DRAFT - please do not review or merge yet

Summary

What changed?

A set of related changes to how the plugin shows a quota to a person.

1. The prompt status line can describe the whole account pool, not just the account serving the last request.

Today the status line names whichever account answered most recently. On one
account that is exactly right. On a pool of several it changes identity as
rotation moves, so the corner of the screen flickers between accounts, and the
only way to learn where the pool actually stands is to keep watching until every
account has gone past.

The new quotaStatus.mode: "overview" renders the pool instead - every account
at once, in a fixed order, so the line changes only when a quota does:

24%: #1 5x 13%, #2 20x 100% 3d 1r, #3 1x 12%

The leading figure is the pool total, and it is a weighted mean. A Pro seat
spent to 50% has given up twenty times the capacity a Business Standard seat
does at 50%, so averaging them as equals describes a pool nobody has. Each
account is summarised by its governing window - the one with the least headroom,
since that is what stops a request. A reset time (3d) is printed only for an
account at or below 25% headroom, which is the threshold the existing
single-account line already uses for the same question. 1r is a banked
rate-limit reset the account could redeem now.

Everything past the total is a switch, so the same pool can read as bare as
24%: 3 accounts or carry allotment badges, reset times, banked counts and the
pool's next recovery (+12% in 3d); how the accounts are arranged, ordered and
named is configurable too, which is section 4. active remains the default
mode: it is what every existing install has today, and on a single account the
two modes say the same thing at different lengths.

The line degrades by preference rather than by length on a narrow terminal -
recovery clause first, then annotations, then the per-account breakdown, leaving
the pool total last. Sorting candidates by length instead would let the shape
change as a percentage crossed from 9% to 10%, which is the flicker this
mode exists to remove.

No extra network traffic. The quota guard already polls every enabled
account on a bounded interval and discards the per-account detail; the new
lib/tui-quota-overview.ts caches that same read as a pool snapshot. The
request path's live per-response push for the serving account is merged back
over it when newer, so the one account a reader can check against their own
activity is not the one that looks stale.

2. Quota percentages can be worded as consumption instead of headroom.

Every quota percentage is currently the headroom still left, matching how Codex
itself reports a quota. quotaDisplay: "used" inverts the wording for people
who track consumption:

5h limit: 88% left     ->  5h limit: 12% used

It covers the TUI status line and quota details dialog, codex-limits, the
standalone limits CLI, the interactive account check, and the macOS quota
notification. The used figure is derived from the free one rather than from the
raw used_percent, so two readings of one window always total 100 - rounding
them independently would let the same window render as 88% left on one surface
and 13% used on another in one session.

3. The status line follows a config edit without a restart.

Both settings above are read by a surface that used to read them once, when
the TUI plugin loaded, and then render that reading for the whole session -
so trying either one meant quitting OpenCode and losing every open session to
move a badge. Every other surface already re-reads the file per call, because
loadPluginConfig re-reads it and only skips re-parsing when the bytes are
unchanged.

The status line now polls that same loader every two seconds, so an edit shows
up in the running TUI. mode is the hard case: the two modes are not two
renderings of one thing, but two pipelines polling different sources. Each is
now a controller behind a small interface, and the node the renderer mounted
stays put while the pipeline underneath it is swapped and the outgoing one
disposed. Readings are compared field by field, since every poll builds a
fresh object and comparing references would re-render the line forever.

Why is this needed?

Anyone running more than one ChatGPT account currently has no way to see the
pool. The status line answers "how much has the account that just replied got
left", which is not the question a person with seven accounts is asking - they
want to know whether the pool as a whole can carry the next hour of work, and
which account is about to come back. Watching a flickering line long enough to
reconstruct that by hand is the workaround this removes.

The weighting matters for the same reason. A plain average over mixed plans
reports a number that no account holds and that moves in the wrong direction
when a large seat is spent. lib/plan-allotment.ts records OpenAI's published
per-seat ratios (Pro 20x, Business Premium / legacy Pro 5x, Plus and Business
Standard 1x) so the total means something; docs/plan-allotments.md documents
where those figures come from. A plan that states no ratio - Free, Go,
Enterprise, and a bare business that names a workspace rather than a seat -
returns "unknown" rather than a guess, so a caller can tell 1x apart from "we
do not know".

Both changes are presentation only and both default to today's behaviour.
Exhaustion, rotation blocks, notification thresholds and the status line's
warning/danger colouring all stay keyed on the percentage remaining, and the
usedPercent / leftPercent fields in JSON output are untouched.

4. The shape of the pool line is configurable, and only from the config file.

One arrangement does not suit every pool, so the line is now described rather
than fixed. layout chooses between one segment per account, one per distinct
percentage, or a bare count. accountNames chooses between #1, the name the
account's owner gave it, and nothing. order arranges by number, by consumption
in either direction, or by which renews first or last. allotment states what
the weighted pool adds up to. resetTimes grows from a boolean into
never/low/always.

aggregate is the one worth showing. On a pool where several seats are spent,
100% 3d, 100% 4d, 100% 5d spends two thirds of its characters repeating a
number that does not change:

72%: 12% 3d, 50% 4d, 100% 3d 1r 4d 5d

A group states its own size - 100% x3 4d 5d - only when its annotations would
not already reveal it, so a group of three where one member has no reset to
print cannot read as a group of two.

mode accepts a list, and the line then alternates between those screens every
rotateMs. A screen with nothing to say is skipped rather than shown blank,
which is what makes the third screen worth leaving in the list permanently:
resets renders only once every account is spent, and lists the banked
credits worth redeeming, latest reset first - redeeming one on an account that
renews by itself tomorrow throws it away, while the account six days out is the
one worth spending it on.

Free resets: 6d 1r damian@nowaker.net, 4d 2r work@example.com

Narrow terminals lose less than they did. 3 accounts steps through 3 acct.
to 3 before the line gives up on the count, the recovery clause drops the word
in before dropping the clause, and an unnamed breakdown is offered as the last
rung above the count - but only while position still identifies an account,
which means number order with nothing missing from the line.

None of quotaStatus is overridable by environment variable, and the
CODEX_AUTH_QUOTA_STATUS* overrides this PR previously added are removed. It is
a display preference belonging to a person rather than to whichever shell
started OpenCode, and modelAccountPools sets the file-only precedent.

5. The line is laid out against the space it actually has.

Two defects, both found by watching a real TUI rather than by reading the code.

The width budget came from the terminal, which is the wrong number whenever a
sidebar is open: at 165 columns the slot had 73 and the line was built for 99,
so the renderer ellipsized it through the middle and destroyed account numbers
either side of the cut. The budget now comes from the prompt row's measured
width.

The label's own width and height are deliberately not measured, and both
were tried and reverted here. The row sizes both boxes by their content with
alignItems: stretch, so the label reports this line's height once this line
grows, and is shrunk to whatever this line did not take once the row is full.
Either reading makes the budget a function of its own output. rows is
therefore a plain ceiling (1-4, default 1) rather than a measurement, and it
costs nothing until the content needs the room, since a candidate that fits on
one row still returns one row.

And the host centres this slot against a model label that wraps to two rows on a
narrow terminal, which put a one-row line on the bottom row and left the top one
empty. The line is now top-aligned, and a second row is one text node with a
newline in it so the renderer measures it and the node sizes itself.

Config surface (all optional, all defaulting to current behaviour):

Testing

• npm run lint
• npm run build
• npm test

npm run lint and npm run typecheck exit 0. npm test on this branch:
3737 passed, 1 skipped, 2 failed - both failures are in
test/index-retry.test.ts and are pre-existing, not caused by this change:
they reproduce on unmodified main under full-suite load and pass when that
file is run alone. Around 230 new tests were added across
test/plan-allotment.test.ts, test/quota-overview.test.ts,
test/tui-quota-overview.test.ts, test/tui-status.test.ts,
test/tui-status-slot.test.ts and test/tui-config-reload.test.ts.

Beyond the suites, every flag was driven through a real OpenCode TUI built from
dist/, in an isolated HOME with a seeded pool cache, by editing the config file
underneath a running session - no restart at any point. That is also where four
defects turned up that no unit test would have caught:

• At 165 columns the line was ellipsized through its middle, because the width
  budget was the terminal's rather than the prompt row's.
• A one-row line sat on the bottom row of a two-row prompt, leaving the top row
  empty.
• Switching screens left the line blank until the next unrelated change: the
  render triggered by the new configuration read the outgoing pipelines, found
  none of them serving the incoming screens, and subscribed to nothing. The swap
  now happens before the signal that re-renders against it.
• An earlier rows: "auto" that took its height from the model label beside it
  latched at two rows permanently, because alignItems: stretch makes that label
  report this line's own height. That is why rows is a plain ceiling now.

The final pass covers all thirteen fields, both quotaDisplay wordings,
maskEmail, the aggregate and count layouts, all five orders, the rotation
between screens, the reset-credit screen on a fully spent pool, the two-row and
three-row renders, and the degradation ladder at eight terminal widths from 110
down to 50 columns.

Compliance Confirmation

• This change stays within the repository scope and OpenAI Terms of Service expectations.
• This change uses official authentication flows only and does not add bypass, scraping, or credential-sharing behavior.
• I updated tests and documentation when the change affected users, maintainers, or repository behavior.

No new network calls, no new endpoints, and no change to authentication. The
pool snapshot reuses the existing /wham/usage read the quota guard already
performs.

On identity: by default the line numbers accounts the way codex-list and
codex-switch already do, and renders no email. Two of the new settings can
print one, and both honour maskEmail. accountNames: "label" prefers a label
set with codex-label and falls back to the local part of the email
(damian@nowaker.net -> damian, or da*** when masking is on). The resets
screen prints the full address, because the whole point of it is to say which
account to redeem a credit on - it shortens to the label and then to #1 on a
narrow terminal, and it appears only when the pool is spent and something is
redeemable. Neither is on by default. The pool cache gained email and label
fields to support this; it is written mode 0600 like every other file the
plugin writes.

Notes

• Linked issue: none
• Follow-up work or rollout notes: docs/plan-allotments.md is the single place
  to revise when OpenAI changes per-seat pricing or ratios; the allotment map is
  deliberately reusable, so a future change could also surface 5x / 20x
  badges in codex-list, codex-status and generated account labels.

This PR is AI generated, but under direct supervision and on request of Nowaker.