Claude's skill system turns a capable AI into a specialized operator. Instead of writing a new prompt every time you want Claude to do something specific, you install a skill once and trigger it with a slash command. The result is faster work, more consistent output, and less mental overhead. These twenty skills are the ones that actually earn their keep — ranked roughly by how often they show up in a real operator's workflow.
1. deep-research
This skill fans out across multiple web searches, fetches source documents, cross-checks claims against each other, and synthesizes a cited report. It is the closest thing to having a research assistant who does not make things up. Use it when you need to understand a market, a competitor, a regulation, or a topic you have never touched before — and you need the output to hold up under scrutiny. Trigger it with a specific, narrow question. If your question is vague, Claude will ask you to narrow it before starting.
2. code-review
Runs a structured review of your current code diff and surfaces correctness bugs, simplification opportunities, and efficiency issues. You control the effort level — low for a quick pass, high or ultra for a thorough audit. Add the --fix flag and Claude applies the findings directly to your working tree. Use it before any merge that touches production logic.
/code-review --fix3. fullstack-dev-skills:nextjs-developer
Loads a Next.js-specialist context that shapes how Claude reasons about routing, server components, API routes, and deployment. If you are building on Next.js and want Claude to stop suggesting patterns that do not fit the framework, this is the skill that fixes that. Pair it with the react-expert skill when you are deep in component architecture.
4. fullstack-dev-skills:python-pro
Activates Python-specialist reasoning — idiomatic patterns, typing, dependency hygiene, and performance trade-offs. Use this when you are writing scripts, building automation, or working on anything data-adjacent. Claude defaults to generic patterns without it; this skill makes the output look like it was written by someone who actually writes Python for a living.
5. fullstack-dev-skills:typescript-pro
Loads strict TypeScript conventions — proper generic usage, discriminated unions, satisfies operator, inference patterns, and common anti-patterns to avoid. If your codebase uses TypeScript and you are tired of Claude producing code that technically compiles but violates every convention your team has, install this.
6. fullstack-dev-skills:postgres-pro
Puts Claude in the mindset of someone who takes Postgres seriously — indexing strategy, query planning, schema design, and avoiding the patterns that destroy performance at scale. Use it when you are writing migrations, designing tables, or debugging slow queries. The one thing to use it for: writing a new schema and asking Claude to identify every indexing decision it made and why.
7. fullstack-dev-skills:fastapi-expert
Shapes Claude's output toward FastAPI best practices — dependency injection, Pydantic v2 models, async patterns, and OpenAPI documentation. Use it when building internal APIs or anything that will eventually expose endpoints to other systems. The skill keeps Claude from reverting to Flask-style thinking when the framework expects something different.
8. fullstack-dev-skills:react-expert
Loads current React conventions — hooks, composition patterns, state management trade-offs, and the difference between what works in demos and what holds up in real apps. Use it when you are building UI components and want Claude to reason about re-renders, prop drilling, and context correctly rather than reaching for the most obvious answer.
9. fullstack-dev-skills:devops-engineer
Activates infrastructure and deployment thinking — CI/CD pipeline design, environment separation, container patterns, and deployment risk management. Use it when you are setting up a new project's deployment pipeline or troubleshooting why something works locally and breaks in production. The one prompt that returns immediate value:
Review my current deployment setup and identify the top three failure points that would cause an outage with no obvious rollback path.10. fullstack-dev-skills:terraform-engineer
Loads Terraform-specific reasoning — module structure, state management, provider patterns, and the common mistakes that turn infrastructure-as-code into infrastructure-as-mystery. Use it when you are writing or auditing Terraform configurations. Claude without this skill writes Terraform that works but that no one will understand six months later.
11. fullstack-dev-skills:secure-code-guardian
Shifts Claude into security-aware mode — OWASP top ten, injection risks, authentication patterns, secrets management, and dependency vulnerabilities. Use it before shipping anything that handles user data, authentication, or external API calls. Do not wait until someone finds the hole.
12. fullstack-dev-skills:database-optimizer
Focuses Claude on query performance and schema efficiency. Paste in a slow query and Claude will explain the execution plan, identify the problem, and propose a fix. Use it when EXPLAIN ANALYZE output looks alarming or when a feature ships fine in development but crawls in production under real load.
Here is my slow query and its EXPLAIN ANALYZE output. Identify the bottleneck and propose the minimum change that fixes it without altering the result set:
[paste query]
[paste EXPLAIN ANALYZE output]13. fullstack-dev-skills:api-designer
Loads API design conventions — RESTful resource modeling, versioning strategy, error response formats, pagination patterns, and the decisions that make an API a pleasure to consume versus a source of endless client-side workarounds. Use it at the design stage, before you have written a single route handler.
14. fullstack-dev-skills:microservices-architect
Activates reasoning about service decomposition, inter-service communication, data ownership, and the failure modes that make microservices painful in practice. Use it when you are deciding whether to split a monolith, designing a new distributed system, or evaluating whether a proposed architecture will actually survive the team that maintains it.
15. fullstack-dev-skills:test-master
Loads testing strategy and execution — unit, integration, end-to-end, test doubles, coverage trade-offs, and how to write tests that catch real bugs instead of just confirming that the code does what you wrote it to do. Use it when you are adding tests to a codebase that has none, or when your existing tests are not catching regressions.
16. fullstack-dev-skills:debugging-wizard
Puts Claude into structured debugging mode — diagnosis before fix, evidence before interpretation. Instead of jumping to a plausible-sounding solution, this skill makes Claude produce a literal evidence report first: file and line, quoted output, no speculation. The fix comes after the diagnosis is confirmed. Use it when you are stuck on a bug that does not make obvious sense.
17. fullstack-dev-skills:graphql-architect
Loads GraphQL schema design, resolver patterns, N+1 prevention with DataLoader, and the real trade-offs between GraphQL and REST for your specific use case. Use it when you are designing a new GraphQL API or troubleshooting why your existing one is slow. The skill keeps Claude grounded in practical patterns rather than theoretical purity.
18. fullstack-dev-skills:prompt-engineer
Activates structured prompt design thinking — system prompt architecture, few-shot examples, output formatting, chain-of-thought structuring, and how to test whether a prompt actually does what you think it does. Use it when you are building an AI feature or automation and need the prompts to be reliable, not just good enough in testing.
I am building a prompt for [describe your use case]. The current prompt is:
[paste prompt]
Identify the three most likely failure modes and suggest specific changes to address each one.19. fullstack-dev-skills:rag-architect
Loads retrieval-augmented generation design — chunking strategy, embedding model selection, vector store trade-offs, retrieval quality evaluation, and the common mistakes that produce a RAG system that retrieves confidently but incorrectly. Use it when you are building anything that involves searching a knowledge base and feeding results into a language model.
20. update-config
This one is different from the rest — it is not about a technical domain, it is about making the whole system work better over time. Use it to configure automated behaviors in Claude Code: hooks that run before or after certain actions, permissions that remove friction, environment variables that belong in settings instead of your head. Any time you find yourself repeating the same instruction at the start of a session, that instruction belongs in your config, and this skill gets it there correctly.
Skills compound. A prompt-engineer skill session spent refining your system prompts makes every downstream automation sharper. A database-optimizer session that rewrites your indexing strategy reduces load that otherwise would have forced an infrastructure upgrade. Install the skills that match your current bottleneck, not the ones that sound impressive.
How to get the most out of these
Trigger a skill at the start of a session, before you start giving Claude work to do. The skill shapes how Claude reasons throughout the session, not just for the first response. If you switch contexts mid-session — for example, from API design to database work — trigger the relevant skill again. Claude does not automatically carry specialist context from one domain to another.
The other pattern that pays off: use the research skill first, then a domain skill. Research what you are building, confirm your understanding, then switch into build mode with the appropriate specialist skill active. You get the breadth of research and the depth of a specialist without mixing them together in a way that produces mediocre results.
- Trigger skills at session start, before giving Claude any work.
- Re-trigger when you switch domains mid-session.
- Pair deep-research with a specialist skill: research first, build second.
- Use update-config to lock in any behavior you find yourself requesting repeatedly.
- When in doubt about which skill to use, describe your task to Claude and ask which skill fits — it knows its own catalog.