Verified September 30, 2026. Every link below was opened and confirmed live on that date.
There is no shortage of Claude guides. There is a shortage of guides that tell you which source to open and in what order.
That is the whole problem with a bookmark folder. Fourteen tabs, no sequence, no opinion. So you open none of them.
This is the same fourteen destinations, with a judgment attached to each. Every one is official and free to read, and nothing below is an affiliate link. The two course sites ask you to register or sign in, and their entries say so.
I spent eighteen months finding out which of these matter and which ones you can leave closed. That judgment is the asset. The links are public.
Pick your route first
Three ways through this list. Pick one, ignore the other two.
Your first week with Claude. Go 2, then 1, then 10. Take a course, learn where the manual lives, then learn what the product does when it breaks. Skip everything else for now.
You are building something. Go 8, then 4, then 6, then 5. Read the thinking, then the working code, then the agent itself, then the skills that run on it.
You are rolling Claude out to a team. Go 2, then 10, then 12, then 3. Certificates give your people a finish line. Support answers what they will ask you. News keeps you ahead of them. The Academy shows them what to take next.
The 14
1. Claude docs, the manual
docs.claude.com
The source of truth for models, pricing, context windows, and every API parameter. It resolves to platform.claude.com, which is where the developer documentation now lives.
My note: this is the one to check when a creator makes a claim about a model. Guides go stale in weeks; this page is maintained, so it is the closer thing to current. If a post tells you which Claude model is best and the page disagrees, the page wins.
Open it when: you need a number, an exact model name, or a capability confirmed.
2. Anthropic Courses, free courses with certificates of completion
anthropic.skilljar.com
Anthropic's course site, now branded Anthropic Courses. More than twenty free, self-paced courses, from Claude 101 to MCP and agent skills. Courses end with a certificate of completion. You register a free account to start.
My note: the highest-return item on this list for anyone starting out, and the best one for something to show. Claude Academy (3) lists Claude 101 at 13 lessons and 2.5 hours, and AI Fluency: Framework and foundations at 14 lessons and 4 hours. The courses are free. Anthropic also sells certification exams, from $99 to $175, through its Partner Academy, built for people at firms in the Claude Partner Network. Before you pay for any other "Claude certification," check that Anthropic is the one issuing it.
Open it when: you want structure instead of scrolling, or you want a credential on your profile by Friday.
3. Claude Academy, the bigger catalog
academy.claude.com
Anthropic's newer learning site, and where anthropic.com/learn now sends you. Free courses, tutorials, and use cases, sorted by product: Claude.ai, Claude Cowork, Claude Code, Claude Tag, and the Claude Platform. Every course shows its lesson count and hours up front. You sign in to save your progress.
My note: this used to be a routing page. It is now the better place to plan. Many courses, Claude 101 among them, sit on both 2 and 3, so browse here, where you can see what each one costs you in hours. One difference to know: 2 issues a certificate of completion, and 3 shows a completion badge. If the certificate is what you want on your profile, take the course on 2.
Open it when: you want to see what exists and how long it takes before you commit an evening to it.
4. The cookbook, working code
github.com/anthropics/claude-cookbooks
Runnable notebooks for the patterns people usually describe in prose. The old anthropic-cookbook address still redirects here.
My note: one of three I reopen most weeks. The difference between reading about a technique and having a working version of it in front of you is most of the learning curve. Copy the notebook, change the inputs, watch what breaks.
Open it when: a concept makes sense in theory and you cannot make it work in practice.
5. The official skills library
github.com/anthropics/skills
Anthropic's own skills, published as source. Read them, install them, modify them. You can use them in Claude Code, in claude.ai on paid plans, or through the API.
My note: read these before you write your own. The value is not the skills themselves. It is seeing how the team that built the model writes instructions for it. That is a free lesson in what Claude responds to.
Open it when: you are about to write a skill, a system prompt, or a set of standing instructions.
6. Claude Code, the repo
github.com/anthropics/claude-code
Install instructions, plugins, worked examples, and the issue tracker for Claude Code.
My note: the issues are the underrated half. When something behaves strangely, someone has already filed it and a maintainer has already explained why. That is faster than any tutorial.
Open it when: Claude Code does something you cannot explain.
7. Claude Code docs, setup through advanced
code.claude.com/docs
Installation, configuration, hooks, subagents, permissions, MCP.
My note: dense, and worth being dense. Read the configuration and permissions sections early. Most of the frustration people report with Claude Code comes from defaults they never changed.
Open it when: you have installed it and want it to fit how you work.
8. Building effective agents
anthropic.com/engineering/building-effective-agents
The essay that set out six agent patterns now widely built on: prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer, autonomous agent.
My note: the single highest-value document on this list if you are building anything. Its most useful line is a warning, not a pattern: add complexity only when a simpler version has measurably failed.
Open it when: before you design a workflow, not after it disappoints you.
9. Claude Code best practices
anthropic.com/engineering/claude-code-best-practices
Working practice from the team that builds and uses the tool daily. Now served from the Claude Code docs; the engineering URL still redirects.
My note: another of the three I reopen. It is the closest thing available to watching someone competent use the tool over their shoulder.
Open it when: you use Claude Code and suspect you are using half of it.
10. Support, the product explained plainly
support.claude.com
Plans, limits, billing, Projects, memory, and what each surface does.
My note: the least glamorous entry and the one that ends the most arguments. Usage limits, plan differences, and feature availability change often. This page is maintained. A creator's screenshot from April is not.
Open it when: you hit a limit, a paywall, or a feature that will not appear for you.
11. Anthropic on YouTube
youtube.com/@anthropic-ai
Launch videos, research explainers, and product walkthroughs from the team.
My note: watch at 1.5x and skip the launch videos. The research explainers and walkthroughs carry the material you cannot get from documentation, which is how the people who built it think about tradeoffs.
Open it when: you learn better from someone working than from a page.
12. The newsroom
anthropic.com/news
Every model release, product launch, research post, and policy announcement, dated.
My note: this is how you stop being three months behind without following anyone. Ten minutes a month here beats a feed. Check the dates on every guide you read against this page, including this one.
Open it when: monthly. Set a recurring reminder and treat it as maintenance.
13. The plugins shelf
claude.com/marketplace/plugins
One-click installs that bundle tools, skills, and integrations for Claude Code and claude.ai. It now sits inside the Claude Marketplace, and the old claude.com/plugins address still redirects here.
My note: browse it once, install two things, ignore the rest. Plugin sprawl is a real cost. In Claude Code, every plugin you add puts its tool names and skill descriptions in front of Claude each session, used or not. The full content loads only when needed, so the cost is small per plugin and real in bulk.
Open it when: you find yourself doing the same manual step for the third time.
14. The interactive prompt tutorial
github.com/anthropics/prompt-eng-interactive-tutorial
A hands-on course in prompting, run as exercises rather than advice. Nine chapters, each with exercises. It needs an API key.
My note: better than any prompting post you will read this year, including the ones with a hundred prompts in them. You write the prompt, you see it fail, you fix it. That sequence is why it holds. One catch: it was written in 2024 for Claude 3 Haiku, which Anthropic retired in April 2026. In the first notebook, change !pip install anthropic to !pip install "anthropic<1", and change the model name to claude-haiku-4-5. The later notebooks reuse both, except the tool-use appendix, which names its own retired model; change it to claude-haiku-4-5 too. The notebooks set a temperature, which the current Python library and Anthropic's newest models no longer accept. Haiku 4.5 still does, and no retirement has been announced for it.
Open it when: your prompts work sometimes and you cannot say why.
One warning about every Claude guide, this one included
Claude changes faster than guides get updated.
Between March and September 2026, every Claude model above Haiku was replaced. Opus and Sonnet both changed again in the two weeks before this page went out. Guides published in April name models that are now legacy. Much of their advice was correct when written and is wrong now, and nothing on the page tells you that.
So two habits are worth more than any list:
Check the publication date on anything you read about Claude. If it is older than about eight weeks, verify the specifics against 1 or 12 before acting on it.
Trust the primary source over the interpretation. Every link above is Anthropic's own. That is deliberate.
This page carries its own verification date at the top. That date tells you how old the checking is. It cannot tell you what changed after it, which is why 1 and 12 stay on this list.
Next verification: November 28, 2026, eight weeks out, the same threshold this page asks you to apply to everyone else.
Eric Melillo writes about AI as operator leverage at ericmelillo.com/newsletter. Twenty years scaling other people's companies, two agencies built and exited, now rebuilding the same systems for operators working alone.