Approved Version v1 Reviewed Jun 20, 2026 Low risk Tool Claude
Claude overview
What Claude is, the surfaces it ships on, and where this guide goes next.
What Claude is
Claude is the family of AI assistants made by Anthropic. As a developer on a delivery team you will mostly meet Claude on three surfaces:
- Claude on the web / API — the chat assistant and the model API, used for prompts, summarization, and structured generation.
- Claude Code — the agentic command-line tool that reads and edits your repository. This is the focus of most of this guide (see Claude Code).
- IDE integrations — Claude Code also runs inside editors (for example VS Code and JetBrains) that wrap the same agent.
When to reach for Claude
- You have a well-scoped coding task and want the agent to make the edits.
- You are reading an unfamiliar codebase and want a guided tour of a file or module.
- You want to draft or refactor something repetitive (tests, migrations, configs) where review is cheap and mistakes are caught quickly.
What this guide covers
The remaining pages walk through the Claude Code surface in order:
- Claude Code — the agent, the loop, the workflow.
- Commands — slash commands, built-in and custom.
- CLAUDE.md — project instructions / memory.
- Skills — reusable, auto-invoked procedures.
- MCP — Model Context Protocol integrations.
- Limitations — what to watch for.
- Security — the do/don’t list for safe use.
Pitfalls
- Treating the chat assistant as the agent. The web chat cannot edit your files; only Claude Code (or an IDE running it) can. Confusing the two leads to copy-paste workflows that defeat the purpose of the agent.
- Skipping CLAUDE.md. Without project instructions the agent guesses at conventions and produces code that fights your existing style. See CLAUDE.md.
- Asserting version numbers from memory. Capabilities move fast. Where a feature is version-dependent, check the official docs rather than trusting a remembered flag.