# Claude Code v2.1.296 adds subagent auto-compact and a larger Read limit

published: 2026-10-12T13:43:03.976Z
topic: frameworks
source: https://github.com/anthropics/claude-code/releases/tag/v2.1.296
url: https://aiagentnews.io/stories/claude-code-v2-1-296-adds-subagent-auto-compact-and-a-larger-read-limit

Anthropic published Claude Code v2.1.296 with new subagent and workflow model settings, plus a Read tool option for large text files. The release also raises the MCP tool description limit and fixes several permission, hook and encoding bugs.

Anthropic published Claude Code v2.1.296 on 9 October. Subagents can now set autoCompactWindow in their frontmatter or in --agents definitions to compact earlier than the main conversation. A new CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL variable runs every workflow agent on one model while other subagents keep their own.

The Read tool gains an allow_large option that lets Claude read a text file past the usual size limits in one call when it needs the whole file and the context has room. The default limit on MCP tool descriptions sent up front, and on MCP server instructions, rises from 2,048 to 4,096 characters.

Fixes include Bash permission checks that had auto-approved some commands assigning BASH_ARGV0, which now prompt for approval. Edit and NotebookEdit also now refuse edits to files that are not valid UTF-8, where they had replaced non-ASCII characters.

## Why it matters if you build agents

Agents that delegate work to subagents can now set a different compaction point for each one, and read large files in a single call. Builders relying on MCP tool descriptions should check their size against the higher default limit. Agents that run Bash commands setting BASH_ARGV0 will now hit approval prompts.