Claude Code v2.1.216: Sandbox Filesystem Toggle, Quadratic Slowdown Fix, and 30+ Bug Fixes

Anthropic shipped Claude Code v2.1.216 on July 20, with a new sandbox toggle and over 30 bug fixes targeting long-session performance, OAuth token handling, worktree isolation, and terminal UX.
Key Addition: sandbox.filesystem.disabled
The new setting sandbox.filesystem.disabled lets you skip filesystem isolation while keeping network egress control. Useful when you trust the code being run but still want to block outbound connections.
Performance Fix
The biggest performance bug: message normalization cost grew quadratically with the number of turns, causing multi-second stalls and slow resumes in long sessions. That's now fixed — normalization is back to linear cost.
Notable Bug Fixes
- Auto mode 401 errors: Fixed auto mode denying commands with "HTTP 401" classifier errors after OAuth token expired or rotated mid-session.
- AskUserQuestion wording: Free-text answers now get neutral wording instead of telling Claude to continue when you asked it to wait or explain first.
- Web session drops: Claude Code on the web no longer re-asks the same question and drops your answer after the session sits idle for a few minutes.
- Background agent resumption: Resumed background agent sessions no longer revert to the default agent — the agent's prompt and tool restrictions are now restored.
- Worktree isolation: Fixed worktree-isolated subagents redirecting git into the shared checkout via
git -C,--git-dir, orGIT_DIR/GIT_WORK_TREE. - Worktree session targeting: Fixed worktree sessions landing in another project's leftover worktree when the working directory didn't match the selected project.
- Background session deletion: Background sessions whose worktree has no git repository are now deletable.
claude daemon stop --any: Fixed a bug that could terminate an unrelated process via a stale legacy daemon lockfile.- Esc-Esc rewind: Esc-Esc at an idle prompt now opens the rewind picker in long-running sessions with background tasks.
- Bash permission checking: Fixed for compound statements with redirects inside
&&lists or negations. - Session deletion: Pressing Ctrl+X twice in the agent list now properly deletes a session, and deleted sessions no longer reappear when their background worker dies.
- Background subagents: Fixed subagents getting cancelled when a high-priority message arrives during their startup window.
- GUI editor UX: Fixed mouse and focus garbage in the terminal while a GUI editor from
/memory,/plan,/keybindings, or Ctrl+G is open;/memoryno longer waits for the editor to close. - Claude-in-Chrome 403 loop: Fixed reconnecting when the session's OAuth token lacks a required scope.
- Symlink writes: Workflow saves and scheduled-task writes no longer follow a symlink at
.claude, which could redirect writes outside the project. - MCP re-authenticate: Fixed revoking working credentials before the new sign-in succeeds; the reconnect needs-auth message in background sessions now points at a usable command.
- Windows network paths: Read-only commands on Windows no longer prompt for permission on network paths.
- Bash non-ASCII parsing: Fixed to match real shell word boundaries.
- PowerShell Unicode: Tool permission validation now handles commands containing invisible Unicode characters.
- Fullscreen mode: Fixed dialogs stretching past the right-hand edge and
/configlist clipping its keyboard-hint footer. - Transcript mode footer: Fixed wrapping on terminals narrower than 104 columns.
- Prometheus metrics: Fixed the endpoint (enabled via
OTEL_METRICS_EXPORTER=prometheus) emitting invalid# UNITlines. - Slash menu updates: Skills and commands changed during a session now appear immediately rather than after restart.
- Plugin skill autocomplete: Plugin skills with a
namefrontmatter field no longer lose their plugin prefix. - Telemetry accuracy: Failed permission-prompt requests no longer count as user rejections; user interrupts are now reported as user aborts instead of rejections.
Improvements
/forkconfirmation is now one line showing the new session's name,claude attachid, and a note when the copy shares your checkout.- Validation of
gitandgharguments in the PowerShell tool is improved. /ultrareviewdiff-too-large error now shows configured limits, measured diff size, and largest contributing files./code-review ultraempty-diff message now names the exact base ref and suggests passing an explicit base.- Spend limit adjustment prompt now shows the server's reason when a change is rejected.
/contextnow warns when the conversation exceeds the context window; a failed/compactdisplays as an error./rewindno longer restores or deletes files through symlinks or hard links at tracked paths.
📖 Read the full source: GitHub Claude-Code
👀 See Also

Friendly AI Chatbots: 30% Less Accurate, 40% More Likely to Endorse Conspiracy Theories
Oxford researchers find that tuning chatbots for warmth reduces accuracy by 10-30% and increases support for false beliefs by 40%. Tested on GPT-4o and Llama.

$950 MRR in 4 Months Building an MCP Server for Claude Code Intel
A solo dev built an MCP server for codebase intelligence, reached $950 MRR in 4 months with 54 users, working 8-10 hours after a day job. No ads, no growth hacking — just Reddit and Medium.

Agent SDK vs Claude CLI: A user's perspective on the practical difference
A Reddit user questions the practical difference between the new Agent SDK for Claude and using Claude CLI to wire Opus 4.7 locally.

Meta's AI Agent Pivot: Zuckerberg Says Progress Not Fast Enough, $145B Spent
Meta CEO Mark Zuckerberg told staff that AI agent development hasn't accelerated as expected. The company spent $145B on AI infrastructure this year and laid off 8,000 employees to reorganize around AI.