@inetafrica/open-claudia 2.6.27 → 2.6.28

package/core/runner.js

@@ -130,6 +130,11 @@ function logTurnUsage(state, usage, costUsd, announce, opts = {}) {
   const backend = state.settings?.backend || "claude";
   const parts = usageParts(usage, backend);
   const usageScope = opts.usageScope || "turn_delta";
+  // `parts.context` is the per-turn SUM of every API call's prefix, inflated
+  // by re-counting the cached prefix once per tool round-trip. When the caller
+  // tracked the true peak single-call prefix, report that as contextTokens and
+  // keep the inflated sum as billedContextTokens for cost attribution.
+  const hasPeak = Number.isFinite(opts.liveContextTokens) && opts.liveContextTokens > 0;
   const record = {
     ts: new Date().toISOString(),
     version: PKG_VERSION,
@@ -142,7 +147,8 @@ function logTurnUsage(state, usage, costUsd, announce, opts = {}) {
     outputTokens: parts.output,
     cacheReadTokens: parts.cacheRead,
     cacheCreationTokens: parts.cacheCreation,
-    contextTokens: parts.context,
+    contextTokens: hasPeak ? opts.liveContextTokens : parts.context,
+    billedContextTokens: parts.context,
     costUsd: typeof costUsd === "number" ? costUsd : 0,
   };
   if (opts.rawUsage && opts.rawUsage !== usage) {
@@ -785,6 +791,11 @@ async function runClaude(prompt, cwd, replyToMsgId, opts = {}) {
   let toolUses = [];
   let currentTool = null;
   let currentToolDetail = "";
+  // True context-window occupancy is the largest single API call's prefix
+  // (input + cache_read + cache_creation), NOT the per-turn sum the result
+  // event reports — that sum re-counts the cached prefix once per tool
+  // round-trip and balloons to millions. Track the peak per-call prefix.
+  let peakContextTokens = 0;
 
   // Hermes-style skill announcements: one chat line when a learned skill
   // is invoked or its SKILL.md is written/patched. Deduped per turn.
@@ -897,6 +908,10 @@ async function runClaude(prompt, cwd, replyToMsgId, opts = {}) {
     const lastNewline = state.streamBuffer.lastIndexOf("\n");
     state.streamBuffer = lastNewline >= 0 ? state.streamBuffer.slice(lastNewline + 1) : state.streamBuffer;
     for (const evt of events) {
+      if (evt.type === "assistant" && evt.message?.usage) {
+        const callPrefix = usageParts(evt.message.usage, settings.backend || "claude").context;
+        if (callPrefix > peakContextTokens) peakContextTokens = callPrefix;
+      }
       if (evt.type === "assistant" && evt.message?.content) {
         for (const block of evt.message.content) {
           if (block.type === "text") assistantText += block.text;
@@ -1003,8 +1018,10 @@ async function runClaude(prompt, cwd, replyToMsgId, opts = {}) {
         // Codex already reports usage at turn.completed. A result usage block,
         // if present, is the same cumulative total and must not be counted again.
         if (evt.usage && settings.backend !== "codex") {
-          applyUsageToState(state, evt.usage, typeof evt.total_cost_usd === "number" ? evt.total_cost_usd : 0, { usageScope: "turn_delta" });
-          logTurnUsage(state, evt.usage, typeof evt.total_cost_usd === "number" ? evt.total_cost_usd : 0, (text) => chatContext.run(store, () => send(text)), { usageScope: "turn_delta" });
+          const peakOpts = { usageScope: "turn_delta" };
+          if (peakContextTokens > 0) peakOpts.liveContextTokens = peakContextTokens;
+          applyUsageToState(state, evt.usage, typeof evt.total_cost_usd === "number" ? evt.total_cost_usd : 0, peakOpts);
+          logTurnUsage(state, evt.usage, typeof evt.total_cost_usd === "number" ? evt.total_cost_usd : 0, (text) => chatContext.run(store, () => send(text)), peakOpts);
         }
         if (typeof evt.total_cost_usd === "number" && settings.backend === "codex") state.sessionUsage.costUsd += evt.total_cost_usd;
         saveState();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@inetafrica/open-claudia",
-  "version": "2.6.27",
+  "version": "2.6.28",
   "description": "Your always-on AI coding assistant — Claude Code, Cursor Agent, and OpenAI Codex via Telegram or Kazee Chat",
   "main": "bot.js",
   "bin": {