@opengeni/codex 0.2.3 → 0.2.7

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@opengeni/codex",
-  "version": "0.2.3",
+  "version": "0.2.7",
   "description": "ChatGPT/Codex subscription auth + transport: device-code login, token refresh, and the Responses-backend fetch. Pure HTTP + transforms; no database dependency.",
   "license": "Apache-2.0",
   "repository": {

package/src/api-client.ts

@@ -3,6 +3,13 @@
 
 import { CODEX_ORIGINATOR, CODEX_RESPONSES_BASE, CODEX_WHAM_BASE } from "./constants";
 import type { CodexFetch } from "./device-code";
+import {
+  parseCodexRateLimitResetConsumeResponse,
+  parseCodexRateLimitResetCreditsDetails,
+  type CodexRateLimitResetConsumeResponse,
+  type CodexRateLimitResetCreditsDetails,
+} from "./reset-credits";
+import { runBoundedCodexOperation } from "./bounded-operation";
 
 export type CodexAuthHeaders = {
   accessToken: string;
@@ -11,6 +18,16 @@ export type CodexAuthHeaders = {
   clientVersion: string;
 };
 
+const CODEX_READ_TIMEOUT_MS = 5_000;
+const RESET_CREDIT_DETAILS_TIMEOUT_MS = 5_000;
+const RESET_CREDIT_CONSUME_TIMEOUT_MS = 10_000;
+
+export type ResetCreditFetchFailureReason =
+  | "http_error"
+  | "invalid_response"
+  | "network_error"
+  | "timeout";
+
 function subscriptionHeaders(a: CodexAuthHeaders): Record<string, string> {
   return {
     Authorization: `Bearer ${a.accessToken}`,
@@ -26,34 +43,126 @@ function subscriptionHeaders(a: CodexAuthHeaders): Record<string, string> {
 export async function fetchCodexModels(
   a: CodexAuthHeaders,
   fetchImpl: CodexFetch = fetch,
+  timeoutMs = CODEX_READ_TIMEOUT_MS,
 ): Promise<{ ok: boolean; status: number; slugs: string[] }> {
-  const res = await fetchImpl(
-    `${CODEX_RESPONSES_BASE}/models?client_version=${encodeURIComponent(a.clientVersion)}`,
-    {
-      method: "GET",
-      headers: subscriptionHeaders(a),
-    },
-  );
-  if (!res.ok) {
-    return { ok: false, status: res.status, slugs: [] };
-  }
-  const body = (await res.json()) as { models?: Array<{ slug?: string }> };
-  const slugs = (body.models ?? [])
-    .map((m) => m.slug)
-    .filter((s): s is string => typeof s === "string");
-  return { ok: true, status: res.status, slugs };
+  const fetched = await runBoundedCodexOperation(async (signal) => {
+    const res = await fetchImpl(
+      `${CODEX_RESPONSES_BASE}/models?client_version=${encodeURIComponent(a.clientVersion)}`,
+      { method: "GET", headers: subscriptionHeaders(a), signal },
+    );
+    if (!res.ok) {
+      await res.arrayBuffer().catch(() => undefined);
+      return { ok: false, status: res.status, slugs: [] as string[] };
+    }
+    const body = (await res.json()) as { models?: Array<{ slug?: string }> };
+    const slugs = (body.models ?? [])
+      .map((model) => model.slug)
+      .filter((slug): slug is string => typeof slug === "string");
+    return { ok: true, status: res.status, slugs };
+  }, timeoutMs);
+  return fetched.ok ? fetched.value : { ok: false, status: 0, slugs: [] };
 }
 
 /** GET /wham/usage — authoritative limits. NB the WHAM base is /backend-api, NOT /codex (spec §1.8a). */
 export async function fetchCodexUsage(
   a: CodexAuthHeaders,
   fetchImpl: CodexFetch = fetch,
+  timeoutMs = CODEX_READ_TIMEOUT_MS,
 ): Promise<{ status: number; payload: unknown }> {
-  const res = await fetchImpl(`${CODEX_WHAM_BASE}/wham/usage`, {
-    method: "GET",
-    headers: subscriptionHeaders(a),
-  });
-  // A 404 may carry a usage-limit body; the route layer normalizes it to a limits state (spec §1.8c).
-  const payload = res.ok || res.status === 404 ? await res.json().catch(() => null) : null;
-  return { status: res.status, payload };
+  const fetched = await runBoundedCodexOperation(async (signal) => {
+    const res = await fetchImpl(`${CODEX_WHAM_BASE}/wham/usage`, {
+      method: "GET",
+      headers: subscriptionHeaders(a),
+      signal,
+    });
+    // A 404 may carry a usage-limit body; the route layer normalizes it to a limits state (spec §1.8c).
+    const payload = res.ok || res.status === 404 ? await res.json().catch(() => null) : null;
+    if (!res.ok && res.status !== 404) await res.arrayBuffer().catch(() => undefined);
+    return { status: res.status, payload };
+  }, timeoutMs);
+  if (!fetched.ok) throw new Error(`Codex usage request ${fetched.reason}`);
+  return fetched.value;
+}
+
+/**
+ * GET /wham/rate-limit-reset-credits — detailed earned reset credits.
+ *
+ * A non-2xx or malformed body returns an explicit non-ok result. The caller may
+ * fall back to the count-only summary embedded in /wham/usage, but must never
+ * invent actionable rows from that count.
+ */
+export async function fetchCodexRateLimitResetCredits(
+  a: CodexAuthHeaders,
+  fetchImpl: CodexFetch = fetch,
+  timeoutMs = RESET_CREDIT_DETAILS_TIMEOUT_MS,
+): Promise<
+  | { ok: true; status: number; details: CodexRateLimitResetCreditsDetails }
+  | { ok: false; status: number; reason: ResetCreditFetchFailureReason }
+> {
+  const fetched = await runBoundedCodexOperation(async (signal) => {
+    const res = await fetchImpl(`${CODEX_WHAM_BASE}/wham/rate-limit-reset-credits`, {
+      method: "GET",
+      headers: subscriptionHeaders(a),
+      signal,
+    });
+    if (!res.ok) {
+      // Drain the body without retaining/logging it. Provider error bodies may
+      // contain account-specific details and are not part of this contract.
+      await res.arrayBuffer().catch(() => undefined);
+      return { ok: false as const, status: res.status, reason: "http_error" as const };
+    }
+    const details = parseCodexRateLimitResetCreditsDetails(await res.json().catch(() => null));
+    return details
+      ? { ok: true as const, status: res.status, details }
+      : { ok: false as const, status: res.status, reason: "invalid_response" as const };
+  }, timeoutMs);
+  return fetched.ok ? fetched.value : { ok: false, status: 0, reason: fetched.reason };
+}
+
+/**
+ * POST /wham/rate-limit-reset-credits/consume with the exact v0.144.6 body.
+ * `idempotencyKey` identifies one logical human redemption and MUST be reused
+ * by the server on retries. Supplying `creditId` is preferred; omission leaves
+ * provider selection in control and is therefore not used by OpenGeni's
+ * human-only flow.
+ */
+export async function consumeCodexRateLimitResetCredit(
+  a: CodexAuthHeaders,
+  input: { idempotencyKey: string; creditId?: string | undefined },
+  fetchImpl: CodexFetch = fetch,
+  timeoutMs = RESET_CREDIT_CONSUME_TIMEOUT_MS,
+): Promise<
+  | { ok: true; status: number; result: CodexRateLimitResetConsumeResponse }
+  | {
+      ok: false;
+      status: number;
+      reason: ResetCreditFetchFailureReason | "invalid_request";
+    }
+> {
+  if (input.idempotencyKey.length === 0 || input.creditId === "") {
+    return { ok: false, status: 0, reason: "invalid_request" };
+  }
+  const fetched = await runBoundedCodexOperation(async (signal) => {
+    const res = await fetchImpl(`${CODEX_WHAM_BASE}/wham/rate-limit-reset-credits/consume`, {
+      method: "POST",
+      headers: {
+        ...subscriptionHeaders(a),
+        "content-type": "application/json",
+      },
+      body: JSON.stringify({
+        redeem_request_id: input.idempotencyKey,
+        ...(input.creditId ? { credit_id: input.creditId } : {}),
+      }),
+      signal,
+    });
+    if (!res.ok) {
+      await res.arrayBuffer().catch(() => undefined);
+      return { ok: false as const, status: res.status, reason: "http_error" as const };
+    }
+    const result = parseCodexRateLimitResetConsumeResponse(await res.json().catch(() => null));
+    return result
+      ? { ok: true as const, status: res.status, result }
+      : { ok: false as const, status: res.status, reason: "invalid_response" as const };
+  }, timeoutMs);
+  return fetched.ok ? fetched.value : { ok: false, status: 0, reason: fetched.reason };
 }

package/src/bounded-operation.ts

@@ -0,0 +1,43 @@
+export type CodexOperationFailureReason = "network_error" | "timeout";
+
+/**
+ * Bound the complete provider operation, including response-body consumption.
+ *
+ * AbortController makes native fetch release its socket, while Promise.race is
+ * the backstop for injected/custom fetch implementations that ignore `signal`.
+ * The losing operation is rejection-handled and can never become an unhandled
+ * promise after the caller has received the timeout result.
+ */
+export async function runBoundedCodexOperation<T>(
+  operation: (signal: AbortSignal) => Promise<T>,
+  timeoutMs: number,
+): Promise<{ ok: true; value: T } | { ok: false; reason: CodexOperationFailureReason }> {
+  if (!Number.isFinite(timeoutMs) || timeoutMs <= 0) {
+    throw new Error("Codex operation timeout must be positive");
+  }
+
+  const controller = new AbortController();
+  let timedOut = false;
+  let timeout: ReturnType<typeof setTimeout> | undefined;
+  const work = operation(controller.signal).then(
+    (value) => ({ ok: true as const, value }),
+    () => ({
+      ok: false as const,
+      reason:
+        timedOut || controller.signal.aborted ? ("timeout" as const) : ("network_error" as const),
+    }),
+  );
+  const deadline = new Promise<{ ok: false; reason: "timeout" }>((resolve) => {
+    timeout = setTimeout(() => {
+      timedOut = true;
+      controller.abort();
+      resolve({ ok: false, reason: "timeout" });
+    }, timeoutMs);
+  });
+
+  try {
+    return await Promise.race([work, deadline]);
+  } finally {
+    if (timeout) clearTimeout(timeout);
+  }
+}

package/src/constants.ts

@@ -31,7 +31,7 @@ export const CODEX_MODEL_ID_PREFIX = "codex/";
 export const CODEX_FALLBACK_MODEL_SLUGS = ["gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"] as const;
 
 // Live Codex model-catalog values for every exposed gpt-5.6 subscription slug.
-// Verified 2026-07-17 against Codex CLI 0.144.5's freshly fetched
+// Verified 2026-07-18 against Codex CLI 0.144.6's freshly fetched
 // ~/.codex/models_cache.json and the matching openai/codex core derivations:
 //   raw context window                = 272,000
 //   effective input window (95%)      = 258,400
@@ -52,11 +52,29 @@ export const CODEX_MODEL_AUTO_COMPACT_TOKEN_LIMIT = Math.floor(
 // 2026-07-09: 0.142.4 filtered every GPT-5.6 slug out of GET /models, while the
 // official Codex 0.144.0+ releases return all three exact slugs above. Keep
 // this pinned to the latest stable Codex release we have verified end-to-end.
-export const CODEX_CLIENT_VERSION = "0.144.5";
+export const CODEX_CLIENT_VERSION = "0.144.6";
 
 export const CODEX_REFRESH_WINDOW_MS = 5 * 60 * 1000; // proactive refresh when within 5 min of exp (spec §1.1)
 export const CODEX_REFRESH_FALLBACK_MS = 8 * 24 * 60 * 60 * 1000; // 8 days when exp is unparseable
 
+// Codex Responses transport deadlines. The OpenAI SDK's own timeout only covers
+// the wait for response headers and erases the underlying timeout class into the
+// bare `Request timed out.` error. Keep the provider-specific budgets here so
+// the transport can enforce and durably report them without enabling the SDK's
+// blind request replay.
+export const CODEX_RESPONSE_HEADERS_TIMEOUT_MS = 4 * 60_000;
+export const CODEX_RESPONSE_STREAM_IDLE_TIMEOUT_MS = 5 * 60_000;
+export const CODEX_RESPONSE_WHOLE_TIMEOUT_MS = 30 * 60_000;
+// Kept as a compatibility-shaped policy field, but automatic replay is disabled
+// until a provider-specific operation receipt can prove non-acceptance or resume
+// the same operation identity. An absent response does not prove that the
+// provider never accepted the request.
+export const CODEX_RESPONSE_NO_BYTE_RETRIES = 0;
+export const CODEX_RESPONSE_RETRY_BACKOFF_MS = 1_000;
+// Must exceed the transport-owned whole-response deadline. This SDK guard is a
+// last-resort envelope; the inner transport emits the typed/durable failure.
+export const CODEX_RESPONSE_SDK_OUTER_TIMEOUT_MS = 35 * 60_000;
+
 // ── Apps / connectors MCP (spec §1.10, §E) ───────────────────────────────────
 // One server-side MCP exposes ALL the user's ChatGPT/Codex connectors
 // (gmail/github/linear/slack/sentry/drive/calendar/…). Streamable HTTP, always.