@providerkit/core 0.15.1 → 0.16.1

package/src/providers/responses.ts

@@ -10,7 +10,7 @@
 // The event names arrive on the SSE `event:` line and are repeated inside each
 // payload's own `type`. The transport yields only `data:` payloads, so this
 // adapter reads `type` — which is what survives, and what gateways agree on.
-import { fileRefused, streamError } from "../errors.ts";
+import { fileRefused, ProviderError, streamCut, streamError } from "../errors.ts";
 import { streamSse, apiUrl } from "../transport.ts";
 import type {
   ChatMessage,
@@ -24,7 +24,7 @@ import type {
   ToolDefinition,
 } from "../types.ts";
 import { toDataUri } from "../types.ts";
-import { isStrictSchema } from "../schema.ts";
+import { isStrictSchema, withoutPatterns } from "../schema.ts";
 import { withConfiguredFallbacks, type ProviderFallbackConfig } from "../fallback.ts";
 
 export interface ResponsesConfig extends ProviderFallbackConfig {
@@ -198,6 +198,7 @@ interface ResponsesUsage {
   input_tokens?: number;
   input_tokens_details?: { cached_tokens?: number };
   output_tokens?: number;
+  output_tokens_details?: { reasoning_tokens?: number };
 }
 
 interface ResponsesItem {
@@ -214,15 +215,66 @@ interface ResponsesEvent {
   type?: string;
   delta?: string;
   item_id?: string;
+  /** xAI keys argument events by the CALL id instead of the item id. */
+  call_id?: string;
+  /** The whole argument string, on `function_call_arguments.done`. */
+  arguments?: string;
   item?: ResponsesItem;
   response?: {
+    /** `completed`, `incomplete` or `failed` — stated on the terminal event,
+     *  and not always the one its `type` implies. */
+    status?: string;
     usage?: ResponsesUsage;
-    incomplete_details?: { reason?: string };
-    error?: { message?: string; code?: string };
+    incomplete_details?: { reason?: string } | null;
+    error?: { message?: string; code?: string } | null;
   };
+  /** The ChatGPT backend's quota snapshot (`codex.rate_limits`). */
+  rate_limits?: CodexRateLimits;
+  credits?: { has_credits?: boolean; unlimited?: boolean };
   error?: { message?: string; code?: string };
   message?: string;
   code?: string;
+  /** The ChatGPT backend's error frame states the HTTP status it stands for,
+   *  and mirrors its `x-codex-*` rate-limit headers, beside `error`. */
+  status_code?: number;
+  headers?: Record<string, unknown>;
+}
+
+interface CodexWindow {
+  used_percent?: number;
+  reset_after_seconds?: number;
+}
+
+interface CodexRateLimits {
+  limit_reached?: boolean;
+  allowed?: boolean;
+  primary?: CodexWindow;
+  secondary?: CodexWindow;
+}
+
+/**
+ * The quota wall a `codex.rate_limits` snapshot reports, if it is one.
+ *
+ * The snapshot rides ahead of the first output on every turn and is telemetry
+ * — a full window the account's credits cover, or that the backend still
+ * `allowed`, lets the turn proceed. It is only the reason when the stream then
+ * closes without a response: that close is a spent window, not a dropped
+ * socket, and retrying it walks every backup into the same wall. The fuller
+ * window binds; cc-proxy took the primary's clock first and told a weekly wall
+ * to come back in a few hours.
+ */
+function quotaWall(event: ResponsesEvent, provider: string): ProviderError | undefined {
+  const limits = event.rate_limits;
+  if (!limits?.limit_reached || limits.allowed) return undefined;
+  if (event.credits?.has_credits || event.credits?.unlimited) return undefined;
+  const windows = [limits.primary, limits.secondary].filter((w): w is CodexWindow => !!w);
+  const binding = windows.sort((a, b) => (b.used_percent ?? 0) - (a.used_percent ?? 0))[0];
+  const after = binding?.reset_after_seconds;
+  return new ProviderError(provider, "quota", `${provider}: the usage limit has been reached`, {
+    ...(after !== undefined && after >= 0
+      ? { retryAfterMs: after * 1000, resetAtMs: Date.now() + after * 1000 }
+      : {}),
+  });
 }
 
 function usageChunk(usage: ResponsesUsage): ProviderChunk {
@@ -235,6 +287,9 @@ function usageChunk(usage: ResponsesUsage): ProviderChunk {
       cachedInputTokens: usage.input_tokens_details?.cached_tokens ?? 0,
       // Reasoning tokens are billed INSIDE output_tokens, not beside them.
       outputTokens: usage.output_tokens ?? 0,
+      ...(usage.output_tokens_details?.reasoning_tokens !== undefined
+        ? { reasoningTokens: usage.output_tokens_details.reasoning_tokens }
+        : {}),
     },
   };
 }
@@ -249,6 +304,15 @@ interface PendingCall {
   streamed: string;
 }
 
+/** The part of an authoritative snapshot the deltas have not already sent.
+ *  Re-emitting it whole concatenates the JSON with itself and every argument
+ *  parse fails; all of it goes when no delta came. */
+function unsentTail(snapshot: string, streamed: string): string {
+  return snapshot.length > streamed.length && snapshot.startsWith(streamed)
+    ? snapshot.slice(streamed.length)
+    : "";
+}
+
 export function createResponsesProvider(config: ResponsesConfig): Provider {
   const baseUrl = config.baseUrl ?? DEFAULT_BASE_URL;
   const id = config.id ?? "openai-responses";
@@ -301,9 +365,20 @@ export function createResponsesProvider(config: ResponsesConfig): Provider {
           type: "function",
           name: tool.name,
           description: tool.description,
-          parameters: tool.inputSchema,
+          parameters: withoutPatterns(tool.inputSchema),
+          // Said either way, never left to the default. Strict demands every
+          // property required and every object closed; on a schema that is not
+          // already shaped that way the backend makes optional arguments
+          // mandatory, and the model starts filling in parameters nobody asked
+          // for (cc-proxy, 0.1.15). Where the schema already qualifies, the
+          // enforcement is free.
+          strict: isStrictSchema(tool.inputSchema),
         }));
+        if (opts.parallelToolCalls === false) request.parallel_tool_calls = false;
       }
+      // The cache affinity key. Without one the backend spreads a
+      // conversation's turns across cache shards, and a re-sent prefix misses.
+      if (opts.sessionId) request.prompt_cache_key = opts.sessionId;
       // Muse Contributor accepts only `auto`. `none` has an exact wire-level
       // equivalent: send no tools. Required/named choices stay explicit and let
       // the endpoint refuse a promise it cannot keep.
@@ -327,13 +402,18 @@ export function createResponsesProvider(config: ResponsesConfig): Provider {
       }
 
       // Tool calls arrive as an item skeleton plus argument deltas; keyed by the
-      // output item id so parallel calls never cross wires. The seam's index is
-      // ours to assign — `output_index` counts reasoning and message items too.
+      // output item id AND the call id, whichever the backend quotes, so
+      // parallel calls never cross wires. The seam's index is ours to assign —
+      // `output_index` counts reasoning and message items too.
       const pending = new Map<string, PendingCall>();
+      const pendingFor = (itemId?: string, callId?: string) =>
+        (itemId ? pending.get(itemId) : undefined) ?? (callId ? pending.get(callId) : undefined);
       let nextIndex = 0;
       // This shape never states a stop reason on a clean finish, so it is
       // inferred from whether the turn produced a function call.
       let sawToolCall = false;
+      let wall: ProviderError | undefined;
+      let produced = false;
 
       for await (const data of streamSse({
         url: apiUrl(baseUrl, config.path ?? DEFAULT_PATH),
@@ -348,6 +428,7 @@ export function createResponsesProvider(config: ResponsesConfig): Provider {
         provider: id,
         ...(opts.signal ? { signal: opts.signal } : {}),
         ...(config.fetchImpl ? { fetchImpl: config.fetchImpl } : {}),
+        ...(opts.onActivity ? { onActivity: opts.onActivity } : {}),
       })) {
         let event: ResponsesEvent;
         try {
@@ -355,6 +436,9 @@ export function createResponsesProvider(config: ResponsesConfig): Provider {
         } catch {
           continue; // a keep-alive or a frame we do not model
         }
+        if (event.type?.endsWith(".delta") || event.type === "response.output_item.added") {
+          produced = true;
+        }
 
         switch (event.type) {
           case "response.output_text.delta":
@@ -371,18 +455,26 @@ export function createResponsesProvider(config: ResponsesConfig): Provider {
 
           case "response.output_item.added": {
             const item = event.item;
-            if (item?.type !== "function_call" || !item.id) break;
+            if (item?.type !== "function_call") break;
+            // OpenAI names the item (`fc_…`) and keys argument deltas by it;
+            // xAI's Grok sends no item id at all and keys them by `call_id`.
+            // Requiring the item id dropped every Grok call here, and the
+            // turn came back as a nameless call with empty arguments.
+            const key = item.id ?? item.call_id;
+            if (!key) break;
             sawToolCall = true;
             const index = nextIndex++;
             // Normally empty here, but a backend that already has the whole
             // call sends it in the skeleton.
             const seeded = item.arguments ?? "";
-            pending.set(item.id, {
+            const call: PendingCall = {
               index,
               id: item.call_id ?? "",
               name: item.name ?? "",
               streamed: seeded,
-            });
+            };
+            pending.set(key, call);
+            if (item.call_id) pending.set(item.call_id, call);
             // Only the fields the skeleton actually states. An empty `id` here
             // is not "unknown", it is a wrong answer: a consumer takes the last
             // stated value, so `""` written into the slot survives the real
@@ -402,18 +494,31 @@ export function createResponsesProvider(config: ResponsesConfig): Provider {
           }
 
           case "response.function_call_arguments.delta": {
-            const call = event.item_id ? pending.get(event.item_id) : undefined;
+            const call = pendingFor(event.item_id, event.call_id);
             if (!call || !event.delta) break;
             call.streamed += event.delta;
             yield { type: "delta", toolCalls: [{ index: call.index, arguments: event.delta }] };
             break;
           }
 
+          // xAI states the finished argument string here and NOT on the
+          // closing item, which carries only `call_id`. Skipped, a call whose
+          // deltas never came reaches the caller with empty arguments.
+          case "response.function_call_arguments.done": {
+            const call = pendingFor(event.item_id, event.call_id);
+            const tail = call ? unsentTail(event.arguments ?? "", call.streamed) : "";
+            if (!call || !tail) break;
+            call.streamed += tail;
+            yield { type: "delta", toolCalls: [{ index: call.index, arguments: tail }] };
+            break;
+          }
+
           case "response.output_item.done": {
             const item = event.item;
             if (item?.type !== "function_call") break;
-            const known = item.id ? pending.get(item.id) : undefined;
+            const known = pendingFor(item.id, item.call_id);
             if (item.id) pending.delete(item.id);
+            if (item.call_id) pending.delete(item.call_id);
             const snapshot = item.arguments ?? "";
 
             if (!known) {
@@ -441,13 +546,8 @@ export function createResponsesProvider(config: ResponsesConfig): Provider {
             }
 
             // The snapshot is authoritative, but the fragments already went
-            // out: re-emitting it whole concatenates the JSON with itself and
-            // every argument parse fails. Send only what the deltas missed —
-            // which is all of it when they never came.
-            const tail =
-              snapshot.length > known.streamed.length && snapshot.startsWith(known.streamed)
-                ? snapshot.slice(known.streamed.length)
-                : "";
+            // out: send only what the deltas missed.
+            const tail = unsentTail(snapshot, known.streamed);
             // `.done` restates the identity, and on a backend that leaves it
             // out of the skeleton this is the only frame that carries it. It
             // rides last so it WINS: the alternative is a caller assembling a
@@ -464,13 +564,34 @@ export function createResponsesProvider(config: ResponsesConfig): Provider {
             break;
           }
 
-          case "response.completed": {
-            const usage = event.response?.usage;
-            if (usage) yield usageChunk(usage);
+          // `response.done` is the same terminal event under the name
+          // OpenCode's gateway uses.
+          case "response.completed":
+          case "response.done": {
+            const response = event.response;
+            // A terminal event names its own outcome, and the type is not
+            // always it: a `completed` that states `failed` or carries an
+            // error failed, and one that states `incomplete` was cut. Read as
+            // a clean stop, either is a broken turn reported as a finished one.
+            if (response?.status === "failed" || response?.error) {
+              throw streamError(id, response.error ?? { message: "response failed" });
+            }
+            if (response?.usage) yield usageChunk(response.usage);
+            if (response?.status === "incomplete") {
+              yield {
+                type: "finish",
+                finishReason: mapIncompleteReason(response.incomplete_details?.reason),
+              };
+              return;
+            }
             yield { type: "finish", finishReason: sawToolCall ? "tool_calls" : "stop" };
             return;
           }
 
+          case "codex.rate_limits":
+            wall = quotaWall(event, id);
+            break;
+
           case "response.incomplete": {
             // A turn that ran out of output tokens still billed for its input,
             // and this event carries usage in the same shape as `completed`.
@@ -491,13 +612,24 @@ export function createResponsesProvider(config: ResponsesConfig): Provider {
 
           case "error":
           case "response.error":
-            throw streamError(id, event.error ?? { message: event.message, code: event.code });
+            // The status and the window clocks ride beside `error`. Without
+            // them a spent Codex window reads as an unnamed in-stream failure:
+            // floored to overload, then retried and walked across every backup
+            // on the same account wall.
+            throw streamError(
+              id,
+              event.error
+                ? { ...event.error, status_code: event.status_code, headers: event.headers }
+                : { message: event.message, code: event.code },
+            );
         }
       }
 
-      // The stream closed without a terminal event. Nothing is lost but the
-      // finish reason and the usage record — every delta, tool-call fragment
-      // included, was already yielded as it arrived.
+      // The stream closed without a terminal event. If the backend's last word
+      // was a spent window and nothing was produced, that is why; otherwise
+      // the stream was cut (see `streamCut`).
+      if (wall && !produced) throw wall;
+      throw streamCut(id);
     },
   };
 

package/src/rate-limit.ts

@@ -57,6 +57,32 @@ const RESET_HEADERS = [
   "x-ratelimit-reset",
 ];
 
+/**
+ * The ChatGPT backend's two windows. Both ride every response — the slots are
+ * "primary" and "secondary", NOT "5h" and "weekly", so the window is named by
+ * its own `-window-minutes` (300 and 10080 in every capture), and its
+ * `-used-percent` says which one is full. Recorded by cc-proxy off a live turn
+ * that spent the 5-hour window, 2026-09.
+ */
+const CODEX_SLOTS = ["primary", "secondary"] as const;
+
+function windowForMinutes(minutes: number | undefined): RateLimitWindow | undefined {
+  return minutes !== undefined && minutes > 0 ? windowForMs(minutes * 60_000) : undefined;
+}
+
+function codexWindow(headers: Headers, slot: (typeof CODEX_SLOTS)[number], now: number) {
+  const name = (suffix: string) => headers.get(`x-codex-${slot}-${suffix}`);
+  const percent = numeric(name("used-percent"));
+  const after = numeric(name("reset-after-seconds"));
+  return {
+    window: windowForMinutes(numeric(name("window-minutes"))),
+    // A percentage here; Anthropic's unified utilization is a fraction.
+    utilization: percent === undefined ? undefined : percent / 100,
+    reset:
+      unixSecondsToMs(name("reset-at")) ?? (after === undefined ? undefined : now + after * 1000),
+  };
+}
+
 /** Seconds since 1970 passed a billion in 2001, so a bare integer above this is
  *  a timestamp and anything below it is a countdown. Vendors disagree on which
  *  they send under the same header name, so magnitude decides. */
@@ -158,10 +184,11 @@ export function parseRateLimitReset(headers: Headers, now = Date.now()): RateLim
       utilization: numeric(headers.get(`${UNIFIED_7D}utilization`)),
       reset: unixSecondsToMs(headers.get(`${UNIFIED_7D}reset`)),
     },
+    ...CODEX_SLOTS.map((slot) => codexWindow(headers, slot, now)),
   ].filter((w) => w.utilization !== undefined && w.reset !== undefined && w.reset > now);
   const binding = windows.sort((a, b) => (b.utilization ?? 0) - (a.utilization ?? 0))[0];
   if (binding) {
-    result.window = binding.window;
+    if (binding.window) result.window = binding.window;
     // The window's own reset outranks retry-after: it names the real horizon,
     // where retry-after names the next polite attempt.
     if (binding.reset !== undefined) result.resetAtMs = binding.reset;
@@ -191,6 +218,20 @@ const WINDOW_FLOOR_MS = 10 * 60_000;
 const FIVE_HOUR_MAX_MS = 5.5 * 3_600_000;
 const WEEKLY_MAX_MS = 7.5 * 86_400_000;
 
+/** The window a span this long belongs to. */
+function windowForMs(ms: number): RateLimitWindow {
+  if (ms <= FIVE_HOUR_MAX_MS) return "5h";
+  if (ms <= WEEKLY_MAX_MS) return "weekly";
+  return "monthly";
+}
+
+/** A wait read off a body, named as a window once it is long enough to be one. */
+function resetAfter(waitMs: number, now: number): RateLimitReset {
+  const result: RateLimitReset = { retryAfterMs: waitMs, resetAtMs: now + waitMs };
+  if (waitMs > WINDOW_FLOOR_MS) result.window = windowForMs(waitMs);
+  return result;
+}
+
 /** A finite number off a JSON field, the string form a relay may stringify it
  *  into included. Bare `Number()` is the trap `numeric()` guards against one
  *  layer up: `Number(null)`, `Number("")`, `Number([])` and `Number(false)` are
@@ -248,34 +289,24 @@ export function parseUsageLimitBody(bodyText: string, now = Date.now()): RateLim
 
     const inSeconds = jsonSeconds(error.resets_in_seconds);
     const atSeconds = jsonSeconds(error.resets_at);
-    const waitMs =
-      inSeconds !== undefined && inSeconds >= 0
-        ? inSeconds * 1000
-        : atSeconds !== undefined && atSeconds > 0
-          ? Math.max(0, atSeconds * 1000 - now)
-          : undefined;
-    if (waitMs !== undefined) {
-      const result: RateLimitReset = { retryAfterMs: waitMs, resetAtMs: now + waitMs };
-      if (waitMs > WINDOW_FLOOR_MS) {
-        result.window =
-          waitMs <= FIVE_HOUR_MAX_MS ? "5h" : waitMs <= WEEKLY_MAX_MS ? "weekly" : "monthly";
-      }
-      return result;
+    // The plain wait some gateways put in the error body instead of a header,
+    // under either name (cc-proxy reads both off the ChatGPT backend's frames).
+    const afterSeconds =
+      jsonSeconds(error.retry_after) ??
+      jsonSeconds(error.retry_after_seconds) ??
+      jsonSeconds((body as Record<string, unknown>).retry_after_seconds);
+    if (inSeconds !== undefined && inSeconds >= 0) return resetAfter(inSeconds * 1000, now);
+    if (atSeconds !== undefined && atSeconds > 0) {
+      return resetAfter(Math.max(0, atSeconds * 1000 - now), now);
+    }
+    if (afterSeconds !== undefined && afterSeconds >= 0) {
+      return resetAfter(afterSeconds * 1000, now);
     }
   }
 
   // If structured JSON did not yield a reset, check prose wording
   const waitMs = parseProseRetryMs(bodyText);
-  if (waitMs !== undefined) {
-    const result: RateLimitReset = { retryAfterMs: waitMs, resetAtMs: now + waitMs };
-    if (waitMs > WINDOW_FLOOR_MS) {
-      result.window =
-        waitMs <= FIVE_HOUR_MAX_MS ? "5h" : waitMs <= WEEKLY_MAX_MS ? "weekly" : "monthly";
-    }
-    return result;
-  }
-
-  return {};
+  return waitMs === undefined ? {} : resetAfter(waitMs, now);
 }
 
 /** Read both places an endpoint can report its reset. A short Retry-After is

package/src/schema.ts

@@ -134,6 +134,57 @@ export function isStrictSchema(node: unknown): boolean {
   );
 }
 
+/** Keywords whose value is a schema, a list of schemas, or a map of them —
+ *  the only places a `pattern` can mean a regex. Everything else (`default`,
+ *  `examples`, `enum`, property NAMES) is data and is left alone. */
+const SCHEMA_SLOTS = [
+  "items",
+  "additionalProperties",
+  "not",
+  "if",
+  "then",
+  "else",
+  "contains",
+  "propertyNames",
+  "unevaluatedItems",
+  "unevaluatedProperties",
+];
+const SCHEMA_LISTS = ["anyOf", "oneOf", "allOf", "prefixItems"];
+const SCHEMA_MAPS = ["properties", "patternProperties", "$defs", "definitions", "dependentSchemas"];
+
+/**
+ * A tool schema with every `pattern` constraint removed.
+ *
+ * The ChatGPT backend compiles tool schemas with OpenAI's own regex dialect and
+ * answers a pattern it cannot read with a 400 for the WHOLE request — strict
+ * or not. Claude Code's Artifact tool carries one, and every request that
+ * offered it failed (cc-proxy #141/#142). Losing the constraint costs a
+ * validation the caller's tool runner repeats anyway; the 400 costs the turn.
+ */
+export function withoutPatterns(node: unknown): unknown {
+  if (!node || typeof node !== "object" || Array.isArray(node)) return node;
+  const { pattern: _pattern, ...schema } = node as Record<string, unknown>;
+  for (const key of SCHEMA_SLOTS) {
+    const value = schema[key];
+    if (value !== undefined) {
+      schema[key] = Array.isArray(value) ? value.map(withoutPatterns) : withoutPatterns(value);
+    }
+  }
+  for (const key of SCHEMA_LISTS) {
+    const value = schema[key];
+    if (Array.isArray(value)) schema[key] = value.map(withoutPatterns);
+  }
+  for (const key of SCHEMA_MAPS) {
+    const value = schema[key];
+    if (value && typeof value === "object" && !Array.isArray(value)) {
+      schema[key] = Object.fromEntries(
+        Object.entries(value).map(([name, child]) => [name, withoutPatterns(child)]),
+      );
+    }
+  }
+  return schema;
+}
+
 /**
  * Sanitize a JSON Schema for Google Gemini's REST API.
  *