The Task tool has been renamed to Agent across permissions, settings, CLI, and hook documentation (with backward compatibility for pre-2.1.63 references)
Memory documentation completely restructured with a new title, improved comparison tables, guidance for large teams, and better organization of CLAUDE.md, rules, and auto memory
New HTTP hook type added, allowing hooks to POST event data to external URLs (web servers, cloud functions, shared audit services)
Three bundled skills now ship with Claude Code: /simplify, /batch, and /debug
Remote Control is now available on Pro plans (previously Max only)
Added a new "CLAUDE.md vs Rules vs Skills" comparison to the feature comparison list, with a full table covering loading behavior, context cost, and best use cases for each. [lines 74-80] [Source]
Added guidance on bundled skills (simplify, batch, debug) that ship with Claude Code. [line 163] [Source]
Added documentation for the new type: "http" hook, which POSTs event data to an HTTP endpoint instead of running a shell command. Useful for shared audit services and external event processing. [line 376] [Source]
New dedicated HTTP hooks section added covering POST behavior, authorization headers with environment variable interpolation, and when to prefer HTTP hooks over command hooks. [lines 680-724] [Source]
Added auto memory (MEMORY.md) to the list of things Claude can access at the start of each session, noting the first 200 lines are loaded automatically. [line ~51]
Session history panel now supports rename and remove actions: hover over a session to reveal options to rename it with a descriptive title or remove it from the list. [line 93] [Source]
Added ENABLE_CLAUDEAI_MCP_SERVERS environment variable documentation: set to false to disable claude.ai MCP servers in Claude Code (enabled by default for logged-in users). [line ~1144]
Page completely restructured with new title "How Claude remembers your project" and a clearer intro roadmap. [line 1] [Source]
New comparison table contrasting CLAUDE.md files vs auto memory across multiple dimensions. [lines 19-27] [Source]
New subsections added for: choosing CLAUDE.md file scope (project/user/repo), setting up a project CLAUDE.md with /init, and writing effective instructions (size limits, structure, specificity). [lines 44-84] [Source]
New "Manage CLAUDE.md for large teams" section covering organization-wide deployment via configuration management, excluding specific files, and team-level controls. [line 237] [Source]
Reorganized .claude/rules/ section with dedicated subsections for path-specific rules, glob patterns, symlinks, and user-level rules. [line 130] [Source]
Subagent permission rules renamed from Task(AgentName) to Agent(AgentName) syntax throughout, including examples for Explore, Plan, and custom agents. [lines 161-178] [Source]
Remote Control is now available as a research preview on both Max and Pro plans (previously described as Max only with Pro "coming soon"). [line 3] [Source]
New "Bundled skills" section added documenting three default skills that ship with every Claude Code installation: /simplify (parallel code review and cleanup), /batch (large-scale parallel codebase changes via git worktrees), and /debug (session troubleshooting). [lines 9-18] [Source]
Task tool officially renamed to Agent tool as of version 2.1.63, with a backward compatibility note that existing Task(...) references in settings and agent definitions still work as aliases. [lines 277-279] [Source]
All code examples updated to use Agent(...) syntax instead of Task(...). [lines 288-309] [Source]
Session history panel now supports rename and remove actions: hover over a session to reveal options to rename it with a descriptive title or remove it from the list. [line 93] [Source]