Agentic Coding Features

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster through autonomous task execution and intelligent workflows.

Agents

Sub-Agents

Spawn specialized AI agents for parallel task execution. Each sub-agent has its own context window, tools, and instructions for handling specific tasks like research, testing, or code review.

/agents command Learn More
Extensions

Plugins

Lightweight packages bundling slash commands, subagents, MCP servers, and hooks. Install with a single command, share via plugin marketplaces, and toggle on/off as needed.

/plugin command Learn More
Extensions

Hooks

Automated shell commands that execute at specific lifecycle points. PreToolUse hooks run before actions, PostToolUse hooks run after. Perfect for quality gates and enforcing standards.

/hooks command Learn More
Core

Extended Thinking

Trigger deeper reasoning with escalating phrases: "think" (4K tokens), "think hard" (10K), "think harder" (32K), "ultrathink" (128K). Ideal for complex architectural decisions.

Planning Mode Learn More
Workflows

Slash Commands

Custom prompt templates stored as Markdown files in .claude/commands/. Support $ARGUMENTS for dynamic inputs. Share with teams via git or globally via ~/.claude/commands/.

.claude/commands/ Learn More
Extensions

MCP Servers

Connect to 1000+ community servers via Model Context Protocol. Access databases, APIs, Google Drive, Figma, Slack, and more. Configure via CLI or .mcp.json files.

claude mcp add Learn More
Agents

Agent Skills

Reusable capability modules that extend Claude's abilities. Create, manage, and share Skills for task-specific workflows. Activates context-aware features automatically.

.claude/skills/ Learn More
Workflows

Headless Mode

Run Claude Code programmatically without interactive UI. Use -p flag for CI/CD pipelines, GitHub Actions, and pre-commit hooks. Output structured JSON for automation.

claude -p "prompt" Learn More
Agents

Claude Agent SDK

Build custom autonomous agents on top of Claude. The SDK that powers Claude Code, now available for finance agents, research assistants, customer support, and more.

Anthropic Official Learn More
Integrations

VS Code Extension

Native VS Code integration with Claude Code. Access agentic coding features directly from your editor with seamless context sharing and inline suggestions.

Official Extension Learn More
Integrations

Slack Integration

Hand off full coding tasks directly from Slack conversations. @mention Claude to convert bug reports or feature discussions into actionable coding sessions with PR generation.

Beta Feature Learn More
Integrations

GitHub Actions

Automate code reviews, issue triage, and migrations in CI/CD pipelines. Use headless mode with structured JSON output for automated workflows and monitoring.

CI/CD Integration Learn More
Core

CLAUDE.md Memory

Persistent memory file automatically loaded into context. Document commands, style guides, testing instructions, and project conventions. Supports hierarchical loading from parent/child directories.

Project Root Browse Templates
Core

Permission Management

Fine-grained control over Claude's tool access. Configure via /permissions command, .claude/settings.json, or --allowedTools CLI flag. Support for "always allow" patterns.

/permissions Learn More
Enterprise

Amazon Bedrock

Deploy Claude Code on AWS infrastructure with enterprise-grade security, compliance, and scalability. Integrates with existing AWS services and IAM policies.

Enterprise Learn More
Enterprise

Google Vertex AI

Run Claude Code on Google Cloud with Vertex AI integration. Enterprise security, data residency controls, and seamless GCP ecosystem integration.

Enterprise Learn More
Enterprise

Sandboxing

Filesystem and network isolation for secure code execution. Run Claude Code in isolated environments with controlled access to system resources and external services.

Security Learn More

The Agentic Feedback Loop

Claude Code operates in a continuous feedback loop that enables autonomous task completion:

1

Gather Context

Read files, search code, understand project structure via agentic and semantic search

2

Take Action

Edit files, run commands, execute tools, generate code using bash and MCP integrations

3

Verify Work

Run tests, lint code, check outputs, visual feedback via screenshots and renders

4

Repeat

Iterate until the goal is achieved, self-correcting along the way with compaction

Essential Commands Reference

Core Commands

/ - Open slash commands menu
/agents - Manage sub-agents
/hooks - Configure lifecycle hooks
/plugin - Install/manage plugins

Extended Thinking

"think" - 4K token budget
"think hard" - 10K token budget
"think harder" - 32K token budget
"ultrathink" - 128K token budget