claude-skills

TypeScript, JavaScript, React, and Node.js coding standards — naming conventions, type safety, error handling, immutability, React patterns, performance, and code quality. Always activate when writing, reviewing, or refactoring TypeScript or JavaScript code; setting up linting or type-checking; enforcing naming or structural conventions; or when the user asks how to structure a component, handle an error, type a value, or write a hook. Also activate proactively when spotting `any`, unsafe mutations, swallowed errors, or missing cleanup in user code.

Self-refinement loops that improve output quality through structured reflection, multi-perspective critique, and persistent memory. Always activate when the user says "reflect on this", "reflect on your last response", "critique this", "review from multiple angles", "/reflect", "/critique", "/memorize", or any variant of asking Claude to review, self-assess, or iterate on its own output. Also activate automatically whenever the user includes the word "reflect" as part of any task instruction (e.g. "implement X then reflect", "draft this and reflect on it"). Proven to increase output quality by 8–21% over one-shot responses. Three modes: Reflect (self-refinement), Critique (multi-judge panel), Memorize (persist insights to memory). Never skip this skill when the user asks for reflection or critique — it is the authoritative handler for all self-improvement loops.

Activates elite-level coding discipline — precise, complete, idiomatic, and verified. Use this skill for ANY coding task: writing functions, implementing features, fixing bugs, refactoring, building scripts, designing classes, writing algorithms, debugging, optimizing, or explaining complex code. Trigger on phrases like "write a", "implement", "code this", "fix", "build", "make a function", "add feature", "script", "help me code", "optimize", "debug", "refactor", "create a class", "how do I", or any time the user wants code written or improved. When in doubt, always apply this skill — precise coding is never the wrong call.

Refresh project intelligence and re-orient on the current codebase. Use this skill whenever the user says things like "catch me up", "what was I working on?", "re-orient yourself", "summarize the project state", "what's in this repo?", "what changed?", "project status", "context refresh", "/warm", or after they mention switching branches, pulling changes, cloning a repo, or starting a new session on an existing project. Trigger even when the request is implicit — if the user dives straight into a task without context and the project state is unknown, run this proactively.

Decision framework and implementation patterns for choosing between regex and LLM when parsing structured text. Always activate when the user is parsing repeating structured formats (quizzes, forms, invoices, tables, documents), deciding how to extract data from text, building a text processing pipeline, or asking about cost/accuracy tradeoffs in text extraction. Also activate when the user is about to send all their text to an LLM — they almost certainly don't need to.

Docker and Docker Compose best practices for local development, container security, networking, volume strategies, and multi-service orchestration. Always activate when the user is writing a Dockerfile, setting up Docker Compose, troubleshooting container issues, designing multi-container architectures, reviewing images for security or size, managing secrets in containers, or migrating to a containerized workflow — even if they don't explicitly ask for "best practices".

Write articles, guides, blog posts, tutorials, newsletter issues, LinkedIn posts, and other long-form content in a distinctive voice derived from supplied examples or brand guidance. Always activate when the user wants polished written content longer than a paragraph — especially when voice consistency, structure, and credibility matter. Also activate when turning notes, transcripts, or bullet-point research into polished prose, and when the user asks to "write", "draft", "rewrite", or "clean up" any piece of long-form content, even if they don't call it an article.

Conduct market research, competitive analysis, investor due diligence, and industry intelligence with source attribution and decision-oriented summaries. Always activate when the user wants market sizing, competitor comparisons, fund research, technology scans, TAM/SAM/SOM estimates, or any research that informs a business decision. Also activate when the user says "research X", "what do you know about Y company", "compare these competitors", "who should I talk to about Z", or "help me think through this market" — even without the word "research".

REST API design patterns including resource naming, HTTP methods, status codes, pagination, filtering, error responses, versioning, rate limiting, idempotency, and security. Always activate when the user is designing or reviewing API endpoints, adding pagination or filtering, implementing error handling, planning versioning, building public or partner-facing APIs, or asks how to structure a route, response, or status code — even if they don't use the word "API design".

Enforce cloud security best practices across AWS, Vercel, Railway, and Cloudflare deployments. Always activate this skill when the user is deploying to a cloud platform, writing or reviewing IaC (Terraform, CloudFormation, Pulumi), configuring IAM roles or policies, managing secrets, setting up CI/CD pipelines, configuring databases or storage, implementing logging/monitoring, or asking about cloud security — even if they don't explicitly mention security. When in doubt, activate. A missed security check is harder to fix than an unrequested one.

Regression testing strategies specifically for AI-assisted development, where the same model writes and reviews code — creating systematic blind spots that only automated tests can catch. Always activate when an AI agent (Claude Code, Cursor, Codex) has modified backend logic or API routes, a bug was found and fixed, sandbox/production path parity needs verification, a bug-check or review command is being run, or the user asks how to prevent AI-introduced regressions. Also activate when the user reports a bug was "fixed" but keeps coming back.

Create reusable, composable CLI scripts and utilities for the Hugging Face API and hf CLI. Builds bash/Python/TSX tools that chain, pipe, and stream HF API data — model search, dataset discovery, model card parsing, trending models, paper lookups, repo management, and more. Always activate when the user wants to query the HF API, build an HF shell script or pipeline, search or filter HF models/datasets, parse model cards, automate Hub tasks, or says anything like "write a script to..." or "build a tool that..." involving Hugging Face — even if they don't say "skill".

Production-ready patterns for building LLM applications — RAG pipelines, agent architectures, prompt engineering, caching, observability, and LLMOps. Always activate when the user is designing an LLM-powered application, implementing RAG or retrieval, building agents with tools, setting up LLMOps monitoring, choosing between agent architectures, or asking about prompt chaining, caching, evaluation, or streaming LLM responses. Also activate when the user asks about prompt injection, model fallback, cost optimization, or embedding model selection.

Add, import, and manage evaluation results in Hugging Face model cards. Supports extracting eval tables from README content, importing benchmark scores from Artificial Analysis API, and running custom model evaluations with vLLM/lighteval/inspect-ai on HF Jobs or locally. Works with the model-index metadata format for leaderboard and Papers with Code integration. Always activate when the user mentions HF model card evaluation, benchmark scores, model-index YAML, evaluation results, leaderboard submission, Artificial Analysis benchmarks, lighteval, inspect-ai, running evals on HF Jobs, or adding eval metrics to a model card — even if they don't say "skill".

Create, manage, query, and transform Hugging Face Hub datasets for LLM fine-tuning, SFT, RLHF, DPO, and training data pipelines. Supports initializing repos, defining configs/system prompts, streaming row updates, SQL-based querying/transformation via DuckDB, and pushing subsets to the Hub. Designed to work alongside the HF MCP server. Always activate when the user mentions Hugging Face datasets, HF Hub, training data creation, dataset preprocessing, pushing to Hub, querying a dataset with SQL, building a fine-tuning dataset, or filtering/transforming HF datasets — even if they don't say "skill".

Anthropic Claude API patterns for Python and TypeScript. Covers the Messages API, streaming, tool use, vision, PDF input, extended thinking, prompt caching, batches, token counting, structured outputs, and agentic loops. Always activate when the user is building with the Claude API or Anthropic SDKs, code imports `anthropic` or `@anthropic-ai/sdk`, or the user asks about tool use, streaming, cost optimization, model selection, or any Anthropic API pattern — even if they don't use the word "API".

PyTorch deep learning patterns and best practices for building robust, efficient, and reproducible training pipelines, model architectures, and data loading. Always activate when the user is writing PyTorch models or training scripts, reviewing deep learning code, debugging training loops or data pipelines, optimizing GPU memory or training speed, setting up reproducible experiments, implementing transfer learning or fine-tuning, or asks anything about nn.Module, DataLoader, autograd, AMP, or torch.compile.

X/Twitter API integration patterns for posting tweets and threads, reading timelines, searching, media upload, pagination, and real-time streaming. Covers OAuth 2.0 and 1.0a authentication, rate limit handling, and content policy compliance. Always activate when the user wants to interact with X programmatically, says "post to X", "tweet this", "X API", "Twitter API", or is building an X bot, scheduler, analytics tool, or content automation workflow.

Deployment workflows, CI/CD pipelines, Docker containerization, health checks, rollback strategies, database migrations, and production readiness for web applications. Always activate when the user is setting up CI/CD, Dockerizing an application, planning a deployment strategy, implementing health checks, preparing a production release, configuring environments, or asking how to deploy, roll back, or migrate a database safely. Also activate proactively when spotting hardcoded secrets, missing health checks, or root-running containers.

Python testing patterns using pytest — fixtures, mocking, parametrization, async tests, and TDD methodology. Always activate when the user is writing, fixing, or designing tests; setting up a test suite; asking about coverage; or working with pytest configuration. Trigger on phrases like "write a test", "add tests for", "how do I test", "mock this", "fixture for", "parametrize", "test coverage", "pytest setup", "TDD", "how do I assert", "test this function", or any request explicitly about testing Python code. Also activate proactively when the user is writing new Python code without any tests — TDD means tests come first.

Run a structured pre-PR verification sweep across build, types, lint, tests, security, and diff review. Always activate this skill when the user says things like "verify my changes", "run checks", "is this ready for PR?", "check my code", "run the verification loop", or after completing a feature, refactor, or significant change. Also activate proactively before any PR creation — even if not explicitly asked. Works for TypeScript, JavaScript, and Python projects.

Build a fully automated AI-powered data collection agent for any public source — job boards, prices, news, GitHub, sports, anything. Scrapes on a schedule, enriches data with a free LLM (Gemini Flash), stores results in Notion/Sheets/Supabase, and learns from user feedback. Runs 100% free on GitHub Actions. Use when the user wants to monitor, collect, or track any public data automatically, or says things like "build a bot that checks...", "monitor X for me", "collect data from...", or "alert me when...".

Enforce a research-before-coding workflow — search for existing tools, libraries, and patterns before writing custom code. Use this skill whenever the user asks to add functionality, integrate a new capability, create a utility or helper, or add a dependency. Trigger on phrases like "add X functionality", "integrate X", "I need a library for", "write a helper to", "implement X from scratch", or any time you're about to write non-trivial code that might already exist as a package or tool. When in doubt, search first — reinventing a solved problem is never the right call.

Enables the assistant to continuously improve by reflecting on tasks, recording mistakes, and retrieving past learnings before solving similar problems. Use this skill when a task fails or produces an error, when the user corrects the assistant, when a better solution is discovered mid-task, when an API or tool behaves unexpectedly, when a repeated workflow could be optimized, or whenever a non-obvious solution is found that is worth remembering. Trigger proactively — before complex tasks, check for relevant past learnings; after complex tasks, check if a learning entry should be recorded.

Security review checklist and implementation patterns for web applications — secrets management, input validation, authentication, authorization, SQL injection, XSS, CSRF, rate limiting, security headers, and sensitive data handling. Always activate when the user is implementing authentication or authorization, handling user input or file uploads, creating API endpoints, working with secrets or credentials, implementing payment features, storing or transmitting sensitive data, or integrating third-party APIs. Also activate proactively when spotting hardcoded secrets, missing auth checks, or unsafe data handling in user code.

Analyze an unfamiliar codebase and generate a structured onboarding guide with architecture map, key entry points, conventions, and a starter CLAUDE.md. Always activate when the user says "onboard me", "help me understand this codebase", "walk me through this repo", "generate a CLAUDE.md", "update the CLAUDE.md", "I just joined this project", or when opening a repo with Claude Code for the first time. Also activate when the user seems lost in an unfamiliar codebase — even without an explicit request.

Act as an expert code reviewer — producing structured feedback with severity tiers, not writing or fixing code. Use this skill when the user wants passive review, analysis, or a quality assessment of existing code without asking for an immediate rewrite. Trigger on: "review this", "look at this code", "check my code", "audit this", "is this safe?", "is this good practice?", "find bugs", "review my PR", "what's wrong here?", "give me feedback on", or any time the user pastes code and wants analysis rather than a fix. Do NOT trigger when the user asks you to fix, implement, refactor, or build something — use elite-coder for that instead.