@fudrouter/fsrouter 0.6.71 → 0.6.73

package/dist/open-sse/utils/stream.js

@@ -1,12 +1,15 @@
 import { translateResponse, initState } from "../translator/index.js";
 import { FORMATS } from "../translator/formats.js";
 import { trackPendingRequest, appendRequestLog } from '../../../dist/lib/usageDb.js';
-import { extractUsage, hasValidUsage, estimateUsage, logUsage, addBufferToUsage, filterUsageForFormat, COLORS } from "./usageTracking.js";
+import { extractUsage, mergeUsage, hasValidUsage, estimateUsage, logUsage, addBufferToUsage, filterUsageForFormat, COLORS } from "./usageTracking.js";
 import { parseSSELine, hasValuableContent, fixInvalidId, formatSSE } from "./streamHelpers.js";
 import { getOpenAIResponsesEventName, isOpenAIResponsesTerminalEvent, formatIncompleteOpenAIResponsesStreamFailure } from "./responsesStreamHelpers.js";
 import { dbg, isDebugEnabled } from "./debugLog.js";
 
+import { SSE_DONE, SSE_HEADERS, SSE_HEADERS_NO_BUFFER } from "./sseConstants.js";
+
 export { COLORS, formatSSE };
+export { SSE_DONE, SSE_HEADERS, SSE_HEADERS_NO_BUFFER };
 
 // sharedEncoder is stateless — safe to share across streams
 const sharedEncoder = new TextEncoder();
@@ -68,6 +71,7 @@ export function createSSEStream(options = {}) {
   let currentOpenAIResponsesEvent = null;
   let openAIResponsesTerminalSeen = false;
   let openAIResponsesDoneSent = false;
+  let streamDoneSent = false;  // track duplicate [DONE] across transform + flush
 
   return new TransformStream({
     transform(chunk, controller) {
@@ -99,35 +103,6 @@ export function createSSEStream(options = {}) {
           let output;
           let injectedUsage = false;
 
-          // Handle raw JSON error disguised as SSE in passthrough mode
-          if (trimmed.startsWith("{")) {
-            try {
-              const parsedJson = JSON.parse(trimmed);
-              if (parsedJson && (parsedJson.error || parsedJson.code !== undefined || parsedJson.msg)) {
-                const errorMsg = parsedJson.error?.message || parsedJson.msg || (typeof parsedJson.error === "string" ? parsedJson.error : JSON.stringify(parsedJson));
-                const errorChunk = {
-                  id: `err_${Date.now()}`,
-                  object: "chat.completion.chunk",
-                  created: Math.floor(Date.now() / 1000),
-                  model: model || "unknown",
-                  choices: [{
-                    index: 0,
-                    delta: { content: `\n\n[Provider Error]: ${errorMsg}\n\n` },
-                    finish_reason: "stop"
-                  }]
-                };
-                output = formatSSE(errorChunk, FORMATS.OPENAI);
-                reqLogger?.appendConvertedChunk?.(output);
-                controller.enqueue(sharedEncoder.encode(output));
-                
-                const doneOutput = "data: [DONE]\n\n";
-                reqLogger?.appendConvertedChunk?.(doneOutput);
-                controller.enqueue(sharedEncoder.encode(doneOutput));
-                continue;
-              }
-            } catch (e) {}
-          }
-
           if (trimmed.startsWith("data:") && trimmed.slice(5).trim() !== "[DONE]") {
             try {
               const parsed = JSON.parse(trimmed.slice(5).trim());
@@ -155,17 +130,25 @@ export function createSSEStream(options = {}) {
                 }
               }
 
+              // Strip empty tool_calls arrays that break AI SDK reasoning tracking.
+              // Some providers (e.g. CodeBuddy CN) include `"tool_calls": []` in
+              // every streaming delta. @ai-sdk/openai-compatible checks
+              // `delta.tool_calls != null` — an empty array passes this check,
+              // causing premature `reasoning-end` on every chunk.
+              if (parsed?.choices) {
+                for (const choice of parsed.choices) {
+                  if (choice.delta?.tool_calls && Array.isArray(choice.delta.tool_calls) && choice.delta.tool_calls.length === 0) {
+                    delete choice.delta.tool_calls;
+                    fieldsInjected = true;
+                  }
+                }
+              }
+
               if (!hasValuableContent(parsed, FORMATS.OPENAI)) {
                 continue;
               }
 
               const delta = parsed.choices?.[0]?.delta;
-              
-              // Map refusal to content for clients that do not support the refusal field
-              if (delta?.refusal && !delta?.content) {
-                delta.content = `[Refusal]: ${delta.refusal}`;
-              }
-              
               const content = delta?.content;
               const reasoning = delta?.reasoning_content;
               if (content && typeof content === "string") {
@@ -179,7 +162,7 @@ export function createSSEStream(options = {}) {
 
               const extracted = extractUsage(parsed);
               if (extracted) {
-                usage = extracted;
+                usage = mergeUsage(usage, extracted);
               }
 
               const isFinishChunk = parsed.choices?.[0]?.finish_reason;
@@ -198,7 +181,12 @@ export function createSSEStream(options = {}) {
                 output = `data: ${JSON.stringify(parsed)}\n`;
                 injectedUsage = true;
               }
-            } catch { }
+            } catch {
+              // Skip non-JSON data lines silently — don't forward garbage to clients.
+              // Upstream providers sometimes return plain-text errors (HTML, rate-limit
+              // messages) in the SSE stream that would break downstream JSON decoders.
+              continue;
+            }
           }
 
           if (!injectedUsage) {
@@ -219,37 +207,6 @@ export function createSSEStream(options = {}) {
 
         const parsed = parseSSELine(trimmed, targetFormat);
         if (!parsed) continue;
-        
-        // Intercept raw JSON error that provider sent instead of SSE
-        if (parsed._isRawError) {
-          const errorMsg = parsed.raw.error?.message || parsed.raw.msg || (typeof parsed.raw.error === "string" ? parsed.raw.error : JSON.stringify(parsed.raw));
-          const errorChunk = {
-            id: `err_${Date.now()}`,
-            object: "chat.completion.chunk",
-            created: Math.floor(Date.now() / 1000),
-            model: model || "unknown",
-            choices: [{
-              index: 0,
-              delta: { content: `\n\n[Provider Error]: ${errorMsg}\n\n` },
-              finish_reason: "stop"
-            }]
-          };
-          const output = formatSSE(errorChunk, sourceFormat);
-          reqLogger?.appendConvertedChunk?.(output);
-          controller.enqueue(sharedEncoder.encode(output));
-          sseEmittedCount++;
-          
-          const doneOutput = "data: [DONE]\n\n";
-          reqLogger?.appendConvertedChunk?.(doneOutput);
-          controller.enqueue(sharedEncoder.encode(doneOutput));
-          if (targetFormat === FORMATS.OPENAI_RESPONSES && sourceFormat === FORMATS.OPENAI_RESPONSES) openAIResponsesDoneSent = true;
-          continue;
-        }
-
-        // Map refusal to content for older clients that do not support refusal field
-        if (parsed.choices?.[0]?.delta?.refusal && !parsed.choices[0].delta.content) {
-          parsed.choices[0].delta.content = `[Refusal]: ${parsed.choices[0].delta.refusal}`;
-        }
 
         // Responses API same-format passthrough: preserve event framing + track terminal state
         const isOpenAIResponsesStream = targetFormat === FORMATS.OPENAI_RESPONSES;
@@ -274,9 +231,12 @@ export function createSSEStream(options = {}) {
             sseEmittedCount++;
           }
 
-          const output = "data: [DONE]\n\n";
-          reqLogger?.appendConvertedChunk?.(output);
-          controller.enqueue(sharedEncoder.encode(output));
+          if (keepsOpenAIResponsesFormat && !streamDoneSent) {
+            const doneOutput = "data: [DONE]\n\n";
+            reqLogger?.appendConvertedChunk?.(doneOutput);
+            controller.enqueue(sharedEncoder.encode(doneOutput));
+          }
+          streamDoneSent = true;
           if (keepsOpenAIResponsesFormat) openAIResponsesDoneSent = true;
           continue;
         }
@@ -320,7 +280,7 @@ export function createSSEStream(options = {}) {
 
         // Extract usage
         const extracted = extractUsage(parsed);
-        if (extracted) state.usage = extracted; // Keep original usage for logging
+        if (extracted) state.usage = mergeUsage(state.usage, extracted); // Keep original usage for logging
 
         // Responses same-format passthrough: re-emit with original event framing
         if (keepsOpenAIResponsesFormat && openAIResponsesEventName) {
@@ -396,14 +356,23 @@ export function createSSEStream(options = {}) {
             usage = estimateUsage(body, totalContentLength, FORMATS.OPENAI);
           }
 
+          if (hasValidUsage(usage)) {
+            logUsage(provider, usage, model, connectionId, apiKey);
+          } else {
+            appendRequestLog({ model, provider, connectionId, tokens: null, status: "200 OK" }).catch(() => { });
+          }
           
           // IMPORTANT: In passthrough mode we still must terminate the SSE stream.
           // Some clients (e.g. OpenClaw) expect the OpenAI-style sentinel:
           //   data: [DONE]\n\n
           // Without it they can hang until timeout and trigger failover.
-          const doneOutput = "data: [DONE]\n\n";
-          reqLogger?.appendConvertedChunk?.(doneOutput);
-          controller.enqueue(sharedEncoder.encode(doneOutput));
+          // Gemini-family clients (Antigravity, Vertex, Gemini) reject this sentinel with 400 syntax errors.
+          const isGeminiFamily = provider === "antigravity" || provider === "gemini" || provider === "vertex";
+          if (!streamDoneSent && !isGeminiFamily) {
+            const doneOutput = "data: [DONE]\n\n";
+            reqLogger?.appendConvertedChunk?.(doneOutput);
+            controller.enqueue(sharedEncoder.encode(doneOutput));
+          }
 
           if (onStreamComplete) {
             onStreamComplete({
@@ -464,15 +433,23 @@ export function createSSEStream(options = {}) {
           openAIResponsesTerminalSeen = true;
         }
 
-        if (!keepsOpenAIResponsesFormat || !openAIResponsesDoneSent) {
+        if (keepsOpenAIResponsesFormat && !openAIResponsesDoneSent && !streamDoneSent) {
           const doneOutput = "data: [DONE]\n\n";
           reqLogger?.appendConvertedChunk?.(doneOutput);
           controller.enqueue(sharedEncoder.encode(doneOutput));
+          openAIResponsesDoneSent = true;
+          streamDoneSent = true;
         }
 
         if (!hasValidUsage(state?.usage) && totalContentLength > 0) {
           state.usage = estimateUsage(body, totalContentLength, sourceFormat);
         }
+
+        if (hasValidUsage(state?.usage)) {
+          logUsage(state.provider || targetFormat, state.usage, model, connectionId, apiKey);
+        } else {
+          appendRequestLog({ model, provider, connectionId, tokens: null, status: "200 OK" }).catch(() => { });
+        }
         
         if (onStreamComplete) {
           onStreamComplete({

package/dist/open-sse/utils/streamHandler.js

@@ -15,16 +15,19 @@ function getTimeString() {
  * @param {string} options.provider - Provider name
  * @param {string} options.model - Model name
  */
-export function createStreamController({ onDisconnect, onError, log, provider, model } = {}) {
+export function createStreamController({ onDisconnect, onError, log, provider, model, reqTag = "" } = {}) {
   const abortController = new AbortController();
   const startTime = Date.now();
   let disconnected = false;
   let abortTimeout = null;
 
-  const logStream = (status) => {
+  // Only abnormal terminations are logged; normal completion is covered by "📊 done".
+  // isError uses errorLine (always shown, ignores LOG_LEVEL) so failures survive quiet levels.
+  const logStream = (symbol, status, isError = false) => {
     const duration = Date.now() - startTime;
-    const p = provider?.toUpperCase() || "UNKNOWN";
-    console.log(`[${getTimeString()}] 🌊 [STREAM] ${p} | ${model || "unknown"} | ${duration}ms | ${status}`);
+    const emit = isError ? log?.errorLine : log?.line;
+    if (emit) emit(reqTag, symbol, `${status} · ${provider}/${model} · ${duration}ms`);
+    else console.log(`[${getTimeString()}] ${symbol} ${provider}/${model} · ${status} · ${duration}ms`);
   };
 
   return {
@@ -38,7 +41,7 @@ export function createStreamController({ onDisconnect, onError, log, provider, m
       if (disconnected) return;
       disconnected = true;
 
-      logStream(`disconnect: ${reason}`);
+      logStream("⚡", `DISCONNECT: ${reason}`);
       dbg("CTRL", `${provider}/${model} | disconnect=${reason} | dur=${Date.now() - startTime}ms`);
 
       // Delay abort to allow cleanup
@@ -49,13 +52,11 @@ export function createStreamController({ onDisconnect, onError, log, provider, m
       onDisconnect?.({ reason, duration: Date.now() - startTime });
     },
 
-    // Call when stream completes normally
+    // Call when stream completes normally (no line here — "📊 done" is authoritative)
     handleComplete: () => {
       if (disconnected) return;
       disconnected = true;
 
-      logStream("complete");
-
       if (abortTimeout) {
         clearTimeout(abortTimeout);
         abortTimeout = null;
@@ -73,11 +74,11 @@ export function createStreamController({ onDisconnect, onError, log, provider, m
       }
 
       if (error.name === "AbortError") {
-        logStream("aborted");
+        logStream("⚡", "ABORTED");
         return;
       }
 
-      logStream(`error: ${error.message}`);
+      logStream("✗", `ERROR: ${error.message}${error.stack ? `\n    ${error.stack}` : ""}`, true);
       onError?.(error);
     },
 
@@ -128,7 +129,10 @@ export function createDisconnectAwareStream(transformStream, streamController, o
         controller.enqueue(value);
       } catch (error) {
         const wasConnected = streamController.isConnected();
-        streamController.handleError(error);
+        // Controller already closed = downstream ended; not an upstream error, skip noisy log.
+        const msg0 = error?.message || "";
+        const isControllerClosed = msg0.includes("already closed") || msg0.includes("Invalid state");
+        if (!isControllerClosed) streamController.handleError(error);
         reader.cancel().catch(() => {});
         writer.abort().catch(() => {});
 
@@ -184,7 +188,7 @@ export function createDisconnectAwareStream(transformStream, streamController, o
  * @param {TransformStream} transformStream - Transform stream for SSE
  * @param {object} streamController - Stream controller from createStreamController
  */
-export function pipeWithDisconnect(providerResponse, transformStream, streamController, onAbortTerminal = null) {
+export function pipeWithDisconnect(providerResponse, transformStream, streamController, onAbortTerminal = null, stallTimeoutMs = STREAM_STALL_TIMEOUT_MS) {
   let stallTimer = null;
   let chunkCount = 0;
   let totalBytes = 0;
@@ -198,10 +202,10 @@ export function pipeWithDisconnect(providerResponse, transformStream, streamCont
     clearStall();
     stallTimer = setTimeout(() => {
       stallTimer = null;
-      dbg(tag, `STALL TIMEOUT ${STREAM_STALL_TIMEOUT_MS}ms | chunks=${chunkCount} | bytes=${totalBytes} | sinceLast=${Date.now() - lastChunkAt}ms`);
+      dbg(tag, `STALL TIMEOUT ${stallTimeoutMs}ms | chunks=${chunkCount} | bytes=${totalBytes} | sinceLast=${Date.now() - lastChunkAt}ms`);
       streamController.handleError?.(new Error("stream stall timeout"));
       streamController.abort?.();
-    }, STREAM_STALL_TIMEOUT_MS);
+    }, stallTimeoutMs);
   };
 
   // Wrap controller so every termination path clears the stall timer.
@@ -218,7 +222,7 @@ export function pipeWithDisconnect(providerResponse, transformStream, streamCont
   };
 
   armStall();
-  dbg(tag, `pipe start | stallTimeout=${STREAM_STALL_TIMEOUT_MS}ms`);
+  dbg(tag, `pipe start | stallTimeout=${stallTimeoutMs}ms`);
 
   const upstreamTap = new TransformStream({
     transform(chunk, controller) {

package/dist/open-sse/utils/streamHelpers.js

@@ -17,19 +17,6 @@ export function parseSSELine(line, format = null) {
     return null;
   }
 
-  // Handle raw JSON errors disguised as SSE
-  if (line.charCodeAt(0) === 123) { // '{' = 123
-    try {
-      const parsedJson = JSON.parse(line.trim());
-      // Check if it looks like an error payload
-      if (parsedJson && (parsedJson.error || parsedJson.code !== undefined || parsedJson.msg)) {
-        return { _isRawError: true, raw: parsedJson };
-      }
-    } catch (e) {
-      // Ignore parse errors, continue to normal SSE check
-    }
-  }
-
   // Standard SSE format: "data: {...}"
   if (line.charCodeAt(0) !== 100) return null; // 'd' = 100
 
@@ -51,13 +38,11 @@ export function hasValuableContent(chunk, format) {
   // OpenAI format
   if (format === FORMATS.OPENAI && chunk.choices?.[0]?.delta) {
     const delta = chunk.choices[0].delta;
-    return !!(delta.content && delta.content !== "" ||
+    return delta.content && delta.content !== "" ||
            delta.reasoning_content && delta.reasoning_content !== "" ||
            delta.tool_calls && delta.tool_calls.length > 0 ||
-           delta.refusal && delta.refusal !== "" ||
            chunk.choices[0].finish_reason ||
-           delta.role ||
-           chunk.usage);
+           delta.role;
   }
 
   // Claude format

package/dist/open-sse/utils/usageTracking.js

@@ -2,9 +2,11 @@
  * Token Usage Tracking - Extract, normalize, estimate and log token usage
  */
 
-import { saveRequestUsage, appendRequestLog } from '../../../dist/lib/usageDb.js';
 import { FORMATS } from "../translator/formats.js";
 
+// Legacy per-chunk usage console line; off by default (superseded by "📊 done")
+const DEBUG_USAGE = process.env.LOG_USAGE_VERBOSE === "1";
+
 // ANSI color codes
 export const COLORS = {
   reset: "\x1b[0m",
@@ -142,6 +144,68 @@ export function normalizeUsage(usage) {
   return normalized;
 }
 
+/**
+ * Canonicalize usage into ONE storage/cost convention so token counts and cost
+ * are consistent across providers:
+ *   prompt_tokens               = total input INCLUDING cache read + cache creation
+ *   cached_tokens               = cache-read portion (subset of prompt_tokens)
+ *   cache_creation_input_tokens = cache-write portion (subset of prompt_tokens)
+ *   completion_tokens, reasoning_tokens, total_tokens
+ *
+ * Discriminator: Claude reports cache_read_input_tokens with a prompt that
+ * EXCLUDES cache, so we fold cache into prompt. OpenAI/Gemini report
+ * cached_tokens already counted inside prompt, so we pass through. Idempotent:
+ * once folded the output carries cached_tokens (not cache_read_input_tokens),
+ * so re-running takes the passthrough branch and does not double-add.
+ *
+ * @param {object} usage - a normalizeUsage()-shaped object
+ * @returns {object|null} canonical token object, or null for invalid input
+ */
+export function canonicalizeUsage(usage) {
+  if (!usage || typeof usage !== "object" || Array.isArray(usage)) return null;
+
+  const num = (v) => (Number.isFinite(Number(v)) ? Number(v) : 0);
+  const completion = num(usage.completion_tokens ?? usage.output_tokens);
+  const reasoning = num(usage.reasoning_tokens);
+  // Fall back to the nested prompt_tokens_details.cache_creation_tokens shape
+  // (buildUsage()'s OpenAI-forwarding format) when the top-level field is
+  // absent, so callers that pass a buildUsage() object through don't silently
+  // drop cache_creation.
+  const cacheCreation = num(usage.cache_creation_input_tokens ?? usage.prompt_tokens_details?.cache_creation_tokens);
+
+  let prompt = num(usage.prompt_tokens ?? usage.input_tokens);
+  let cached;
+
+  // Claude path: prompt excludes cache; cache_read_input_tokens and/or
+  // cache_creation_input_tokens are separate. A cache-miss "first write" only
+  // carries cache_creation_input_tokens (no cache_read_input_tokens yet), so
+  // check both fields — otherwise a first-write request falls through to the
+  // OpenAI passthrough branch below and cache_creation never gets folded in.
+  // Guard on the absence of `cached_tokens`: our own canonical output always
+  // sets that key (even to 0), so re-running canonicalizeUsage on an already-
+  // folded result takes the passthrough branch instead of folding again.
+  if (usage.cached_tokens === undefined &&
+      (usage.cache_read_input_tokens !== undefined || usage.cache_creation_input_tokens !== undefined)) {
+    cached = num(usage.cache_read_input_tokens);
+    prompt = prompt + cached + cacheCreation;
+  } else {
+    // OpenAI/Gemini path (or already-canonical input): prompt already includes cached_tokens.
+    cached = num(usage.cached_tokens);
+  }
+
+  const result = {
+    prompt_tokens: prompt,
+    completion_tokens: completion,
+    // Recompute rather than pass through: when the fold branch ran above,
+    // an upstream total_tokens (cache-exclusive) would otherwise be stale.
+    total_tokens: prompt + completion,
+    cached_tokens: cached,
+    cache_creation_input_tokens: cacheCreation,
+  };
+  if (reasoning > 0) result.reasoning_tokens = reasoning;
+  return result;
+}
+
 /**
  * Check if usage has valid token data
  * Valid = has at least one token field with value > 0
@@ -172,6 +236,19 @@ export function hasValidUsage(usage) {
 export function extractUsage(chunk) {
   if (!chunk || typeof chunk !== "object") return null;
 
+  // Claude format (message_start event): carries input_tokens + cache_read +
+  // cache_creation. message_delta later carries only the final output_tokens,
+  // so callers must MERGE (mergeUsage), not overwrite, to keep cache counts.
+  if (chunk.type === "message_start" && chunk.message?.usage && typeof chunk.message.usage === "object") {
+    const u = chunk.message.usage;
+    return normalizeUsage({
+      prompt_tokens: u.input_tokens || 0,
+      completion_tokens: u.output_tokens || 0,
+      cache_read_input_tokens: u.cache_read_input_tokens,
+      cache_creation_input_tokens: u.cache_creation_input_tokens
+    });
+  }
+
   // Claude format (message_delta event)
   if (chunk.type === "message_delta" && chunk.usage && typeof chunk.usage === "object") {
     return normalizeUsage({
@@ -233,6 +310,27 @@ export function extractUsage(chunk) {
   return null;
 }
 
+// Field-wise max-merge of two usage objects. Anthropic splits usage across
+// events: message_start has real input+cache (output is a placeholder 1),
+// message_delta has the real cumulative output (input/cache absent). Max keeps
+// the meaningful value from each without clobbering. Idempotent for other
+// providers that emit a single complete usage object.
+export function mergeUsage(prev, next) {
+  if (!prev) return next || null;
+  if (!next) return prev;
+  const merged = { ...prev };
+  for (const [k, v] of Object.entries(next)) {
+    // typeof NaN === "number" — guard with Number.isFinite so one malformed
+    // chunk can't poison the whole accumulation (Math.max(x, NaN) is NaN).
+    if (typeof v === "number" && Number.isFinite(v)) {
+      merged[k] = Math.max(typeof merged[k] === "number" ? merged[k] : 0, v);
+    } else if (v && typeof v === "object") {
+      merged[k] = v; // nested details objects: take latest
+    }
+  }
+  return merged;
+}
+
 /**
  * Estimate input tokens from request body
  * Calculate total body size for more accurate estimation
@@ -306,6 +404,10 @@ export function estimateUsage(body, contentLength, targetFormat = FORMATS.OPENAI
 export function logUsage(provider, usage, model = null, connectionId = null, apiKey = null) {
   if (!usage || typeof usage !== "object") return;
 
+  // Console output moved to the unified "📊 done" line (streamingHandler). Kept as
+  // a no-op hook so callers stay unchanged; usage persistence happens via saveUsageStats.
+  if (!DEBUG_USAGE) return;
+
   const p = provider?.toUpperCase() || "UNKNOWN";
 
   // Support both formats:
@@ -333,15 +435,4 @@ export function logUsage(provider, usage, model = null, connectionId = null, api
   if (reasoning) msg += ` | reasoning=${reasoning}`;
 
   console.log(msg);
-
-  // Save to usage DB
-  const tokens = {
-    prompt_tokens: inTokens,
-    completion_tokens: outTokens,
-    cache_read_input_tokens: cacheRead || 0,
-    cache_creation_input_tokens: cacheCreation || 0,
-    reasoning_tokens: reasoning || 0
-  };
-  saveRequestUsage({ model, provider, connectionId, tokens, apiKey: apiKey || undefined }).catch(() => { });
-  appendRequestLog({ model, provider, connectionId, tokens, status: "200 OK" }).catch(() => { });
 }

package/open-sse/config/appConstants.js

@@ -1,44 +1,10 @@
 import { platform, arch } from "os";
-import { proxyAwareFetch } from "../utils/proxyFetch.js";
-
-// === Kimchi CLI version (dynamic, cached from GitHub releases) ===
-const KIMCHI_FALLBACK_VERSION = "0.1.39";
-let _kimchiVersionCache = null;
-let _kimchiVersionFetching = null;
-const KIMCHI_VERSION_TTL_MS = 3600_000; // 1h
-
-async function fetchKimchiVersion() {
-  try {
-    const res = await proxyAwareFetch("https://api.github.com/repos/getkimchi/kimchi/releases/latest", {
-      headers: { Accept: "application/vnd.github+json" },
-      signal: AbortSignal.timeout(5000),
-    });
-    if (!res.ok) return KIMCHI_FALLBACK_VERSION;
-    const body = await res.json();
-    const tag = body?.tag_name || "";
-    const ver = tag.replace(/^v/, "");
-    return ver || KIMCHI_FALLBACK_VERSION;
-  } catch {
-    return KIMCHI_FALLBACK_VERSION;
-  }
-}
-
-export async function getKimchiVersion() {
-  if (_kimchiVersionCache && Date.now() - _kimchiVersionCache.ts < KIMCHI_VERSION_TTL_MS) return _kimchiVersionCache.ver;
-  if (_kimchiVersionFetching) return _kimchiVersionFetching;
-  _kimchiVersionFetching = fetchKimchiVersion().finally(() => { _kimchiVersionFetching = null; });
-  const ver = await _kimchiVersionFetching;
-  _kimchiVersionCache = { ver, ts: Date.now() };
-  return ver;
-}
+import { PROVIDERS, PROVIDER_OAUTH } from "./providers.js";
+import { ANTIGRAVITY_IDE_USER_AGENT } from "../providers/shared.js";
 
-export function getKimchiVersionSync() {
-  return _kimchiVersionCache?.ver || KIMCHI_FALLBACK_VERSION;
-}
-
-// === Gemini CLI ===
-export const GEMINI_CLI_VERSION = "0.34.0";
-export const GEMINI_CLI_API_CLIENT = "google-genai-sdk/1.41.0 gl-node/v22.19.0";
+// === Gemini CLI === derive từ registry gemini-cli.transport
+export const GEMINI_CLI_VERSION = PROVIDERS["gemini-cli"]?.cliVersion;
+export const GEMINI_CLI_API_CLIENT = PROVIDERS["gemini-cli"]?.apiClient;
 
 // Map Node arch to Gemini CLI arch string (x64/x86/arm64/...)
 function geminiCLIArch() {
@@ -52,11 +18,13 @@ export function geminiCLIUserAgent(model = "unknown") {
 }
 
 // === GitHub Copilot ===
+// Derive từ registry github.transport.copilot
+const _ghCopilot = PROVIDERS.github?.copilot || {};
 export const GITHUB_COPILOT = {
-  VSCODE_VERSION: "1.110.0",
-  COPILOT_CHAT_VERSION: "0.38.0",
-  USER_AGENT: "GitHubCopilotChat/0.38.0",
-  API_VERSION: "2025-04-01",
+  VSCODE_VERSION: _ghCopilot.vscodeVersion,
+  COPILOT_CHAT_VERSION: _ghCopilot.chatVersion,
+  USER_AGENT: _ghCopilot.userAgent,
+  API_VERSION: _ghCopilot.apiVersion,
 };
 
 // === Antigravity enums ===
@@ -92,7 +60,7 @@ export function getPlatformEnum() {
 }
 
 export function getPlatformUserAgent() {
-  return `antigravity/1.104.0 ${platform()}/${arch()}`;
+  return ANTIGRAVITY_IDE_USER_AGENT;
 }
 
 export const CLIENT_METADATA = {
@@ -162,7 +130,7 @@ export const AG_DEFAULT_TOOLS = new Set([
 
 // Antigravity chat/stream headers
 export const ANTIGRAVITY_HEADERS = {
-  "User-Agent": `antigravity/1.107.0 ${platform()}/${arch()}`
+  "User-Agent": ANTIGRAVITY_IDE_USER_AGENT
 };
 
 // Cloud Code Assist API
@@ -188,43 +156,19 @@ export const LOAD_CODE_ASSIST_METADATA = {
 export const CLAUDE_SYSTEM_PROMPT = "You are Claude Code, Anthropic's official CLI for Claude.";
 export const ANTIGRAVITY_DEFAULT_SYSTEM = "You are Antigravity, a powerful agentic AI coding assistant designed by the Google Deepmind team working on Advanced Agentic Coding.You are pair programming with a USER to solve their coding task. The task may require creating a new codebase, modifying or debugging an existing codebase, or simply answering a question.**Absolute paths only****Proactiveness**";
 
-// Proactive token refresh lead times per provider (ms)
-export const REFRESH_LEAD_MS = {
-  codex:       5 * 24 * 60 * 60 * 1000, // 5 days
-  claude:       4 * 60 * 60 * 1000,     // 4 hours
-  iflow:       24 * 60 * 60 * 1000,     // 24 hours
-  qwen:        20 * 60 * 1000,          // 20 minutes
-  "kimi-coding": 5 * 60 * 1000,         // 5 minutes
-  antigravity:  5 * 60 * 1000,          // 5 minutes
-};
+// Derive từ registry oauth.refreshLeadMs
+export const REFRESH_LEAD_MS = Object.fromEntries(
+  Object.entries(PROVIDER_OAUTH).filter(([, o]) => o.refreshLeadMs).map(([id, o]) => [id, o.refreshLeadMs])
+);
 
 // OAuth endpoints
 export const OAUTH_ENDPOINTS = {
-  google: {
-    token: "https://oauth2.googleapis.com/token",
-    auth: "https://accounts.google.com/o/oauth2/auth"
-  },
-  openai: {
-    token: "https://auth.openai.com/oauth/token",
-    auth: "https://auth.openai.com/oauth/authorize"
-  },
-  anthropic: {
-    token: "https://api.anthropic.com/v1/oauth/token",
-    auth: "https://api.anthropic.com/v1/oauth/authorize"
-  },
-  qwen: {
-    token: "https://qwen.ai/api/v1/oauth2/token",
-    auth: "https://qwen.ai/api/v1/oauth2/device/code"
-  },
-  iflow: {
-    token: "https://iflow.cn/oauth/token",
-    auth: "https://iflow.cn/oauth"
-  },
-  github: {
-    token: "https://github.com/login/oauth/access_token",
-    auth: "https://github.com/login/oauth/authorize",
-    deviceCode: "https://github.com/login/device/code"
-  }
+  google:    { token: "https://oauth2.googleapis.com/token", auth: "https://accounts.google.com/o/oauth2/auth" },
+  openai:    { token: PROVIDER_OAUTH["codex"]?.tokenUrl, auth: PROVIDER_OAUTH["codex"]?.authorizeUrl },
+  anthropic: { token: PROVIDER_OAUTH["claude"]?.tokenUrl, auth: "https://api.anthropic.com/v1/oauth/authorize" }, // ≠ claude.authorizeUrl (claude.ai login) — keep
+  qwen:      { token: PROVIDER_OAUTH["qwen"]?.tokenUrl, auth: PROVIDER_OAUTH["qwen"]?.deviceCodeUrl },
+  iflow:     { token: PROVIDER_OAUTH["iflow"]?.tokenUrl, auth: PROVIDER_OAUTH["iflow"]?.authorizeUrl },
+  github:    { token: PROVIDER_OAUTH["github"]?.tokenUrl, auth: PROVIDER_OAUTH["github"]?.authorizeUrl, deviceCode: PROVIDER_OAUTH["github"]?.deviceCodeUrl },
 };
 
 // Generate Kimi OAuth custom headers