Anthropic: Claude Code v2.1.215 and v2.1.216 Fix Performance, an OAuth Error, and Agent Worktree Isolation
Anthropic released Claude Code v2.1.215 and v2.1.216, which remove automatic triggering of the /verify and /code-review skills, fix quadratic performance degradation in long sessions, an OAuth 401 error, and a worktree isolation leak in subagents, and also fix message loss on cloud container restarts and symlink following inside the .claude directory that could write outside the project.
This article was generated using artificial intelligence from primary sources.
Skills no longer trigger automatically
Claude Code (Anthropic’s CLI coding agent) v2.1.215, released on July 19, 2026, removes the automatic triggering of the /verify and /code-review skills after code changes. Previously, Claude Code would invoke these checks on its own; now the user must run them explicitly with a command.
The change builds on v2.1.214 and aims to reduce unwanted automatic actions during agent operation, giving the user more control over the flow of a session.
What does v2.1.216 bring?
v2.1.216, released July 20, 2026, delivers five fixes. The first corrects quadratic performance degradation described as “message normalization cost” in long sessions. The second resolves auto mode rejecting commands after OAuth token expiry (HTTP 401) — an OAuth token is the credential that authorizes the agent to access the Anthropic API.
The third fixes worktree-isolated subagents that were redirecting git operations into a shared checkout instead of their own isolated directory (a worktree is a separate working copy of a git repository). The fourth prevents symlink following inside the .claude directory that could write outside the project. The fifth fixes the loss of an in-flight message on cloud session container restarts.
Why does the quadratic degradation fix matter?
Without this fix, longer Claude Code sessions spent increasingly more time processing messages as the session grew — the cost scaled quadratically instead of linearly with the number of messages. That slowed the agent down precisely in the more demanding, long-running tasks where speed matters most.
Security fixes: worktree and symlinks
The worktree isolation and symlink-following fixes have a security dimension: they prevent an isolated subagent from accidentally modifying a shared git checkout, or a symlink inside .claude from enabling writes outside the project’s boundaries. Both fixes reduce the risk of unwanted filesystem changes during automated operation, compared to the behavior in v2.1.214 where these gaps were present.
Frequently Asked Questions
- What changes in Claude Code v2.1.215?
- Version v2.1.215, released July 19, 2026, removes automatic triggering of the /verify and /code-review skills — they now require an explicit command invocation.
- What are the main fixes in v2.1.216?
- Version v2.1.216, released July 20, 2026, fixes quadratic performance degradation in long sessions, an OAuth 401 error in auto mode, a worktree isolation leak, and message loss on container restarts.
Sources
📬 AI news in your inbox
A daily digest built your way — pick topics, sources and cadence. One-click unsubscribe.
Related news
arXiv:2607.16122: CRAFT Method Clusters Grading Rubrics to Reveal Weaknesses in Language Models
CNCF: Goodput Metric Reveals Nearly 10× Worse p95 Latency in Throughput-Optimized LLM Serving
GitHub: Code Quality Reaches General Availability with CodeQL Analysis and Copilot Autofix Fixes