v2.1.193

Source: Anthropic Claude Code Date: 2026-06-25 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.193

Summary

Claude Code v2.1.193 ships a cluster of agentic infrastructure improvements: a new autoMode.classifyAllShell setting routes all shell commands through the auto-mode classifier (not just ACE patterns), denial reasons now surface in transcripts and the /permissions panel, and background agents no longer receive the “end your response” instruction — enabling genuine concurrent task execution. Observable quality-of-life fixes close out the agent panel’s sibling-hiding bug and the phantom “general-purpose (resumed)” subagent on main-turn backgrounding.

Implications

• Agent lifecycle / orchestration thread: the removal of the “end your response” instruction for background agents is a meaningful primitive — concurrent multi-agent execution becomes a first-class pattern, not a workaround.
• Claude Code security surface: autoMode.classifyAllShell expanding classifier coverage to all shell commands tightens the auto-mode trust boundary; paired with the denial-tracking visibility, this is an audit-trail investment ahead of enterprise deployments.
• Anthropic distribution machine: v2.1.193 is incremental but directionally consistent with the fleet-hardening pattern tracked since v2.1.143 — each release adds a governance or observability primitive.