claudish 7.19.0 → 7.19.1

package/dist/index.js

@@ -651,7 +651,7 @@ var init_onepassword_config = __esm(() => {
 });
 
 // src/version.ts
-var VERSION = "7.19.0";
+var VERSION = "7.19.1";
 
 // src/logger.ts
 var exports_logger = {};
@@ -31745,6 +31745,14 @@ var init_web_search_detector = __esm(() => {
   WEB_SEARCH_NAMES = new Set(["web_search", "brave_web_search", "tavily_search"]);
 });
 
+// src/handlers/shared/stream-parsers/message-start-usage.ts
+function messageStartUsage(priorInputTokens) {
+  return {
+    input_tokens: priorInputTokens && priorInputTokens > 0 ? priorInputTokens : 100,
+    output_tokens: 1
+  };
+}
+
 // src/handlers/shared/stream-parsers/openai-sse.ts
 function validateToolArguments(toolName, argsStr, toolSchemas, textContent) {
   const result = validateAndRepairToolCall(toolName, argsStr, toolSchemas, textContent);
@@ -31774,7 +31782,7 @@ function createStreamingState() {
     accumulatedText: ""
   };
 }
-function createStreamingResponseHandler(c, response, adapter, target, middlewareManager, onTokenUpdate, toolSchemas, toolNameMap) {
+function createStreamingResponseHandler(c, response, adapter, target, middlewareManager, onTokenUpdate, toolSchemas, toolNameMap, priorInputTokens) {
   log(`[Streaming] ===== HANDLER STARTED for ${target} =====`);
   let isClosed = false;
   let ping = null;
@@ -31803,7 +31811,7 @@ data: ${JSON.stringify(d)}
           model: target,
           stop_reason: null,
           stop_sequence: null,
-          usage: { input_tokens: 100, output_tokens: 1 }
+          usage: messageStartUsage(priorInputTokens)
         }
       });
       send("ping", { type: "ping" });
@@ -31916,7 +31924,10 @@ data: ${JSON.stringify(d)}
           send("message_delta", {
             type: "message_delta",
             delta: { stop_reason: stopReason, stop_sequence: null },
-            usage: { output_tokens: state.usage?.completion_tokens || 0 }
+            usage: {
+              ...state.usage?.prompt_tokens ? { input_tokens: state.usage.prompt_tokens } : {},
+              output_tokens: state.usage?.completion_tokens || 0
+            }
           });
           send("message_stop", { type: "message_stop" });
         }
@@ -31927,7 +31938,7 @@ data: ${JSON.stringify(d)}
           } else {
             const estimatedOutputTokens = Math.ceil(state.accumulatedText.length / 4);
             log(`[Streaming] No usage data from provider, estimating: ~${estimatedOutputTokens} output tokens`);
-            onTokenUpdate(100, estimatedOutputTokens);
+            onTokenUpdate(priorInputTokens || 100, estimatedOutputTokens);
           }
         }
         if (!isClosed) {
@@ -37545,7 +37556,7 @@ data: ${JSON.stringify(data)}
           model: opts.modelName,
           stop_reason: null,
           stop_sequence: null,
-          usage: { input_tokens: 100, output_tokens: 1 }
+          usage: messageStartUsage(opts.priorInputTokens)
         }
       });
       send("ping", { type: "ping" });
@@ -37589,7 +37600,10 @@ data: ${JSON.stringify(data)}
           send("message_delta", {
             type: "message_delta",
             delta: { stop_reason: hasToolCalls ? "tool_use" : "end_turn", stop_sequence: null },
-            usage: { output_tokens: outputTokens }
+            usage: {
+              ...inputTokens > 0 ? { input_tokens: inputTokens } : {},
+              output_tokens: outputTokens
+            }
           });
           send("message_stop", { type: "message_stop" });
         }
@@ -37792,7 +37806,7 @@ data: ${JSON.stringify(data)}
           model: opts.modelName,
           stop_reason: null,
           stop_sequence: null,
-          usage: { input_tokens: 100, output_tokens: 1 }
+          usage: messageStartUsage(opts.priorInputTokens)
         }
       });
       send("ping", { type: "ping" });
@@ -37813,7 +37827,10 @@ data: ${JSON.stringify(data)}
           send("message_delta", {
             type: "message_delta",
             delta: { stop_reason: "end_turn", stop_sequence: null },
-            usage: { output_tokens: completionTokens }
+            usage: {
+              ...promptTokens > 0 ? { input_tokens: promptTokens } : {},
+              output_tokens: completionTokens
+            }
           });
           send("message_stop", { type: "message_stop" });
         }
@@ -37965,7 +37982,7 @@ data: ${JSON.stringify(data)}
           model: opts.modelName,
           stop_reason: null,
           stop_sequence: null,
-          usage: { input_tokens: 100, output_tokens: 1 }
+          usage: messageStartUsage(opts.priorInputTokens)
         }
       });
       send("ping", { type: "ping" });
@@ -38255,6 +38272,7 @@ class TokenTracker {
   sessionTotalCost = 0;
   sessionInputTokens = 0;
   sessionOutputTokens = 0;
+  lastInputTokens = 0;
   modelNameOverride;
   quotaRemaining;
   constructor(port, config2) {
@@ -38271,10 +38289,11 @@ class TokenTracker {
     this.quotaRemaining = fraction;
   }
   rewrite() {
-    this.writeFile(this.sessionInputTokens, this.sessionOutputTokens);
+    this.writeFile(this.getLastInputTokens(), this.sessionOutputTokens);
   }
   update(inputTokens, outputTokens) {
     this.sessionInputTokens = inputTokens;
+    this.lastInputTokens = inputTokens;
     this.sessionOutputTokens += outputTokens;
     const pricing = this.getPricing();
     const cost = inputTokens / 1e6 * pricing.inputCostPer1M + outputTokens / 1e6 * pricing.outputCostPer1M;
@@ -38283,6 +38302,7 @@ class TokenTracker {
   }
   accumulateBoth(inputTokens, outputTokens) {
     this.sessionInputTokens += inputTokens;
+    this.lastInputTokens = this.sessionInputTokens;
     this.sessionOutputTokens += outputTokens;
     const pricing = this.getPricing();
     const cost = this.sessionInputTokens / 1e6 * pricing.inputCostPer1M + this.sessionOutputTokens / 1e6 * pricing.outputCostPer1M;
@@ -38291,6 +38311,7 @@ class TokenTracker {
   }
   updateWithDelta(inputTokens, outputTokens) {
     let incrementalInputTokens;
+    this.lastInputTokens = inputTokens;
     if (inputTokens >= this.sessionInputTokens) {
       incrementalInputTokens = inputTokens - this.sessionInputTokens;
       this.sessionInputTokens = inputTokens;
@@ -38306,10 +38327,11 @@ class TokenTracker {
     const pricing = this.getPricing();
     const cost = incrementalInputTokens / 1e6 * pricing.inputCostPer1M + outputTokens / 1e6 * pricing.outputCostPer1M;
     this.sessionTotalCost += cost;
-    this.writeFile(Math.max(inputTokens, this.sessionInputTokens), this.sessionOutputTokens, pricing.isEstimate);
+    this.writeFile(inputTokens, this.sessionOutputTokens, pricing.isEstimate);
   }
   updateWithActualCost(inputTokens, outputTokens, actualCost) {
     this.sessionInputTokens = inputTokens;
+    this.lastInputTokens = inputTokens;
     this.sessionOutputTokens += outputTokens;
     if (typeof actualCost === "number" && actualCost > 0) {
       this.sessionTotalCost += actualCost;
@@ -38325,6 +38347,7 @@ class TokenTracker {
   updateLocal(inputTokens, outputTokens) {
     if (inputTokens > 0) {
       this.sessionInputTokens = inputTokens;
+      this.lastInputTokens = inputTokens;
     }
     this.sessionOutputTokens += outputTokens;
     this.writeFile(this.sessionInputTokens, this.sessionOutputTokens);
@@ -38338,6 +38361,9 @@ class TokenTracker {
   getInputTokens() {
     return this.sessionInputTokens;
   }
+  getLastInputTokens() {
+    return this.lastInputTokens || this.sessionInputTokens;
+  }
   getOutputTokens() {
     return this.sessionOutputTokens;
   }
@@ -38879,16 +38905,18 @@ class ComposedHandler {
       }
     };
     const streamFormat = this.provider.overrideStreamFormat?.() ?? this.explicitAdapter?.getStreamFormat() ?? this.modelAdapter?.getStreamFormat() ?? this.getAdapter().getStreamFormat();
+    const priorInputTokens = this.tokenTracker.getLastInputTokens();
     switch (streamFormat) {
       case "openai-sse":
-        return createStreamingResponseHandler(c, response, adapter, this.bareModelName, this.middlewareManager, onTokenUpdate, claudeRequest.tools, toolNameMap);
+        return createStreamingResponseHandler(c, response, adapter, this.bareModelName, this.middlewareManager, onTokenUpdate, claudeRequest.tools, toolNameMap, priorInputTokens);
       case "openai-responses-sse":
         return createResponsesStreamHandler(c, response, {
           modelName: this.bareModelName,
           onTokenUpdate,
           toolNameMap: adapter.getToolNameMap(),
           contextWindow: lookupModelForProvider(this.bareModelName, this.provider.name),
-          onApiError
+          onApiError,
+          priorInputTokens
         });
       case "anthropic-sse":
         return createAnthropicPassthroughStream(c, response, {
@@ -38908,13 +38936,15 @@ class ComposedHandler {
           middlewareManager: this.middlewareManager,
           onTokenUpdate,
           onToolCall,
-          unwrapResponse: this.options.unwrapGeminiResponse
+          unwrapResponse: this.options.unwrapGeminiResponse,
+          priorInputTokens
         });
       }
       case "ollama-jsonl":
         return createOllamaJsonlStream(c, response, {
           modelName: this.bareModelName,
-          onTokenUpdate
+          onTokenUpdate,
+          priorInputTokens
         });
       default:
         throw new Error(`Unknown stream format: ${streamFormat}`);
@@ -71147,7 +71177,8 @@ __export(exports_claude_runner, {
   isProxyAuthMode: () => isProxyAuthMode,
   computeMainThreadContextWindow: () => computeMainThreadContextWindow,
   checkClaudeInstalled: () => checkClaudeInstalled,
-  buildClaudishSettingsOverlay: () => buildClaudishSettingsOverlay
+  buildClaudishSettingsOverlay: () => buildClaudishSettingsOverlay,
+  MIN_AUTO_COMPACT_WINDOW: () => MIN_AUTO_COMPACT_WINDOW
 });
 import { spawn as spawn4 } from "child_process";
 import {
@@ -71477,11 +71508,13 @@ async function runClaudeWithProxy(config3, proxyUrl, onCleanup) {
       env.ANTHROPIC_AUTH_TOKEN = "placeholder-token-not-used-proxy-handles-auth";
       if (!process.env[ENV.CLAUDE_CODE_AUTO_COMPACT_WINDOW]) {
         const autoCompactWindow = await computeMainThreadContextWindow(config3);
-        if (autoCompactWindow > 0) {
+        if (autoCompactWindow >= MIN_AUTO_COMPACT_WINDOW) {
           env[ENV.CLAUDE_CODE_AUTO_COMPACT_WINDOW] = String(autoCompactWindow);
           if (!config3.quiet) {
             console.error(`[claudish] Auto-compact window: ${autoCompactWindow.toLocaleString()} tokens ` + "(Claude Code compacts before the backend's real limit)");
           }
+        } else if (autoCompactWindow > 0 && !config3.quiet) {
+          console.error(`[claudish] Model's real context window (${autoCompactWindow.toLocaleString()}) is below ` + `Claude Code's ${MIN_AUTO_COMPACT_WINDOW.toLocaleString()}-token auto-compact floor \u2014 ` + "leaving CLAUDE_CODE_AUTO_COMPACT_WINDOW unset so native auto-compaction stays on.");
         }
       }
     }
@@ -71666,7 +71699,7 @@ async function checkClaudeInstalled() {
   const binary = await findClaudeBinary();
   return binary !== null;
 }
-var restoreTerminal = null;
+var restoreTerminal = null, MIN_AUTO_COMPACT_WINDOW = 200000;
 var init_claude_runner = __esm(() => {
   init_model_catalog();
   init_config();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claudish",
-  "version": "7.19.0",
+  "version": "7.19.1",
   "description": "Run Claude Code with any model - OpenRouter, Ollama, LM Studio & local models",
   "type": "module",
   "main": "./dist/index.js",
@@ -60,10 +60,10 @@
     "ai"
   ],
   "optionalDependencies": {
-    "@claudish/magmux-darwin-arm64": "7.19.0",
-    "@claudish/magmux-darwin-x64": "7.19.0",
-    "@claudish/magmux-linux-arm64": "7.19.0",
-    "@claudish/magmux-linux-x64": "7.19.0"
+    "@claudish/magmux-darwin-arm64": "7.19.1",
+    "@claudish/magmux-darwin-x64": "7.19.1",
+    "@claudish/magmux-linux-arm64": "7.19.1",
+    "@claudish/magmux-linux-x64": "7.19.1"
   },
   "author": "Jack Rudenko <i@madappgang.com>",
   "license": "MIT",