feat: add a Session Cost Rate widget - #671

What

A new Session Cost Rate widget shows the session's cost per hour: Rate: $5.00/hr. It sits next to Session Cost in the picker's Session category.

Why

Session Cost shows what a session has cost so far, but not how fast it's spending. A rate makes long and short sessions comparable, and on usage-based plans, where session cost is close to what's actually billed, it reads as the agent's hourly rate.

How

• Time base: by default it divides by the time Claude spent working, the cost.total_api_duration_ms Claude Code sends with the status payload. That reads as a rate for the hours it actually worked. A t keybind in the line editor switches to clock time (cost.total_duration_ms, the whole session including time spent waiting for the user), which reads as the burn rate. The editor row shows (active time) or (clock time).
• First minute: it shows nothing until a full minute of the chosen time has passed, since a rate over the first seconds swings wildly ($0.50 in 30 seconds is $60/hr).
• Format: the cost number format applies, and the unit stays in raw value mode ($5.00/hr).
• Data: only the status payload, so it works on every account type and needs no fetch.
• Preview: a fixed sample, Rate: $4.90/hr (the Session Cost sample over 30 minutes).

Demo

A status payload with $2.50 over a 2-hour session in which Claude worked for 30 minutes: the third widget uses active time ($2.50 ÷ 0.5 h), the fourth clock time ($2.50 ÷ 2 h).

Powerline

Plain

Switching between active and clock time with t in the line editor:

Powerline

Plain

Testing

• New widget tests: the rate over active time and clock time, raw value mode, nothing before a full minute (both time bases) and a rate at exactly one minute, missing cost or duration, the cost number format, the preview sample, and the t toggle (keybind labels, editor modifier, metadata).
• On today's main (with docs: explain an empty status line in one folder (workspace trust) #606 and feat(git-is-fork): render forks as an editable glyph #617 merged in locally): bun test gives 2790 pass, 0 fail (Bun 1.4.2), and bun run lint passes.
• Under Node (Vitest): the new tests pass, 8 of 8.
• Built CLI under Bun 1.4.2 and Node 26.10.0 with a payload carrying cost and durations: all runtimes give identical output in plain and Powerline modes; it differs from main only by the new widgets. The TUI opens and exits cleanly under both.