Feature request: "agent-hours" — a labor metric for agent teams (sum of active agent time)

Tokens and cost measure spend; they don't measure work capacity. I propose agent-hours: the sum of active wall-clock time across ALL agents — main sessions, subagents, workflow workers — where "active" = transcript timestamps merged with a ~10-minute gap threshold (idle waits don't count).

Why it's interesting:

• It's a labor unit — stable across models and pricing changes, unlike tokens/$.
• It makes parallelism visible: one hour with 20 agents ≠ one hour with 1 agent.
• It reads like a team metric: "my agents worked N hours today" is instantly meaningful to non-engineers too.

Real numbers (computed from my ~/.claude/projects transcripts with a small script):

• 891 agent-hours in 37 days — an average of 24 agent-hours per calendar day, i.e. one agent effectively working for me around the clock, including while I sleep.
• Record day: 60.2 agent-hours in one calendar day (~7.5 full work shifts).
• Orchestrator/subagent split: 40% / 60%, 130 distinct agents on the record day.

The data is already all there in the transcript JSONLs — the metric is one aggregation away.

Suggestion: expose agent-hours in /usage and the statusline API (daily/weekly totals, orchestrator vs subagent split). Happy to share my computing script and dashboard if useful.