codex-multi-auth 2.19.1 → 2.21.0

package/dist/lib/prompts/codex.js

@@ -381,83 +381,83 @@ export function prewarmCodexInstructions(models = []) {
 /**
  * Tool remapping instructions for host runtime tools
  */
-export const TOOL_REMAP_MESSAGE = `<user_instructions priority="0">
-<environment_override priority="0">
-YOU ARE IN A DIFFERENT ENVIRONMENT. These instructions override ALL previous tool references.
-</environment_override>
-
-<tool_replacements priority="0">
-<critical_rule priority="0">
-apply_patch/applyPatch are Codex names, but host tool names vary by version.
-- Inspect the actual tool list before editing.
-- If edit exists: use edit for precise in-place string replacements and hashline edits.
-- If edit is absent and apply_patch exists: use apply_patch for those precise/hashline edits in this plugin.
-- For diff-style or multi-line structural edits: use patch if available, otherwise use apply_patch.
-- In this plugin, edit/apply_patch support hashline refs (lineRef, endLineRef, operation, content)
-</critical_rule>
-
-<critical_rule priority="0">
-UPDATE_PLAN DOES NOT EXIST -> USE "todowrite" INSTEAD
-- NEVER use: update_plan, updatePlan
-- ALWAYS use: todowrite for ALL task/plan operations
-- Use todoread to read current plan
-- Before plan operations: Verify you're using "todowrite", NOT "update_plan"
-</critical_rule>
-</tool_replacements>
-
-<available_tools priority="0">
-File Operations:
-  - write  - Create new files (if available)
-  - edit   - Modify existing files with string replacement (version-dependent)
-  - oldString must be literal text from the current file; never pass unresolved placeholders like \${TARGET_SNIPPET}
-  - apply_patch - May be the edit/patch tool name in newer host builds (version-dependent)
-  - hashline_read - Read lines with hashline refs (L<line>#<hash>) for deterministic edits
-  - patch  - Apply diff patches (version-dependent)
-  - read   - Read file contents
-
-Search/Discovery:
-  - grep   - Search file contents
-  - glob   - Find files by pattern
-  - list   - List directories (if available)
-
-Execution:
-  - bash   - Run shell commands
-
-Network:
-  - webfetch - Fetch web content
-
-Task Management:
-  - todowrite - Manage tasks/plans (REPLACES update_plan)
-  - todoread  - Read current plan
-</available_tools>
-
-<substitution_rules priority="0">
-Base instruction says:    You MUST use instead:
-apply_patch           ->   patch (preferred if available), otherwise edit/apply_patch based on actual tool list
-update_plan           ->   todowrite
-read_plan             ->   todoread
-absolute paths        ->   relative paths
-</substitution_rules>
-
-<verification_checklist priority="0">
-Before file/plan modifications:
-1. Am I using the actual available edit tool name (edit, patch, or apply_patch)?
-2. Am I using "todowrite" NOT "update_plan"?
-3. Is this tool in the approved list above?
-4. Am I using relative paths?
-
-If ANY answer is NO -> STOP and correct before proceeding.
-</verification_checklist>
-
-<hashline_workflow priority="0">
-When hashline_read is available:
-1. call hashline_read on the file
-2. edit/apply_patch using lineRef/endLineRef + operation + content
-3. use oldString/newString only as fallback, and only with literal file text (no template placeholders)
-</hashline_workflow>
-
-<safety_rules priority="0">
-- Never run destructive git commands (\`git reset --hard\`, \`git checkout --\`) unless explicitly requested by the user.
-- Never call \`request_user_input\` unless collaboration mode is explicitly Plan mode.
-</safety_rules>
+export const TOOL_REMAP_MESSAGE = `<user_instructions priority="0">
+<environment_override priority="0">
+YOU ARE IN A DIFFERENT ENVIRONMENT. These instructions override ALL previous tool references.
+</environment_override>
+
+<tool_replacements priority="0">
+<critical_rule priority="0">
+apply_patch/applyPatch are Codex names, but host tool names vary by version.
+- Inspect the actual tool list before editing.
+- If edit exists: use edit for precise in-place string replacements and hashline edits.
+- If edit is absent and apply_patch exists: use apply_patch for those precise/hashline edits in this plugin.
+- For diff-style or multi-line structural edits: use patch if available, otherwise use apply_patch.
+- In this plugin, edit/apply_patch support hashline refs (lineRef, endLineRef, operation, content)
+</critical_rule>
+
+<critical_rule priority="0">
+UPDATE_PLAN DOES NOT EXIST -> USE "todowrite" INSTEAD
+- NEVER use: update_plan, updatePlan
+- ALWAYS use: todowrite for ALL task/plan operations
+- Use todoread to read current plan
+- Before plan operations: Verify you're using "todowrite", NOT "update_plan"
+</critical_rule>
+</tool_replacements>
+
+<available_tools priority="0">
+File Operations:
+  - write  - Create new files (if available)
+  - edit   - Modify existing files with string replacement (version-dependent)
+  - oldString must be literal text from the current file; never pass unresolved placeholders like \${TARGET_SNIPPET}
+  - apply_patch - May be the edit/patch tool name in newer host builds (version-dependent)
+  - hashline_read - Read lines with hashline refs (L<line>#<hash>) for deterministic edits
+  - patch  - Apply diff patches (version-dependent)
+  - read   - Read file contents
+
+Search/Discovery:
+  - grep   - Search file contents
+  - glob   - Find files by pattern
+  - list   - List directories (if available)
+
+Execution:
+  - bash   - Run shell commands
+
+Network:
+  - webfetch - Fetch web content
+
+Task Management:
+  - todowrite - Manage tasks/plans (REPLACES update_plan)
+  - todoread  - Read current plan
+</available_tools>
+
+<substitution_rules priority="0">
+Base instruction says:    You MUST use instead:
+apply_patch           ->   patch (preferred if available), otherwise edit/apply_patch based on actual tool list
+update_plan           ->   todowrite
+read_plan             ->   todoread
+absolute paths        ->   relative paths
+</substitution_rules>
+
+<verification_checklist priority="0">
+Before file/plan modifications:
+1. Am I using the actual available edit tool name (edit, patch, or apply_patch)?
+2. Am I using "todowrite" NOT "update_plan"?
+3. Is this tool in the approved list above?
+4. Am I using relative paths?
+
+If ANY answer is NO -> STOP and correct before proceeding.
+</verification_checklist>
+
+<hashline_workflow priority="0">
+When hashline_read is available:
+1. call hashline_read on the file
+2. edit/apply_patch using lineRef/endLineRef + operation + content
+3. use oldString/newString only as fallback, and only with literal file text (no template placeholders)
+</hashline_workflow>
+
+<safety_rules priority="0">
+- Never run destructive git commands (\`git reset --hard\`, \`git checkout --\`) unless explicitly requested by the user.
+- Never call \`request_user_input\` unless collaboration mode is explicitly Plan mode.
+</safety_rules>
 </user_instructions>`;

package/dist/lib/request/helpers/model-map.d.ts

@@ -6,7 +6,7 @@
  * tool-surface metadata so routing logic stays consistent across the request
  * transformer, prompt selection, and CLI diagnostics.
  */
-import type { ModelReasoningEffort, WireReasoningEffort } from "../../constants.js";
+import { type ModelReasoningEffort, type WireReasoningEffort } from "../../constants.js";
 export type { ModelReasoningEffort, WireReasoningEffort };
 export type PromptModelFamily = "gpt-5-codex" | "codex-max" | "codex" | "gpt-5.2" | "gpt-5.1";
 /**

package/dist/lib/request/helpers/model-map.js

@@ -6,6 +6,10 @@
  * tool-surface metadata so routing logic stays consistent across the request
  * transformer, prompt selection, and CLI diagnostics.
  */
+// The effort union lives in the leaf constants module so the base types layer
+// (`lib/types.ts`) can depend on it without importing this file, which would
+// close a cycle through `lib/schemas.ts`. Re-exported here for existing callers.
+import { LUNA_RESERVE_MODEL, } from "../../constants.js";
 /**
  * All supported model families
  * Used for per-family account rotation and rate limit tracking
@@ -301,6 +305,13 @@ export const MODEL_PROFILES = {
         supportedReasoningEfforts: GPT_6_LUNA_EFFORTS,
         capabilities: TOOL_CAPABILITIES.full,
     },
+    [LUNA_RESERVE_MODEL]: {
+        normalizedModel: LUNA_RESERVE_MODEL,
+        promptFamily: "gpt-5.2",
+        defaultReasoningEffort: "medium",
+        supportedReasoningEfforts: GPT_6_LUNA_EFFORTS,
+        capabilities: TOOL_CAPABILITIES.full,
+    },
     [DAYBREAK_BLUE_MODEL]: {
         normalizedModel: DAYBREAK_BLUE_MODEL,
         promptFamily: "gpt-5.2",
@@ -392,6 +403,7 @@ function addGpt6Aliases() {
     // user's model generation without asking.
     addEffortAliases(GPT_6_SOL_MODEL, GPT_6_SOL_MODEL, GPT_6_SOL_EFFORTS);
     addEffortAliases(GPT_6_LUNA_MODEL, GPT_6_LUNA_MODEL, GPT_6_LUNA_EFFORTS);
+    addEffortAliases(LUNA_RESERVE_MODEL, LUNA_RESERVE_MODEL, GPT_6_LUNA_EFFORTS);
     addEffortAliases(GPT_6_1_SOL_MODEL, GPT_6_1_SOL_MODEL, GPT_6_1_SOL_EFFORTS);
     addEffortAliases(GPT_6_1_FLAGSHIP_ALIAS, GPT_6_1_SOL_MODEL, GPT_6_1_SOL_EFFORTS);
 }

package/dist/lib/request/response-outcome.d.ts

@@ -13,7 +13,7 @@ export declare class ResponseOutcome {
             param?: string;
         };
     } | undefined;
-    constructor(requireTerminal: boolean);
+    constructor(requireTerminal: boolean | (() => boolean));
     observe(value: unknown): void;
     /**
      * `response.incomplete` (e.g. max_output_tokens) is a delivered response, not an

package/dist/lib/request/response-outcome.js

@@ -38,7 +38,10 @@ export class ResponseOutcome {
      * type survives only as the usage annotation in `errorCode`.
      */
     finish() {
-        const missingTerminal = this.requireTerminal && !this.terminal;
+        const requireTerminal = typeof this.requireTerminal === "function"
+            ? this.requireTerminal()
+            : this.requireTerminal;
+        const missingTerminal = requireTerminal && !this.terminal;
         const success = !missingTerminal && (!this.terminal || this.terminal === "completed" || this.terminal === "incomplete");
         return {
             success,

package/dist/lib/runtime/account-status.js

@@ -1,4 +1,5 @@
 import { getQuotaKey } from "../accounts/rate-limits.js";
+import { LUNA_RESERVE_MODEL } from "../constants.js";
 export function resolveActiveIndex(storage, family = "codex") {
     const total = storage.accounts.length;
     if (total === 0)
@@ -58,7 +59,7 @@ function laterBound(a, b) {
 export function getAccountRecoveryBoundsForFamily(account, now, family, model) {
     const times = account.rateLimitResetTimes;
     const rateLimitAtMs = times
-        ? laterBound(activeBound(times[getQuotaKey(family)], now), model ? activeBound(times[getQuotaKey(family, model)], now) : null)
+        ? laterBound(model === LUNA_RESERVE_MODEL ? null : activeBound(times[getQuotaKey(family)], now), model ? activeBound(times[getQuotaKey(family, model)], now) : null)
         : null;
     return {
         rateLimitAtMs,

package/dist/lib/runtime/luna-reserve-usage.d.ts

@@ -0,0 +1,9 @@
+import { type LunaReserveSnapshot } from "../luna-reserve.js";
+import type { AccountStorageV3 } from "../storage.js";
+export type LunaReserveUsageByAccount = Record<number, LunaReserveSnapshot>;
+/**
+ * Read Luna Reserve usage through the same isolated native app-server bridge
+ * used for reset credits. Failures are per-account: one stale/revoked account
+ * never prevents other rows from reporting their Reserve percentage.
+ */
+export declare function refreshLunaReserveUsage(storage: AccountStorageV3, now?: () => number): Promise<LunaReserveUsageByAccount>;

package/dist/lib/runtime/luna-reserve-usage.js

@@ -0,0 +1,68 @@
+import { AccountManager } from "../accounts.js";
+import { LUNA_RESERVE_MODEL } from "../constants.js";
+import { parseLunaReserveRateLimits, } from "../luna-reserve.js";
+import { codexCliAccountIdFor } from "../auth/token-utils.js";
+import { isRecord } from "../utils.js";
+import { workspaceModelScopes } from "./workspace-model-scopes.js";
+import { nativeRateLimitsRpc } from "./native-rate-limits.js";
+import { ensureFreshAccessToken } from "./rotation-token-refresh.js";
+/**
+ * Read Luna Reserve usage through the same isolated native app-server bridge
+ * used for reset credits. Failures are per-account: one stale/revoked account
+ * never prevents other rows from reporting their Reserve percentage.
+ */
+export async function refreshLunaReserveUsage(storage, now = Date.now) {
+    const manager = new AccountManager(undefined, storage);
+    const result = {};
+    try {
+        for (let index = 0; index < storage.accounts.length; index += 1) {
+            const stored = storage.accounts[index];
+            const account = manager.getAccountByIndex(index);
+            if (!stored || !account || account.enabled === false)
+                continue;
+            const scopes = workspaceModelScopes(account);
+            const targetScope = account.workspaces?.length
+                ? scopes.find((scope) => scope.selected && scope.routable)
+                : scopes.find((scope) => scope.bound && scope.routable);
+            if (!targetScope)
+                continue;
+            try {
+                const fresh = await ensureFreshAccessToken({
+                    accountManager: manager,
+                    account,
+                    family: "gpt-5.2",
+                    model: LUNA_RESERVE_MODEL,
+                    now: now(),
+                    tokenRefreshSkewMs: 60_000,
+                    tokenInvalidationCooldownMs: 300_000,
+                });
+                if (!fresh.ok)
+                    continue;
+                const auth = {
+                    accessToken: fresh.accessToken,
+                    accountId: targetScope.accountId,
+                    expiresAt: fresh.account.expires ?? 0,
+                    codexCliMirror: account.codexCliMirror,
+                };
+                const sentAccountId = codexCliAccountIdFor(auth, fresh.accessToken) ?? targetScope.accountId;
+                const reply = await nativeRateLimitsRpc(auth, "account/rateLimits/read", {
+                    supportsLunaReserve: true,
+                    excludeResetCreditDetails: true,
+                });
+                if (isRecord(reply) &&
+                    typeof reply.accountId === "string" &&
+                    reply.accountId !== sentAccountId) {
+                    continue;
+                }
+                result[index] = parseLunaReserveRateLimits(reply, now());
+            }
+            catch {
+                // Best-effort usage reporting; callers still emit normal quota.
+            }
+        }
+        return result;
+    }
+    finally {
+        await manager.flushPendingSave();
+    }
+}

package/dist/lib/runtime/model-discovery-status.d.ts

@@ -1,6 +1,6 @@
 import { z } from "zod";
-import { type AccountStorageV3 } from "../storage.js";
-import { type ApiRouteCredential, updateApiModelDiscovery } from "../api-route-store.js";
+import { type AccountStorageV3, loadAccounts } from "../storage.js";
+import { type ApiRouteCredential, loadApiRoutes, updateApiModelDiscovery } from "../api-route-store.js";
 declare const schema: z.ZodObject<{
     apiConfigurationUnavailable: z.ZodOptional<z.ZodBoolean>;
     version: z.ZodLiteral<1>;
@@ -90,9 +90,13 @@ export declare function refreshAndPrintModelInventory(log: (message: string) =>
     }>;
     fetchImpl?: typeof fetch;
     loadInventory?: typeof loadModelInventory;
+    loadAccounts?: typeof loadAccounts;
+    loadApiRoutes?: typeof loadApiRoutes;
     now?: () => number;
     /** Only an explicit `check capabilities` bypasses the 15-minute probe cache. */
     forceProbes?: boolean;
+    /** Installed `codex --version` probe, used when the inventory stores no clientVersion. */
+    detectClientVersion?: () => Promise<string | null>;
 }): Promise<boolean>;
 /** Compare stable credential/workspace scopes, keeping the last successful baseline through outages. */
 export declare function withInventoryChanges(current: ModelInventory, previous: ModelInventory | null): ModelInventory;

package/dist/lib/runtime/model-discovery-status.js

@@ -15,6 +15,7 @@ import { loadApiRoutes, updateApiModelDiscovery, } from "../api-route-store.js";
 import { ApiModelRuntime, boundedJson } from "./api-model-runtime.js";
 import { AccountModelCatalog } from "./account-model-catalog.js";
 import { ensureFreshAccessToken } from "./rotation-token-refresh.js";
+import { CLIENT_VERSION_PATTERN, detectInstalledCodexCliVersion, } from "../codex-cli/version.js";
 import { createCodexHeaders } from "../request/headers.js";
 import { CODEX_BASE_URL } from "../constants.js";
 import { getCodexMultiAuthDir } from "../runtime-paths.js";
@@ -103,7 +104,7 @@ export async function discoverModelInventory(storage, routes, options = {}) {
             throw Error("Identity unavailable");
         const url = new URL(`${CODEX_BASE_URL}/codex/models`);
         const version = options.clientVersion;
-        if (version && /^[0-9A-Za-z.+_-]{1,80}$/.test(version))
+        if (version && CLIENT_VERSION_PATTERN.test(version))
             url.searchParams.set("client_version", version);
         const response = await fetcher(url, {
             headers: createCodexHeaders({
@@ -250,6 +251,11 @@ export async function refreshAndPrintModelInventory(log, deps = {}) {
     try {
         const load = deps.loadInventory ?? loadModelInventory;
         const previous = await load();
+        // The catalog requires client_version (issue #733): the stored inventory
+        // version wins, otherwise probe the installed `codex --version` so a fresh
+        // inventory seeds one and saves it for the next check.
+        const detectClientVersion = async () => (await (deps.detectClientVersion ?? detectInstalledCodexCliVersion)().catch(() => null)) ?? undefined;
+        const clientVersion = previous?.clientVersion;
         const binding = await (deps.getBinding ?? getAppBindStatus)();
         if (binding.running && binding.state?.nativeOpenai) {
             const started = (deps.now ?? Date.now)();
@@ -266,8 +272,9 @@ export async function refreshAndPrintModelInventory(log, deps = {}) {
             url.searchParams.set("refresh_capabilities", "1");
             if (deps.forceProbes)
                 url.searchParams.set("force_probes", "1");
-            if (previous?.clientVersion)
-                url.searchParams.set("client_version", previous.clientVersion);
+            const proxyClientVersion = clientVersion ?? (await detectClientVersion());
+            if (proxyClientVersion)
+                url.searchParams.set("client_version", proxyClientVersion);
             const clientApiKey = binding.state.clientApiKey;
             const current = await withCheckProgress("Refreshing workspace catalogs and API/ZDR probes", async () => {
                 const response = await (deps.fetchImpl ?? fetch)(url, {
@@ -294,8 +301,9 @@ export async function refreshAndPrintModelInventory(log, deps = {}) {
         }
         const value = await withCheckProgress("Discovering workspace models and API/ZDR capabilities", async () => {
             let apiConfigurationUnavailable = false;
-            const value = await discoverModelInventory(await loadAccounts(), await loadApiRoutes().catch(() => { apiConfigurationUnavailable = true; log("API configuration unavailable; checking subscription models only."); return []; }), {
-                clientVersion: previous?.clientVersion,
+            const value = await discoverModelInventory(await (deps.loadAccounts ?? loadAccounts)(), await (deps.loadApiRoutes ?? loadApiRoutes)().catch(() => { apiConfigurationUnavailable = true; log("API configuration unavailable; checking subscription models only."); return []; }), {
+                clientVersion: clientVersion ?? (await detectClientVersion()),
+                fetchImpl: deps.fetchImpl,
                 updateApiDiscovery: updateApiModelDiscovery,
                 forceProbes: deps.forceProbes === true,
                 persistProbes: true,

package/dist/lib/runtime/reset-credit-routing.js

@@ -1,9 +1,10 @@
 import { extractAccountId } from '../accounts.js';
+import { LUNA_RESERVE_MODEL } from '../constants.js';
 import { parseModelRoute } from '../model-route-policy.js';
 import { subscriptionQuotaPreference } from './subscription-quota-order.js';
 import { resetSnapshotQuota } from './reset-credits.js';
 export async function recoverResetQuota(options) {
-    if (!options.native || options.pinned || parseModelRoute(options.model).kind !== 'oauth')
+    if (options.model === LUNA_RESERVE_MODEL || !options.native || options.pinned || parseModelRoute(options.model).kind !== 'oauth')
         return false;
     const family = options.family ?? 'codex';
     const candidates = [...options.scopes].flatMap(([index, scopes]) => { const account = options.manager.getAccountByIndex(index); return account && !account.authInvalidatedAt ? scopes.map(scope => ({ account, scope })) : []; });
@@ -50,6 +51,8 @@ export async function recoverResetQuota(options) {
 }
 /** Apply a confirmed earned reset once per workspace/model observation. */
 export function applyConfirmedReset(options) {
+    if (options.model === LUNA_RESERVE_MODEL)
+        return false;
     const s = options.snapshot;
     if (!s || options.lastRedemptionAt === undefined || s.updatedAt < options.lastRedemptionAt || s.ordinaryUsageAllowed !== true || s.updatedAt <= (options.previous?.updatedAt ?? 0) || options.now - s.updatedAt > 60000 || s.updatedAt > options.now)
         return false;

package/dist/lib/runtime/rotation-server-types.d.ts

@@ -26,6 +26,8 @@ export interface RuntimeRotationProxyStatus {
     totalRequests: number;
     upstreamRequests: number;
     retries: number;
+    /** Number of ordinary GPT-6 Luna requests retried through Luna Reserve. */
+    lunaReserveFallbacks?: number;
     rotations: number;
     streamsStarted: number;
     lastError: string | null;

package/dist/lib/runtime/runtime-observability.d.ts

@@ -35,6 +35,8 @@ export interface RuntimeObservabilitySnapshot {
     responsesRequests: number;
     authRefreshRequests: number;
     diagnosticProbeRequests: number;
+    /** Number of ordinary Luna requests that switched to the separately metered Reserve model. */
+    lunaReserveFallbacks: number;
     poolExhaustionCooldownUntil: number | null;
     serverBurstCooldownUntil: number | null;
     lastAccountIndex?: number | null;

package/dist/lib/runtime/runtime-observability.js

@@ -27,6 +27,7 @@ function createDefaultSnapshot() {
         responsesRequests: 0,
         authRefreshRequests: 0,
         diagnosticProbeRequests: 0,
+        lunaReserveFallbacks: 0,
         poolExhaustionCooldownUntil: null,
         serverBurstCooldownUntil: null,
         lastAccountIndex: null,