v1.5.1

Fixed

• Live plan limits and subscription detection now work on macOS. Claude Code
  stores its OAuth credentials in the login Keychain on macOS, not in
  ~/.claude/.credentials.json — so the live-limits fetch found no token and
  silently fell back to the on-disk cache (which can sit hours stale). The
  extension now reads the token from the Keychain (the same
  Claude Code-credentials item Claude Code itself writes) read-only, and
  still prefers the credentials file when one is present. Windows and Linux are
  unchanged.

v1.5.0

Fixed

• 1-hour cache writes were undercounted by 37.5%. Anthropic bills prompt-cache writes
  with the 1-hour TTL at 2× input, versus 1.25× for the 5-minute TTL — and Claude Code
  sessions use the 1-hour TTL. The extension priced all cache writes at the 5-minute
  rate. It now splits the 1-hour portion out of every usage record and prices it at 2×.
  Your cache-write cost figures will step up after this release — the new numbers are
  the correct ones, not an increase in your actual usage.
• Sonnet 5 is priced correctly ($2 / $10 per MTok introductory pricing, through
  2026-08-31) — it previously fell through to the $3 / $15 default, overstating Sonnet 5
  usage by ~50%. Model matching is also more robust: Bedrock-style prefixed ids
  (anthropic. / us.anthropic.) now match, and an unlisted or future id is priced by
  its family (a future Opus prices as Opus) instead of silently billing at the mid-tier
  default.
• Totals no longer dip or zero out on transient file errors. Three separate causes,
  all fixed to keep the last good data and retry: a session file locked mid-write (common
  on Windows) was cached as empty until its next change; a failed listing of the log
  directory purged every parsed file and published zeros; and a failed stat on one file
  silently dropped its contribution for a pass.
• Today's bar no longer disappears from the Daily chart for the rest of the month
  after a DST spring-forward in timezones where the clocks skip midnight (e.g. Havana,
  Santiago, the Azores).
• Case-distinct projects on Linux no longer merge. /dev/API and /dev/api are
  different directories on a case-sensitive filesystem and now group separately in
  By project; Windows and macOS keep folding case, matching their filesystems.
• The status bar's four items are now distinguishable in the right-click "Hide" menu.
  They used to appear as four identical unnamed entries sharing one visibility setting;
  each now has its own name and remembers its own visibility.
• "Explain with AI" can no longer get stuck on "Analyzing your usage…" — an
  unexpected failure now surfaces as an error box in the panel.
• A relocated CLAUDE_CONFIG_DIR is honored everywhere. The log reader, the limits
  cache, credentials, and the shared multi-window cache all resolve from the same root;
  previously the live-limits fetch honored the override while the log and cache readers
  kept reading ~/.claude, splitting the two sources.

Changed

• Much lighter on long, heavy histories. The active session file is no longer re-read
  and re-parsed in full on every assistant turn — only the newly appended lines are read
  and folded into the parsed cache (with truncation/rewrites detected and re-parsed in
  full). Cold startup streams files in chunks instead of loading each as one giant string,
  and yields to the editor while parsing. A refresh that finds nothing changed reuses the
  previous result instead of rebuilding the record set, repeated strings (model ids,
  paths, session ids) are stored once, and the dashboard webview is released while hidden
  and rebuilt on reveal — with the last AI explanation preserved — instead of keeping a
  full browser context alive in memory.
• The limit colors are now yours. The status-bar sunburst and the dashboard bars turn
  yellow at your first predictiveAlerts.warnThresholds value and red at your last
  (75% / 90% by default), so the colors always agree with your warning notifications. An
  explicit warning from Anthropic can still escalate a color, never downgrade one — and
  the status bar and dashboard can no longer disagree with each other.
• The 5-hour pace meter is visible again for the whole window. 1.4.0 stopped it from
  crying wolf, but overshot: it hid the projection for the first half of every window and
  whenever you paused for a few minutes, so most of the time there was simply no estimate —
  the meter looked broken. It now shows from early in the window and stays put while you're
  idle. Early on, the projection is damped toward your current utilization instead of
  hidden: a thin early slice extrapolates wildly (5h 15% at 30 min in would read a raw
  → 150%), so the meter is shrunk toward where you actually are and firms up to the plain
  projection by the time the window is half over. Past the half-mark the number is exactly
  what 1.4.0 showed. The breach popup is unchanged — it still fires only once the window
  is at least half over, you're actively burning, and you're clear of the on-pace band — so
  an always-on meter never brings back false "you'll breach" warnings.

v1.4.0

Fixed

• Output tokens (and cost) were undercounted by ~20%. Claude Code writes an assistant
  message to its transcript once per streamed content block; the interim lines carry a
  placeholder output_tokens and only the final line reports the settled figure. The
  extension was keeping the first line of each message and discarding the rest, so it
  banked the placeholder and threw the real output count away. It now takes the final,
  settled usage. Your output-token and cost figures will step up after this release —
  the new numbers are the correct ones, not an increase in your actual usage.

Changed

• The 5-hour pace meter no longer cries wolf. It was a bare extrapolation
  (utilization ÷ elapsed), which is a knife-edge at exactly the moment it matters: being
  a point or two ahead of an even pace rendered as a confident → 104%, and one idle
  minute later as → 99%. Three changes:
  • It waits until the window is half over. The meter was shown from 10% elapsed, but
    the breach warning already refused to fire before 50% because the projection isn't
    reliable before then — so the extension was printing a number it was too suspicious of
    to warn on. Both now share one gate. Early in a window you see just 5h 15%.
  • New · on pace reading. Within ten points of 100% the projection is noise, so the
    meter now says 5h 53% · on pace instead of committing to a number that flips on the
    next refresh. Clearly under or over pace still shows the figure (5h 23% → 45%,
    5h 95% → 119%). The breach warning clears the same band, so a popup can no longer
    contradict an "on pace" meter.
  • It hides when you stop. The meter ignored whether you were still burning, so a
    heavy start followed by an hour of idling kept projecting a breach. It now disappears
    when you're idle, like the weekly ETA and the warnings already did.
• The burn readout says what it actually measures. Burn: 1.6M tok/min was ~95% cache
  reads — a long session re-reads its whole context every turn — so the headline tracked
  context size × turn count rather than anything you consume, and it sat under the "Claude
  plan limits" heading as though it drove the pace projection (it never did). It is now its
  own tooltip block, split into new work versus context re-reads, with the rate labeled as
  an API-equivalent estimate on a subscription, matching the dashboard:
  49.5k new tok/min · 849.1k/min context re-read · ≈ $0.80/min at API rates.

v1.3.1

Changed

• Per-table totals. Replaced the all-time grand-total strip (added in 1.3.0) with a
  totals row at the foot of each dashboard breakdown — By model, By project, and By
  branch now sum their Messages, Tokens, and Cost columns, in the same format as the rows
  above. The All Time card already serves as the grand total, so the separate strip was
  redundant; per-section totals are more useful.

v1.3.0

Added

• Grand total — a strip beneath the Today / This Month / All Time cards in the
  dashboard, summing your all-time spend, messages, and active time in one place.
  Active time reuses the session active-time measure (idle gaps longer than 15 minutes
  are excluded), so it reflects time actually spent working, not wall-clock. On a
  subscription the spend is labeled "≈ API" to match the rest of the dashboard.

Fixed

• Plan limits now stay live across multiple editor windows. With more than one
  window open (and especially on a Max plan), each window used to poll Anthropic's
  usage endpoint independently; the extra traffic could get a window rate-limited
  (HTTP 429) and stuck showing a stale "Updated Nd ago" reading until you reloaded it.
  A successful fetch is now published to a small shared file that every window reads,
  so all open windows share one fetch (~one request between them instead of one each),
  and a rate-limited window shows a sibling window's fresh data instead of falling back
  to a stale cache. The shared file holds only the usage figures and a timestamp — never
  your credentials — and lives alongside Claude's own files (honoring CLAUDE_CONFIG_DIR)
  so Cursor and VS Code on the same account coordinate too.

v1.2.1

Full Changelog: v1.2.0...v1.2.1

v1.2.0

Added

• Advisor — a new panel in the dashboard with ranked, money-quantified tips to
  cut waste, computed entirely from your local usage data (no network, no API key,
  no prompt text). Toggle with advisor.enabled (on by default). It surfaces:
  • Routing routine turns to a cheaper model — when a period's spend includes a
    run of short turns on an expensive model, it estimates what those same turns
    would have cost one tier down (e.g. Opus → Sonnet) and shows the saving.
  • Long sessions re-processing context uncached — sessions with many turns but
    low prompt-cache reuse, where context was paid for at full input price instead
    of read back at ~1/10th.
  • Sessions running near the context limit — a habit of peaking above 80%
    context, where Claude has to compact or drop earlier context to keep going.
  • A month-end spend forecast at your current pace.
  • Subscription-aware framing — on a Pro/Max plan (flat fee, no per-token billing)
    the dollar figures are labeled estimated API-equivalent usage (a gauge, not a
    bill) and the advice — local cards and the AI explanation alike — is framed around
    your 5-hour / weekly session limits instead of a bill. The Today / This Month /
    All Time cost cards also carry a subtle "≈ API-equivalent" caption.
• Explain with AI (opt-in) — a button on the Advisor panel that turns the
  computed signals into written, prioritized coaching. It calls Anthropic's Messages
  API with your own API key (stored in VS Code Secret Storage via the new Set
  Anthropic API Key command; never the Claude Code OAuth token). Only usage metadata
  is sent by default; prompt text is included only if you enable
  advisor.ai.includePrompts and confirm each time. The model is configurable with
  advisor.ai.model (default Sonnet). Needs a pay-as-you-go Anthropic API key (a
  Pro/Max subscription doesn't include API credit); the call surfaces Anthropic's
  actual error — e.g. a 400 "credit balance is too low" with a one-click link to
  billing — instead of a bare status code.

v1.1.1

Changed

• Clarified the new status-bar pace meter in the README and Marketplace overview:
  5h 42% → 53% is the projected end-of-window utilization at your current pace
  (here, on track to finish the 5-hour window at 53%). Spelled out in the hero
  caption and the Predictive alerts feature description so it reads clearly before
  install. No functional change.

v1.1.0

Added

• Predictive alerts — the extension now forecasts when you'll hit a plan limit
  and warns you before you do, so you don't get cut off mid-session. On by default;
  turn the whole feature off with predictiveAlerts.enabled. It runs entirely on
  the usage figures already shown and makes no network calls of its own.
  • Status-bar pace forecast, computed from your average consumption over the
    current window so far — so each window is judged on its own timescale (the 5-hour
    window over hours, the 7-day window over days) and a single percentage tick can't
    manufacture an alarming number. The 5-hour window shows a live pace meter: the
    projected end-of-window utilization at your current rate (5h 24% → 48%). The
    weekly window shows a time-to-limit ETA only when you're on a trustworthy
    track to breach it before reset. Toggle each with predictiveAlerts.showFiveHourEta
    and predictiveAlerts.showWeeklyEta.
  • Proactive warnings. A notification fires when a window crosses a configurable
    utilization threshold — 75% and 90% by default (predictiveAlerts.warnThresholds)
    — and, with predictiveAlerts.predictBreach on, when your current pace projects
    you'll hit a limit before it resets. Each warning fires once per window, re-arms
    when the window resets, and only fires while you're actually burning, so an idle
    window never nags.
  • Burn rate in the status-bar tooltip: tokens/min and cost/min over a
    configurable trailing window (predictiveAlerts.windowMinutes, default 15).
  • Model-cost advisor (predictiveAlerts.modelAdvisor.enabled, off by
    default): when recent turns show heavy spend on an expensive model for little
    output, a one-time, dismissible hint suggests a cheaper model for routine work.

v1.0.5

Added

• Pay-as-you-go extra usage monitoring (showExtraUsage, off by default).
  When your account has extra usage enabled, your spend beyond plan limits is
  shown as extra <spent> / <cap> in the status bar, with detail in the tooltip
  and an Extra usage section in the dashboard. Read from Anthropic's usage
  response (extra_usage: minor units + currency); nothing is shown when your
  account has extra usage disabled.