@fudrouter/fsrouter 0.6.72 → 0.6.74

package/dist/open-sse/config/kiroConstants.js

@@ -15,9 +15,29 @@
  * fiction. The suffix is stripped before the request leaves this process.
  */
 
+import { extractThinking } from "../translator/concerns/thinkingUnified.js";
+import { effortToBudget } from "../translator/concerns/thinking.js";
+
 export const KIRO_AGENTIC_SUFFIX = "-agentic";
 export const KIRO_THINKING_SUFFIX = "-thinking";
 
+// Public default CodeWhisperer profile ARNs (us-east-1), keyed by auth method.
+// Used when an account cannot resolve its own profileArn. Builder ID and social
+// (Google/GitHub) sign-ins map to different shared profiles.
+export const KIRO_DEFAULT_PROFILE_ARNS = {
+  "builder-id": "arn:aws:codewhisperer:us-east-1:638616132270:profile/AAAACCCCXXXX",
+  social: "arn:aws:codewhisperer:us-east-1:699475941385:profile/EHGA3GRVQMUK",
+};
+
+// Back-compat single default (Builder ID).
+export const KIRO_DEFAULT_PROFILE_ARN = KIRO_DEFAULT_PROFILE_ARNS["builder-id"];
+
+/** Resolve the shared default profileArn for a given auth method. */
+export function resolveDefaultProfileArn(authMethod) {
+  const social = authMethod === "google" || authMethod === "github";
+  return social ? KIRO_DEFAULT_PROFILE_ARNS.social : KIRO_DEFAULT_PROFILE_ARNS["builder-id"];
+}
+
 export const KIRO_THINKING_BUDGET_DEFAULT = 16000;
 
 export const KIRO_AGENTIC_SYSTEM_PROMPT = `
@@ -72,61 +92,100 @@ REMEMBER: When in doubt, write LESS per operation. Multiple small operations > o
 `.trim();
 
 /**
- * Detect whether an inbound request is asking for reasoning / thinking output.
+ * Resolve the Kiro thinking budget requested by a client.
  *
- * Sources of intent (any one is enough):
- *   - HTTP header `Anthropic-Beta: ...interleaved-thinking...`
- *   - JSON `thinking.type === "enabled"` (Claude Messages API)
- *   - JSON `reasoning_effort` in {low, medium, high, auto} (OpenAI o1/o3)
- *   - JSON `reasoning.effort` in {low, medium, high, auto} (OpenAI Responses)
- *   - System prompt contains `<thinking_mode>enabled</thinking_mode>` or
- *     `<thinking_mode>interleaved</thinking_mode>` (AMP / Cursor)
- *   - Model name contains `thinking` or `-reason`
+ * Reuses the shared thinkingUnified parser (extractThinking) so every client
+ * shape (Claude output_config.effort / thinking.budget_tokens, OpenAI
+ * reasoning_effort / reasoning.effort, Gemini, Qwen) maps consistently. Explicit
+ * `none`/`off`/disabled wins and returns null (no prefix injected).
+ * buildThinkingSystemPrefix performs Kiro's final 1..32000 clamp.
  *
- * @param {object} body OpenAI-shaped request body (post-translation)
+ * @param {object} body OpenAI/Claude-shaped request body
  * @param {object} [headers] Original inbound HTTP headers (case-insensitive)
- * @param {string} [model] Model id the caller asked for (post-strip ok)
- * @returns {boolean}
+ * @param {string} [model] Model id the caller asked for
+ * @returns {number|null} budget to inject, or null when thinking is disabled
  */
-export function isThinkingEnabled(body, headers, model) {
+export function resolveKiroThinkingBudget(body, headers, model) {
+  const cfg = extractThinking(body);
+  if (cfg) {
+    if (cfg.mode === "none") return null;
+    if (cfg.mode === "budget") return cfg.budget;
+    if (cfg.mode === "level") return effortToBudget(cfg.level) ?? KIRO_THINKING_BUDGET_DEFAULT;
+    return KIRO_THINKING_BUDGET_DEFAULT;
+  }
+
   if (headers) {
     const beta = pickHeader(headers, "anthropic-beta");
     if (typeof beta === "string" && beta.toLowerCase().includes("interleaved-thinking")) {
-      return true;
+      return KIRO_THINKING_BUDGET_DEFAULT;
     }
   }
 
-  if (body && typeof body === "object") {
-    const thinking = body.thinking;
-    if (thinking && typeof thinking === "object" && thinking.type === "enabled") {
-      const budget = Number(thinking.budget_tokens);
-      if (!Number.isFinite(budget) || budget > 0) {
-        return true;
-      }
-    }
-
-    const effort = body.reasoning_effort
-      ?? (body.reasoning && typeof body.reasoning === "object" ? body.reasoning.effort : null);
-    if (typeof effort === "string") {
-      const v = effort.toLowerCase();
-      if (v && v !== "none" && (v === "low" || v === "medium" || v === "high" || v === "auto")) {
-        return true;
-      }
-    }
-
-    if (containsThinkingModeTag(body)) {
-      return true;
-    }
-  }
+  if (containsThinkingModeTag(body)) return KIRO_THINKING_BUDGET_DEFAULT;
 
   if (typeof model === "string" && model) {
     const m = model.toLowerCase();
-    if (m.includes("thinking") || m.includes("-reason")) {
-      return true;
-    }
+    if (m.includes("thinking") || m.includes("-reason")) return KIRO_THINKING_BUDGET_DEFAULT;
   }
 
-  return false;
+  return null;
+}
+
+export function extractKiroEffortLevel(body) {
+  const effort =
+    body?.output_config?.effort ??
+    body?.reasoning_effort ??
+    (typeof body?.reasoning === "object" ? body.reasoning?.effort : null);
+  if (typeof effort !== "string") return null;
+  const normalized = effort.toLowerCase();
+  if (normalized === "none" || normalized === "off" || normalized === "disabled") return null;
+  if (normalized === "xhigh" || normalized === "max") return "high";
+  if (["low", "medium", "high"].includes(normalized)) return normalized;
+  return null;
+}
+
+export function buildKiroAdditionalModelRequestFields(body) {
+  const effort = extractKiroEffortLevel(body);
+  if (!effort) return undefined;
+  // Mirrors Kiro CLI/KAS buildEffortRequestFields("output_config").
+  return {
+    thinking: { type: "adaptive", display: "summarized" },
+    output_config: { effort },
+  };
+}
+
+export function supportsKiroAdditionalModelRequestFields(model) {
+  if (typeof model !== "string") return false;
+  const normalized = model.toLowerCase().replace(/-/g, ".");
+  if (!normalized.includes("claude")) return false;
+  const match = normalized.match(/(?:^|[/.])claude(?:[/.][a-z]+)*[/.](\d+)(?:[/.](\d+))?(?:[/.]|$)/);
+  if (!match) return false;
+  const [, majorText, minorText] = match;
+  const major = Number(majorText);
+  const minor = minorText === undefined ? null : Number(minorText);
+  const dateSuffixMinor = minor !== null && minor >= 1000;
+  // Kiro rejected additionalModelRequestFields on legacy 4.5 models in live smoke.
+  // Default future Claude/Kiro models to supported so new model releases do not
+  // need a code allowlist update.
+  return !(major < 4 || (major === 4 && (minor === null || minor <= 5 || dateSuffixMinor)));
+}
+
+export function buildKiroAdditionalModelRequestFieldsForModel(body, model) {
+  if (!supportsKiroAdditionalModelRequestFields(model)) return undefined;
+  return buildKiroAdditionalModelRequestFields(body);
+}
+
+/**
+ * Detect whether an inbound request is asking for reasoning / thinking output.
+ * Thin wrapper over resolveKiroThinkingBudget (single source of truth).
+ *
+ * @param {object} body OpenAI-shaped request body (post-translation)
+ * @param {object} [headers] Original inbound HTTP headers (case-insensitive)
+ * @param {string} [model] Model id the caller asked for (post-strip ok)
+ * @returns {boolean}
+ */
+export function isThinkingEnabled(body, headers, model) {
+  return resolveKiroThinkingBudget(body, headers, model) !== null;
 }
 
 /**
@@ -260,9 +319,3 @@ function containsTagInText(text) {
   return text.includes("<thinking_mode>enabled</thinking_mode>")
     || text.includes("<thinking_mode>interleaved</thinking_mode>");
 }
-
-// Added from 9router - missing exports
-export function resolveDefaultProfileArn(authMethod) {
-  const social = authMethod === "google" || authMethod === "github";
-  return social ? KIRO_DEFAULT_PROFILE_ARNS.social : KIRO_DEFAULT_PROFILE_ARNS["builder-id"];
-}

package/dist/open-sse/config/runtimeConfig.js

@@ -65,6 +65,8 @@ export const GEMINI_NATIVE_TTS_FETCH_TIMEOUT_MS = envMs("GEMINI_NATIVE_TTS_FETCH
 export const DEFAULT_MAX_TOKENS = 64000;
 export const DEFAULT_MIN_TOKENS = 32000;
 
+export const TOKEN_SAVER_HEADER = "x-9router-token-saver";
+
 // Retry config for 429 responses (legacy - kept for backward compatibility)
 export const RETRY_CONFIG = {
   maxAttempts: 2,

package/dist/open-sse/config/ttsModels.js

@@ -96,10 +96,12 @@ export const TTS_MODELS_CONFIG = {
   },
   gemini: {
     models: [
+      { id: "gemini-3.1-flash-tts-preview", name: "Gemini 3.1 Flash TTS", type: "tts" },
       { id: "gemini-2.5-flash-preview-tts", name: "Gemini 2.5 Flash TTS", type: "tts" },
       { id: "gemini-2.5-pro-preview-tts",   name: "Gemini 2.5 Pro TTS",   type: "tts" },
     ],
     voices: {
+      "gemini-3.1-flash-tts-preview": GEMINI_VOICES,
       "gemini-2.5-flash-preview-tts": GEMINI_VOICES,
       "gemini-2.5-pro-preview-tts":   GEMINI_VOICES,
     },