the corresponding minimum scan matrix, and asks the human whether to proceed, helping expose rushed assumptions before CI. The hook cannot verify that scans were actually run and is not a linter, test runner, or replacement for CI; its enforcement is the human confirmation moment.

oss-contribution-harness gives Claude Code a forward-only workflow for deduplicating candidate work, studying repository norms, reproducing the issue, routing security findings, drafting the change, and planning follow-up. Every network-visible action requires fresh human confirmation, and the workflow also throttles contribution volume; it expects authenticated gh plus git and jq.

The skill asks which Apple platform, framework, and interface you are building before loading only the relevant HIG references. It applies platform-correct navigation, semantic colors, Dynamic Type, SF Symbols, touch targets, VoiceOver, and motion guidance across SwiftUI, UIKit, AppKit, and Apple-style web or Figma mockups.

Use JARVIS-Starter to bootstrap a new Claude Code project or adopt an existing repository through gap analysis. It adds project-scoped hooks and a `.jarvis/` workspace that maintain wiki context, recommend models, recall past decisions, track focus, and watch for hardcoded secrets or `.env` leaks without replacing existing documentation. The assistant is aimed at users who prefer plain-language development, and each added behavior remains optional.

Preflight sends a plan, specification, or RFC to three to five independent domain experts before implementation begins. It combines their findings into a ranked report with a verdict, must-fix items, explicit decision cards, cross-confirmed concerns, and gaps no panelist covered. A human gate confirms the proposed roster first; the workflow is intended for complex multi-domain designs, not trivial plans, debugging, or review of code that already exists.

Research Backlog fetches unstarted issues for a Linear team and assigns a separate research agent to each ticket. The agents investigate the technical domain, official documentation, prior art, best practices, and likely gotchas, then post structured comments back to Linear. It finishes with a summary table and requires a Linear API key plus `jq`.

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.

Give it an HTML file or folder to capture desktop, mobile, and optionally tablet viewports, then combine the normalized panels into one PNG or JPEG. A temporary local HTTP server preserves relative CSS, scripts, images, and fetch calls, while full-page capture, batch processing, separate device files, and custom backgrounds are available.

Use this skill when a converted or hand-built WordPress theme traps copy inside PHP templates and leaves forms or styling brittle. It guides Customizer refactors, high-priority CSS overrides, generator-script fixes, database-versus-template decisions, and form handoffs to contact pages or n8n. The approach relies on WordPress core rather than paid plugins or page builders and includes a small PHP brace-balance checker for safer edits.

The guide walks a 10–50 person company through moving medical leave, vacation requests, employee records, and approvals out of chat and ad hoc spreadsheets. Employees submit from AppSheet, supervisors approve or reject through email buttons, approval history is recorded, and HR reports from Google Sheets; the first build is mostly configuration and copied formulas.