fix(usage): poll the plan limits even when the chat is not on screen

[serialexperimentslainnnn]

The bug

The plan-limit poll was gated on isShowing:

Timer(USAGE_POLL_MS) { if (isShowing) requestUsage() }   // before

A collapsed tool window, or a chat tab that was not the selected one, asked for nothing at all. And a quota window is not this plugin's state — other sessions, other devices and claude.ai spend the same windows, and a reset is a wall-clock event that owes nothing to the IDE. So the figure on screen was whatever the last probe happened to catch, and it moved again only when something else triggered one: a turn, or opening the dashboard. "It only updates when I talk to the agent" is exactly what a visibility-gated poll looks like from outside.

The gate is gone and the period is 30 s. The event-driven refreshes (turn edges, rate_limit_event, dashboard open, session ready) are unchanged. What the gate saved was one control request per half minute against a process that is already running.

Added

The countdown, in the chat. Reset time: 4h 18m on its own line under each window's bar, sentence in the tooltip:

Current session [###-----] 6.0%     All models [###-----] 18.0%    Fable [###-----] 17.0%
Reset time: 4h 18m                  Reset time: 70h 38m            Reset time: 70h 37m

A percentage alone does not say whether it is urgent: 90% with eight minutes left and 90% with six hours left are different situations, and only the dashboard was answering that. Under the bar rather than beside it, because the row is already three items wide per window and a fourth made the countdown the first thing to be squeezed out — the one case where it matters most. One implementation (CC.resetIn/resetInShort in app-core), now shared with the dashboard card, and RateLimitInfo.resetsAtIso() puts the epoch→ISO conversion on the model so an event-sourced window and a report-sourced one are interchangeable to both surfaces.

Diagnostics

Every get_usage poll logs the reply verbatim at INFO. The derived per-window lines cannot say whether a number is stale — a window the reply omits leaves no line, and one carried forward looks fresh. It earned its place immediately: a live capture showed two of three consecutive polls returning the header-seeded shape (five_hour/seven_day only, no limits[], resets_at rounded to .000Z), which is the degraded reply 5.1.0's merge exists for — and confirmed the binary does not cache the endpoint.

Process

docs/RELEASE_PROCEDURE.md §6 now states that the release PR is closed to late commits, cosmetic ones above all: a commit pushed onto an open release PR invalidates every gate that passed on it, and main publishes.

Gates — clean locally (clean … --rerun-tasks)

Gate	Result
test	781 tests, 0 failures
koverVerify / detekt / spotlessCheck	pass
verifyPlugin	Compatible IC-251 → IU-262
buildPlugin	0 node_modules, META-INF/{LICENSE,THIRD-PARTY-NOTICES.md} present
npm test / lint / format	113/113, clean
npm audit --omit=dev	0 vulnerabilities

Manually validated in IntelliJ IDEA 2026.2 against a live session — including the case that failed: tool window collapsed and a different tab selected.

🤖 Generated with Claude Code