AI Best Practices -- A Practical Guide from Anthropic Courses
Language: English | Turkish
This repo is a distilled summary of 10 courses from Anthropic's Skilljar training platform. The goal is not to pile up theory; it is to give someone who wants to use AI better than average, without spending hours watching videos a consistent, practical reference they can keep close.
What Is This Repo Based On?
The concepts were extracted from notes across these 10 courses:
| Course | What It Teaches |
|---|---|
| AI Capabilities and Limitations | How models work (next-token prediction), what they cannot do (hallucination, knowledge cutoff, working memory), and how to diagnose which mechanism caused a mistake. |
| AI Fluency Framework and Foundations | The 4D Framework: Delegation, Description, Discernment, Diligence. A universal frame for responsible, effective AI work. |
| Claude 101 | Claude's core capabilities: Projects, Artifacts, Connectors, Research mode, custom instructions, memory. |
| Claude Code 101 | The agentic loop in the terminal: explore -> plan -> code -> commit. CLAUDE.md, context management, plan mode. |
| Claude Code in Action | Claude Code in real projects: MCP server setup, UI testing with Playwright, GitHub Actions, hooks, SDK. |
| Introduction to Claude Cowork | Cowork as a delegation environment that handles files, scheduled tasks, and plugins. |
| Introduction to agent skills | Skills: reusable, automatically triggered instruction packages built around SKILL.md. |
| Introduction to subagents | Subagents: specialized helpers that run in a separate context without polluting the main chat. |
| Introduction to Model Context Protocol | MCP fundamentals: tools, resources, prompts, and client-server architecture. |
| Model Context Protocol Advanced Topics | Sampling, logging, transport (stdio / HTTP / streamable), roots, progress notifications. |
Language Structure
The main repository structure is English. Turkish is kept as an optional mirrored version under tr/:
- English entry point:
README.md - Turkish entry point:
tr/README.md - English docs stay in the root folders:
concepts/,checker/,starter-template/ - Turkish docs mirror the same structure under
tr/
How To Use This Repo
There are three layers:
- Skimmable: how-to-write-better.md -- all concepts in one file, with real-world examples, without going too deep. Read it over a tea break.
- Detailed: concepts/ -- focused documents for each concept, deep enough without drowning you.
- Applied:
- checker/ -- a drop-in folder that audits an existing project against AI best practices.
- starter-template/ -- a fill-in template for starting a new project from zero.
Navigator
Quick Start
- how-to-write-better.md -- A skimmable walkthrough of all concepts + examples
- prompt-examples.md -- Real-world prompt + follow-up examples you can copy and adapt
Concept Deep Dives
- concepts/prompt-engineering.md -- Prompt writing, few-shot examples, chain-of-thought, XML tagging
- concepts/4d-framework.md -- Delegation, Description, Discernment, Diligence
- concepts/ai-capabilities-and-limits.md -- What models actually do, hallucination, context windows
- concepts/context-management.md -- Managing the context window well: compact, clear, @file
- concepts/claude-md.md -- CLAUDE.md: how to write project memory
- concepts/skills.md -- Agent Skills: SKILL.md, frontmatter, progressive disclosure
- concepts/subagents.md -- When to use subagents, and when not to
- concepts/mcp.md -- Model Context Protocol: tools, resources, transport
- concepts/hooks.md -- Automation with pre/post-tool-use hooks
- concepts/slash-commands.md -- Custom slash commands
- concepts/modes-chat-cowork-code.md -- Which mode fits which job
- concepts/projects-and-memory.md -- Projects, Artifacts, memory, custom instructions
- concepts/connectors.md -- Google Drive, Slack, Gmail integrations
- concepts/file-organization.md -- Folder structure and organization for AI-ready projects
- concepts/workflow-patterns.md -- Sustainable workflows, iteration, scheduled tasks
Applied
- checker/ -- Audit an existing project
- starter-template/ -- Start a new project by filling in the template
This repo is designed as a living document. As the concepts change, the related files should be updated. Each file starts with a short "what this is for / when to read it" section.