Platform
Observability
Run history and cost tracking.
Observability shows what ran, why it ran, what it cost, and whether it succeeded. Use it to debug bad runs, compare agents, and track agent spend over time.
What each run records
Each agent run stores the config snapshot used at creation time, input source, attribution, status, status reason, sandbox, output, token usage, model, and cost. Cost is broken down into token spend, sandbox CPU, sandbox memory, and Ellipsis fee.
Run steps
Run steps record the timeline of the Claude Code session: assistant output, tool calls, tool results, token details, cost, duration, and stop reasons when available.