@cortexkit/opencode-antigravity-auth 2.2.0 → 3.0.0

package/dist/src/plugin/quota.js

@@ -14,9 +14,10 @@
  * with `force: true` — manual quota screens must always refresh, even if
  * the background manager has backed off.
  */
-import { aggregateGeminiCliQuota, aggregateQuota, aggregateQuotaSummary, createQuotaManager, defaultKeyOf, fetchAvailableModels, fetchGeminiCliQuota, fetchQuotaSummary, getHealthTracker, } from '@cortexkit/antigravity-auth-core';
+import { aggregateGeminiCliQuota, aggregateQuota, aggregateQuotaSummary, createQuotaManager, defaultKeyOf, fetchAvailableModels, fetchGeminiCliQuota, fetchQuotaSummary, fetchWithActiveTimeout, getHealthTracker, rowRefKey, sameRowRef, VAULT_ACCOUNT_REMEDY, } from '@cortexkit/antigravity-auth-core';
 import { ANTIGRAVITY_ENDPOINT_FALLBACKS, ANTIGRAVITY_PROVIDER_ID, buildGeminiCliUserAgent, } from '../constants';
 import { buildSidebarMachineStateFromAccounts, isAccountCurrent, setSidebarMachineState, } from '../sidebar-state';
+import { createLocalAccountCredentials, } from './accounts';
 import { accessTokenExpired, formatRefreshParts, parseRefreshParts, } from './auth';
 import { logQuotaFetch, logQuotaStatus } from './debug';
 import { buildAntigravityHarnessUserAgent } from './fingerprint';
@@ -41,7 +42,20 @@ export { classifyQuotaGroup, createQuotaManager, defaultKeyOf, } from '@cortexki
  * no-op.
  */
 export function createOpenCodeQuotaManager(client, providerId = ANTIGRAVITY_PROVIDER_ID, options = {}) {
-    const fetchAccountQuota = makeFetchAccountQuota(client, providerId, options.fetchVia);
+    const fetchPoolFileQuota = makeFetchAccountQuota(client, providerId, options.fetchVia);
+    const fetchAccountQuota = async (account, request) => {
+        const target = account;
+        const ref = target.rowRef;
+        if (options.mayRun && !(await options.mayRun()))
+            throw new Error(VAULT_ACCOUNT_REMEDY);
+        if (ref !== undefined) {
+            if (!options.fetchStoreAccountQuota)
+                throw new Error('Store quota authorization is unavailable');
+            const captured = { ...target, rowRef: ref };
+            return options.fetchStoreAccountQuota(captured, request);
+        }
+        return fetchPoolFileQuota(account, request);
+    };
     const manager = createQuotaManager({
         fetchAccountQuota,
         keyOf: options.keyOf ?? defaultKeyOf,
@@ -49,16 +63,59 @@ export function createOpenCodeQuotaManager(client, providerId = ANTIGRAVITY_PROV
         maxBackoffMs: options.maxBackoffMs,
         fetchTimeoutMs: options.fetchTimeoutMs,
     });
-    const originalRefreshAccount = manager.refreshAccount;
-    const originalRefreshAccounts = manager.refreshAccounts;
     const getAccountsForSidebar = options.getAccountsForSidebar;
     const getActiveIndexByFamily = options.getActiveIndexByFamily;
+    return withSidebarPushAfterRefresh(manager, options.publishSidebar ??
+        (getAccountsForSidebar
+            ? (account) => pushSidebarQuotaSnapshot(getAccountsForSidebar, manager.getBackoffUntil(account), getActiveIndexByFamily)
+            : undefined));
+}
+/**
+ * Build one location's quota manager. Backoff, in-flight dedupe and the
+ * per-account cache belong to the returned manager alone, so a second
+ * location never shares another location's backoff or cached results.
+ * Refreshes push a sidebar snapshot exactly like `createOpenCodeQuotaManager`
+ * does, but to the location's own state file and through its own logger.
+ */
+export function createLocationQuotaManager(options) {
+    if (typeof options?.logger?.debug !== 'function') {
+        throw new TypeError('createLocationQuotaManager requires a location logger');
+    }
+    if (typeof options.fetchAccountQuota !== 'function') {
+        throw new TypeError('createLocationQuotaManager requires a fetchAccountQuota function');
+    }
+    const manager = createQuotaManager({
+        fetchAccountQuota: options.fetchAccountQuota,
+        keyOf: options.keyOf ?? defaultKeyOf,
+        baseBackoffMs: options.baseBackoffMs,
+        maxBackoffMs: options.maxBackoffMs,
+        fetchTimeoutMs: options.fetchTimeoutMs,
+    });
+    const sidebar = options.sidebar;
+    const logger = options.logger;
+    return withSidebarPushAfterRefresh(manager, sidebar
+        ? (account) => pushSidebarQuotaSnapshot(sidebar.getAccounts, manager.getBackoffUntil(account), sidebar.getActiveIndexByFamily, {
+            write: sidebar.write,
+            logger,
+            healthScore: sidebar.healthScore ?? null,
+            now: sidebar.now,
+        })
+        : undefined);
+}
+/**
+ * Wrap a quota manager so every refresh started before disposal pushes one
+ * sidebar snapshot after it settles, and so `dispose()` waits for those
+ * pushes. A refresh that starts after disposal pushes nothing.
+ */
+function withSidebarPushAfterRefresh(manager, push) {
+    const originalRefreshAccount = manager.refreshAccount;
+    const originalRefreshAccounts = manager.refreshAccounts;
     let disposed = false;
     const inFlight = new Set();
     const pushAfterRefresh = async (account) => {
-        if (!getAccountsForSidebar)
+        if (!push)
             return;
-        await pushSidebarQuotaSnapshot(getAccountsForSidebar, manager.getBackoffUntil(account), getActiveIndexByFamily).catch(() => {
+        await push(account).catch(() => {
             // Sidebar persistence remains best-effort when lock contention
             // outlives its retry budget.
         });
@@ -157,13 +214,22 @@ export async function checkAccountsQuota(accounts, client, providerId = ANTIGRAV
  * On lock contention the error is logged-and-swallowed so a quota dialog
  * never fails just because the sidebar file is busy.
  */
-export async function pushSidebarQuotaSnapshot(getAccounts, backoffUntil = 0, getActiveIndexByFamily) {
+export async function pushSidebarQuotaSnapshot(getAccounts, backoffUntil = 0, getActiveIndexByFamily, options = {}) {
     const accounts = getAccounts();
     if (!accounts || accounts.length === 0)
         return;
     const activeByFamily = getActiveIndexByFamily?.() ?? null;
+    // `undefined` keeps the process-wide health tracker the OpenCode 1 plugin
+    // uses; `null` means the caller has no health source and the sidebar shows
+    // its default.
+    const healthScore = options.healthScore === undefined
+        ? (index) => getHealthTracker().getScore(index)
+        : options.healthScore;
+    const now = options.now ?? Date.now;
+    const write = options.write ??
+        ((state) => setSidebarMachineState(state, options.stateFile === undefined ? {} : { stateFile: options.stateFile }));
     try {
-        await setSidebarMachineState(buildSidebarMachineStateFromAccounts(accounts.map((entry) => ({
+        await write(buildSidebarMachineStateFromAccounts(accounts.map((entry) => ({
             index: entry.index,
             label: entry.label,
             enabled: entry.enabled,
@@ -172,17 +238,21 @@ export async function pushSidebarQuotaSnapshot(getAccounts, backoffUntil = 0, ge
                 : false,
             coolingDownUntil: entry.coolingDownUntil,
             cachedQuota: entry.cachedQuota,
+            cachedQuotaUpdatedAt: entry.cachedQuotaUpdatedAt,
             cachedQuotaAccountId: entry.cachedQuotaAccountId,
             currentQuotaAccountId: entry.currentQuotaAccountId,
-            healthScore: getHealthTracker().getScore(entry.index),
+            ...(healthScore ? { healthScore: healthScore(entry.index) } : {}),
             tier: entry.tier,
         })), {
-            checkedAt: Date.now(),
+            checkedAt: now(),
             quotaBackoffUntil: backoffUntil > 0 ? backoffUntil : undefined,
         }));
     }
     catch (error) {
-        log.debug('sidebar-quota-write-failed', { error: String(error) });
+        ;
+        (options.logger ?? log).debug('sidebar-quota-write-failed', {
+            error: String(error),
+        });
     }
 }
 /**
@@ -218,6 +288,428 @@ async function fetchLegacyModelsFallback(options) {
         };
     }
 }
+/**
+ * Fetches both quota payloads for one account with a bearer and project the
+ * caller already resolved: the windowed quota summary (with the legacy
+ * model-list fallback) and the Gemini CLI quota. Shared by the OpenCode 1
+ * fetcher and `createAuthorizedFetchAccountQuota`.
+ */
+async function fetchQuotaPayloads(options) {
+    const { accessToken, managedProjectId, projectId, fetchVia } = options;
+    // Two independent payload contracts: the windowed summary
+    // (with legacy fallback) and the gemini-CLI quota. They share
+    // access + project but target different endpoints, so the
+    // two 10s timeouts ran back-to-back for ~20s per account on
+    // modal open. Run them concurrently; either rejection is
+    // handled by its own branch and the result is still merged.
+    const fetchSummaryPayload = (async () => {
+        try {
+            const summaryResult = await fetchQuotaSummary({
+                accessToken,
+                managedProjectId,
+                projectId,
+                endpoints: ANTIGRAVITY_ENDPOINT_FALLBACKS,
+                userAgent: buildAntigravityHarnessUserAgent(),
+                timeoutMs: 10_000,
+                ...(fetchVia ? { fetchVia } : {}),
+            });
+            return {
+                result: aggregateQuotaSummary(summaryResult.summary),
+                fellBackToLegacy: summaryResult.fellBackToLegacy ?? false,
+            };
+        }
+        catch {
+            return {
+                result: await fetchLegacyModelsFallback({
+                    accessToken,
+                    projectId,
+                    fetchVia,
+                }),
+                fellBackToLegacy: true,
+            };
+        }
+    })();
+    // CLI fetch is independent of the summary fetch. A CLI failure must
+    // NOT kill the summary result, but it also must not be laundered into
+    // "No Gemini CLI quota available" (a permanent-looking status) when
+    // the real cause is a transient network error. Capture the error
+    // message separately so the annotated result can carry it.
+    let geminiCliFetchError;
+    const fetchGeminiCliPayload = fetchGeminiCliQuota({
+        accessToken,
+        projectId,
+        endpoints: ANTIGRAVITY_ENDPOINT_FALLBACKS,
+        userAgent: buildGeminiCliUserAgent(),
+        timeoutMs: 10_000,
+        ...(fetchVia ? { fetchVia } : {}),
+    }).catch((error) => {
+        geminiCliFetchError = error instanceof Error ? error.message : String(error);
+        options.logger.debug('fetchGeminiCliQuota failed', {
+            error: geminiCliFetchError,
+        });
+        return { buckets: undefined };
+    });
+    const [summary, geminiCliResponse] = await Promise.all([
+        fetchSummaryPayload,
+        fetchGeminiCliPayload,
+    ]);
+    return {
+        quota: summary.result,
+        geminiCliQuota: annotateGeminiCliQuota(geminiCliResponse, geminiCliFetchError),
+        fellBackToLegacy: summary.fellBackToLegacy,
+    };
+}
+/**
+ * The Gemini CLI quota summary for a `retrieveUserQuota` answer. An empty or
+ * missing answer carries an error: the fetch error when the request failed
+ * (a transient failure, not "no CLI quota configured"), else the generic
+ * "no quota" text.
+ */
+function annotateGeminiCliQuota(response, fetchError) {
+    const result = aggregateGeminiCliQuota(response);
+    if (response.buckets !== undefined && response.buckets.length > 0) {
+        return result;
+    }
+    return {
+        ...result,
+        error: fetchError ??
+            (result.models.length === 0
+                ? 'No Gemini CLI quota available'
+                : undefined),
+    };
+}
+/** Stops an outbound quota request after cancellation or a vault authorization refusal. */
+class VaultQuotaCheckStoppedError extends Error {
+    name = 'VaultQuotaCheckStoppedError';
+    constructor() {
+        super('the quota check was stopped before this request was sent');
+    }
+}
+/** Refuses a second request using the same vault authorization. */
+class VaultAdmissionReusedError extends Error {
+    name = 'VaultAdmissionReusedError';
+    constructor() {
+        super('a vault admission authorizes exactly one physical request');
+    }
+}
+/**
+ * Makes exactly one physical quota request for `ref` under a new served
+ * receipt from `source.send`. Inside the dispatch, `build` runs one of the
+ * core quota builders with that receipt's access token and project, so the
+ * Authorization header and the request project both come from the same
+ * receipt and are built after it was served.
+ *
+ * The builder reads a clone of the response; the original Response goes back
+ * to `send`. `send` needs the original to read its status: it reports a 401
+ * against that receipt's record version and may retry once, under a new
+ * receipt of the same credential and account, when the vault now serves a
+ * different record version (not necessarily a higher one; a restore can
+ * serve an older one). The retried request is built from the retry's
+ * receipt. A builder that tries a second request under one receipt is
+ * stopped before that request is sent.
+ */
+async function vaultQuotaAttempt(options, 
+/**
+ * True once the whole check must send nothing more (a receipt was refused
+ * elsewhere in the check, or it was aborted). Checked again right before
+ * the physical request, because serving this receipt may have taken a
+ * while.
+ */
+isStopped, build) {
+    const transport = options.transport ??
+        ((url, init, extra) => fetchWithActiveTimeout(url, { ...init, signal: extra.signal ?? options.signal }, { timeoutMs: extra.timeoutMs }));
+    let attempt = {};
+    let dispatched = false;
+    try {
+        const response = await options.source.send(options.ref, async (admission, signal) => {
+            dispatched = true;
+            let used = false;
+            let served;
+            const fetchVia = async (url, init, extra) => {
+                // Marked before the request goes out, so a failed or still
+                // pending first request cannot let a second one use this receipt.
+                if (used)
+                    throw new VaultAdmissionReusedError();
+                used = true;
+                if (isStopped())
+                    throw new VaultQuotaCheckStoppedError();
+                served = await transport(url, init, {
+                    ...extra,
+                    signal: signal ?? options.signal,
+                });
+                // The builder reads a clone; `send` gets the original Response,
+                // whose status it needs for 401 reporting.
+                return served.clone();
+            };
+            attempt = {};
+            try {
+                attempt.value = await build({
+                    accessToken: admission.accessToken,
+                    projectId: admission.projectId,
+                }, fetchVia);
+            }
+            catch (error) {
+                attempt.error = error;
+            }
+            if (served === undefined)
+                throw attempt.error ?? new Error('no request');
+            attempt.status = served.status;
+            if (attempt.value !== undefined) {
+                attempt.attribution = options.source.attribution(admission);
+            }
+            return served;
+        }, { site: 'quota', signal: options.signal });
+        await response.body?.cancel().catch(() => { });
+    }
+    catch (error) {
+        if (!dispatched)
+            return { error, refused: true };
+        attempt.error ??= error;
+    }
+    return attempt;
+}
+/**
+ * Fetches a vault account's quota: the windowed quota summary (with the
+ * legacy model-list fallback) and the Gemini CLI quota, from the same core
+ * builders and endpoint list the local check uses.
+ *
+ * Every physical request (each endpoint, each fallback, and the model
+ * list's retry without a project after a 403) is one `source.send` with a
+ * new served receipt, and that request's bearer and project come from that
+ * receipt alone. Nothing is authorized once and reused, and no local refresh
+ * token, token exchange, project cache or earlier grant is used. A refused
+ * receipt or an abort stops the check with no further request. Recording
+ * the reading (with the returned attributions and the source's
+ * `commitState`) is the caller's.
+ */
+export async function fetchVaultAccountQuota(options) {
+    // Shared by the summary and Gemini CLI halves, which run concurrently: a
+    // refusal in one stops requests in the other too, including one whose
+    // receipt was being served when the refusal happened.
+    const stopped = { value: false };
+    const isStopped = () => stopped.value || options.signal.aborted;
+    const attempt = async (build) => {
+        if (isStopped()) {
+            return { error: new VaultQuotaCheckStoppedError(), refused: true };
+        }
+        const result = await vaultQuotaAttempt(options, isStopped, build);
+        if (result.refused || options.signal.aborted)
+            stopped.value = true;
+        return result;
+    };
+    const fetchSummary = async () => {
+        for (const endpoint of ANTIGRAVITY_ENDPOINT_FALLBACKS) {
+            // An answer that cannot be aggregated counts as a failed request.
+            const result = await attempt(async (grant, fetchVia) => aggregateQuotaSummary((await fetchQuotaSummary({
+                accessToken: grant.accessToken,
+                projectId: grant.projectId,
+                endpoints: [endpoint],
+                userAgent: buildAntigravityHarnessUserAgent(),
+                timeoutMs: 10_000,
+                fetchVia,
+            })).summary));
+            if (result.value !== undefined) {
+                return {
+                    result: result.value,
+                    fellBackToLegacy: false,
+                    ...(result.attribution ? { attribution: result.attribution } : {}),
+                };
+            }
+            if (result.refused)
+                break;
+        }
+        // Fetch the legacy model list. After a 403, retry once without a project
+        // as a separate request with a newly authorized vault credential, rather
+        // than reusing the authorization from the rejected request.
+        for (const endpoint of ANTIGRAVITY_ENDPOINT_FALLBACKS) {
+            for (const withProject of [true, false]) {
+                const result = await attempt((grant, fetchVia) => fetchAvailableModels({
+                    accessToken: grant.accessToken,
+                    projectId: withProject ? grant.projectId : '',
+                    endpoints: [endpoint],
+                    userAgent: buildAntigravityHarnessUserAgent(),
+                    timeoutMs: 10_000,
+                    fetchVia,
+                }));
+                if (result.value !== undefined) {
+                    return result.value.models
+                        ? {
+                            result: aggregateQuota(result.value.models),
+                            fellBackToLegacy: true,
+                            ...(result.attribution
+                                ? { attribution: result.attribution }
+                                : {}),
+                        }
+                        : {
+                            result: {
+                                groups: {},
+                                modelCount: 0,
+                                error: 'Failed to fetch Antigravity quota (legacy fallback)',
+                            },
+                            fellBackToLegacy: true,
+                        };
+                }
+                if (result.refused) {
+                    return {
+                        result: {
+                            groups: {},
+                            modelCount: 0,
+                            error: 'Failed to fetch Antigravity quota',
+                        },
+                        fellBackToLegacy: true,
+                    };
+                }
+                // Only a 403 for the request with a project is retried, without one.
+                if (!withProject || result.status !== 403)
+                    break;
+            }
+        }
+        return {
+            result: {
+                groups: {},
+                modelCount: 0,
+                error: 'Failed to fetch Antigravity quota',
+            },
+            fellBackToLegacy: true,
+        };
+    };
+    const fetchGeminiCli = async () => {
+        const errors = [];
+        for (const endpoint of ANTIGRAVITY_ENDPOINT_FALLBACKS) {
+            const result = await attempt((grant, fetchVia) => fetchGeminiCliQuota({
+                accessToken: grant.accessToken,
+                projectId: grant.projectId,
+                endpoints: [endpoint],
+                userAgent: buildGeminiCliUserAgent(),
+                timeoutMs: 10_000,
+                fetchVia,
+            }));
+            if (result.value !== undefined) {
+                return {
+                    response: result.value,
+                    ...(result.attribution ? { attribution: result.attribution } : {}),
+                };
+            }
+            errors.push(result.error instanceof Error
+                ? result.error.message
+                : String(result.error));
+            if (result.refused)
+                break;
+        }
+        const error = errors.join('; ') || 'fetchGeminiCliQuota failed';
+        options.logger.debug('vault fetchGeminiCliQuota failed', { error });
+        return { response: { buckets: undefined }, error };
+    };
+    const [summary, cli] = await Promise.all([fetchSummary(), fetchGeminiCli()]);
+    return {
+        quota: summary.result,
+        geminiCliQuota: annotateGeminiCliQuota(cli.response, cli.error),
+        fellBackToLegacy: summary.fellBackToLegacy,
+        ...(summary.attribution ? { quotaAttribution: summary.attribution } : {}),
+        ...(cli.attribution ? { geminiCliAttribution: cli.attribution } : {}),
+    };
+}
+/** Raised inside a check when a request's local grant is no longer current. */
+class LocalQuotaGrantStaleError extends Error {
+    name = 'LocalQuotaGrantStaleError';
+}
+/**
+ * A quota fetcher for LOCAL credentials whose bearer and project come from
+ * the caller's `authorize` callback instead of the account's stored refresh
+ * token. The network requests and result shape are the OpenCode 1
+ * fetcher's (`fetchQuotaPayloads`), sent through a transport that calls the
+ * authorization's `confirmSend` before every physical request.
+ *
+ * - A disabled account is answered without calling `authorize`.
+ * - A refused authorization, an authorization that is not `local`, or a
+ *   throw from `authorize` is answered as an error result without any
+ *   request.
+ * - Once `confirmSend` reports `stale`, no further request of the check is
+ *   sent and the whole check is answered as an error, so a reading taken
+ *   for a replaced or removed credential is never applied.
+ * - The result never carries `updatedAccount`: any credential or project
+ *   change belongs to the `authorize` callback, which knows which stored row
+ *   it may write to.
+ */
+export function createAuthorizedFetchAccountQuota(options) {
+    return async (account, signal) => {
+        const base = { index: 0, email: account.email };
+        if (account.enabled === false) {
+            return { ...base, status: 'disabled', disabled: true };
+        }
+        if (signal.aborted) {
+            return { ...base, status: 'error', disabled: false, error: 'aborted' };
+        }
+        let staleReason = null;
+        try {
+            // Called before the first await, so the callback sees the account as
+            // handed in; a synchronous throw lands in the catch below.
+            const authorizing = options.authorize(account, signal);
+            const authorization = await authorizing;
+            if (authorization.status !== 'authorized') {
+                return {
+                    ...base,
+                    status: 'error',
+                    disabled: false,
+                    error: `quota check refused (${authorization.reason})`,
+                };
+            }
+            if (authorization.domain !== 'local') {
+                return {
+                    ...base,
+                    status: 'error',
+                    disabled: false,
+                    error: 'quota check refused (unsupported credential domain)',
+                };
+            }
+            if (signal.aborted) {
+                return { ...base, status: 'error', disabled: false, error: 'aborted' };
+            }
+            const send = options.transport?.(signal) ??
+                ((url, init, extra) => fetchWithActiveTimeout(url, { ...init, signal: extra.signal ?? signal }, { timeoutMs: extra.timeoutMs }));
+            const guarded = async (url, init, extra) => {
+                if (staleReason !== null)
+                    throw new LocalQuotaGrantStaleError(staleReason);
+                const check = await authorization.confirmSend(signal);
+                if (check.status !== 'current') {
+                    staleReason = check.reason;
+                    throw new LocalQuotaGrantStaleError(check.reason);
+                }
+                // `confirmSend` (the check that the account still holds this exact
+                // credential and access token) has just finished; the request goes
+                // out at once, with this quota check's abort signal.
+                return send(url, init, { ...extra, signal });
+            };
+            const payloads = await fetchQuotaPayloads({
+                accessToken: authorization.accessToken,
+                managedProjectId: authorization.managedProjectId,
+                projectId: authorization.projectId,
+                fetchVia: guarded,
+                logger: options.logger,
+            });
+            if (staleReason !== null) {
+                return {
+                    ...base,
+                    status: 'error',
+                    disabled: false,
+                    error: `quota check refused (stale grant: ${staleReason})`,
+                };
+            }
+            return {
+                ...base,
+                status: 'ok',
+                disabled: false,
+                quota: payloads.quota,
+                geminiCliQuota: payloads.geminiCliQuota,
+            };
+        }
+        catch (error) {
+            const message = error instanceof Error ? error.message : String(error);
+            options.logger.debug('authorized quota check failed', { error: message });
+            return { ...base, status: 'error', disabled: false, error: message };
+        }
+    };
+}
 function makeFetchAccountQuota(client, providerId, fetchVia) {
     return async (account, signal) => {
         const index = 0;
@@ -264,78 +756,16 @@ function makeFetchAccountQuota(client, providerId, fetchVia) {
             // account record. The real managedProjectId lives on the persisted
             // account, not in the packed refresh string.
             const managedProjectId = authParts.managedProjectId ?? account.managedProjectId;
-            // Two independent payload contracts: the windowed summary
-            // (with legacy fallback) and the gemini-CLI quota. They share
-            // access + project but target different endpoints, so the
-            // two 10s timeouts ran back-to-back for ~20s per account on
-            // modal open. Run them concurrently; either rejection is
-            // handled by its own branch and the result is still merged.
-            const fetchSummaryPayload = (async () => {
-                try {
-                    const summaryResult = await fetchQuotaSummary({
-                        accessToken: auth.access ?? '',
-                        managedProjectId,
-                        projectId: projectContext.effectiveProjectId,
-                        endpoints: ANTIGRAVITY_ENDPOINT_FALLBACKS,
-                        userAgent: buildAntigravityHarnessUserAgent(),
-                        timeoutMs: 10_000,
-                        ...(fetchVia ? { fetchVia } : {}),
-                    });
-                    return {
-                        result: aggregateQuotaSummary(summaryResult.summary),
-                        fellBackToLegacy: summaryResult.fellBackToLegacy ?? false,
-                    };
-                }
-                catch {
-                    return {
-                        result: await fetchLegacyModelsFallback({
-                            accessToken: auth.access ?? '',
-                            projectId: projectContext.effectiveProjectId,
-                            fetchVia,
-                        }),
-                        fellBackToLegacy: true,
-                    };
-                }
-            })();
-            // CLI fetch is independent of the summary fetch. A CLI failure must
-            // NOT kill the summary result, but it also must not be laundered into
-            // "No Gemini CLI quota available" (a permanent-looking status) when
-            // the real cause is a transient network error. Capture the error
-            // message separately so the annotated result can carry it.
-            let geminiCliFetchError;
-            const fetchGeminiCliPayload = fetchGeminiCliQuota({
+            const payloads = await fetchQuotaPayloads({
                 accessToken: auth.access ?? '',
+                managedProjectId,
                 projectId: projectContext.effectiveProjectId,
-                endpoints: ANTIGRAVITY_ENDPOINT_FALLBACKS,
-                userAgent: buildGeminiCliUserAgent(),
-                timeoutMs: 10_000,
-                ...(fetchVia ? { fetchVia } : {}),
-            }).catch((error) => {
-                geminiCliFetchError =
-                    error instanceof Error ? error.message : String(error);
-                log.debug('fetchGeminiCliQuota failed', { error: geminiCliFetchError });
-                return { buckets: undefined };
+                fetchVia,
+                logger: log,
             });
-            const [summary, geminiCliResponse] = await Promise.all([
-                fetchSummaryPayload,
-                fetchGeminiCliPayload,
-            ]);
-            quotaResult = summary.result;
-            fellBackToLegacy = summary.fellBackToLegacy;
-            const geminiCliQuotaResult = aggregateGeminiCliQuota(geminiCliResponse);
-            const annotated = geminiCliResponse.buckets === undefined ||
-                geminiCliResponse.buckets.length === 0
-                ? {
-                    ...geminiCliQuotaResult,
-                    error: 
-                    // A real fetch exception is a transient failure, not a
-                    // "no CLI configured" scenario — propagate the actual message.
-                    geminiCliFetchError ??
-                        (geminiCliQuotaResult.models.length === 0
-                            ? 'No Gemini CLI quota available'
-                            : undefined),
-                }
-                : geminiCliQuotaResult;
+            quotaResult = payloads.quota;
+            fellBackToLegacy = payloads.fellBackToLegacy;
+            const annotated = payloads.geminiCliQuota;
             for (const [family, groupQuota] of Object.entries(quotaResult.groups)) {
                 const remainingPercent = (groupQuota.remainingFraction ?? 0) * 100;
                 logQuotaStatus(account.email, index, remainingPercent, family);
@@ -424,30 +854,55 @@ async function persistRotatedRefresh(client, providerId, auth) {
 /**
  * Build a per-account tier-loader callback for the background poller.
  *
- * Calls `loadManagedProject` (loadCodeAssist) directly, bypassing the
- * `ensureProjectContext` cache that fast-paths on `managedProjectId` and
- * never returns a tier for existing accounts. One call per account per 24 h.
+ * Calls `loadManagedProject` (loadCodeAssist) directly. The
+ * `ensureProjectContext` cache answers from a known `managedProjectId`
+ * without asking Google, so for an account that already has a project it
+ * never returns a tier; the tier is only in loadCodeAssist's answer. One
+ * call per account per 24 h.
  *
- * Uses the same token-refresh infrastructure as `makeFetchAccountQuota` so
- * an expired access token does not silently fail the tier lookup.
+ * An expired access token is refreshed the same way `makeFetchAccountQuota`
+ * refreshes it, so the lookup does not fail just because the token aged.
  *
- * `loadManagedProject` uses the production TLS transport (`fetchWithAgyCliTransport`)
- * and is not interceptable via `fetchVia` -- the same design constraint applies to
- * `ensureProjectContext`. Tier lookup is best-effort; any failure resolves `null`.
+ * `loadManagedProject` sends through the native TLS transport
+ * (`fetchWithAgyCliTransport`), not through a generic `fetch`, so a
+ * `fetchVia` override cannot intercept it; `ensureProjectContext` has the
+ * same limit. Tier lookup is best-effort; any failure resolves `null`.
  */
-export function makeTierLoader(client, providerId) {
+export function makeTierLoader(client, providerId, options = {}) {
     return async (account) => {
         try {
-            let auth = buildAuthFromAccount(account);
-            if (accessTokenExpired(auth)) {
-                const refreshed = await refreshAccessToken(auth, client, providerId);
-                if (!refreshed)
+            if (options.mayRun && !(await options.mayRun()))
+                return null;
+            let accessToken;
+            const target = account;
+            if (target.rowRef !== undefined) {
+                if (!options.authorizeStore)
                     return null;
-                auth = refreshed;
+                const authorization = await options.authorizeStore(target);
+                if (authorization.status !== 'authorized' ||
+                    authorization.domain !== 'local')
+                    return null;
+                if (options.mayRun && !(await options.mayRun()))
+                    return null;
+                if ((await authorization.confirmSend(new AbortController().signal))
+                    .status !== 'current')
+                    return null;
+                accessToken = authorization.accessToken;
+            }
+            else {
+                let auth = buildAuthFromAccount(account);
+                if (accessTokenExpired(auth)) {
+                    const refreshed = await refreshAccessToken(auth, client, providerId);
+                    if (!refreshed)
+                        return null;
+                    auth = refreshed;
+                }
+                accessToken = auth.access;
             }
-            const accessToken = auth.access;
             if (!accessToken)
                 return null;
+            if (options.mayRun && !(await options.mayRun()))
+                return null;
             const payload = await loadManagedProject(accessToken);
             if (!payload?.currentTier?.id)
                 return null;
@@ -465,4 +920,279 @@ export function makeTierLoader(client, providerId) {
         }
     };
 }
+/**
+ * The quota groups a reading carries, in one canonical text: groups by
+ * name, each with its model count, remaining fraction, reset time and
+ * windows in order; absent and null are the same. The in-memory groups a
+ * check records and the groups the store keeps for them give the same text,
+ * so the two can be compared exactly.
+ */
+function canonicalQuotaGroups(groups) {
+    const entries = Object.entries(groups ?? {})
+        .filter((entry) => entry[1] !== undefined)
+        .sort(([left], [right]) => (left < right ? -1 : left > right ? 1 : 0))
+        .map(([name, group]) => [
+        name,
+        group.modelCount,
+        group.remainingFraction ?? null,
+        group.resetTime ?? null,
+        group.windows?.map((window) => [
+            window.window,
+            window.remainingFraction ?? null,
+            window.resetTime ?? null,
+        ]) ?? null,
+    ]);
+    return JSON.stringify(entries);
+}
+/**
+ * The manager's account holding exactly `ref`. A RowRef names one
+ * credential version of one row (row id, credential epoch and recorded
+ * identity), so the lookup succeeds only while that row still holds that
+ * very credential.
+ */
+function heldAccount(manager, ref) {
+    return manager
+        .getAccounts()
+        .find((candidate) => candidate.ref !== undefined && sameRowRef(candidate.ref, ref));
+}
+/**
+ * Quota checks for one plugin account pool whose accounts come from the
+ * account store. One check, in order:
+ * 1. Capture the target's RowRef before any await, and take the manager's
+ *    account holding exactly that ref; if none does, the check is refused.
+ * 2. If that account's access token has expired, refresh it through the
+ *    repository, which commits the new token only to the row still holding
+ *    that ref. Neither a token copied from an earlier read nor the host
+ *    client is used.
+ * 3. Resolve the account's project from the local project cache; a newly
+ *    resolved project is recorded through the manager on the account's ref.
+ * 4. Before every physical quota request, check that the manager still
+ *    holds the same account object, at the same ref, enabled, with the same
+ *    access token; otherwise no further request of the check is sent.
+ * 5. Hand the reading to the manager only while it still holds that exact
+ *    ref; otherwise discard it. The manager's write can still be refused by
+ *    the store, so `checkStoreQuota` reads the store back and counts a
+ *    check only when the store holds this check's whole reading.
+ *
+ * Only accounts whose refresh token is stored locally use this service;
+ * vault-held accounts never do.
+ */
+export function createStoreQuotaService(options) {
+    const { manager, logger } = options;
+    const credentials = options.credentials ??
+        createLocalAccountCredentials(manager, { repository: options.repository });
+    const authorize = async (account) => {
+        const ref = account.rowRef;
+        if (ref === undefined) {
+            return { status: 'refused', reason: 'the account has no store ref' };
+        }
+        if (options.mayRun && !(await options.mayRun()))
+            return { status: 'refused', reason: VAULT_ACCOUNT_REMEDY };
+        const held = heldAccount(manager, ref);
+        if (held === undefined) {
+            return { status: 'refused', reason: 'the account changed or was removed' };
+        }
+        let auth = manager.toAuthDetails(held);
+        if (accessTokenExpired(auth)) {
+            const refreshed = await credentials.refresh(held);
+            if (refreshed === undefined) {
+                return { status: 'refused', reason: 'token refresh failed' };
+            }
+            auth = refreshed;
+        }
+        // A refresh keeps the credential epoch but can record the provider's
+        // account identity on the ref and always brings a new access token, so
+        // the grant is the account's ref and token as they are now, after it.
+        const grantRef = held.ref;
+        const accessToken = auth.access;
+        if (grantRef === undefined || !accessToken) {
+            return { status: 'refused', reason: 'no access token' };
+        }
+        if (options.mayRun && !(await options.mayRun()))
+            return { status: 'refused', reason: VAULT_ACCOUNT_REMEDY };
+        const project = await credentials.ensureProject(auth);
+        if (options.mayRun && !(await options.mayRun()))
+            return { status: 'refused', reason: VAULT_ACCOUNT_REMEDY };
+        if (project.auth.refresh !== auth.refresh) {
+            // The resolved project belongs to this credential: the manager records
+            // only the project fields, fenced on this account's ref, so they can
+            // never land on another row or a replaced credential.
+            manager.updateFromAuth(held, { ...project.auth, access: accessToken });
+        }
+        const managedProjectId = parseRefreshParts(project.auth.refresh).managedProjectId ??
+            held.managedProjectId;
+        return {
+            status: 'authorized',
+            domain: 'local',
+            accessToken,
+            projectId: project.effectiveProjectId,
+            ...(managedProjectId !== undefined ? { managedProjectId } : {}),
+            async confirmSend() {
+                try {
+                    if (options.mayRun && !(await options.mayRun()))
+                        return { status: 'stale', reason: VAULT_ACCOUNT_REMEDY };
+                    if (held.ref === undefined || !sameRowRef(held.ref, grantRef)) {
+                        return { status: 'stale', reason: 'the credential changed' };
+                    }
+                    await credentials.assertGrantCurrent({
+                        account: held,
+                        accessToken,
+                        ref: grantRef,
+                    });
+                    if (options.mayRun && !(await options.mayRun()))
+                        return { status: 'stale', reason: VAULT_ACCOUNT_REMEDY };
+                    return { status: 'current' };
+                }
+                catch (error) {
+                    return {
+                        status: 'stale',
+                        reason: error instanceof Error ? error.message : String(error),
+                    };
+                }
+            },
+        };
+    };
+    const fetchAccountQuota = createAuthorizedFetchAccountQuota({
+        authorize: (account) => authorize(account),
+        logger,
+        transport: options.transport,
+    });
+    const quotaManager = createLocationQuotaManager({
+        logger,
+        fetchAccountQuota,
+        // Backoff and cache are per credential: a re-authenticated row starts
+        // fresh instead of inheriting its predecessor's state.
+        keyOf: (account) => {
+            const ref = account.rowRef;
+            return ref === undefined ? defaultKeyOf(account) : rowRefKey(ref);
+        },
+        baseBackoffMs: options.baseBackoffMs,
+        maxBackoffMs: options.maxBackoffMs,
+        fetchTimeoutMs: options.fetchTimeoutMs,
+        sidebar: options.sidebar,
+    });
+    const record = async (target, result) => {
+        if (result.status !== 'ok' || result.quota === undefined) {
+            return { status: 'failed', result };
+        }
+        // A completed poll does not authorize deferred local metadata writes after enrollment.
+        if (options.mayRun && !(await options.mayRun()))
+            return { status: 'discarded', result };
+        const held = target.rowRef === undefined
+            ? undefined
+            : heldAccount(manager, target.rowRef);
+        if (held === undefined)
+            return { status: 'discarded', result };
+        manager.updateQuotaCache(held.index, result.quota.groups, held.parts.refreshToken);
+        // Captured now: the account's fields may change before the store is
+        // read back, and a later reading must not stand in for this one.
+        return {
+            status: 'recorded',
+            result,
+            reading: {
+                cachedQuotaAccountId: held.cachedQuotaAccountId,
+                cachedQuotaUpdatedAt: held.cachedQuotaUpdatedAt,
+                groups: canonicalQuotaGroups(result.quota.groups),
+            },
+        };
+    };
+    const refreshAccount = async (target, refreshOptions = {}) => {
+        const index = Math.max(0, manager
+            .getAccountsForQuotaCheck()
+            .findIndex((candidate) => candidate.rowRef !== undefined &&
+            target.rowRef !== undefined &&
+            sameRowRef(candidate.rowRef, target.rowRef)));
+        const result = await quotaManager.refreshAccount(target, {
+            index,
+            force: refreshOptions.force === true,
+        });
+        return record(target, result);
+    };
+    return {
+        quotaManager,
+        fetchAccountQuota,
+        authorizeLocalCheck: authorize,
+        refreshAccount,
+        async checkStoreQuota(refs) {
+            // Targets are taken before any await, so each names the credential
+            // the manager held when the action started.
+            const targets = manager.getAccountsForQuotaCheck();
+            const outcomes = await Promise.all(refs.map((ref) => {
+                const target = targets.find((candidate) => candidate.rowRef !== undefined &&
+                    sameRowRef(candidate.rowRef, ref));
+                return target === undefined
+                    ? undefined
+                    : refreshAccount(target, { force: true });
+            }));
+            // Wait for queued quota writes before counting an account as checked.
+            // If a failed write cannot be attributed to an account, no account's
+            // quota reading is confirmed.
+            let failedRows = new Set();
+            try {
+                await manager.saveToDisk();
+            }
+            catch (error) {
+                // AccountManagerPersistError includes a flush report, but the public
+                // core entry does not export its constructor. Identify it by name and
+                // require at least one failed row ID. Only accounts outside the
+                // reported failure list can still count as checked; missing or empty
+                // failure details confirm none.
+                const report = error instanceof Error && error.name === 'AccountManagerPersistError'
+                    ? error.report
+                    : undefined;
+                const failedIds = (report?.failures ?? []).map((failure) => failure.rowId);
+                failedRows =
+                    report === undefined ||
+                        failedIds.length === 0 ||
+                        failedIds.some((id) => id === undefined)
+                        ? 'all'
+                        : new Set(failedIds.filter((id) => id !== undefined));
+                logger.debug('quota readings were not all persisted', {
+                    error: error instanceof Error ? error.message : String(error),
+                });
+            }
+            const read = await options.repository.read();
+            let checked = 0;
+            refs.forEach((ref, position) => {
+                const outcome = outcomes[position];
+                if (outcome?.status !== 'recorded')
+                    return;
+                if (failedRows === 'all' || failedRows.has(ref.id))
+                    return;
+                const row = read.status === 'ready'
+                    ? read.rows.find((candidate) => sameRowRef(candidate.ref, ref))
+                    : undefined;
+                if (row?.quota.status !== 'present')
+                    return;
+                const stored = row.quota.quota;
+                // The store must hold exactly this check's reading: same identity
+                // stamp, same time and the same groups. An older reading stored with
+                // the same time, or another check's reading of the same row, does
+                // not count.
+                if (stored.cachedQuotaAccountId ===
+                    outcome.reading.cachedQuotaAccountId &&
+                    stored.cachedQuotaUpdatedAt ===
+                        outcome.reading.cachedQuotaUpdatedAt &&
+                    canonicalQuotaGroups(stored.cachedQuota) === outcome.reading.groups) {
+                    checked += 1;
+                }
+            });
+            return { checked, notChecked: refs.length - checked };
+        },
+        async refreshAll(refreshOptions = {}) {
+            const targets = manager.getAccountsForQuotaCheck();
+            const results = await quotaManager.refreshAccounts(targets, {
+                indexFor: (account) => targets.indexOf(account),
+                force: refreshOptions.force === true,
+            });
+            return Promise.all(results.map((result, position) => {
+                const target = targets[position];
+                return target === undefined
+                    ? { status: 'failed', result }
+                    : record(target, result);
+            }));
+        },
+        dispose: () => quotaManager.dispose(),
+    };
+}
 //# sourceMappingURL=quota.js.map