Claude Code Digest — 2026-03-31 08:29:07
What the docs reveal
The End of Manual Team Configurations
Claude Code now strictly manages team configuration files. If you manually tweak a teams.json file, the system overwrites your changes. Anthropic explicitly dropped support for project-level team configurations, choosing to ignore manual JSON edits completely.
State synchronization forced this engineering trade-off. When developers edit configuration files while autonomous agents execute tasks, the file state and the runtime state diverge. Resolving these live state conflicts creates heavy engineering overhead and unpredictable agent behavior. Anthropic chose reliability over maximal user control. They locked the team configuration state behind the CLI, turning teams.json into an internal implementation detail rather than a public developer interface.
Adjust your repository management. Stop tracking teams.json in version control. Delete any deployment scripts that generate or manipulate team configurations. You must treat team orchestration as an ephemeral runtime process handled entirely by the Claude Code engine.
Subagent Inheritance Drives Composability
To preserve role reusability without exposing internal configuration files, Anthropic introduced subagent inheritance. When you spawn a team, you now reference an existing subagent definition. The new teammate automatically inherits the system prompt, tool access, and model configuration from that blueprint.
This shift reveals Anthropic's product direction: modular composability. The Claude Code team wants developers to stop hand-wiring static, monolithic agent structures. By restricting team definitions and forcing reliance on subagent definitions, Claude Code enforces a superior architectural pattern. You build isolated, testable components. A standalone "database-optimizer" agent now translates directly into a reusable team role with zero configuration drift.
Migrate your workflow to a registry model. Build a library of single-purpose subagent definitions within your codebase. Give each subagent a narrow scope and specific tool access. When a complex task requires multi-agent coordination, orchestrate a team dynamically by deploying these pre-defined definitions. Your subagent files now serve as the permanent, version-controllable source of truth for all team behavior.