Claude Code — Agent Flow Playground

Simple question (no tools) Text only — 1 turn Read & edit a file 2 tools, permission check — 2 turns Search, read, edit, test 4 tools, parallel reads — 3 turns Dangerous command denied Permission denied path Sub-agent spawned Agent delegation — background Context overflow recovery Compaction + retry Slash command: /commit Command → skill expansion → query MCP tool call (GitHub) External server integration Web search & fetch WebSearchTool + WebFetchTool Plan mode workflow Plan → approve → execute Worktree isolation Agent in isolated git worktree Model fallback (529) Overloaded → fallback model Tool error & recovery Bash fails → Claude adapts Memory creation User asks Claude to remember Hook lifecycle Pre/post tool hooks firing Multi-agent swarm Coordinator + workers