CloudCodeTree LogoCloudCodeTree
AI NewsTutorialsAbout
CloudCodeTree Logo
CloudCodeTree
  • AI News
  • Tutorials
  • About
← Back to AI News
Claude Code 2.1.210 Hardens Worktree Isolation and Agent Injection Guards; Anthropic Launches Claude Science

Claude Code 2.1.210 Hardens Worktree Isolation and Agent Injection Guards; Anthropic Launches Claude Science

Chris Harper

2 min read

Jul 15, 2026 · 04:08 UTC

AI
News
Claude Code
Agents

TL;DR: Claude Code 2.1.210 patches a silent worktree isolation bug and hardens the Agent tool against prompt injection; Anthropic also launches Claude Science, an AI workbench for scientists.

Claude Code v2.1.210 (July 14)

Two fixes that matter for agentic workflows:

Worktree isolation bug fixed. Subagents launched with isolation: 'worktree' were running git-mutating commands (commits, checkouts, resets) against the main repo instead of their isolated copy. If you use parallel subagents with worktree isolation for file editing, this was silently corrupting your working tree. Upgrade before running any multi-agent refactor.

Agent tool injection hardening. The Agent tool now resists indirect prompt injection via content that subagents read — files, web pages, tool output. Any agent that processes untrusted external content (crawled pages, user-supplied docs) should consider this a meaningful safety improvement.

Also in 2.1.210: a live elapsed-time counter on collapsed tool summary lines for long-running calls; auto mode's permission classifier defaults to Sonnet 5 for external sessions.

Claude Science (beta)

Anthropic launched Claude Science, an integrated AI research workbench for scientists. It ships with 60+ curated skills for genomics, single-cell RNA analysis, proteomics, structural biology, and cheminformatics; handles compute scaling from a laptop to hundreds of GPUs automatically; and produces reproducible artifacts — code, output, and plain-language explanation bundled together. Available now on Claude Pro, Max, Team, and Enterprise.

Why it matters: The worktree isolation fix is a correctness fix for anyone building parallel subagent pipelines — silent bugs in isolation guarantees are the hardest category to catch in production. Claude Science signals Anthropic moving Claude into deep domain-specific professional tooling beyond software engineering.

Sources: Claude Code changelog · Claude Science — Anthropic