@livx.cc/agentx 0.99.74 → 0.99.75

package/dist/{Agent-Bs2ufrIU.d.ts → Agent-BBlKI8uD.d.ts}

@@ -308,6 +308,13 @@ interface RunResult {
         cacheCreationTokens?: number;
         cacheReadTokens?: number;
     };
+    /** Present iff `finishReason === 'budget'`: the weighted spend vs `maxTokens`, and the last step's prompt
+     *  size — a prompt near the limit means the CONVERSATION is too big for one run (compact / new chat). */
+    budget?: {
+        spent: number;
+        limit: number;
+        stepPrompt: number;
+    };
     /** True if ANY turn's usage was estimated (provider gave none) rather than exact — lets the UI mark cost `~`. */
     usageEstimated?: boolean;
     /** Set by a router (`--model auto`) whose turn spans several models: the turn's USD at each model's own rate. */

package/dist/cli.d.ts

@@ -1,5 +1,5 @@
 #!/usr/bin/env bun
-import { m as RouterState, H as Hooks, n as RunResult, R as ReasoningEffort, A as Agent } from './Agent-Bs2ufrIU.js';
+import { m as RouterState, H as Hooks, n as RunResult, R as ReasoningEffort, A as Agent } from './Agent-BBlKI8uD.js';
 import { IFilesystem } from '@livx.cc/wcli/core';
 import { M as Message, U as UserQuestion, H as HostBridge, c as ContentPart, o as MessageContent } from './tools-lC3jn60o.js';
 
@@ -297,6 +297,11 @@ declare function jsonResult(res: RunResult, session: SessionData): {
     sessionId: string;
     costUsd?: number | undefined;
     routing?: Record<string, unknown> | undefined;
+    budget?: {
+        spent: number;
+        limit: number;
+        stepPrompt: number;
+    } | undefined;
     text: string;
     steps: number;
     tools: number;

package/dist/cli.js

@@ -4876,6 +4876,7 @@ var Agent = class _Agent {
     let steps = 0;
     const usage = { promptTokens: 0, completionTokens: 0, totalTokens: 0, cacheCreationTokens: 0, cacheReadTokens: 0 };
     let usageEstimated = false;
+    let spent = 0, prevPrompt = -1, stepPrompt = 0;
     const start = Date.now();
     this.parkedMs = 0;
     this.ctx.needsInput = void 0;
@@ -4886,12 +4887,12 @@ var Agent = class _Agent {
     let closureNudged = false;
     let lastProgressAt = start;
     const kill = (finishReason) => {
-      log8.warn(`kill-switch: ${finishReason} (steps=${steps}, tokens=${usage.totalTokens}, budgetTokens=${Math.round(usage.totalTokens - 0.9 * usage.cacheReadTokens)}, ms=${Date.now() - start - this.parkedMs}${finishReason === "timeout" ? `, idle=${Date.now() - lastProgressAt - this.parkedMs}ms` : ""}${this.parkedMs ? ` +${this.parkedMs} parked` : ""})`);
+      log8.warn(`kill-switch: ${finishReason} (steps=${steps}, tokens=${usage.totalTokens}, budgetTokens=${Math.round(spent)}, ms=${Date.now() - start - this.parkedMs}${finishReason === "timeout" ? `, idle=${Date.now() - lastProgressAt - this.parkedMs}ms` : ""}${this.parkedMs ? ` +${this.parkedMs} parked` : ""})`);
       this.jobs.killAll(`run ended: ${finishReason}`, true);
       const claims = this.survivingClaims();
-      return { text: lastAssistantText(this.transcript), steps, finishReason, messages: this.transcript, usage, usageEstimated, ...claims.length ? { fabricatedToolClaims: claims } : {} };
+      return { text: lastAssistantText(this.transcript), steps, finishReason, messages: this.transcript, usage, usageEstimated, ...finishReason === "budget" ? { budget: { spent: Math.round(spent), limit: o.maxTokens, stepPrompt } } : {}, ...claims.length ? { fabricatedToolClaims: claims } : {} };
     };
-    const overBudget = () => !!o.maxTokens && usage.totalTokens - 0.9 * usage.cacheReadTokens >= o.maxTokens;
+    const overBudget = () => !!o.maxTokens && spent >= o.maxTokens;
     const canNudge = () => steps < o.maxSteps && !overBudget();
     while (true) {
       if (o.signal?.aborted) return kill("aborted");
@@ -5045,6 +5046,14 @@ var Agent = class _Agent {
         usage.totalTokens += res.usage.totalTokens ?? 0;
         usage.cacheCreationTokens += res.usage.cacheCreationTokens ?? 0;
         usage.cacheReadTokens += res.usage.cacheReadTokens ?? 0;
+        const prompt = res.usage.promptTokens ?? 0, cacheRead = res.usage.cacheReadTokens ?? 0;
+        if (prevPrompt < 0) {
+          const u = sent.map((m) => m.role).lastIndexOf("user");
+          prevPrompt = u > 0 ? estimateTokens(sent.slice(0, u)) : 0;
+        }
+        const reused = Math.min(prompt, Math.max(cacheRead, prevPrompt));
+        spent += (res.usage.totalTokens ?? 0) - 0.9 * reused;
+        prevPrompt = stepPrompt = prompt;
       }
       const delegatedTools = res.delegatedTools ?? [];
       if (delegatedTools.length) {
@@ -16942,6 +16951,7 @@ function jsonResult(res, session) {
     steps: res.steps,
     tools: res.messages.slice(lastUser).filter((m) => m.role === "tool").length,
     usage: res.usage,
+    ...res.budget ? { budget: res.budget } : {},
     ...res.routing ? { costUsd: res.costUsd, routing: res.routing } : {},
     // cast: tsup dts hoists a modelRouter import above the shebang
     sessionId: session.meta.id,