upstream: official claude code updates 2026-05-30

docs/tips/fast-mode.md

@@ -14,6 +14,13 @@ this is the most common misconception i see. people assume fast mode = dumber mo
 
 fast mode (lower thinking budget) now coexists with effort levels (`--effort low|medium|high|xhigh|max`). effort controls throughput, model selection, and reasoning depth across the session. fast mode is a narrow toggle on output speed. they compose: `--effort low --fast` minimizes both reasoning and output latency. `--effort max --fast` may behave unexpectedly -- max effort expects time to think, fast mode tries to skip it. avoid that combination.
 
+
+
+### auto mode availability (v2.1.158+)
+
+auto mode (if enabled via `CLAUDE_CODE_ENABLE_AUTO_MODE=1`) is now available on Bedrock, Vertex, and Foundry for Opus 4.7 and 4.8, in addition to the main Claude Code service. auto mode changes how models are selected and tool calls are approved, so verify your settings on enterprise platforms.
+
+
 ## why i don't use it
 
 never use fast mode. i mean it. the only scenario where fast mode makes sense is if you're at a hackathon with 30 minutes left before demo, or you're someone who literally doesn't care about burning through usage. fast mode can easily run up over a hundred dollars of usage in half an hour.

docs/tips/hooks-v2.md

@@ -87,6 +87,13 @@ use this to warn before stopping a session with active background work, or to lo
 
 MCP tool handlers can now be invoked from hooks using the `mcp_tool` type with event-driven logic. this lets you intercept and react to MCP calls without spinning up a shell or http process.
 
+
+
+### plugin-defined hooks (v2.1.157+)
+
+plugins loaded from `.claude/skills/` can now include hooks in their manifests just like manually registered hooks. hooks from plugins fire in the same lifecycle events (PreToolUse, PostToolUse, etc.) and follow the same timeout/cost rules as command/http/prompt hooks.
+
+
 ### updatedToolOutput (PostToolUse, v2.1.121+)
 
 PostToolUse hooks can now replace tool output before claude sees it. return `{"hookSpecificOutput": {"PostToolUse": {"updatedToolOutput": "your replacement text"}}}` to modify what claude receives from the tool.

docs/tips/subagents.md

@@ -40,6 +40,13 @@ use worktree isolation for:
 
 skip it for read-only research. worktree setup adds overhead you don't need when the agent is just reading files.
 
+
+
+### switching worktrees mid-session (v2.1.157+)
+
+use `EnterWorktree` within a session to switch between Claude-managed worktrees without ending the session. this lets one agent spawn others in different worktrees and coordinate across them.
+
+
 ## the scout pattern
 
 send a cheap model to explore, then a capable model to act. each subagent is its own billing stream, so model choice matters.

docs/worktrees.md

@@ -59,6 +59,20 @@ three use cases:
 
 the claude code desktop app has a built-in UI for managing worktrees. you can see all active worktrees, their branches, and their status in one view. it's significantly better than managing worktrees from the CLI.
 
+
+
+### worktree cleanup (v2.1.157+)
+
+worktrees managed by Claude are now left unlocked when the agent finishes. this means you can clean them up with standard git commands:
+
+```bash
+git worktree remove <name>
+git worktree prune  # clean up stale entries
+```
+
+no special cleanup steps needed.
+
+
 ### when to use desktop vs CLI
 
 | scenario | desktop | CLI |