Mark the pace on agent limit meters

[mojek]

The Agents panel shows how much of each limit is used and when it resets, but not whether that usage is sustainable. 87% of a weekly window reads very differently on day one than on day six.

This adds a pace tick to each limit meter: the share of the window that has already elapsed, i.e. where usage would sit if the allowance were spent evenly until the reset. Usage to the right of the tick runs out before the reset. The reset line spells it out as well:

Resets in 3d 3h · pace 55% (+32 ahead)

• The span comes from the existing windowSpanMs() label parsing (5-hour, 7-day, monthly, Codex 5h window). It is kept on the limit record as spanMs.
• When the span is unknown, or the reset is further away than the parsed span allows (the misparsed (1M context) case already called out in limitWindow()), no tick is drawn and the reset line is unchanged.
• The tick has a surface-colored halo so it stays readable over both the fill and the empty track.

Testing

• test/shell.d/agent-usage-claude-limits-test.sh: record-shape expectations updated for spanMs, plus new cases for session and weekly pace, a fresh window, a missing reset time and a misparsed span.
• ./test/all: the only failures are the three omarchy-pkgs checkout checks that need that repository cloned next to this one.
• Verified in the running shell on Omarchy 4.0.3 (Claude Max and Codex windows).

🤖 Generated with Claude Code

https://claude.ai/code/session_01M2z9bzhcU7sA3PnHWoMSRL