Core AIBeginnerPractice
Prompt engineering
Write explicit goals, context, constraints, examples and output formats.
Fast AI · Practical knowledge
AI Skills Directory: the practical guide to modern AI systems
Learn the core skills behind AI assistants, agents and production systems—from prompting and retrieval to MCP, evaluation, security and AI search optimization. Each entry explains what it does, when it helps and where to read the primary documentation.
Scope note: this is a curated starting point, not a claim to list every community skill or vendor feature. Product capabilities and documentation change; follow the linked primary sources for current details.
Recommended learning path
- 1. Prompting and structured outputs
- 2. Tool calling and API boundaries
- 3. Evaluation and regression tests
- 4. RAG and source-grounded answers
- 5. Security, permissions and monitoring
- 6. Agents only where they add value
Searchable skills catalog
Explore skills by task, level and category
A broad practical catalog of AI capabilities, engineering practices and upstream skill libraries. Each entry explains what it does, when to use it and the production controls that make it safer to operate.
71 skills shown · 71 catalog entries
Core AIIntermediatePractice
Structured outputs
Constrain responses to a JSON schema or typed contract.
Core AIAdvancedPractice
Context engineering
Select, order and compress context relevant to the current task.
Core AIAdvancedPractice
Model routing
Route tasks to models based on capability, latency, privacy and cost.
Core AIIntermediatePractice
Multimodal understanding
Interpret images, audio, video or mixed media alongside text.
Core AIAdvancedPractice
Fine-tuning
Adapt a model with permitted examples for a repeatable behavior gap.
Agents & orchestrationAdvancedPractice
AI agent design
Give an agent a bounded goal, tools, state, stop conditions and escalation rules.
Agents & orchestrationIntermediatePractice
Function and tool calling
Let a model request narrowly defined application functions.
Agents & orchestrationAdvancedPractice
Multi-agent orchestration
Coordinate specialized agents with explicit handoffs and shared contracts.
Agents & orchestrationIntermediatePractice
Human-in-the-loop
Pause for approval when actions are costly, irreversible or sensitive.
Agents & orchestrationIntermediateTooling
Model Context Protocol (MCP)
Expose tools and resources through a shared protocol for compatible clients.
Agents & orchestrationAdvancedPractice
Workflow and state-machine design
Represent long tasks as explicit states and transitions rather than opaque loops.
Automation & integrationAdvancedTooling
Browser automation
Automate browser navigation, extraction and controlled UI interactions.
Automation & integrationIntermediatePractice
API integration
Connect tools to REST or GraphQL services with typed contracts.
Automation & integrationAdvancedPractice
Event-driven automation
Trigger work from webhooks, queues, schedules or business events.
Automation & integrationAdvancedTooling
MCP server development
Build an MCP server that exposes narrowly scoped tools and resources.
Knowledge & memoryAdvancedPractice
RAG architecture
Retrieve evidence from approved documents and ground model responses in it.
Knowledge & memoryIntermediatePractice
Embeddings and semantic search
Use vector representations to retrieve semantically related material.
Knowledge & memoryAdvancedPractice
Hybrid search and reranking
Combine keyword and vector retrieval, then rerank candidate results.
Knowledge & memoryAdvancedPractice
Persistent memory
Store durable preferences or task state with provenance and update rules.
Knowledge & memoryAdvancedPractice
Knowledge graphs
Represent entities and relationships explicitly for multi-hop queries.
Evaluation & observabilityIntermediatePractice
Evaluation datasets
Build representative examples with expected outcomes and failure cases.
Evaluation & observabilityAdvancedPractice
LLM-as-judge calibration
Use model-based grading with rubrics, calibration and human spot checks.
Evaluation & observabilityAdvancedPractice
Tracing and observability
Record agent steps, tool calls, latency, errors and cost for debugging.
Evaluation & observabilityIntermediatePractice
Regression and integration testing
Test expected outputs and end-to-end interactions after each change.
Evaluation & observabilityIntermediatePractice
Cost and latency optimization
Measure token usage, caching, model selection and end-to-end response time.
Security & governanceAdvancedPractice
Prompt-injection defense
Treat user input and retrieved content as untrusted data, not privileged instructions.
Security & governanceIntermediatePractice
Secrets and permission management
Limit access to credentials, tools, files and external actions.
Security & governanceAdvancedPractice
Supply-chain and skill review
Inspect skill scripts, dependencies, licenses and provenance before installation.
Security & governanceAdvancedPractice
Privacy and data governance
Control collection, access, retention and processing of personal or confidential data.
SEO, AEO & growthIntermediatePractice
Technical SEO
Audit crawlability, indexability, canonical URLs, metadata and internal links.
SEO, AEO & growthIntermediatePractice
AI search optimization (AEO/GEO)
Make pages helpful, clear, evidence-backed and easy for readers and crawlers to interpret.
SEO, AEO & growthIntermediatePractice
Keyword and intent research
Group queries by intent, demand, competition and business relevance.
SEO, AEO & growthIntermediatePractice
Content briefs and editorial QA
Create evidence-led briefs with audience needs, sources and quality criteria.
SEO, AEO & growthAdvancedPractice
Conversion-rate optimization
Improve page clarity, calls to action and user journeys using observed behavior.
SEO, AEO & growthIntermediatePractice
Analytics and attribution
Connect acquisition signals to meaningful user actions and conversions.
Software engineeringIntermediateRepository
Frontend engineering
Build accessible, responsive interfaces using component-based UI patterns.
Software engineeringAdvancedRepository
Backend and database design
Design validated APIs, durable schemas, indexes and transactional operations.
Software engineeringIntermediateRepository
Code review and refactoring
Find correctness, maintainability, security and performance issues in code.
Software engineeringIntermediatePractice
Test-driven development
Write failing tests before implementing behavior, then refactor safely.
Software engineeringIntermediateRepository
Accessibility and UX audits
Review keyboard navigation, semantics, contrast and assistive-technology behavior.
Software engineeringAdvancedRepository
Performance engineering
Profile rendering, network requests, caching and resource use.
Cloud & operationsAdvancedRepository
CI/CD and release engineering
Automate checks, builds, previews and controlled releases.
Cloud & operationsAdvancedPractice
Cloud deployment
Configure hosting, runtime settings, networking and deployment environments.
Cloud & operationsAdvancedRepository
Containers and infrastructure as code
Make runtime environments and infrastructure reproducible.
Cloud & operationsAdvancedPractice
Incident response and recovery
Detect failures, triage impact and restore a known-good service state.
Data & researchIntermediateRepository
SQL and data analysis
Query, validate and explain structured data with reproducible analysis.
Data & researchIntermediateRepository
Research and source verification
Find primary sources, compare claims and record citations and uncertainty.
Documents & mediaIntermediateRepository
Document processing and OCR
Extract, transform and validate information from PDFs and office documents.
Documents & mediaIntermediateRepository
Spreadsheet and presentation generation
Produce structured spreadsheets and presentations from verified source data.
Documents & mediaAdvancedRepository
Image, audio and video workflows
Generate or transform media and extract useful information from audiovisual sources.
Content & communicationBeginnerRepository
Technical writing and documentation
Explain systems, procedures and APIs for a defined audience.
Content & communicationIntermediateRepository
Translation and localization
Adapt language, terminology and UX for a target locale.
Business workflowsIntermediateRepository
Customer support automation
Classify requests, retrieve policy evidence and draft consistent responses.
Business workflowsIntermediateRepository
Sales and CRM workflows
Summarize account context, prepare follow-ups and maintain structured deal data.
Business workflowsIntermediatePractice
Project planning and task management
Convert goals into milestones, owners, dependencies and measurable acceptance criteria.
Skill developmentIntermediateTooling
Skill authoring
Package reusable instructions, examples and optional resources in a documented skill.
Skill developmentIntermediateTooling
Skill discovery and installation
Find upstream skills and install them into supported agent environments.
Skill developmentAdvancedPractice
Skill compatibility and portability
Check whether instructions, tools and file layouts work across agent products.
Skill developmentAdvancedPractice
Skill registry maintenance
Maintain catalog metadata, upstream links, category tags and freshness status.
Specialized AIAdvancedPractice
Robotics and physical-world agents
Connect perception and planning to real-world devices under strict safety constraints.
Specialized AIAdvancedPractice
Scientific and domain-specific AI
Apply AI to specialized fields with domain evidence and expert review.
Security & governanceAdvancedPractice
Responsible AI and policy compliance
Assess bias, transparency, harmful outputs and compliance requirements.
SEO, AEO & growthIntermediatePractice
AI search and content extraction
Structure content so people and machines can locate key facts and supporting evidence.
Business workflowsIntermediatePractice
Email and lifecycle automation
Trigger relevant messages from verified user actions and lifecycle states.
SEO, AEO & growthAdvancedPractice
Experiment design
Test hypotheses with measurable outcomes and clear decision rules.
Software engineeringAdvancedPractice
API documentation and SDK generation
Generate and maintain API references and client examples from contracts.
Security & governanceAdvancedPractice
Authentication and authorization
Design login, sessions, tokens, roles and resource-level access controls.
Cloud & operationsAdvancedPractice
Queues and durable background jobs
Run long or asynchronous tasks outside a user request lifecycle.
Cloud & operationsIntermediatePractice
Caching and rate limiting
Reduce repeated work and protect services from overload.
Cloud & operationsAdvancedPractice
Release and rollback validation
Confirm a release works and can be safely reverted.
Catalog policy: skills and frameworks are not interchangeable. Repository links are upstream references, not an endorsement or security audit. Check license, compatibility, revision, scripts and dependencies before installation. This directory does not execute third-party skills on the website.
Production integration: from skill repository to reliable workflow
A skill is a reusable instruction package—not automatically a running service or a drop-in software dependency. Use this controlled pipeline before adopting one in a real product.
1. Discover and classify
Record the upstream URL, publisher, purpose, supported agents, version and last-checked date.
2. Review license and provenance
Confirm redistribution and commercial-use terms; pin a commit or release rather than tracking an unreviewed moving branch.
3. Inspect content and code
Review SKILL.md, scripts, dependencies, network access, file access and commands for unsafe behavior.
4. Adapt to your runtime
Check the target agent's skill format, tool names, permissions, environment requirements and compatibility.
5. Test in isolation
Use a sandbox and synthetic data; test expected cases, adversarial inputs, failures and missing dependencies.
6. Add approval boundaries
Require explicit human approval for publishing, payments, destructive operations and sensitive data access.
7. Release with observability
Version the skill, log outcomes safely, monitor cost and errors, and keep a rollback path.
8. Refresh and retire
Run scheduled link and version checks, re-review upstream changes, and remove stale or unsafe entries.
9. Measure real outcomes
Track task success, defect rate, latency, cost and user value—not the number of skills installed.
For FastSEOHub, this change implements the public discovery and learning layer. It does not silently install third-party code into the production server or grant skills access to user accounts.
Real GitHub skill sources
Explore actual AI agent skill libraries
These are source repositories and community indexes containing reusable agent instructions, often packaged as SKILL.md files. Open the original repository to inspect each skill, its license, dependencies and installation instructions. This directory links to upstream projects; it does not claim to own, copy or security-audit their contents.
Official · documents, design, development
Anthropic Agent Skills
Examples for document workflows, frontend design, MCP servers, testing and reusable skill creation.
Open source repository ↗Official · coding agent workflows
OpenAI Skills for Codex
Reusable skills and workflows for Codex. Check each skill folder for current scope and license.
Open source repository ↗Official · web development
Vercel Agent Skills
Skills for frontend work, React and Next.js practices, design and modern web app workflows.
Open source repository ↗Official · Google ecosystem
Google Skills
Explore Google-published skill resources and the supported products documented in the repository.
Open source repository ↗Official · developer ecosystem
Microsoft Skills
Browse Microsoft-published skills and related tooling; confirm product and platform requirements upstream.
Open source repository ↗Official · Azure
Microsoft Azure Skills
Cloud and Azure-focused workflows for development and platform tasks.
Open source repository ↗Official · edge and Workers
Cloudflare Skills
Skills related to Cloudflare development and edge application workflows.
Open source repository ↗Official · AWS
AWS Agent Toolkit for AWS
AWS-focused agent toolkit and skill resources. Review the README before installing or granting cloud access.
Open source repository ↗Official · programmatic video
Remotion Skills
Reusable guidance for creating video with Remotion and React.
Open source repository ↗Tooling · discovery and installation
Vercel Skills CLI
CLI and ecosystem resources for finding and installing skills into supported coding agents.
Open source repository ↗Community index · broad catalog
VoltAgent Awesome Agent Skills
A large community-curated index spanning many vendors and use cases. Listings are not security endorsements.
Open source repository ↗Community collection
sickn33 Agentic Awesome Skills
A broad collection of agent skills. Inspect provenance and licensing for each included project.
Open source repository ↗Community · GitHub Copilot
GitHub Awesome Copilot
Community resources and custom instructions, prompts and agent workflows for GitHub Copilot.
Open source repository ↗Community · engineering workflow
Matt Pocock Skills
Developer workflow skills covering planning, testing, architecture and code quality practices.
Open source repository ↗Open standard
Agent Skills Specification
Learn the portable format and conventions used by SKILL.md-based agent skills.
Open source repository ↗Community · AI applications
Awesome LLM Apps
Examples and projects for LLM applications and agentic systems; not every project is a drop-in skill.
Open source repository ↗Safety note: a skill is executable-adjacent content, not automatically safe software. Before use, inspect scripts and dependencies, avoid exposing secrets, grant minimum permissions, and test in a sandbox. Community indexes may contain stale, duplicated or unreviewed entries.
Skill coverage map: major capability areas
Use this map to discover the kinds of skills available across the linked libraries. A topic is a capability area, not a claim that every linked provider publishes a dedicated skill for it.
AI foundations
Prompt design, instruction hierarchy, context windows, token budgeting, structured outputs, JSON schemas, model selection.
Agent architecture
Planning, task decomposition, tool use, function calling, agent loops, state machines, handoffs, multi-agent orchestration.
Memory and knowledge
Persistent memory, retrieval-augmented generation, embeddings, vector databases, hybrid search, reranking, knowledge graphs.
Coding and software delivery
Repository exploration, code generation, refactoring, debugging, code review, TDD, test generation, documentation, migrations.
Frontend and product design
React, Next.js, accessibility, responsive design, design systems, UI prototyping, visual QA, performance optimization.
Backend and APIs
API design, authentication, OAuth, webhooks, databases, queues, caching, rate limits, schema validation, integrations.
Cloud and DevOps
CI/CD, containers, infrastructure as code, deployment, observability, incident response, rollback, cloud cost control.
Data and analytics
SQL, data cleaning, spreadsheets, dashboards, statistical analysis, ETL, data quality, visualization and reporting.
Research and knowledge work
Web research, source verification, document synthesis, citation checks, literature review, fact checking and summarization.
Content and communication
Writing, editing, translation, localization, brand voice, email, presentations, reports, documentation and editorial QA.
SEO and AI search
Technical SEO, crawling, indexing, internal links, schema markup, content briefs, query intent, AEO/GEO and search analytics.
Marketing and growth
Keyword research, landing pages, conversion optimization, campaign analysis, product positioning, email funnels and experimentation.
Security and privacy
Threat modeling, prompt injection, secret handling, permissions, dependency review, privacy checks, audit logs and secure defaults.
Evaluation and reliability
Golden datasets, regression tests, LLM-as-judge calibration, tracing, latency measurement, failure analysis and release gates.
Multimodal and media
Image understanding, OCR, speech, transcription, audio generation, video workflows, diagrams and media asset processing.
Business and productivity
Project planning, task tracking, CRM workflows, meeting notes, customer support, operations and repeatable SOPs.
Documents and office files
PDF extraction, DOCX creation, spreadsheets, presentations, structured reports, document conversion and template-based output.
Automation and browsers
Browser automation, form filling, UI testing, scheduled jobs, event-driven workflows and human approval checkpoints.
Foundations
Prompt engineering
Write clear instructions, provide relevant context, define constraints and specify an output format. Test prompts against representative examples instead of relying on one impressive response.
Useful for: reliable assistants, content workflows and structured extraction.
Read primary documentationStructured outputs
Ask models for schema-constrained data such as JSON, then validate the result in code. A valid structure does not automatically make the values true, so validate facts separately.
Useful for: APIs, forms, data extraction and automation.
Read primary documentationContext management
Select and organize the information the model needs for the current task. Remove irrelevant context, preserve source references and make limits explicit.
Useful for: long documents, multi-step work and lower-noise answers.
Read primary documentationAgents and tools
AI agent design
Give an agent a bounded goal, a small set of tools, observable progress, stopping conditions and a human escalation path. Start with a deterministic workflow before adding autonomous planning.
Useful for: research, support operations and multi-step tasks.
Read primary documentationTool / function calling
Let a model request specific application functions with validated arguments. Your application—not the model—must enforce permissions, validate inputs and execute side effects.
Useful for: search, databases, calendars and business APIs.
Read primary documentationModel Context Protocol (MCP)
Use a standard protocol to expose tools and resources to compatible AI clients. Treat every tool as a security boundary and grant only the minimum required access.
Useful for: connecting assistants to repositories, files and internal systems.
Read primary documentationMulti-agent coordination
Split work into roles only when the tasks genuinely differ, define the hand-off format and reconcile conflicting results. More agents can increase latency, cost and failure points.
Useful for: research plus review, coding plus testing, or specialist pipelines.
Read primary documentationKnowledge and memory
Retrieval-augmented generation (RAG)
Retrieve relevant passages from trusted documents and provide them as evidence for an answer. Measure retrieval quality and require citations to retrieved sources where appropriate.
Useful for: knowledge bases, policy assistants and documentation search.
Read primary documentationEmbeddings and semantic search
Represent text as vectors to find conceptually related content. Combine semantic similarity with keyword search and filters when exact names, codes or dates matter.
Useful for: document discovery and recommendation systems.
Read primary documentationPersistent memory
Store only information that is useful across sessions, with clear scope, provenance, update rules and deletion controls. Do not treat inferred or stale memories as confirmed facts.
Useful for: personalized assistants and long-running workflows.
Read primary documentationQuality and safety
Evaluation and testing
Build a representative test set, define measurable success criteria and compare versions before release. Track factuality, task completion, latency, cost and failure cases.
Useful for: preventing regressions in prompts, models and agent workflows.
Read primary documentationGuardrails and prompt-injection defense
Treat user content and retrieved documents as untrusted data. Validate tool arguments, restrict permissions, separate instructions from data and require approval for consequential actions.
Useful for: agents that browse, read files or modify systems.
Read primary documentationHuman review and observability
Log meaningful events without collecting unnecessary sensitive data. Provide traceable outputs, clear error states and review checkpoints for high-impact or irreversible operations.
Useful for: production reliability and incident investigation.
Read primary documentationModels and deployment
Model selection
Choose a model using a task-specific benchmark, not brand reputation alone. Compare quality, speed, context limits, availability, data terms and total operating cost.
Useful for: balancing quality, latency and budget.
Read primary documentationFine-tuning
Use fine-tuning when examples show a repeatable behavior gap that prompting and retrieval do not solve. Keep a held-out evaluation set and avoid training on data without the right to use it.
Useful for: consistent formats, specialized styles and repeated task patterns.
Read primary documentationMultimodal AI
Design workflows that process combinations of text, images, audio or video. Test modality-specific limitations and verify extracted details against the original input.
Useful for: document understanding, accessibility and media analysis.
Read primary documentationProduction deployment and cost control
Use timeouts, retries with limits, rate limiting, caching where safe, secrets management and usage monitoring. Add a fallback path and test what happens when a provider is unavailable.
Useful for: reliable customer-facing AI features.
Read primary documentationAI search and website growth
AI search optimization (AEO / GEO)
Make useful pages easy to crawl, understand and cite: answer specific questions directly, show expertise and sources, use descriptive headings and maintain accurate facts. No tactic guarantees inclusion or citations in an AI answer.
Useful for: websites seeking qualified organic and AI-assisted discovery.
Read primary documentationTechnical SEO for AI-related content
Ensure important pages are indexable, use canonical URLs, meaningful internal links, descriptive titles and valid structured data where appropriate. Check actual Search Console evidence rather than assuming a page is indexed.
Useful for: discoverability, crawl health and search performance diagnosis.
Read primary documentationContent extraction and answer design
Put the main answer near the beginning, define specialist terms, use concise steps and tables where they help, and support claims with primary sources. Do not create thin pages solely to target variations of a keyword.
Useful for: people-first guides that are easy to scan and reference.
Read primary documentationFrequently asked questions
What does “AI skill” mean in this directory?
Here, an AI skill is a reusable capability or engineering practice for designing, connecting, evaluating, securing or deploying AI systems. It is not a claim that every model or agent supports every capability.
Is this a complete list of every AI skill or framework?
No single directory can reliably contain every skill, framework and provider-specific feature because the ecosystem changes quickly. This is a curated foundation covering major concepts, with links to primary documentation for deeper and current details.
Which skills should a beginner learn first?
Start with prompt engineering, structured outputs, tool calling, basic evaluation and security. Learn RAG when the system needs external knowledge, and introduce multi-agent coordination only when a simpler workflow is insufficient.
Can AI skills guarantee more website traffic?
No. Useful content can improve discoverability, but traffic depends on demand, competition, technical accessibility, content quality and distribution. Measure impressions, clicks, qualified visits and conversions instead of assuming that publishing a guide guarantees rankings.