auditability

There is no free()

Claude Code's extended-thinking output is a summary, not the reasoning that drove the agent, so it cannot serve as an audit trail of agent logic.