Learn how Claude Code works as a terminal-based agentic assistant. Covers the agentic loop, built-in tools, session management, context windows, and safety features for developers.
You type a single command. Then Claude reads your files, runs your tests, finds the bug, fixes it, and verifies everything works. No copy-pasting between tools. No manual navigation. Just a task and a result.
That kind of end-to-end automation is not magic. It is a specific design: an agentic loop backed by real tools that can act on your system. Understanding how this works will help you use Claude Code more effectively and trust it with bigger tasks.
This post breaks it all down in plain language.
When you give Claude a task, it works through three phases: gather context, take action, and verify results. These phases repeat until the task is done.
A simple question might only need the first phase. A bug fix will cycle through all three multiple times. A large refactor might involve dozens of loops.
Here is the key part: Claude decides what each step requires based on what it learned from the previous step. It is not following a fixed script. It is reasoning as it goes.
You are part of the loop too. You can interrupt at any point, give more context, or ask Claude to try a different approach. Claude works on its own but stays responsive to you.
Claude Code is not just a chat interface. It has real tools that let it act on your project. Without tools, a language model can only respond with text. With tools, it can read files, edit code, run commands, and browse the web.
| Category | What Claude Can Do |
|---|---|
| File operations | Read files, edit code, create new files, rename and reorganize |
| Search | Find files by pattern, search content with regex, explore codebases |
| Execution | Run shell commands, start servers, run tests, use git |
| Web | Search the web, fetch documentation, look up error messages |
| Code intelligence | See type errors and warnings, jump to definitions, find references |
Here is a realistic example. You say "fix the failing tests." Claude might:
Each step returns new information that feeds into the next decision. That is the agentic loop working in practice.
When you run claude inside a directory, Claude Code gets access to:
Because Claude sees your whole project, it can work across it. This is different from inline code assistants that only see the current file open in your editor.
Claude Code runs in three environments depending on your setup:
| Environment | Where Code Runs | Use Case |
|---|---|---|
| Local | Your machine | Default. Full access to your files and tools |
| Cloud | Anthropic-managed VMs | Offload tasks, work on repos you do not have locally |
| Remote Control | Your machine, via browser | Use the web UI while keeping everything local |
You can access Claude Code through the terminal, the desktop app, VS Code, JetBrains, claude.ai, Slack, and CI/CD pipelines like GitHub Actions.
The interface changes how you see and interact with Claude, but the underlying agentic loop is the same everywhere.
Claude Code saves your conversation locally as you work. Each message, tool use, and result is stored in a JSONL file under ~/.claude/projects/.
The important thing to know: each new session starts fresh. There is no automatic carry-over from your last conversation.
Claude handles persistence through two features:
(Learn more about these features in our dedicated guide on how Claude Code remembers things between sessions.)
You can pick up where you left off:
# Resume the last session
claude --continue
# Choose which session to resume
claude --resumeYou can also fork a session to explore a different direction without losing your original:
# Fork the current session into a new one
claude --fork-sessionResuming appends new messages to the same session. Forking copies the history into a new session, leaving the original untouched.
Claude sees whichever branch you are currently on. When you switch branches, Claude sees the new branch's files. Your conversation history stays the same across branch switches.
For parallel work, use git worktrees. Each worktree is a separate directory, which means you can run separate Claude sessions in each one.
Claude's context window holds your conversation history, file contents, command outputs, CLAUDE.md, auto memory, and system instructions. As you work, this fills up.
Claude compacts automatically when context gets full. It clears older tool outputs first, then summarizes the conversation if needed. Your requests and key code snippets are preserved. Detailed instructions from early in the session may be lost.
Best practice: Put persistent rules in CLAUDE.md instead of relying on conversation history.
To control what gets preserved during compaction, add a "Compact Instructions" section to your CLAUDE.md:
## Compact Instructions
Always preserve:
- The API design decisions we agreed on
- The test-first approach for all new features
- The folder structure conventions in src/You can also run /compact with a focus:
/compact focus on the API changesRun /context to see what is currently using space in your context window.
Claude has two safety layers built in.
Before Claude edits any file, it snapshots the current contents. If something goes wrong, press Escape twice to rewind. Or just ask Claude to undo.
Checkpoints are separate from git and only cover file changes. Actions that affect remote systems (databases, APIs, deployments) cannot be checkpointed, which is why Claude asks before running commands with external side effects.
Press Shift+Tab to cycle through permission modes:
| Mode | What It Does |
|---|---|
| Default | Claude asks before file edits and shell commands |
| Auto-accept edits | Claude edits files and runs common filesystem commands without asking |
| Plan mode | Claude uses read-only tools only, creates a plan for you to approve |
| Auto mode | Claude evaluates all actions with background safety checks (research preview) |
You can also allow specific trusted commands in .claude/settings.json so Claude does not ask each time:
{
"allowedCommands": ["npm test", "git status", "git diff"]
}Start with a specific prompt. Vague prompts work, but specific ones often succeed on the first try with fewer corrections.
# Vague
Fix the login bug
# Specific
The checkout flow is broken for users with expired cards.
Check src/payments/ for the issue, especially token refresh.
Write a failing test first, then fix it.Give Claude something to verify against. Claude performs better when it can check its own work.
Implement validateEmail.
Test cases: 'user@example.com' -> true, 'invalid' -> false, 'user@.com' -> false.
Run the tests after.Use plan mode for complex tasks. Press Shift+Tab twice to enter plan mode. Ask Claude to analyze first, review the plan, then let it implement.
Read src/auth/ and understand how we handle sessions.
Then create a plan for adding OAuth support.Interrupt freely. If Claude is going down the wrong path, type your correction and press Enter. You do not have to wait for it to finish.
Delegate, do not dictate. Give context and direction, then trust Claude to figure out the details. You do not need to specify which files to read or which commands to run.
1. What is the agentic loop in Claude Code?
It is a three-phase cycle: gather context, take action, and verify results. Claude repeats these phases, chaining tool uses together, until your task is complete.
2. What tools does Claude Code have built in?
File operations, search, shell execution, web browsing, and code intelligence (with the right plugins). You can extend these with MCP servers, skills, and subagents.
3. Does Claude remember my previous sessions?
No. Each session starts fresh. Use CLAUDE.md for persistent instructions, and Claude's auto memory feature for learnings that carry over automatically.
4. How do I resume a session I was working in earlier?
Run claude --continue to pick up the last session, or claude --resume to choose from a list of past sessions.
5. What happens when the context window fills up?
Claude compacts automatically. It clears older tool outputs first, then summarizes the conversation. Put persistent rules in CLAUDE.md so they do not get lost during compaction.
6. Can I undo changes Claude makes to my files?
Yes. Claude snapshots every file before editing it. Press Escape twice to rewind, or ask Claude to undo. Note: this only covers file changes, not actions on remote systems.
7. How do I stop Claude from asking for permission every time?
Add trusted commands to .claude/settings.json under allowedCommands. You can also switch to auto-accept edits mode with Shift+Tab.
8. What is CLAUDE.md and why does it matter?
It is a markdown file in your project where you store instructions, conventions, and context for Claude. It loads at the start of every session, so anything you put there persists across conversations.
9. Can I run Claude Code in a CI/CD pipeline?
Yes. Claude Code supports GitHub Actions and similar pipelines. The same agentic loop runs in CI environments.
10. What is the difference between resuming and forking a session?
Resuming appends new messages to the same session under the same ID. Forking copies the conversation history into a new session, leaving the original unchanged so you can explore a different direction safely.
A practical guide to Claude Code best practices covering context management, prompt writing, CLAUDE.md setup, parallel sessions, and how to avoid the most common mistakes developers make.
Learn how to use CLAUDE.md files and Claude Code's auto memory feature to persist project instructions, coding standards and personal preferences across sessions.
Learn what Claude Cowork is, how it works, and how to get started using it on Claude Desktop to automate complex, multi-step tasks on your Mac or Windows PC.
Explore how Anthropic is accelerating the future of AI scaling in April 2026. Read our summary covering the release of Claude Opus 4.7, new global offices in Sydney and Japan, and enhanced election safeguards.
A clear, beginner-friendly breakdown of Claude Opus 4.7's new features, breaking changes, and behavior updates, including high-res image support, task budgets, adaptive thinking, and migration tips.
