How do I run all Claude Code workflow agents on one model?

Claude Code 2.1.296 adds the CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL environment variable. Set it to a model and every agent started by a workflow runs on that model, while other subagents keep the model they were given. This lets you run large workflows on a cheaper or faster model without editing each agent definition.

Can a Claude Code subagent compact its context earlier than the main session?

Claude Code 2.1.296 adds autoCompactWindow to subagent frontmatter and to --agents definitions. A subagent with this setting auto-compacts its context earlier than the main conversation's window. This helps long-running subagents stay inside a smaller working context without changing compaction for the main session.

What security fixes are in Claude Code 2.1.296?

Claude Code 2.1.296 fixes Bash permission checks that auto-approved some commands assigning and then using the BASH_ARGV0 shell variable; these now ask for approval. On Windows, rm -rf on a user folder in Git Bash now asks in bypass permissions mode, and secret redaction in shared transcripts and debug logs catches more values.

Does Claude Code 2.1.296 change how MCP tools are loaded?

Claude Code 2.1.296 raises the default limit on MCP tool descriptions sent up front, and on MCP server instructions, from 2,048 to 4,096 characters. It also fixes headless sessions starting a folder's .mcp.json or plugin MCP server that had been switched off for that folder, and stdio MCP servers on Windows now get 300 ms to exit before being killed.

Anthropic · 2026-10-09 · major

Claude Code 2.1.296 — one model for workflow agents, earlier subagent compaction

Claude Code 2.1.296 adds CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL to run every workflow agent on one model, lets subagents set their own autoCompactWindow, and fixes managed hooks that denied a call but did not end the turn.

Claude Code lets you pin workflow agents to one model and lets each subagent compact on its own schedule.

Quick facts

Version	v2.1.296
Maker	Anthropic
Released	2026-10-09
Workflow agents	CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL
Subagents	autoCompactWindow in frontmatter and --agents
MCP descriptions	Default limit 4,096 chars (was 2,048)

What is it?

A new environment variable in Claude Code 2.1.296, CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL, runs every workflow agent on one model while other subagents keep their own. Subagent frontmatter and --agents definitions also accept autoCompactWindow, so a subagent can compact its context earlier than the main conversation.

How does it work?

The Read tool gains an allow_large option to read a big text file in one call when the context has room, and CLAUDE_CODE_OVERLOADED_RETRY_MAX_DELAY_MS sets a longer backoff for overloaded (529) retries. The Claude apps gateway gets a code key in managed.policies[] that applies CLI settings to Claude Desktop's Code tab. The default limit on MCP tool descriptions and server instructions sent up front doubles from 2,048 to 4,096 characters.

Why does it matter?

Admins who enforce rules through managed hooks get a fix: a PreToolUse hook that denies with "continue": false now refuses the call without ending the turn. The release also closes a Bash permission gap around BASH_ARGV0, stops Edit from mangling non-UTF-8 files, and corrects token counts for Haiku 5.5 behind some gateways.

Who is it for?

Claude Code users, workflow and subagent authors, admins running managed settings or the Claude apps gateway

Frequently asked questions

How do I run all Claude Code workflow agents on one model?: Claude Code 2.1.296 adds the CLAUDE_CODE_WORKFLOW_SUBAGENT_MODEL environment variable. Set it to a model and every agent started by a workflow runs on that model, while other subagents keep the model they were given. This lets you run large workflows on a cheaper or faster model without editing each agent definition.
Can a Claude Code subagent compact its context earlier than the main session?: Claude Code 2.1.296 adds autoCompactWindow to subagent frontmatter and to --agents definitions. A subagent with this setting auto-compacts its context earlier than the main conversation's window. This helps long-running subagents stay inside a smaller working context without changing compaction for the main session.
What security fixes are in Claude Code 2.1.296?: Claude Code 2.1.296 fixes Bash permission checks that auto-approved some commands assigning and then using the BASH_ARGV0 shell variable; these now ask for approval. On Windows, rm -rf on a user folder in Git Bash now asks in bypass permissions mode, and secret redaction in shared transcripts and debug logs catches more values.
Does Claude Code 2.1.296 change how MCP tools are loaded?: Claude Code 2.1.296 raises the default limit on MCP tool descriptions sent up front, and on MCP server instructions, from 2,048 to 4,096 characters. It also fixes headless sessions starting a folder's .mcp.json or plugin MCP server that had been switched off for that folder, and stdio MCP servers on Windows now get 300 ms to exit before being killed.

Try it

npm i -g @anthropic-ai/claude-code@2.1.296