CLAUDE CODE

Hide skills from Claude's auto-selection until manually triggered.

Pre-approve tools before a skill executes so it runs without prompts.

Run a skill in an isolated context via fork mode.

Changes to skill files are detected and reloaded automatically.

Spawn isolated workers with independent context windows.

Researcher, auditor, reviewer, and other ready-made subagent types.

Create reusable subagent definitions at project or user level.

Configure model, tools, MCP, skills, memory, and scoping.

Limit which tools a subagent can access.

Route specific MCP servers only to specific subagents.

Auto-memory that persists across multiple subagent invocations.

Prevent bloating the main conversation with research or exploration.

Continue a subagent's work across sessions.

Event-driven automation with 20+ lifecycle events.

Fires when a session begins; load env vars and initialize state.

Fires when a session terminates.

Fires before Claude processes user input; can validate or block.

Fires when a slash command expands; can block or inject context.

Fires before any tool executes. Allow, deny, defer, or modify the call.

Fires after a successful tool call. Good for feedback and follow-ups.