bench-opus-4-8-ultra

Analyze Claude Code session jsonl logs to report token usage, estimated cost, per-agent and per-PR breakdowns, timing, and the cache-read/write split. This skill should be used when the user asks how many tokens or how much money a workspace/session/project consumed, "how was this built", how many agents or sub-agents ran, how long it took, or wants a 0-to-1 rundown of an implementation reconstructed from the session transcripts.