0sec-cli 0.21.4 → 0.22.1

package/chunks/{chunk-BACXRTQ4.js → chunk-AV2LG2L7.js}

@@ -2,9 +2,8 @@
 import { createRequire as __0CreateRequire } from "node:module";
 const require = __0CreateRequire(import.meta.url);
 import {
-  VERSION,
-  cloudStateDir
-} from "./chunk-PC6RCKQV.js";
+  VERSION
+} from "./chunk-KAU6NLGP.js";
 
 // packages/core/dist/agent/feature-presets.js
 var FP_MOAT_FLAGS = [
@@ -309,18 +308,6 @@ var features = {
   get cloudSurface() {
     return env("ZERO_FEATURE_CLOUD_SURFACE", false);
   },
-  /**
-   * #978 (ADR-060) — agent fan-out. When ON, the agent gets the `start_scan`
-   * tool: it can dispatch CHILD scans (via the same POST /scans the UI uses)
-   * that run independently and report up the scan tree — the recursive
-   * sub-agent orchestration. Default OFF: fan-out multiplies scans/cost, so it
-   * stays opt-in even though the orchestrator enforces budget + a tree-level
-   * cap (max children/depth). Enable with ZERO_FEATURE_AGENT_FANOUT=1.
-   * Getter so the CLI `--features` flag is honored at dispatch time.
-   */
-  get agentFanout() {
-    return env("ZERO_FEATURE_AGENT_FANOUT", false);
-  },
   // ── Phase-2 offensive-engine feature flags (dev-live-engine-recovery) ──
   // Each gates a tool that RUNS/BUILDS untrusted code or WEAPONIZES. They are
   // deny-by-default and mirror `cloudSurface` exactly: the env flag is only one
@@ -448,20 +435,6 @@ var features = {
   get dynamicTriageRouting() {
     return env("ZERO_FEATURE_DYNAMIC_TRIAGE", false);
   },
-  /**
-   * Opt-in cloud-sink webhook integration (`packages/core/src/cloud-sink.ts`).
-   * When enabled AND the user has set ZERO_CLOUD_SINK + ZERO_CLOUD_SCAN_ID,
-   * every finding and the final scan report are POSTed to the configured
-   * remote endpoint in real time.
-   *
-   * Default ON so the env-var trio is sufficient to enable streaming, but the
-   * flag exists so operators can force-disable the integration in environments
-   * where outbound HTTP from the scanner is not desired (e.g. air-gapped CI).
-   * Disable via ZERO_FEATURE_CLOUD_SINK=0.
-   */
-  get cloudSink() {
-    return env("ZERO_FEATURE_CLOUD_SINK", true);
-  },
   /**
    * Pre-recon CVE check (`packages/core/src/pre-recon-cve.ts`).
    * In white-box mode (`--repo` set), runs `npm audit` / `pip-audit`
@@ -1050,514 +1023,11 @@ function _resetDiagnosticsForTests() {
   diag.resetForTests();
 }
 
-// packages/core/dist/cloud/credentials.js
-import { readFileSync, statSync } from "node:fs";
-import { join } from "node:path";
-var DEFAULT_CLOUD_HOST = "https://cloud.0.security";
-var CloudAuthMissingError = class extends Error {
-  constructor(message) {
-    super(message);
-    this.name = "CloudAuthMissingError";
-  }
-};
-var CloudAuthError = class extends Error {
-  status;
-  constructor(message, status) {
-    super(message);
-    this.status = status;
-    this.name = "CloudAuthError";
-  }
-};
-function loadCloudCredentials(opts = {}) {
-  const env2 = opts.env ?? process.env;
-  const warn = opts.warn ?? ((m) => process.stderr.write(`${m}
-`));
-  const envTok = env2["ZERO_CLOUD_TOKEN"]?.trim() || env2["0SEC_CLOUD_TOKEN"]?.trim();
-  if (envTok) {
-    const envHost = normaliseHost(env2["ZERO_CLOUD_HOST"]?.trim() ?? env2["0SEC_CLOUD_HOST"]?.trim() ?? DEFAULT_CLOUD_HOST);
-    if (!env2["ZERO_CLOUD_TOKEN"]?.trim()) {
-      warn("[0 cloud] using legacy 0SEC_CLOUD_* credentials; re-run `0 auth login` to migrate to ZERO_CLOUD_*.");
-    }
-    return { host: envHost, token: envTok, source: "env" };
-  }
-  const path = join(cloudStateDir(opts.homeDir, env2), "cloud.env");
-  let raw;
-  try {
-    raw = readFileSync(path, "utf-8");
-  } catch (err) {
-    const code = err.code;
-    if (code === "ENOENT") {
-      throw new CloudAuthMissingError(`0-cloud credentials not found. Run \`${env2["ZERO_DEV_SOURCE_ROOT"]?.trim() ? "0dev" : "0"} auth login\`.`);
-    }
-    throw err;
-  }
-  try {
-    const st = statSync(path);
-    const mode = st.mode & 511;
-    if (mode !== 384) {
-      warn(`[0 cloud] WARNING: ${path} mode is ${mode.toString(8).padStart(3, "0")} (expected 600). Run: chmod 600 ${path}`);
-    }
-  } catch {
-  }
-  const parsed = parseEnvFile(raw);
-  const fileTok = parsed["ZERO_CLOUD_TOKEN"]?.trim() || parsed["0SEC_CLOUD_TOKEN"]?.trim();
-  if (!fileTok) {
-    throw new CloudAuthMissingError(`0-cloud credentials in ${path} are incomplete: ZERO_CLOUD_TOKEN is required.`);
-  }
-  if (!parsed["ZERO_CLOUD_TOKEN"]?.trim()) {
-    warn("[0 cloud] using legacy 0SEC_CLOUD_* credentials; re-run `0 auth login` to migrate to ZERO_CLOUD_*.");
-  }
-  const fileHost = normaliseHost(parsed["ZERO_CLOUD_HOST"]?.trim() ?? parsed["0SEC_CLOUD_HOST"]?.trim() ?? env2["ZERO_CLOUD_HOST"]?.trim() ?? env2["0SEC_CLOUD_HOST"]?.trim() ?? DEFAULT_CLOUD_HOST);
-  return { host: fileHost, token: fileTok, source: "file" };
-}
-function normaliseHost(host) {
-  let h = host;
-  if (!/^https?:\/\//.test(h)) {
-    throw new CloudAuthMissingError(`ZERO_CLOUD_HOST must be an http(s) URL (got ${JSON.stringify(host)}).`);
-  }
-  while (h.endsWith("/"))
-    h = h.slice(0, -1);
-  return h;
-}
-function parseEnvFile(raw) {
-  const out = {};
-  const lines = raw.split(/\r?\n/);
-  for (let i = 0; i < lines.length; i++) {
-    const line = lines[i];
-    const trimmed = line.trim();
-    if (trimmed.length === 0)
-      continue;
-    if (trimmed.startsWith("#"))
-      continue;
-    const eq = trimmed.indexOf("=");
-    if (eq <= 0) {
-      throw new CloudAuthMissingError(`Malformed cloud.env at line ${i + 1}: expected KEY=VALUE, got ${JSON.stringify(line)}`);
-    }
-    const key = trimmed.slice(0, eq).trim();
-    const value = trimmed.slice(eq + 1).trim();
-    if (!/^[A-Z0-9_]+$/.test(key)) {
-      throw new CloudAuthMissingError(`Malformed cloud.env at line ${i + 1}: invalid key ${JSON.stringify(key)}`);
-    }
-    out[key] = value;
-  }
-  return out;
-}
-
-// packages/core/dist/cloud/client.js
-var CloudError = class extends Error {
-  status;
-  path;
-  code;
-  constructor(message, status, path, code) {
-    super(message);
-    this.status = status;
-    this.path = path;
-    this.code = code;
-    this.name = "CloudError";
-  }
-};
-var CloudUnauthorizedError = class extends CloudError {
-  constructor(path) {
-    super(`0-cloud auth rejected (HTTP 401) on ${path}. Run \`0 auth login\` to refresh.`, 401, path);
-    this.name = "CloudUnauthorizedError";
-  }
-};
-var CloudForbiddenError = class extends CloudError {
-  constructor(path) {
-    super(`0-cloud forbidden (HTTP 403) on ${path}. Token lacks scope for this resource.`, 403, path);
-    this.name = "CloudForbiddenError";
-  }
-};
-var CloudNetworkError = class extends CloudError {
-  constructor(message, path) {
-    super(`0-cloud network error on ${path}: ${message}`, void 0, path);
-    this.name = "CloudNetworkError";
-  }
-};
-function isUsageAccount(raw) {
-  if (!raw || typeof raw !== "object")
-    return false;
-  const obj = raw;
-  if (obj.schemaVersion !== "usage-v2")
-    return false;
-  if (!isUtcDate(obj.snapshotAt))
-    return false;
-  if (!obj.scope || typeof obj.scope !== "object")
-    return false;
-  if (typeof obj.scope.orgId !== "string")
-    return false;
-  if (typeof obj.state !== "string")
-    return false;
-  if (!["ready", "disabled", "unavailable", "restricted"].includes(obj.state))
-    return false;
-  if (obj.reason !== null && typeof obj.reason !== "string")
-    return false;
-  const plan = obj.plan;
-  if (!plan || typeof plan !== "object")
-    return false;
-  if (typeof plan.id !== "string" && plan.id !== null)
-    return false;
-  if (plan.id !== null && !["pro", "gold", "enterprise"].includes(plan.id))
-    return false;
-  if (typeof plan.name !== "string" && plan.name !== null)
-    return false;
-  if (!isUsd(plan.monthlyPriceUsd))
-    return false;
-  const included = obj.included;
-  if (!included || typeof included !== "object")
-    return false;
-  if (typeof included.state !== "string")
-    return false;
-  if (!["active", "exhausted", "none", "unavailable"].includes(included.state))
-    return false;
-  if (included.usedPercent !== null && typeof included.usedPercent !== "number")
-    return false;
-  if (included.usedPercent !== null && (!Number.isFinite(included.usedPercent) || included.usedPercent < 0 || included.usedPercent > 100))
-    return false;
-  if (included.resetsAt !== null && !isUtcDate(included.resetsAt))
-    return false;
-  const prepaid = obj.prepaid;
-  if (!prepaid || typeof prepaid !== "object")
-    return false;
-  if (!isUsd(prepaid.balanceUsd))
-    return false;
-  if (typeof prepaid.fallbackEnabled !== "boolean")
-    return false;
-  if (typeof obj.canManageBilling !== "boolean")
-    return false;
-  const admission = obj.admission;
-  if (!admission || typeof admission !== "object")
-    return false;
-  if (typeof admission.eligible !== "boolean")
-    return false;
-  if (admission.reason !== null && typeof admission.reason !== "string")
-    return false;
-  return true;
-}
-function isUsd(value) {
-  return value === null || typeof value === "string" && /^\d+(?:\.\d{1,9})?$/.test(value);
-}
-function isUtcDate(value) {
-  return typeof value === "string" && /^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(?:\.\d+)?Z$/.test(value) && Number.isFinite(Date.parse(value));
-}
-function healthPath(host) {
-  try {
-    const hostname = new URL(host).hostname.toLowerCase();
-    if (hostname === "cloud.0.ai" || hostname === "cloud.0.security") {
-      return "/api/health";
-    }
-  } catch {
-  }
-  return "/health";
-}
-var CloudClient = class {
-  host;
-  token;
-  fetchImpl;
-  constructor(opts) {
-    this.host = opts.host;
-    this.token = opts.token;
-    this.fetchImpl = opts.fetchImpl ?? fetch;
-  }
-  /**
-   * Verify cloud reachability through its health route. The hosted dashboard
-   * uses `/api/health`; a self-hosted receiver uses `/health`.
-   */
-  async pingHealth() {
-    return this.getJson(healthPath(this.host));
-  }
-  /**
-   * Fetch the hosted inference model catalog — available models, pricing,
-   * wire API protocol, and context limits. Used at runtime for model
-   * selection and by the hosted provider to determine per-model capabilities.
-   * Returns the raw list response; the caller caches/filters as needed.
-   */
-  async getInferenceModels() {
-    return this.getJson("/api/inference/v1/models");
-  }
-  /**
-   * Fetch the organization's credit account — usage-v2 shape including
-   * plan, included allowance, prepaid balance, and admission status.
-   *
-   * Returns `null` when the response is a recognised HTTP 200 (customer is
-   * authenticated) but the payload is missing, legacy, or structurally
-   * unrecognised — not an auth failure. HTTP 401/403 still throw the
-   * existing typed errors so the caller can distinguish a credential
-   * problem from unsupported credit data.
-   *
-   * Monetary amounts are decimal strings (no Number coercion); the
-   * caller preserves them for exact display.
-   */
-  async getInferenceAccount() {
-    const raw = await this.getJson("/api/inference/account");
-    if (!isUsageAccount(raw))
-      return null;
-    const { plan, included, prepaid, admission } = raw;
-    return {
-      schemaVersion: raw.schemaVersion,
-      snapshotAt: raw.snapshotAt,
-      scope: { orgId: raw.scope.orgId },
-      state: raw.state,
-      reason: raw.reason,
-      plan: {
-        id: plan.id,
-        name: plan.name,
-        monthlyPriceUsd: plan.monthlyPriceUsd
-      },
-      included: {
-        state: included.state,
-        usedPercent: included.usedPercent,
-        resetsAt: included.resetsAt
-      },
-      prepaid: {
-        balanceUsd: prepaid.balanceUsd,
-        fallbackEnabled: prepaid.fallbackEnabled
-      },
-      canManageBilling: raw.canManageBilling,
-      admission: { eligible: admission.eligible, reason: admission.reason }
-    };
-  }
-  /**
-   * Fetch request-level usage metadata for the operator's hosted
-   * inference sessions. Returns lightweight metadata records (model,
-   * tokens, provider, timestamp) — no prompt/response payload.
-   */
-  async getInferenceUsage() {
-    return this.getJson("/api/inference/usage");
-  }
-  // ── Audit-skills helpers (#audit-skills) ──
-  /** List all audit skills for the authenticated organization. */
-  async listAuditSkills() {
-    return this.getJson("/api/audit-skills");
-  }
-  /**
-   * Get a single audit skill with its revision history and project
-   * assignments.
-   */
-  async getAuditSkill(id) {
-    return this.getJson(`/api/audit-skills/${encodeURIComponent(id)}`);
-  }
-  /** Create a new markdown-based audit skill. */
-  async createAuditSkill(input) {
-    return this.postJson("/api/audit-skills", input);
-  }
-  /** Import an audit skill from a GitHub repository. */
-  async importAuditSkillFromGithub(input) {
-    return this.postJson("/api/audit-skills/import", input);
-  }
-  /** Create a new revision of an audit skill (CAS — 409 on stale expectedRevision). */
-  async createAuditSkillRevision(id, input) {
-    return this.postJson(`/api/audit-skills/${encodeURIComponent(id)}/revisions`, input);
-  }
-  /** Re-fetch the skill from its original GitHub source. */
-  async syncAuditSkill(id, expectedRevision) {
-    return this.postJson(`/api/audit-skills/${encodeURIComponent(id)}/sync`, { expectedRevision });
-  }
-  /** Pin an audit skill revision to a project for future scans. */
-  async assignAuditSkill(id, projectId, revisionId) {
-    return this.postJson(`/api/audit-skills/${encodeURIComponent(id)}/projects/${encodeURIComponent(projectId)}`, { revisionId });
-  }
-  /** Unpin an audit skill from a project (future scans no longer use it). */
-  async unassignAuditSkill(id, projectId) {
-    return this.deleteJson(`/api/audit-skills/${encodeURIComponent(id)}/projects/${encodeURIComponent(projectId)}`);
-  }
-  /** Archive an audit skill (disables future bindings, preserves history). */
-  async archiveAuditSkill(id) {
-    await this.deleteJson(`/api/audit-skills/${encodeURIComponent(id)}`);
-  }
-  /**
-   * List audit skills available to a specific project, along with current
-   * assignments for that project. The project must belong to the caller's org.
-   */
-  async listAuditSkillsByProject(projectId) {
-    return this.getJson(`/api/audit-skills/by-project/${encodeURIComponent(projectId)}`);
-  }
-  /**
-   * Generic JSON DELETE helper with the same error mapping as getJson/postJson.
-   * Used by `0 service disconnect` to remove scan schedules.
-   */
-  async deleteJson(path) {
-    const url = `${this.host}${path}`;
-    let res;
-    try {
-      res = await this.fetchImpl(url, {
-        method: "DELETE",
-        headers: { ...this.headers(), "Content-Type": "application/json" }
-      });
-    } catch (err) {
-      const msg = err instanceof Error ? err.message : String(err);
-      throw new CloudNetworkError(this.scrub(msg), path);
-    }
-    if (!res.ok) {
-      let code;
-      try {
-        const parsed = await res.json();
-        const raw = typeof parsed?.error === "object" ? parsed.error?.code : void 0;
-        if (typeof raw === "string" && raw.length > 0)
-          code = raw;
-      } catch {
-      }
-      this.throwForStatus(res.status, path, code);
-    }
-    if (res.status === 204)
-      return void 0;
-    return await res.json();
-  }
-  /**
-   * Generic JSON POST helper with the same error mapping as getJson.
-   * Used by `0 connect` to enqueue scans and schedules.
-   */
-  async postJson(path, body) {
-    const url = `${this.host}${path}`;
-    let res;
-    try {
-      res = await this.fetchImpl(url, {
-        method: "POST",
-        headers: { ...this.headers(), "Content-Type": "application/json" },
-        body: JSON.stringify(body)
-      });
-    } catch (err) {
-      const msg = err instanceof Error ? err.message : String(err);
-      throw new CloudNetworkError(this.scrub(msg), path);
-    }
-    if (!res.ok) {
-      let code;
-      try {
-        const parsed = await res.json();
-        const raw = typeof parsed?.error === "object" ? parsed.error?.code : void 0;
-        if (typeof raw === "string" && raw.length > 0)
-          code = raw;
-      } catch {
-      }
-      this.throwForStatus(res.status, path, code);
-    }
-    return await res.json();
-  }
-  /**
-   * Generic JSON GET helper. Public so future modules (scans, findings)
-   * can reuse the same error mapping without duplicating it. Not exported
-   * past the package boundary — see ./index.ts.
-   */
-  async getJson(path) {
-    const url = `${this.host}${path}`;
-    let res;
-    try {
-      res = await this.fetchImpl(url, {
-        method: "GET",
-        headers: this.headers()
-      });
-    } catch (err) {
-      const msg = err instanceof Error ? err.message : String(err);
-      throw new CloudNetworkError(this.scrub(msg), path);
-    }
-    if (!res.ok) {
-      let code;
-      try {
-        const body = await res.json();
-        const raw = body?.error?.code;
-        if (typeof raw === "string" && raw.length > 0)
-          code = raw;
-      } catch {
-      }
-      this.throwForStatus(res.status, path, code);
-    }
-    return await res.json();
-  }
-  /** Throw a typed error for a non-2xx status, carrying the gateway's body `code`. */
-  throwForStatus(status, path, code) {
-    if (status === 401)
-      throw new CloudUnauthorizedError(path);
-    if (status === 403)
-      throw new CloudForbiddenError(path);
-    throw new CloudError(`0-cloud request failed (HTTP ${status}${code ? ` ${code}` : ""}) on ${path}.`, status, path, code);
-  }
-  /**
-   * Throw a typed error for non-2xx responses. Public so direct callers
-   * (e.g. an integration test driving raw fetch) can reuse the mapping.
-   */
-  assertOk(res, path) {
-    if (res.ok)
-      return;
-    if (res.status === 401)
-      throw new CloudUnauthorizedError(path);
-    if (res.status === 403)
-      throw new CloudForbiddenError(path);
-    throw new CloudError(`0-cloud request failed (HTTP ${res.status}) on ${path}.`, res.status, path);
-  }
-  // ── internals ──
-  headers() {
-    return {
-      Authorization: `Bearer ${this.token}`,
-      Accept: "application/json",
-      "User-Agent": `@0/cli/${VERSION}`
-    };
-  }
-  /**
-   * Strip anything that looks like our own token from a string. The
-   * cloud token may be interpolated into a TLS-layer error message in
-   * exotic failure modes — we redact it to keep the no-leak invariant
-   * local to this module.
-   */
-  scrub(s) {
-    if (!this.token)
-      return s;
-    return s.split(this.token).join("[REDACTED]");
-  }
-};
-
 // packages/core/dist/runtime/llm-api.js
-import { randomUUID } from "node:crypto";
-import { appendFileSync, existsSync, readFileSync as readFileSync2, renameSync, writeFileSync } from "node:fs";
+import { createHash, randomUUID } from "node:crypto";
+import { appendFileSync, existsSync, readFileSync, renameSync, writeFileSync } from "node:fs";
 import { homedir } from "node:os";
-import { join as join2 } from "node:path";
-
-// packages/core/dist/runtime/hosted-request-queue.js
-var MAX_IN_FLIGHT = 4;
-var endpoints = /* @__PURE__ */ new Map();
-function acquireHostedRequestSlot(endpoint, token, signal) {
-  signal?.throwIfAborted();
-  let accounts = endpoints.get(endpoint);
-  if (!accounts)
-    endpoints.set(endpoint, accounts = /* @__PURE__ */ new Map());
-  let queue = accounts.get(token);
-  if (!queue)
-    accounts.set(token, queue = { active: 0, waiting: [] });
-  const accountQueues = accounts;
-  const current = queue;
-  return new Promise((resolve, reject) => {
-    const abort = () => {
-      const index = current.waiting.indexOf(waiter);
-      if (index !== -1)
-        current.waiting.splice(index, 1);
-      reject(signal.reason);
-    };
-    const waiter = { grant: () => {
-      signal?.removeEventListener("abort", abort);
-      current.active++;
-      let released = false;
-      resolve(() => {
-        if (released)
-          return;
-        released = true;
-        current.active--;
-        const next = current.waiting.shift();
-        if (next)
-          next.grant();
-        else if (current.active === 0) {
-          accountQueues.delete(token);
-          if (accountQueues.size === 0)
-            endpoints.delete(endpoint);
-        }
-      });
-    } };
-    if (current.active < MAX_IN_FLIGHT)
-      waiter.grant();
-    else {
-      current.waiting.push(waiter);
-      signal?.addEventListener("abort", abort, { once: true });
-    }
-  });
-}
+import { join } from "node:path";
 
 // packages/core/dist/runtime/prompt-cache.js
 var MAX_CACHE_BREAKPOINTS = 4;
@@ -2085,8 +1555,7 @@ function parseLlmFallbackChain(env2 = process.env) {
     xai: true,
     opencode: true,
     copilot: true,
-    google: true,
-    hosted: true
+    google: true
   };
   for (const part of raw.split(",")) {
     const trimmed = part.trim();
@@ -2192,24 +1661,6 @@ function resolveFailoverProvider(provider, model, env2 = process.env, apiKey) {
         return void 0;
       return { apiKey: "", baseUrl: CODE_ASSIST_ENDPOINT, wireApi: "google_generate_content" };
     }
-    case "hosted": {
-      try {
-        const creds = loadCloudCredentials({
-          env: env2,
-          warn: () => {
-          }
-        });
-        return {
-          apiKey: creds.token,
-          baseUrl: `${creds.host}/api/inference/v1`,
-          wireApi: "chat_completions"
-        };
-      } catch (err) {
-        if (err instanceof CloudAuthMissingError)
-          return void 0;
-        throw err;
-      }
-    }
   }
 }
 var fallbackChainCache;
@@ -2235,6 +1686,12 @@ var KIMI_DEFAULT_MODEL = "k3";
 var QWEN_DEFAULT_BASE_URL = "https://token-plan.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1";
 var QWEN_DEFAULT_MODEL = "qwen3.8-max";
 var CODEX_API_ENDPOINT = "https://chatgpt.com/backend-api/codex/responses";
+var CODEX_CLIENT_VERSION = "0.153.0";
+var CODEX_PROTOCOL_HEADERS = {
+  "OpenAI-Beta": "responses=experimental",
+  originator: "0",
+  version: CODEX_CLIENT_VERSION
+};
 var CODEX_OAUTH_ISSUER = "https://auth.openai.com";
 var CODEX_OAUTH_CLIENT_ID = "app_EMoamEEZ73f0CkXaXp7hrann";
 var CODEX_DEFAULT_MODEL = "gpt-5.5";
@@ -2258,13 +1715,13 @@ function readChatGptCodexEnv(env2 = process.env) {
   };
 }
 function resolveChatGptCodexAuthPath(env2 = process.env) {
-  return env2["ZERO_CHATGPT_AUTH_FILE"] ?? join2(env2.HOME ?? homedir(), ".codex", "auth.json");
+  return env2["ZERO_CHATGPT_AUTH_FILE"] ?? join(env2.HOME ?? homedir(), ".codex", "auth.json");
 }
 function persistChatGptCodexAuthFile(authPath, tokens, usedRefreshToken) {
   try {
     let existing = {};
     try {
-      existing = JSON.parse(readFileSync2(authPath, "utf8"));
+      existing = JSON.parse(readFileSync(authPath, "utf8"));
     } catch {
       existing = {};
     }
@@ -2296,7 +1753,7 @@ function readChatGptCodexAuthFile(env2 = process.env) {
   if (!existsSync(authPath))
     return void 0;
   try {
-    const auth = JSON.parse(readFileSync2(authPath, "utf8"));
+    const auth = JSON.parse(readFileSync(authPath, "utf8"));
     const tokens = auth.tokens;
     if (!tokens)
       return void 0;
@@ -2642,7 +2099,7 @@ function parseCodexAzureConfig(env2 = process.env) {
   if (!existsSync(configPath))
     return {};
   try {
-    const content = readFileSync2(configPath, "utf8");
+    const content = readFileSync(configPath, "utf8");
     const azureSectionMatch = content.match(/\[model_providers\.azure\]([\s\S]*?)(?:\n\[|$)/);
     const activeProviderMatch = content.match(/^\s*model_provider\s*=\s*"([^"]+)"/m);
     const baseUrlMatch = azureSectionMatch?.[1]?.match(/base_url\s*=\s*"([^"]+)"/);
@@ -2698,7 +2155,7 @@ function providerForModel(model, env2) {
     return env2["ZERO_GEMINI_ACCESS_TOKEN"] || env2["ZERO_GEMINI_OAUTH_REFRESH_TOKEN"] ? "google" : void 0;
   }
   if (/^gpt-|^o[1-4](?:[-_]|$)/.test(m)) {
-    if (env2["ZERO_CHATGPT_ACCESS_TOKEN"] || env2["ZERO_CHATGPT_OAUTH_REFRESH_TOKEN"])
+    if (readChatGptCodexEnv(env2) || readChatGptCodexAuthFile(env2))
       return "chatgpt-codex";
     if (env2.OPENAI_API_KEY)
       return "openai";
@@ -2726,8 +2183,7 @@ var DEFAULT_PROVIDER_MODELS = {
   xai: XAI_DEFAULT_MODEL,
   opencode: OPENCODE_DEFAULT_MODEL,
   copilot: COPILOT_DEFAULT_MODEL,
-  google: GEMINI_DEFAULT_MODEL,
-  hosted: ""
+  google: GEMINI_DEFAULT_MODEL
 };
 var AUTO_MODEL_SENTINEL = "auto";
 function detectProvider(configApiKey, preferredModel, env2, configProvider) {
@@ -2748,11 +2204,11 @@ function detectProvider(configApiKey, preferredModel, env2, configProvider) {
       throw new Error(`${source} is unsupported: ${pinnedProviderRaw}`);
     }
     const provider = pinnedProviderRaw;
-    const model = preferredModel ?? env2["ZERO_MODEL"] ?? (configProvider !== void 0 || provider === "hosted" ? DEFAULT_PROVIDER_MODELS[provider] : void 0);
-    if (model === void 0 || model === "" && provider !== "hosted") {
+    const model = preferredModel ?? env2["ZERO_MODEL"] ?? (configProvider !== void 0 ? DEFAULT_PROVIDER_MODELS[provider] : void 0);
+    if (model === void 0 || model === "") {
       throw new Error(`${source} requires an explicit model`);
     }
-    if (configApiKey && (provider === "hosted" || provider === "chatgpt-codex")) {
+    if (configApiKey && provider === "chatgpt-codex") {
       throw new Error(`${source}=${provider} requires its own authentication, not RuntimeConfig.apiKey`);
     }
     const resolved = resolveFailoverProvider(provider, model, env2, configApiKey);
@@ -2906,21 +2362,6 @@ function detectProvider(configApiKey, preferredModel, env2, configProvider) {
     default:
       break;
   }
-  try {
-    const hostedCreds = loadCloudCredentials({
-      env: env2,
-      warn: () => {
-      }
-    });
-    return {
-      provider: "hosted",
-      apiKey: hostedCreds.token,
-      baseUrl: `${hostedCreds.host}/api/inference/v1`,
-      defaultModel: "",
-      wireApi: "chat_completions"
-    };
-  } catch {
-  }
   const chatGptAccess = env2["ZERO_CHATGPT_ACCESS_TOKEN"];
   const chatGptRefresh = env2["ZERO_CHATGPT_OAUTH_REFRESH_TOKEN"];
   const chatGptAuthFile = !chatGptAccess && !chatGptRefresh ? readChatGptCodexAuthFile(env2) : void 0;
@@ -3084,6 +2525,9 @@ var LlmApiRuntime = class _LlmApiRuntime {
   apiKey;
   baseUrl;
   model;
+  pricingModelSource = "";
+  pricingProviderSource;
+  pricingModelKey = "";
   wireApi;
   reasoningEffort;
   azureConfig;
@@ -3092,19 +2536,12 @@ var LlmApiRuntime = class _LlmApiRuntime {
   fallbackChain;
   /** Index into fallbackChain — which entry to try next. */
   fallbackIndex;
-  /** Resolve and validate the hosted model and wire protocol once per runtime. */
-  hostedCatalogPromise = null;
-  /** Catalog ceiling, resolved before hosted inference is submitted. */
-  hostedMaxOutputTokens;
   constructor(config, inherited) {
     if (inherited) {
       const timeout = config.timeout ?? inherited.config.timeout ?? 12e4;
       if (!Number.isFinite(timeout) || timeout <= 0) {
         throw new Error("Subagent timeout must be a positive finite number");
       }
-      if (inherited.provider === "hosted" && inherited.hostedMaxOutputTokens === void 0) {
-        throw new Error("Hosted model catalog must resolve before creating a subagent");
-      }
       const model = config.model ?? inherited.model;
       const modelChanged = model !== inherited.model;
       this.config = {
@@ -3138,10 +2575,6 @@ var LlmApiRuntime = class _LlmApiRuntime {
         this.geminiAuthState = inherited.geminiAuthState;
       this.fallbackChain = [];
       this.fallbackIndex = 0;
-      if (!modelChanged) {
-        this.hostedMaxOutputTokens = inherited.hostedMaxOutputTokens;
-        this.hostedCatalogPromise = inherited.hostedCatalogPromise;
-      }
       return;
     }
     this.applyConfiguration(config);
@@ -3201,66 +2634,59 @@ var LlmApiRuntime = class _LlmApiRuntime {
       });
     }
   }
+  /** Non-secret identity for preferences scoped to the captured connection. */
+  connectionIdentity() {
+    let identity;
+    if (this.provider === "chatgpt-codex") {
+      const state = this.codexAuthState;
+      if (!state)
+        return void 0;
+      if (state.accountId) {
+        identity = [this.provider, "account", state.accountId];
+      } else {
+        const credential = state.refreshToken || state.accessToken;
+        if (!credential)
+          return void 0;
+        identity = [this.provider, "credential", credential];
+      }
+    } else if (this.provider === "google") {
+      const credential = this.geminiAuthState?.refreshToken || this.geminiAuthState?.accessToken;
+      if (!credential)
+        return void 0;
+      identity = [this.provider, this.baseUrl, credential, firstNonEmptyEnv(this.env, "GOOGLE_CLOUD_PROJECT", "ZERO_GEMINI_PROJECT") ?? ""];
+    } else {
+      if (!this.apiKey)
+        return void 0;
+      identity = [this.provider, this.baseUrl, this.apiKey];
+    }
+    return createHash("sha256").update(JSON.stringify(identity)).digest("hex");
+  }
   /** Discover models using this runtime's captured account, including after a separate login changes. */
   async codexModelCatalog(signal) {
     const state = this.codexAuthState;
     if (this.provider !== "chatgpt-codex" || !state)
       throw new Error("No active Codex subscription");
-    const { loadCodexModelCatalog } = await import("./codex-models-TYYV4JNN.js");
+    const { loadCodexModelCatalog } = await import("./codex-models-NHQW45SO.js");
     return loadCodexModelCatalog({ signal, resolveCredentials: () => refreshChatGptCodexAuthState(state) });
   }
-  /** Check account admission and resolve the service model without inference.
-   * Each explicit check refreshes admission; failed discovery is never cached.
-   */
-  async prepare() {
-    while (this.provider === "hosted") {
-      const config = this.config;
-      const client = new CloudClient({
-        host: this.baseUrl.replace(/\/api\/inference\/v1$/, ""),
-        token: this.apiKey
-      });
-      try {
-        const account = await client.getInferenceAccount();
-        if (this.config !== config)
-          continue;
-        if (!account) {
-          throw new CloudError("0cloud account availability could not be read. Check again or review your account in /connect.", void 0, "/api/inference/account", "unsupported_account_data");
-        }
-        if (!account.admission.eligible) {
-          const reason = account.admission.reason ?? account.reason ?? "account_restricted";
-          throw new CloudError(account.state === "unavailable" ? `0cloud account availability could not be checked (${reason}). Check again or review your account in /connect.` : `0cloud account access is restricted (${reason}). Review your account in /connect or contact your organization owner.`, void 0, "/api/inference/account", reason);
-        }
-        await this.ensureHostedModel();
-        if (this.config === config)
-          return;
-      } catch (error) {
-        if (this.config !== config)
-          continue;
-        throw error;
-      }
-    }
-  }
   /**
    * Mutate the live selection in place so the NEXT turn (the engine reads
    * `config.runtime` per turn) and the NEXT `forkForSubagent` pick up the new
-   * model / provider / role map with zero session teardown. No-op-safe:
-   * undefined fields leave the corresponding state unchanged.
+   * model / provider / role map with zero session teardown. Undefined fields
+   * preserve the current state, except a provider change resets an omitted model.
    */
   reconfigure(sel) {
     const providerChanged = sel.provider !== void 0 && sel.provider !== this.provider;
-    if (providerChanged || this.provider === "hosted" && (sel.env !== void 0 || sel.provider === "hosted")) {
+    if (providerChanged) {
       const merged = {
         ...this.config,
-        ...providerChanged && sel.provider === "hosted" && sel.model === void 0 ? { model: "" } : {},
         apiKey: void 0,
         provider: sel.provider ?? this.provider,
-        ...sel.model !== void 0 ? { model: sel.model } : {},
+        model: sel.model,
         ...sel.agentModels !== void 0 ? { agentModels: sel.agentModels } : {},
         ...sel.singleModel !== void 0 ? { singleModel: sel.singleModel } : {},
         ...sel.env !== void 0 ? { env: sel.env } : {}
       };
-      this.hostedCatalogPromise = null;
-      this.hostedMaxOutputTokens = void 0;
       this.applyConfiguration(merged);
       return;
     }
@@ -3283,15 +2709,13 @@ var LlmApiRuntime = class _LlmApiRuntime {
           this.wireApi = openAICompatibleWireApi(this.env, "AZURE_OPENAI_WIRE_API", this.azureConfig.wireApi);
         this.applyModelWireApi();
         this.reasoningEffort = void 0;
-        this.hostedCatalogPromise = null;
-        this.hostedMaxOutputTokens = void 0;
       }
     }
   }
-  /** Exact deployments requiring Responses for tools, shared by roots and forks. */
+  /** Modern OpenAI tool calls use Responses, shared by roots and forks. */
   applyModelWireApi() {
     const normalizedModel = this.model.toLowerCase();
-    if (this.wireApi === "chat_completions" && (this.provider === "azure" && normalizedModel === "gpt-5.6-sol" || this.provider === "openai" && normalizedModel === "gpt-5.6-luna")) {
+    if (this.wireApi === "chat_completions" && (this.provider === "azure" && normalizedModel === "gpt-5.6-sol" || this.provider === "openai" && (normalizedModel === "gpt-5.6-luna" || /^gpt-(?:[6-9]|\d{2,})(?:[-.]|$)/.test(normalizedModel)))) {
       this.wireApi = "responses";
     }
   }
@@ -3300,18 +2724,14 @@ var LlmApiRuntime = class _LlmApiRuntime {
    * Providers whose credentials are present in THIS runtime's environment
    * (`this.env`, never process-global). A provider counts as accessible iff
    * `resolveFailoverProvider` — the same auth-presence check the cross-provider
-   * failover chain uses — can build a connection for it, so auth-only providers
-   * (chatgpt-codex OAuth) and cloud-hosted are covered by the identical rule.
+   * failover chain uses — including auth-only providers (chatgpt-codex OAuth).
    */
   accessibleProviders() {
     const out = [];
     for (const provider of Object.keys(DEFAULT_PROVIDER_MODELS)) {
       const probeModel = DEFAULT_PROVIDER_MODELS[provider] || "probe";
-      try {
-        if (resolveFailoverProvider(provider, probeModel, this.env))
-          out.push(provider);
-      } catch {
-      }
+      if (resolveFailoverProvider(provider, probeModel, this.env))
+        out.push(provider);
     }
     return out;
   }
@@ -3348,7 +2768,6 @@ var LlmApiRuntime = class _LlmApiRuntime {
     if (!Number.isFinite(timeoutMs) || timeoutMs <= 0) {
       throw new Error("Subagent timeout must be a positive finite number");
     }
-    await this.ensureHostedModel();
     const roleModel = selection?.role !== void 0 && this.config.agentModels && Object.hasOwn(this.config.agentModels, selection.role) ? this.config.agentModels[selection.role] : void 0;
     const autoInEffect = roleModel === AUTO_MODEL_SENTINEL || this.config.autoRoute === true && roleModel === void 0;
     const configuredModel = roleModel === AUTO_MODEL_SENTINEL ? void 0 : roleModel;
@@ -3364,70 +2783,15 @@ var LlmApiRuntime = class _LlmApiRuntime {
       throw new Error(autoInEffect ? `Subagent model "${selectedModel}" is not reachable: its provider has no configured credentials.` : `Subagent model "${selectedModel}" is not operator-approved. Configure agentModels before selecting it.`);
     }
     const child = new _LlmApiRuntime({ type: "api", timeout: timeoutMs, model }, this);
-    await child.ensureHostedModel();
     return child;
   }
-  /** The server catalog is authoritative even when a model was selected explicitly. */
-  async ensureHostedModel() {
-    while (this.provider === "hosted") {
-      let pending = this.hostedCatalogPromise;
-      if (!pending) {
-        const requestedModel = this.model;
-        const client = new CloudClient({
-          host: this.baseUrl.replace(/\/api\/inference\/v1$/, ""),
-          token: this.apiKey
-        });
-        const request = client.getInferenceModels().then((catalog) => {
-          if (this.hostedCatalogPromise !== request)
-            return;
-          const selected = requestedModel ? catalog.data.find((model) => model.id === requestedModel) : catalog.data[0];
-          if (!selected) {
-            throw new Error(requestedModel ? `Hosted model "${requestedModel}" is unavailable. Run \`0 models\` for available models.` : "No hosted models are available. Run `0 models` to check service availability.");
-          }
-          this.model = selected.id;
-          this.wireApi = selected.wire_api;
-          this.hostedMaxOutputTokens = selected.max_output_tokens;
-        });
-        this.hostedCatalogPromise = pending = request;
-      }
-      try {
-        await pending;
-      } catch (error) {
-        if (this.hostedCatalogPromise !== pending)
-          continue;
-        this.hostedCatalogPromise = null;
-        throw error;
-      }
-      if (this.hostedCatalogPromise === pending)
-        return;
-    }
-  }
-  /** All audits and nested workers on this hosted credential share admission. */
-  async acquireHostedSlot(signal) {
-    while (this.provider === "hosted") {
-      const config = this.config;
-      const release = await acquireHostedRequestSlot(this.baseUrl, this.apiKey, signal);
-      if (this.config === config)
-        return release;
-      release();
-      await this.ensureHostedModel();
-    }
-    return void 0;
-  }
   /**
    * A hard dollar ceiling needs a provider-enforced bound on the next response.
    * ChatGPT Codex OAuth rejects `max_output_tokens`, so it cannot support that
    * contract; callers must fail closed before making a metered comparison call.
    */
   get outputTokenLimit() {
-    return this.provider === "chatgpt-codex" ? void 0 : this.effectiveOutputTokens;
-  }
-  /** Hosted requests honor both the catalog ceiling and the local hard cap. */
-  get effectiveOutputTokens() {
-    if (this.provider === "hosted" && this.hostedMaxOutputTokens !== void 0) {
-      return Math.min(NATIVE_COMPLETION_TOKEN_LIMIT, this.hostedMaxOutputTokens);
-    }
-    return NATIVE_COMPLETION_TOKEN_LIMIT;
+    return this.provider === "chatgpt-codex" ? void 0 : NATIVE_COMPLETION_TOKEN_LIMIT;
   }
   /**
    * Whether this provider uses OpenAI-compatible chat/completions format.
@@ -3439,7 +2803,7 @@ var LlmApiRuntime = class _LlmApiRuntime {
    * inert default — it is intentionally unused for these two providers.
    */
   get isOpenAICompat() {
-    return this.provider === "openrouter" || this.provider === "openai" || this.provider === "azure" || this.provider === "deepseek" || this.provider === "qwen" || this.provider === "xai" || this.provider === "copilot" || this.provider === "hosted" || this.provider === "opencode" && (this.wireApi === "chat_completions" || this.wireApi === "responses") || // chatgpt-codex always speaks Responses API; treat it as
+    return this.provider === "openrouter" || this.provider === "openai" || this.provider === "azure" || this.provider === "deepseek" || this.provider === "qwen" || this.provider === "xai" || this.provider === "copilot" || this.provider === "opencode" && (this.wireApi === "chat_completions" || this.wireApi === "responses") || // chatgpt-codex always speaks Responses API; treat it as
     // OpenAI-compat for body-shape branching purposes (the Responses
     // wire-API code paths below already key on `wireApi === "responses"`
     // and produce a body codex's backend accepts as-is).
@@ -3485,12 +2849,24 @@ var LlmApiRuntime = class _LlmApiRuntime {
   resolvedModel() {
     return this.model;
   }
+  /** Price discovered IDs on their actual route without changing the wire model. */
+  resolvedPricingModel() {
+    if (this.provider === "chatgpt-codex" || this.provider === "copilot" || this.provider === "google")
+      return this.model;
+    if (this.pricingModelSource !== this.model || this.pricingProviderSource !== this.provider) {
+      this.pricingModelSource = this.model;
+      this.pricingProviderSource = this.provider;
+      this.pricingModelKey = `${this.provider}/${this.model}`;
+    }
+    return this.pricingModelKey;
+  }
   /** Build the appropriate headers for the configured provider. */
   buildHeaders() {
     if (this.provider === "chatgpt-codex") {
       return {
         "Content-Type": "application/json",
-        originator: "0",
+        ...CODEX_PROTOCOL_HEADERS,
+        "x-codex-routing-hint": `model=${this.model}`,
         "User-Agent": `0/${VERSION}`
       };
     }
@@ -3753,12 +3129,10 @@ var LlmApiRuntime = class _LlmApiRuntime {
         return "GitHub Copilot";
       case "google":
         return "Google Gemini (Code Assist)";
-      case "hosted":
-        return "0.security Cloud";
     }
   }
   noKeyError() {
-    return "No provider credential found. Set one of:\n  env ZERO_CHATGPT_OAUTH_REFRESH_TOKEN=... 0 <command> (ChatGPT Codex subscription auth)\n  export OPENROUTER_API_KEY=sk-or-...   (OpenRouter \u2014 many models, one key)\n  export DEEPSEEK_API_KEY=...           (DeepSeek \u2014 direct Flash 0731 inference)\n  export ANTHROPIC_API_KEY=sk-ant-...    (Anthropic \u2014 direct Claude access)\n  export AZURE_OPENAI_API_KEY=...        (Azure OpenAI \u2014 reuse your Codex Azure provider)\n  export OPENAI_API_KEY=sk-...           (OpenAI \u2014 direct GPT access)\n  export Z_AI_API_KEY=...                (Z.ai GLM \u2014 flat-rate Coding Plan, Anthropic-compatible)\n  export KIMI_API_KEY=...                (Moonshot Kimi K3 \u2014 flat-rate coding, Anthropic-compatible)\n  export QWEN_API_KEY=...                (Alibaba Qwen \u2014 Token Plan sub, OpenAI-compatible)\n  export XAI_API_KEY=...                 (xAI Grok \u2014 OpenAI-compatible)\n  export OPENCODE_API_KEY=...            (OpenCode Zen \u2014 multi-wire gateway)\n  export ZERO_COPILOT_GITHUB_TOKEN=...   (GitHub Copilot \u2014 device-code OAuth token)\n  Run `0 login`                     (0 hosted inference)";
+    return "No provider credential found. Set one of:\n  env ZERO_CHATGPT_OAUTH_REFRESH_TOKEN=... 0 <command> (ChatGPT Codex subscription auth)\n  export OPENROUTER_API_KEY=sk-or-...   (OpenRouter \u2014 many models, one key)\n  export DEEPSEEK_API_KEY=...           (DeepSeek \u2014 direct Flash 0731 inference)\n  export ANTHROPIC_API_KEY=sk-ant-...    (Anthropic \u2014 direct Claude access)\n  export AZURE_OPENAI_API_KEY=...        (Azure OpenAI \u2014 reuse your Codex Azure provider)\n  export OPENAI_API_KEY=sk-...           (OpenAI \u2014 direct GPT access)\n  export Z_AI_API_KEY=...                (Z.ai GLM \u2014 flat-rate Coding Plan, Anthropic-compatible)\n  export KIMI_API_KEY=...                (Moonshot Kimi K3 \u2014 flat-rate coding, Anthropic-compatible)\n  export QWEN_API_KEY=...                (Alibaba Qwen \u2014 Token Plan sub, OpenAI-compatible)\n  export XAI_API_KEY=...                 (xAI Grok \u2014 OpenAI-compatible)\n  export OPENCODE_API_KEY=...            (OpenCode Zen \u2014 multi-wire gateway)\n  export ZERO_COPILOT_GITHUB_TOKEN=...   (GitHub Copilot \u2014 device-code OAuth token)";
   }
   getConfigurationDiagnostics() {
     if (!this.apiKey && this.provider !== "chatgpt-codex") {
@@ -3851,7 +3225,6 @@ Missing: ${missing.join("; ")}
       this.apiKey = cfg.apiKey;
       this.baseUrl = cfg.baseUrl;
       this.wireApi = cfg.wireApi;
-      this.hostedCatalogPromise = null;
       diag.warn("failover_engaged", `${reason} \u2014 failover to ${entry.provider} (${entry.model})`, { reason, provider: entry.provider, model: entry.model });
       return true;
     }
@@ -3865,9 +3238,8 @@ Missing: ${missing.join("; ")}
    * hammering the limit in lockstep).
    *
    * The body factory is valid only for the current provider and wire protocol.
-   * A null result signals failover: the caller resolves the hosted catalog and
-   * rebuilds the complete request under its existing timeout/cancellation signal.
-   *
+   * A null result signals failover so the caller rebuilds the complete request
+   * under its existing timeout/cancellation signal.
    * Retry + failover caps documented on `retryBackoffMs` / `llm429MaxRetries`.
    *
    * Headers are re-resolved per attempt (via ensureFreshHeaders → OAuth
@@ -3888,9 +3260,6 @@ Missing: ${missing.join("; ")}
         });
       } catch (error) {
         abort?.throwIfCancelled();
-        if (this.provider === "hosted") {
-          throw new Error("0 hosted request outcome is unknown. Automatic replay is disabled; check your inference usage before retrying.", { cause: error });
-        }
         const cause = error instanceof Error ? error.cause : void 0;
         const causeCode = cause && typeof cause === "object" && "code" in cause && typeof cause.code === "string" ? cause.code : "unknown";
         const causeHost = cause && typeof cause === "object" && "hostname" in cause && typeof cause.hostname === "string" ? `@${cause.hostname}` : "";
@@ -3917,13 +3286,6 @@ Missing: ${missing.join("; ")}
       if (res.ok || !isRetryableHttpStatus(res.status)) {
         return res;
       }
-      if (this.provider === "hosted" && res.status === 429 && res.headers.get("x-0-retry-safe") !== "1") {
-        return res;
-      }
-      if (this.provider === "hosted" && res.status >= 500) {
-        await res.body?.cancel();
-        throw new Error(`0 hosted request returned HTTP ${res.status}; its outcome may be unknown. Automatic replay is disabled; check your inference usage before retrying.`);
-      }
       abort?.throwIfCancelled();
       const is429 = res.status === 429;
       let bodyText;
@@ -4014,7 +3376,6 @@ Missing: ${missing.join("; ")}
   }
   // ── Legacy Runtime interface (single-prompt) ──
   async execute(prompt, context) {
-    await this.ensureHostedModel();
     const start = Date.now();
     if (!this.apiKey && this.provider !== "chatgpt-codex" && this.provider !== "google") {
       return {
@@ -4026,15 +3387,11 @@ Missing: ${missing.join("; ")}
       };
     }
     const systemPrompt = context?.systemPrompt ?? "";
-    let releaseHostedSlot = this.provider === "hosted" ? await this.acquireHostedSlot() : void 0;
     const controller = new AbortController();
     const timer = setTimeout(() => controller.abort(), this.config.timeout || 12e4);
     try {
       let res;
       do {
-        if (this.provider === "hosted" && !releaseHostedSlot) {
-          releaseHostedSlot = await this.acquireHostedSlot(controller.signal);
-        }
         if (this.isOpenAICompat && this.wireApi === "chat_completions") {
           const messages = [];
           if (systemPrompt) {
@@ -4043,7 +3400,7 @@ Missing: ${missing.join("; ")}
           messages.push({ role: "user", content: prompt });
           res = await this.postWithRetry(() => JSON.stringify({
             model: this.model,
-            [this.maxTokensParamKey]: this.effectiveOutputTokens,
+            [this.maxTokensParamKey]: NATIVE_COMPLETION_TOKEN_LIMIT,
             messages,
             // See executeNative: explicit reasoning_effort passthrough only.
             ...this.reasoningEffort ? { reasoning_effort: this.reasoningEffort } : {}
@@ -4064,7 +3421,7 @@ Missing: ${missing.join("; ")}
           res = await this.postWithRetry(() => JSON.stringify({
             model: this.model,
             input,
-            ...isCodex ? { store: false } : { max_output_tokens: this.effectiveOutputTokens }
+            ...isCodex ? { store: false } : { max_output_tokens: NATIVE_COMPLETION_TOKEN_LIMIT }
           }), controller.signal);
         } else if (this.isGeminiCodeAssist) {
           const request = {
@@ -4091,8 +3448,6 @@ Missing: ${missing.join("; ")}
         } else {
           throw new Error(`execute: provider ${this.provider} is not mapped to a wire`);
         }
-        if (!res)
-          await this.ensureHostedModel();
       } while (!res);
       clearTimeout(timer);
       const body = await res.text();
@@ -4162,8 +3517,6 @@ Missing: ${missing.join("; ")}
         durationMs: Date.now() - start,
         error: timedOut ? `${this.providerLabel} API request timed out` : `${this.providerLabel} API error: ${msg}`
       };
-    } finally {
-      releaseHostedSlot?.();
     }
   }
   // ── Native Runtime interface (structured messages + tool_use) ──
@@ -4177,14 +3530,7 @@ Missing: ${missing.join("; ")}
       error: `${this.providerLabel} request cancelled by operator`
     };
   }
-  /**
-   * Public entry point. Runs {@link executeNativeAttempt} and, ONLY for a
-   * transient empty stream (see {@link shouldRetryNativeStream}), re-issues the
-   * whole request up to {@link llmStreamMaxAttempts} times with a short backoff.
-   * Every other outcome — success, a real API error, a timeout, an operator
-   * cancellation — is returned from the first attempt untouched, preserving the
-   * existing behaviour exactly.
-   */
+  /** Public entry point. Retries a transient empty stream with bounded backoff. */
   async executeNative(system, messages, tools, callbacks, signal) {
     const maxAttempts = llmStreamMaxAttempts();
     let result;
@@ -4205,7 +3551,6 @@ Missing: ${missing.join("; ")}
     return result;
   }
   async executeNativeAttempt(system, messages, tools, callbacks, signal) {
-    await this.ensureHostedModel();
     const start = Date.now();
     if (!this.apiKey && this.provider !== "chatgpt-codex" && this.provider !== "google") {
       return {
@@ -4217,25 +3562,12 @@ Missing: ${missing.join("; ")}
     }
     if (signal?.aborted)
       return this.cancelledResult(start);
-    let releaseHostedSlot;
-    if (this.provider === "hosted") {
-      try {
-        releaseHostedSlot = await this.acquireHostedSlot(signal);
-      } catch (error) {
-        if (signal?.aborted)
-          return this.cancelledResult(start);
-        throw error;
-      }
-    }
     const controller = new AbortController();
     const timer = setTimeout(() => controller.abort(), this.config.timeout || 12e4);
     const call = composeCallAbort(controller.signal, signal);
     try {
       let res;
       do {
-        if (this.provider === "hosted" && !releaseHostedSlot) {
-          releaseHostedSlot = await this.acquireHostedSlot(call.signal);
-        }
         if (this.isOpenAICompat && this.wireApi === "chat_completions") {
           const chatMessages = [];
           chatMessages.push({ role: "system", content: system });
@@ -4283,7 +3615,7 @@ Missing: ${missing.join("; ")}
           }
           const body = {
             model: this.model,
-            [this.maxTokensParamKey]: this.effectiveOutputTokens,
+            [this.maxTokensParamKey]: NATIVE_COMPLETION_TOKEN_LIMIT,
             messages: chatMessages
           };
           if (this.reasoningEffort) {
@@ -4351,7 +3683,7 @@ Missing: ${missing.join("; ")}
           const body = {
             model: this.model,
             input,
-            ...isCodex ? { store: false, instructions: system } : { max_output_tokens: this.effectiveOutputTokens },
+            ...isCodex ? { store: false, instructions: system } : { max_output_tokens: NATIVE_COMPLETION_TOKEN_LIMIT },
             ...reasoningEffort ? {
               reasoning: {
                 effort: reasoningEffort,
@@ -4410,10 +3742,8 @@ Missing: ${missing.join("; ")}
             }
           }
           res = await this.postWithRetry(() => JSON.stringify({ ...body, stream: true, model: this.model }), call.signal, call);
-          if (!res) {
-            await this.ensureHostedModel();
+          if (!res)
             continue;
-          }
           if (!res.ok) {
             const responseText2 = await res.text();
             clearTimeout(timer);
@@ -4522,8 +3852,6 @@ Missing: ${missing.join("; ")}
         } else {
           throw new Error(`executeNative: provider ${this.provider} is not mapped to a wire`);
         }
-        if (!res)
-          await this.ensureHostedModel();
       } while (!res);
       const responseText = await res.text();
       clearTimeout(timer);
@@ -4734,7 +4062,6 @@ Missing: ${missing.join("; ")}
         error: timedOut ? `${this.providerLabel} API request timed out` : `${this.providerLabel} API error: ${msg}`
       };
     } finally {
-      releaseHostedSlot?.();
       call.dispose();
     }
   }
@@ -5083,18 +4410,11 @@ export {
   isDiagnosticsClaimed,
   recentDiagnostics,
   _resetDiagnosticsForTests,
-  DEFAULT_CLOUD_HOST,
-  CloudAuthMissingError,
-  CloudAuthError,
-  loadCloudCredentials,
-  CloudError,
-  CloudUnauthorizedError,
-  CloudForbiddenError,
-  CloudNetworkError,
-  CloudClient,
   QuotaExhaustedError,
   OperatorAbortError,
   parseUsageLimitReached,
+  CODEX_CLIENT_VERSION,
+  CODEX_PROTOCOL_HEADERS,
   LOOP_SERVER_COMPACTION_TOKENS,
   getChatGptCodexAccessToken,
   LlmApiRuntime