Claude docs changes for July 27th, 2026 [diff]

Executive Summary

• Subagent nesting is on by default again: subagents can now spawn subagents of their own up to three layers deep (CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH default raised from 1 to 3 as of v2.1.219), reversing the v2.1.217 change that turned nesting off by default.
• The workflowSizeGuideline setting's default changed from unrestricted to medium (fewer than 15 agents), and as of v2.1.219 it can be set directly in any settings file, taking precedence over the /config Dynamic workflow size row.
• Usage-policy-refusal and cybersecurity-safeguard error messages were shortened and now name the specific model that flagged the request; non-interactive mode drops the /feedback mention in favor of a plain "Learn more:" link.

Claude Code changes

Changed documents

agent-sdk/subagents [Source]

• Subagents can now spawn subagents of their own by default, up to three layers below the main conversation; CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH still controls the limit, with 1 turning nesting off. Documents the intermediate default of 1 used in v2.1.217-v2.1.218 before v2.1.219 raised it to 3. [lines 174-177] [Source]

claude-platform-on-aws [Source]

• Removed the enterprise sales/pricing banner ("Deploying Claude Code across your organization?") from the top of the page. [line 1] [Source]

env-vars [Source]

• CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH default changed from 1 to 3, matching the new default-on nesting behavior; set 1 to turn nesting off again. [line 246] [Source]

errors [Source]

• Usage policy refusal message shortened to <model> can't help with this. Start a new session to continue., naming the specific model (or Claude when none is recorded); non-interactive mode now ends with Learn more: and a link instead of mentioning /feedback. Documents the prior wording used before v2.1.219. [lines 998-1004] [Source]
• Cybersecurity safeguard message shortened to <model>'s safeguards flagged this message, pointing to the Cyber Verification Program; non-interactive mode now ends with Learn more: and a link instead of dropping the /feedback sentence. Adds a version history of the three message wordings used since before v2.1.203. [lines 1015-1030] [Source]
• "Matched no tools in this session" and the zero-tools refusal now describe the Agent tool being withheld because a subagent is at the nesting depth limit, rather than because nesting is off entirely — consistent with nesting now being on by default. [lines 1264 and 1279] [Source]

settings [Source]

• workflowSizeGuideline moved out of the global-config-only table and into the main settings table: its default changed from unrestricted to medium, it can now be set directly in any settings file as of v2.1.219 (taking precedence over /config, which hides the row once a settings file sets the key), and on v2.1.202-v2.1.218 it must still be set via /config. [line 314] [Source]

sub-agents [Source]

• "Let subagents spawn their own subagents" rewritten: nesting is on by default up to three layers, Agent is withheld only once a subagent hits the depth limit, and CLAUDE_CODE_MAX_SUBAGENT_SPAWN_DEPTH=1 turns nesting off rather than being the default. Adds a version history table for the v2.1.172-v2.1.216 (5 layers, unconfigurable) and v2.1.217-v2.1.218 (default 1) eras. [lines 757-779] [Source]

workflows [Source]

• "Set a size guideline" rewritten: the default is now medium (previously unrestricted), /config shows medium (default) until a value is chosen, and as of v2.1.219 the workflowSizeGuideline settings-file key takes precedence over /config and hides that row. The Large workflow warning threshold still follows only a guideline you explicitly set. [lines 306-333] [Source]

API changes

No significant documentation changes today; only Cloudflare email-obfuscation link updates in build-with-claude/files and build-with-claude/thinking.