Claude Code sends full background agent transcripts to its model

An analysis of Claude Code's architecture found that each background agent request serializes and ships the agent's full history to the model, including file paths, command output and environment variables. This inflates token usage and latency and raises data-leakage and compliance concerns for sensitive codebases.

• Transcript includes file paths, content hashes, commands and environment variables
• Resending history raises token usage and time-to-first-token
• No state-compression layer forces the model to reprocess past steps
• Data leaves for provider servers, raising compliance questions