@hystudio/dsh-quota-pill 0.2.11 → 0.2.13

package/AGENTS.md

@@ -16,10 +16,10 @@ For a fresh DSH Desktop:
 
 1. If `dsh` is not on PATH: DSH Desktop → Settings → **Manage dsh Command**.
 2. Install a **pinned** build, never the bare name:
-   - GUI: **Plugins → Install** → `@hystudio/dsh-quota-pill@0.2.11`
-   - CLI: `dsh plugin add @hystudio/dsh-quota-pill@0.2.11 --profile desktop`
-   - Until 0.2.11 is on npm:
-     `github:598934321/dsh-quota-pill#feat/user-adapters-0.2.11`
+   - GUI: **Plugins → Install** → `@hystudio/dsh-quota-pill@0.2.12`
+   - CLI: `dsh plugin add @hystudio/dsh-quota-pill@0.2.12 --profile desktop`
+   - If pnpm's minimumReleaseAge rejects the fresh 0.2.12 release, use the
+     branch until it ages: `github:598934321/dsh-quota-pill#master`
 3. Confirm the dependency and the bundle are present:
 
    ```bash
@@ -34,7 +34,7 @@ For a fresh DSH Desktop:
 
 **Why pinned:** pnpm's `minimumReleaseAge` can resolve an unpinned install to the
 broken 0.2.2 (its client half fails to import and the web GUI shows "Failed to
-load plugins"). 0.2.11 is the first version with all providers, Open Design
+load plugins"). 0.2.12 is the first version with all providers, Open Design
 Control Keys, sidecar auto-start, and user-defined adapters.
 
 The host half starts the sidecar when nothing is listening on :8319, so a normal
@@ -77,7 +77,7 @@ DSH account menu, then re-run step 5.
 
 ## 1. Verify the sidecar runs
 
-The 0.2.11 host plugin starts it automatically when :8319 is free. Give DSH a few
+The 0.2.12 host plugin starts it automatically when :8319 is free. Give DSH a few
 seconds, then check. If the host half could not start it (or you opted out with
 `QUOTA_PILL_NO_AUTOSTART=1`), run it manually:
 
@@ -105,7 +105,7 @@ CLIProxyAPI". That is expected on a fresh machine — tell the human that
 `kimi` / `opencode` / `commandcode` work without it, and ask whether they want
 the OAuth three as well.
 
-**Auto-start is the default from 0.2.11**; do not add a service unless the human
+**Auto-start is the default from 0.2.12**; do not add a service unless the human
 asks for one or DSH is headless. If you do add one, set
 `QUOTA_PILL_NO_AUTOSTART=1` so the plugin and the service do not race, and use
 the **absolute** Node path (launchd's PATH does not include nvm/homebrew):

package/README.md

@@ -71,25 +71,26 @@ in [`src/user-adapters.js`](src/user-adapters.js) and in AGENTS.md §2b.
 
 **DSH Desktop (recommended).** Open **Plugins → Install**, paste a spec, enable
 the bundle. **Pin the version**: an unpinned install can resolve 0.2.2, which
-breaks web boot. 0.2.11 is the first version with every provider, Open Design
+breaks web boot. 0.2.12 is the first version with every provider, Open Design
 Control Keys, sidecar auto-start, and **user-defined adapters** for providers the
 package does not ship.
 
 ```text
-@hystudio/dsh-quota-pill@0.2.11
+@hystudio/dsh-quota-pill@0.2.12
 ```
 
-Until 0.2.11 is on npm, install the branch:
+0.2.12 is published on npm. If pnpm's minimumReleaseAge rejects this fresh
+release, install the branch until it ages:
 
 ```text
-github:598934321/dsh-quota-pill#feat/user-adapters-0.2.11
+github:598934321/dsh-quota-pill#master
 ```
 
 **CLI.** If `dsh` is not on PATH, open DSH Desktop → Settings → **Manage dsh
 Command** first. Then:
 
 ```bash
-dsh plugin add @hystudio/dsh-quota-pill@0.2.11
+dsh plugin add @hystudio/dsh-quota-pill@0.2.12
 ```
 
 > The npm package is `@hystudio/...` while the repo lives at `598934321/...` —
@@ -239,9 +240,15 @@ directly, current plans only reveal allowance in `x-ratelimit-*` response header
 of a one-token completion, and unified-billing accounts have no ceiling at all
 (reported as `payg`).
 
-**Antigravity** shares one quota pool across all Gemini models, so it reports a
-single aggregate row — per-model rows would all be the same number and the label
-would flicker between models you do not even have configured.
+**Antigravity** has three independent quota families, and the pill shows one row
+per family: Gemini models (5-hour + weekly), Gemini 4 Argon (its own weekly
+bucket) and Claude / GPT (5-hour + weekly). The backend decides which catalogue
+and which quota groups it reports from the client label in the `User-Agent`: a
+generic UA gets a legacy answer where every Gemini model is a constant 100% and
+`gemini-4-argon-*` does not exist at all. The adapter therefore sends the
+Antigravity CLI's own UA (`antigravity/cli/<version> (aidev_client; …)`), reads
+`retrieveUserQuotaSummary` and falls back to per-model `fetchAvailableModels`.
+Set `QUOTA_PILL_AGY_UA` to override the UA if a future CLI changes it.
 
 **Command Code** does not expose a monthly ceiling. The adapter derives it from
 `monthlyCap = weekly.cap × 2`, an identity verified against two different plans
@@ -263,6 +270,11 @@ then writes `OPENDESIGN_CONTROL_KEY` and `OPENDESIGN_WORKSPACE_ID` into
 `~/.dsh/.credentials.yaml`. The workspace id is optional when using a Control
 Key — it is otherwise resolved from `/workspaces/current`.
 
+Windows with a real reset period (5-hour / weekly / monthly) always render as a
+percentage — including the 100% seen right after a reset — so a bucket never
+switches units as it is spent. The credit counts stay in the row's tooltip;
+only credit packs without a period render as credits.
+
 The older console-session route (`__Secure-better-auth.session_token` cookie,
 browser-session-only `/billing/coding-plan-usage`) is still supported as a
 fallback; that cookie is HttpOnly, so copy it from devtools → Application →

package/client.js

@@ -172,7 +172,7 @@ window.__ModuleLoader__.load({
      */
     function adaptationPrompt(model, provider, lang) {
       const m = String(model || '(unknown)');
-      const p = String(provider && provider !== 'unknown' ? provider : '(unknown)');
+      const p = String(provider && provider !== 'unknown' ? provider : 'unknown');
       if (lang === 'zh') {
         return [
           '你是 DSH quota-pill 的 provider 适配 agent。当前模型 ' + m + ' 的 provider（' + p + '）未被插件识别，读取不到余额。',
@@ -430,6 +430,12 @@ window.__ModuleLoader__.load({
       '7d': ['每周', 'Weekly'],
       monthly: ['每月', 'Monthly'],
       'model-pool': ['模型池', 'Model pool'],
+      // Antigravity 的两个独立额度族（Gemini / Claude·Opus）。必须登记成已知
+      // 窗口：否则 providerEntries 会认为它们不是额度周期，再补一行汇总，
+      // 同一份额度显示两次。
+      'agy-gemini': ['Gemini', 'Gemini'],
+      'agy-argon': ['Gemini 4', 'Gemini 4'],
+      'agy-opus': ['Claude / Opus', 'Claude / Opus'],
       // 积分类的固定三行。放进这里是为了让 isKnownWindow 认得它们——
       // 否则 providerEntries 会再前置一条汇总行，变成四行。
       'credits-all': ['全部积分', 'All credits'],
@@ -535,6 +541,21 @@ window.__ModuleLoader__.load({
       return !!WINDOW_LABEL[String(key).replace(/^USAGE_PERIOD_TYPE_/i, '').toLowerCase()];
     }
 
+    /**
+     * 「周期性额度窗口」= 有真实重置周期的桶（5 小时 / 每周 / 每月）。
+     *
+     * 这类窗口的百分比永远有信息量：刚重置就是 100%，用掉就是 94%，两种状态
+     * 都值得看。它们不能走 pctIsInformative 的「满格就显示积分」规则——那条
+     * 规则是为「没有周期的积分包」写的，用在 opendesign 的 5h 桶上会让它在
+     * 刚重置时显示「50k 积分」、用掉之后又变回百分比，同一个窗口两种单位。
+     */
+    const PERIODIC_WINDOWS = new Set(['5h', 'fivehour', 'rolling', 'weekly', '7d', 'monthly']);
+
+    function isPeriodicWindow(key) {
+      const k = String(key || '').replace(/^USAGE_PERIOD_TYPE_/i, '').toLowerCase();
+      return PERIODIC_WINDOWS.has(k) || /^weekly-\d+$/.test(k);
+    }
+
     /**
      * 免费渠道池的两种汇总口径。它们不是「额度周期」，所以不加「额度剩余」后缀，
      * 否则会读成「最紧窗口额度剩余」这种别扭的说法。
@@ -641,6 +662,8 @@ window.__ModuleLoader__.load({
           pct: Number.isFinite(pct) ? pct : null,
           reset: (w.credits && w.resetAt) ? fmtExpiry(w.resetAt, lang) : fmtResetAt(w.resetAt, lang),
           credits: w.credits || null,
+          // 有重置周期的窗口一律优先显示百分比（见 isPeriodicWindow）。
+          periodic: isPeriodicWindow(w.key),
         });
       }
       // 积分类用的是固定三行，三个键都算「已知窗口」，所以不会再前置汇总行——
@@ -1385,7 +1408,7 @@ window.__ModuleLoader__.load({
     }
 
     function WindowRow(props) {
-      const { label, pct, reset, detail, credits, lang, stale, money: isMoney, compact } = props;
+      const { label, pct, reset, detail, credits, lang, stale, money: isMoney, compact, periodic } = props;
       // `detail` is a generic secondary line (a payg bonus, an aggregate hint);
       // `reset` is the window-reset countdown. Rows in a normal group reserve
       // the line even when empty so their numbers stay aligned inside the group;
@@ -1402,7 +1425,8 @@ window.__ModuleLoader__.load({
       //   · 百分比恒 100%   → 显示积分（这种渠道百分比什么都说明不了）
       // 判据见 pctIsInformative；两种方式都保留，只是不同渠道各取其一。
       const hasCredits = credits && typeof credits.remain === 'number';
-      const showPct = isNum && pctIsInformative(pct, credits);
+      // 周期窗口：百分比永远是主读数；积分只在 tooltip 里作为补充。
+      const showPct = isNum && (periodic || pctIsInformative(pct, credits));
       const showCredits = hasCredits && !showPct;
       // data-stale：这份读数是「该 provider 上次的值」，正在后台刷新。
       // 视觉上降饱和 + 缓慢呼吸，数据到了再撤掉（见 CSS）。
@@ -1458,7 +1482,7 @@ window.__ModuleLoader__.load({
         h('div', { className: 'quota-pill-group-head' }, name),
         // 组内行 key 带组名，避免不同 provider 的同名窗口（都是 5h/weekly）
         // 被 React 当成同一个组件实例，导致 tween 跨 provider 插值。
-        entries.map((e) => h(WindowRow, { key: `${name}|${e.key}`, label: e.label, pct: e.pct, reset: e.reset, detail: e.detail, compact: e.compact, credits: e.credits, lang, stale, money: e.money })),
+        entries.map((e) => h(WindowRow, { key: `${name}|${e.key}`, label: e.label, pct: e.pct, reset: e.reset, detail: e.detail, compact: e.compact, credits: e.credits, periodic: e.periodic, lang, stale, money: e.money })),
       );
     }
 
@@ -1533,7 +1557,7 @@ window.__ModuleLoader__.load({
             // key 带 provider：换 provider 时强制重挂载，让 tween 从「空」开始，
             // 而不是把 A 的 5h 平滑滑到 B 的 5h——那是两个不同账户的额度，
             // 中间的数字不代表任何真实状态。
-            entries.map((e) => h(WindowRow, { key: `${providerName}|${e.key}`, label: e.label, pct: e.pct, reset: e.reset, credits: e.credits, lang, stale, money: e.money })),
+            entries.map((e) => h(WindowRow, { key: `${providerName}|${e.key}`, label: e.label, pct: e.pct, reset: e.reset, credits: e.credits, periodic: e.periodic, lang, stale, money: e.money })),
           ),
           h('dl', { className: 'quota-pill-stats' },
             data.account ? statRow(pick(lang, 'account'), String(data.account)) : null,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hystudio/dsh-quota-pill",
-  "version": "0.2.11",
+  "version": "0.2.13",
   "private": false,
   "type": "module",
   "description": "Quota pill for DeepSeek Harness: remaining quota for every LLM subscription you have, in one collapsible pill",

package/src/adapters-cliproxy.js

@@ -422,67 +422,210 @@ export async function codex() {
 
 /* --------------------------------------------------------------------- agy */
 /**
- * Antigravity (Google Cloud Code) — all Gemini models share ONE pool.
+ * Antigravity (Google Cloud Code) reports THREE independent quota families,
+ * each with its own limits:
  *
- * That is why this adapter deliberately reports no per-model rows: all 22
- * remainingFraction values are identical, so picking "the tightest" would just
- * be "first in dict order" and the label would flicker between model names that
- * are not even what the user has configured. The pool total is real, so only
- * the aggregate is reported.
+ *   - Gemini models   - 5-hour + weekly buckets  ("Gemini Models" group)
+ *   - Gemini 4 Argon  - its own weekly bucket    ("Gemini 4 Argon" group)
+ *   - Claude / GPT    - 5-hour + weekly buckets  ("Claude and GPT models")
+ *
+ * The backend decides which catalogue AND which quota groups it reports from
+ * the client label in the User-Agent. With a generic UA ("antigravity/1.0") it
+ * serves a legacy catalogue: no gemini-4-argon-*, and every Gemini model comes
+ * back with remainingFraction 1 and a resetTime that merely tracks "now + 5h"
+ * - a permanent 100% that hides real usage. The Antigravity CLI's own UA
+ * ("antigravity/cli/<version> (aidev_client; ...)") gets the real catalogue and
+ * the real groups. Verified against the CLI's captured traffic on 2026-10-26
+ * (same token: generic UA -> 23 models, no argon, no Gemini group; CLI UA ->
+ * 29 models, argon at ~91%, three groups).
+ *
+ * Preferred source is retrieveUserQuotaSummary: one request, one group per
+ * family, each with its buckets. Per-model fetchAvailableModels stays as a
+ * fallback if that RPC changes shape.
+ */
+const AGY_CLI_UA = process.env.QUOTA_PILL_AGY_UA
+  || 'antigravity/cli/1.4.7 (aidev_client; os_type=darwin; arch=arm64; cl=1006008713; auth_method=consumer)';
+const AGY_PROJECT = { project: 'aicode-consumers' };
+
+/**
+ * One entry per quota family the pill shows. The key is the window key the
+ * client renders; keep it stable - the label table and the spend history are
+ * keyed by it.
+ */
+const AGY_FAMILIES = [
+  { key: 'agy-gemini', label: 'Gemini' },
+  { key: 'agy-argon', label: 'Gemini 4' },
+  { key: 'agy-opus', label: 'Claude / Opus' },
+];
+
+/** Family a model id belongs to (decides which number is the headline one). */
+function agyFamilyForModel(model) {
+  // Model ids arrive bare from the all page and as provider/model on /quota.
+  const id = String(model || '').split('/').pop();
+  if (/argon/i.test(id)) return 'agy-argon';
+  if (/^(claude|gpt)/i.test(id)) return 'agy-opus';
+  if (/gemini/i.test(id)) return 'agy-gemini';
+  return null;
+}
+
+/** Family a quota-summary group belongs to. */
+function agyFamilyForGroup(name) {
+  const n = String(name || '');
+  if (/argon/i.test(n)) return 'agy-argon';
+  if (/claude|gpt/i.test(n)) return 'agy-opus';
+  if (/gemini/i.test(n)) return 'agy-gemini';
+  return null;
+}
+
+/**
+ * Collapse quota-summary groups into one reading per family. The tightest
+ * bucket decides the percentage (a 5-hour bucket at 0 blocks even when the
+ * weekly one still has 46%), and that bucket's reset is what the user can act
+ * on.
+ */
+function agyFamiliesFromSummary(data) {
+  const out = new Map();
+  for (const group of (data && data.groups) || []) {
+    const key = agyFamilyForGroup(group && group.displayName);
+    if (!key) continue;
+    const fam = AGY_FAMILIES.find((f) => f.key === key);
+    let frac = null;
+    let resetAt = null;
+    const buckets = [];
+    for (const b of (group && group.buckets) || []) {
+      const v = b && b.remainingFraction;
+      if (typeof v !== 'number') continue;
+      buckets.push({ id: b.bucketId || null, window: b.window || null, frac: v, resetAt: b.resetTime || null });
+      if (frac === null || v < frac) {
+        frac = v;
+        resetAt = b.resetTime || resetAt;
+      }
+    }
+    if (frac === null) continue;
+    out.set(key, { key, label: fam.label, frac, resetAt, buckets });
+  }
+  return out;
+}
+
+/** Fallback: collapse per-model buckets into one reading per family. */
+function agyFamiliesFromModels(models) {
+  const out = new Map();
+  for (const [mid, info] of Object.entries(models || {})) {
+    const key = agyFamilyForModel(mid);
+    if (!key) continue;
+    const q = (info && info.quotaInfo) || {};
+    const frac = q.remainingFraction;
+    if (typeof frac !== 'number') continue;
+    const fam = AGY_FAMILIES.find((f) => f.key === key);
+    const cur = out.get(key) || { key, label: fam.label, frac: 1, resetAt: null, buckets: [] };
+    cur.buckets.push({ id: mid, window: null, frac, resetAt: q.resetTime || null });
+    if (frac < cur.frac) {
+      cur.frac = frac;
+      cur.resetAt = q.resetTime || cur.resetAt;
+    } else if (!cur.resetAt && q.resetTime) {
+      cur.resetAt = q.resetTime;
+    }
+    out.set(key, cur);
+  }
+  return out;
+}
+
+/** Shape one adapter payload from the families that were actually reported. */
+function agyPayload(cred, fams, model, source) {
+  const wanted = agyFamilyForModel(model);
+  // Headline: the family the current model uses. Without a model (the all
+  // page) Gemini leads, which is also the number quota failover tools compare
+  // against - do not change that silently.
+  const primary = (wanted && fams.get(wanted))
+    || fams.get('agy-gemini')
+    || fams.values().next().value;
+  const remaining = round2(primary.frac * 100);
+  const windows = AGY_FAMILIES
+    .filter((f) => fams.has(f.key))
+    .map((f) => {
+      const v = fams.get(f.key);
+      return { key: v.key, remainPct: round2(v.frac * 100), resetAt: v.resetAt || null };
+    });
+  return {
+    kind: 'subscription-quota',
+    provider: 'agy',
+    account: cred.account || cred.name,
+    remainingPct: remaining,
+    usedPct: round2(100 - remaining),
+    window: primary.key,
+    resetAt: primary.resetAt || null,
+    windows,
+    raw: {
+      source,
+      families: Object.fromEntries([...fams].map(([k, v]) => [k, {
+        remainPct: round2(v.frac * 100),
+        resetAt: v.resetAt || null,
+        buckets: v.buckets,
+      }])),
+    },
+  };
+}
+
+/**
+ * Read one Antigravity login. Both endpoints need the CLI User-Agent - see the
+ * comment above: without it the backend answers from a legacy catalogue whose
+ * Gemini quota is a permanent 100%.
  */
-export async function agy() {
+async function agyReading(cred, model, errors) {
+  const post = (url) => curlJson(url, {
+    method: 'POST',
+    // googleapis is unreachable directly on some networks; CLIProxyAPI users
+    // almost always have a local proxy configured.
+    proxy: process.env.QUOTA_PILL_PROXY || defaultProxy(),
+    token: cred.token,
+    body: AGY_PROJECT,
+    headers: { 'Content-Type': 'application/json', 'User-Agent': AGY_CLI_UA },
+    timeoutMs: 25000,
+  });
+
+  const summary = await post('https://cloudcode-pa.googleapis.com/v1internal:retrieveUserQuotaSummary');
+  if (summary && summary.ok) {
+    const fams = agyFamiliesFromSummary(summary.data);
+    if (fams.size) return agyPayload(cred, fams, model, 'v1internal:retrieveUserQuotaSummary');
+    errors.push(cred.name + ': retrieveUserQuotaSummary had no known groups');
+  } else {
+    errors.push(cred.name + ': retrieveUserQuotaSummary ' + ((summary && summary.error) || 'request failed'));
+  }
+
+  const available = await post('https://cloudcode-pa.googleapis.com/v1internal:fetchAvailableModels');
+  if (!available || !available.ok) {
+    errors.push(cred.name + ': ' + ((available && available.error) || 'request failed'));
+    return null;
+  }
+  const fams = agyFamiliesFromModels((available.data && available.data.models) || {});
+  if (!fams.size) {
+    errors.push(cred.name + ': fetchAvailableModels had no quotaInfo');
+    return null;
+  }
+  return agyPayload(cred, fams, model, 'v1internal:fetchAvailableModels');
+}
+
+export async function agy(model) {
   const usable = usableFiles('antigravity');
   if (!usable.length) return noCredential('agy', 'antigravity');
 
   const errors = [];
   let best = null;
-
   for (const cred of usable) {
-    const res = await curlJson(
-      'https://cloudcode-pa.googleapis.com/v1internal:fetchAvailableModels',
-      {
-        method: 'POST',
-        // googleapis is unreachable directly on some networks; CLIProxyAPI
-        // users almost always have a local proxy configured.
-        proxy: process.env.QUOTA_PILL_PROXY || defaultProxy(),
-        token: cred.token,
-        body: {},
-        headers: { 'Content-Type': 'application/json', 'User-Agent': 'antigravity/1.0' },
-        timeoutMs: 25000,
-      },
-    );
-    if (!res || !res.ok) {
-      errors.push(`${cred.name}: ${(res && res.error) || 'request failed'}`);
-      continue;
-    }
-    const models = (res.data && res.data.models) || {};
-    const fractions = [];
-    for (const [mid, info] of Object.entries(models)) {
-      const q = (info && info.quotaInfo) || {};
-      const frac = q.remainingFraction;
-      if (typeof frac === 'number') fractions.push([mid, frac, q.resetTime]);
-    }
-    if (!fractions.length) {
-      errors.push(cred.name + ': no quotaInfo');
-      continue;
-    }
-    fractions.sort((a, b) => a[1] - b[1]);
-    const [sampleModel, frac, reset] = fractions[0];
-    const remaining = round2(frac * 100);
-    const candidate = {
-      kind: 'subscription-quota',
-      provider: 'agy',
-      account: cred.account || cred.name,
-      remainingPct: remaining,
-      usedPct: round2(100 - remaining),
-      window: 'model-pool',
-      resetAt: reset || null,
-      raw: { accounts: usable.length, poolModels: fractions.length, sampleModel },
-    };
-    if (!best || candidate.remainingPct > best.remainingPct) best = candidate;
+    const reading = await agyReading(cred, model, errors);
+    if (!reading) continue;
+    // Several antigravity logins: report the one with the most headroom in its
+    // tightest family, so a half-spent account cannot mask a full one.
+    const tightest = Math.min(...reading.windows.map((w) => w.remainPct));
+    if (!best || tightest > best.tightest) best = { reading, tightest };
   }
 
-  if (best) return best;
+  if (best) {
+    if (usable.length > 1) {
+      best.reading.raw = { ...best.reading.raw, accounts: usable.length };
+    }
+    return best.reading;
+  }
   return { kind: 'probe-failed', provider: 'agy', error: errors.join('; ') || 'no quota' };
 }