AI Coding News: July 21, 2026 — Claude Code v2.1.216 Fixes a Quadratic Slowdown, Adds a Sandbox Opt-Out
Just one release lands today, but it's a substantial one: Claude Code v2.1.216, pushed out late on July 20, bundles more than 40 fixes and one new opt-out setting. The headline items are a fix for a quadratic slowdown that turned long-running sessions into multi-second stalls, and a new sandbox.filesystem.disabled setting for teams whose environments don't get along with Claude Code's filesystem isolation. Codex CLI, GitHub Copilot CLI, and OpenCode all stayed on their current stable versions today — nothing new to report from Anthropic's usual co-travelers this cycle.
Claude Code v2.1.216: Fixing a Quadratic Slowdown in Long Sessions
The most consequential fix in this build addresses message-normalization cost that grew quadratically with the number of turns in a session. In practice, that meant anyone running Claude Code for hours — the kind of multi-turn agentic workflow the tool has been increasingly built around — would hit multi-second stalls mid-conversation and slow resumes when picking a session back up. Quadratic-cost bugs like this are easy to miss in testing because they're invisible in short sessions and only bite once a session accumulates enough turns, which is exactly the usage pattern Claude Code has been optimizing for all month with its subagent caps and background-session work.
A New Sandbox Opt-Out: sandbox.filesystem.disabled
The release also adds a sandbox.filesystem.disabled setting that lets teams skip Claude Code's filesystem isolation layer while keeping network egress control intact. It's aimed at environments — custom mounts, unusual container setups, or existing sandboxing already in place — where the filesystem virtualization conflicts with tooling Claude Code doesn't control, without forcing teams to give up network restrictions entirely just to get there. It's an escape hatch, not a new default: filesystem isolation stays on unless you deliberately opt out.
Security and Session-Isolation Fixes Worth Knowing About
Several smaller fixes tighten session isolation and security boundaries. Worktree-isolated subagents could previously redirect git operations back to the shared checkout via git -C, --git-dir, or environment variables — effectively stepping outside the isolation the worktree was supposed to provide — which is now closed. claude daemon stop --any could potentially terminate unrelated processes; it's now scoped correctly. Auto mode was incorrectly denying commands that failed with an "HTTP 401" classifier error after an OAuth token expired, instead of treating it as a token-refresh case. Resumed background agent sessions were reverting to the default agent instead of restoring the original session's prompt and tool restrictions — now fixed, which matters for anyone relying on a custom agent's restrictions staying in force across a resume. MCP re-authenticate flows were also revoking working credentials before a new sign-in completed, which could leave a session without any valid MCP credentials if the new sign-in failed.
Why It Matters
This is Claude Code's fourth release in five days, and the pattern across all of them is consistent: harden the boundaries around long-running, background, and multi-agent sessions rather than ship new headline features. Together with the subagent-spawn caps from v2.1.212 and the permission-bypass fixes from v2.1.214, v2.1.216 continues closing gaps that only show up once you run Claude Code unattended for extended stretches — exactly the mode more teams are adopting it in. If you run multi-hour sessions, worktree-isolated subagents, or background agents that get resumed later, update to v2.1.216: the quadratic-slowdown fix alone is worth it, and the worktree git-redirect fix closes a real isolation gap if you rely on worktrees for untrusted or parallel agent work.
Sources & Further Reading
Comments
Share your thoughts and join the conversation
Leave a Comment
Keep reading.
Daily Dev Stack Release Audit — July 28, 2026: Redis, MongoDB, and Nuxt Rush Out Critical Security Patches
AI Coding News: July 28, 2026 — Model Context Protocol Ships Its Final 2026-07-28 Spec
