claude-code

You can delegate large repository edits to a non-interactive Claude Code CLI session, passing a context handoff prompt with goals, scope, and constraints to execute long-horizon coding tasks and return a reconciled summary of diffs.

Subagent-style code delegation uses isolated headless sessions to handle complex implementations like feature slices or batch file edits. This mechanism allows the main orchestrator agent to stay lean while completing tool-using, long-horizon reasoning over the codebase.

To run non-interactive Claude Code execution for repo refactoring, invoke the CLI with explicit model selection and permission auto mode. You provide a context handoff prompt covering the goal, decisions, scope, constraints, and verification commands.

Yes, Claude Code CLI supports automated test verification for batch file edits. You embed an explicit verification command within the context handoff prompt to validate results and reduce wasted edits before the session returns the final diff summary.

You should use an isolated headless session for deep code exploration when your orchestrator risks context overload from iterative tool usage. It handles bounded, long-horizon coding passes that require complex implementations and multi-step reasoning without bloating the main agent.