Claude Code 2.1: Skills, Hot Reload, and the Future of Terminal AI
By Jordan Patel
What Changed in 2.1.0
Claude Code 2.1.0 is the biggest update since the tool launched. Skills are now first-class citizens with their own lifecycle, contexts, and hot reload support.
Skills: The Key Feature
Skills are reusable modules you invoke with slash commands. They run in forked contexts so they don't pollute your main session.
Creating a Skill
Add a markdown file to your .claude/skills/ directory with frontmatter defining the skill's behavior, hooks, and agent configuration.
Hot Reload
Edit a skill file and it reloads instantly. No restart needed. This makes iterating on custom workflows as fast as editing a config file.
Custom Agent Support
You can now define custom agents in skill frontmatter. Each agent gets its own context, tools, and permissions. This enables team-specific workflows where different agents handle different responsibilities.
/teleport to claude.ai
The /teleport command sends your current terminal session to claude.ai/code. You can continue working in the browser with full session history. This bridges terminal and web workflows seamlessly.
Other Improvements
- Shift+Enter for newlines with zero setup
- Agents continue working when you deny a tool use
- Wildcard tool permissions (e.g., Bash(-h))
- Multi-language response configuration
- 1,096 commits of stability improvements
Security Updates
- Writes to .claude/skills blocked in sandbox mode
- Improved heredoc parsing to prevent command smuggling
- Better bash permission matching
Getting Started
Run claude update to get version 2.1.0. Create your first skill in .claude/skills/ and invoke it with a slash command. The hot reload makes iteration nearly instant.