micro-models-agent 0.61.1 → 0.62.0

package/dist/main.js

@@ -3188,6 +3188,25 @@ Apply a matching solution from these results. If none is relevant — do NOT rep
     "repl.cost": "Total cost: {cost}",
     "repl.cost_breakdown": "By provider: {breakdown}",
     "repl.tokens": "Tokens used: {tokens}",
+    "repl.cache": "Cache: {hit}% hit · saved {saved}",
+    "repl.cache_nosave": "Cache: {hit}% hit",
+    "repl.prefix": "Prefix stable: {stable}% · broke: {cause}",
+    "cache.cause.system": "system prompt",
+    "cache.cause.tools": "tool set",
+    "cache.cause.history": "history",
+    "cache.cause.volatile": "volatile content",
+    "cache.cause.unknown": "unknown",
+    "cli.session_usage": "API usage: {prompt} prompt + {completion} completion = {total} tokens",
+    "cli.session_cache": "Cache: {hit}% hit ({cached} cached / {uncached} uncached)",
+    "cli.usage": "Show provider balance/usage (OpenRouter)",
+    "cli.usage_unsupported": 'Provider "{provider}" does not expose an API balance. Only OpenRouter does; OpenCode Zen/Go show it in the web dashboard.',
+    "cli.usage_no_key": "No API key configured for the active provider.",
+    "cli.usage_error": "Failed to fetch balance: {error}",
+    "cli.usage_key_usage": "Key usage: {usage}",
+    "cli.usage_key_limit": "Key limit: {limit} · remaining {remaining}",
+    "cli.usage_balance": "Balance: {balance}",
+    "cli.usage_account": "Account: {credits} purchased · {used} used",
+    "cli.usage_empty": "No balance information returned.",
     "repl.ctrl_c_interrupt": `
 [Ctrl+C] Stopping agent... (press again to force)`,
     "exec.stop_directive": 'STOP. Step {stepId} ("{description}") took {iterations} iterations with no progress. DO NOT continue this step. Immediately call: plan update step={stepId} status=done (if code works despite warnings) OR plan update step={stepId} status=skipped note="reason". Do NOT make any other tool calls before updating the plan.',
@@ -4005,6 +4024,25 @@ var init_ru = __esm(() => {
     "repl.cost": "Итого потрачено: {cost}",
     "repl.cost_breakdown": "По провайдерам: {breakdown}",
     "repl.tokens": "Потрачено токенов: {tokens}",
+    "repl.cache": "Кеш: {hit}% попаданий · сэкономлено {saved}",
+    "repl.cache_nosave": "Кеш: {hit}% попаданий",
+    "repl.prefix": "Префикс стабилен: {stable}% · сломалось: {cause}",
+    "cache.cause.system": "системный промпт",
+    "cache.cause.tools": "набор инструментов",
+    "cache.cause.history": "история",
+    "cache.cause.volatile": "волатильное содержимое",
+    "cache.cause.unknown": "неизвестно",
+    "cli.session_usage": "API-использование: {prompt} prompt + {completion} completion = {total} токенов",
+    "cli.session_cache": "Кеш: {hit}% попаданий ({cached} из кеша / {uncached} новых)",
+    "cli.usage": "Показать баланс/расход провайдера (OpenRouter)",
+    "cli.usage_unsupported": 'Провайдер "{provider}" не отдаёт баланс по API. Это умеет только OpenRouter; OpenCode Zen/Go показывают его в веб-дашборде.',
+    "cli.usage_no_key": "Для активного провайдера не настроен API-ключ.",
+    "cli.usage_error": "Не удалось получить баланс: {error}",
+    "cli.usage_key_usage": "Расход по ключу: {usage}",
+    "cli.usage_key_limit": "Лимит ключа: {limit} · осталось {remaining}",
+    "cli.usage_balance": "Баланс: {balance}",
+    "cli.usage_account": "Аккаунт: куплено {credits} · израсходовано {used}",
+    "cli.usage_empty": "Провайдер не вернул данных о балансе.",
     "repl.ctrl_c_interrupt": `
 [Ctrl+C] Остановка агента... (ещё раз — принудительно)`,
     "exec.stop_directive": 'STOP. Step {stepId} ("{description}") took {iterations} iterations with no progress. DO NOT continue this step. Immediately call: plan update step={stepId} status=done (if code works despite warnings) OR plan update step={stepId} status=skipped note="reason". Do NOT make any other tool calls before updating the plan.',
@@ -5893,6 +5931,73 @@ class StreamState {
   }
 }
 
+// src/llm/cache-usage.ts
+function num(value) {
+  return typeof value === "number" && Number.isFinite(value) ? value : undefined;
+}
+function isRecord(value) {
+  return typeof value === "object" && value !== null;
+}
+function parseCacheUsage(usage, format, estimatedPromptTokens) {
+  if (!isRecord(usage))
+    return;
+  switch (format) {
+    case "openai": {
+      const details = usage.prompt_tokens_details;
+      if (!isRecord(details))
+        return;
+      const cached = num(details.cached_tokens);
+      if (cached === undefined)
+        return;
+      const prompt = num(usage.prompt_tokens) ?? cached;
+      const write = num(details.cache_write_tokens) ?? 0;
+      return {
+        cachedTokens: cached,
+        cacheWriteTokens: write,
+        uncachedTokens: Math.max(0, prompt - cached),
+        source: "api"
+      };
+    }
+    case "deepseek": {
+      const hit = num(usage.prompt_cache_hit_tokens);
+      const miss = num(usage.prompt_cache_miss_tokens);
+      if (hit === undefined && miss === undefined)
+        return;
+      return {
+        cachedTokens: hit ?? 0,
+        cacheWriteTokens: 0,
+        uncachedTokens: miss ?? 0,
+        source: "api"
+      };
+    }
+    case "anthropic": {
+      const read = num(usage.cache_read_input_tokens);
+      const creation = num(usage.cache_creation_input_tokens);
+      if (read === undefined && creation === undefined)
+        return;
+      return {
+        cachedTokens: read ?? 0,
+        cacheWriteTokens: creation ?? 0,
+        uncachedTokens: num(usage.input_tokens) ?? 0,
+        source: "api"
+      };
+    }
+    case "ollama": {
+      const evaluated = num(usage.prompt_eval_count);
+      if (evaluated === undefined || estimatedPromptTokens === undefined)
+        return;
+      return {
+        cachedTokens: Math.max(0, estimatedPromptTokens - evaluated),
+        cacheWriteTokens: 0,
+        uncachedTokens: evaluated,
+        source: "derived"
+      };
+    }
+    default:
+      return;
+  }
+}
+
 // src/llm/openai-compat.ts
 var exports_openai_compat = {};
 __export(exports_openai_compat, {
@@ -5911,6 +6016,14 @@ function buildRequestBody(opts) {
   };
   if (opts.maxTokens !== undefined)
     body.max_tokens = opts.maxTokens;
+  if (opts.cachePrompt)
+    body.cache_prompt = true;
+  if (opts.promptCacheKey)
+    body.prompt_cache_key = opts.promptCacheKey;
+  if (opts.sessionId)
+    body.session_id = opts.sessionId;
+  if (opts.stream && opts.streamUsage)
+    body.stream_options = { include_usage: true };
   const strategy = opts.reasoningStrategy ?? "openai-effort";
   const level = opts.reasoningEffort;
   if (strategy === "openai-effort" && level && level !== "default") {
@@ -5944,10 +6057,14 @@ class OpenAICompatProvider {
   debug;
   getSessionId;
   userAgent;
+  cache;
+  cacheReport;
   constructor(config) {
     this.config = config;
     this.model = config.model;
     this.contextWindow = config.contextWindow ?? 32768;
+    this.cache = config.cache;
+    this.cacheReport = config.cache?.report ?? "openai";
     this.tokenCounter = new TokenCounter;
     this.retryConfig = config.retry ?? {
       maxRetries: 3,
@@ -6050,7 +6167,8 @@ class OpenAICompatProvider {
       stream: true,
       maxTokens,
       reasoningEffort: options?.reasoningEffort,
-      reasoningStrategy: options?.reasoningStrategy
+      reasoningStrategy: options?.reasoningStrategy,
+      ...this.cacheHints()
     });
     this.debug?.("LLM stream request", {
       baseUrl: this.config.baseUrl,
@@ -6158,7 +6276,8 @@ class OpenAICompatProvider {
                 st.usage = {
                   promptTokens: parsed.usage.prompt_tokens ?? 0,
                   completionTokens: parsed.usage.completion_tokens ?? 0,
-                  totalTokens: parsed.usage.total_tokens ?? 0
+                  totalTokens: parsed.usage.total_tokens ?? 0,
+                  cache: parseCacheUsage(parsed.usage, this.cacheReport)
                 };
                 this.debug?.("LLM stream usage", { ...st.usage });
               } else if (parsed.error) {
@@ -6246,6 +6365,15 @@ class OpenAICompatProvider {
     }
     return st.sawDone;
   }
+  cacheHints() {
+    const sessionId = this.getSessionId?.();
+    return {
+      cachePrompt: this.cache?.requestCachePrompt === true,
+      promptCacheKey: this.cache?.requestPromptCacheKey ? sessionId : undefined,
+      sessionId: this.cache?.requestSessionId ? sessionId : undefined,
+      streamUsage: this.cache?.requestStreamUsage === true
+    };
+  }
   buildRequestSetup(signal) {
     const headers = {
       "Content-Type": "application/json",
@@ -6255,7 +6383,8 @@ class OpenAICompatProvider {
       headers["Authorization"] = `Bearer ${this.config.apiKey}`;
     }
     const sessionId = this.getSessionId?.();
-    if (sessionId)
+    const sendSessionHeader = this.cache ? this.cache.sessionHeader : true;
+    if (sessionId && sendSessionHeader)
       headers["x-opencode-session"] = sessionId;
     const controller = new AbortController;
     let timedOut = false;
@@ -6294,7 +6423,8 @@ class OpenAICompatProvider {
       stream: false,
       maxTokens: options?.maxTokens ?? this.config.maxCompletionTokens ?? 4096,
       reasoningEffort: options?.reasoningEffort,
-      reasoningStrategy: options?.reasoningStrategy
+      reasoningStrategy: options?.reasoningStrategy,
+      ...this.cacheHints()
     });
     const { headers, abortSignal, cleanup, isTimeout } = this.buildRequestSetup(signal);
     try {
@@ -6347,7 +6477,8 @@ class OpenAICompatProvider {
           usage: {
             promptTokens: data.usage.prompt_tokens ?? 0,
             completionTokens: data.usage.completion_tokens ?? 0,
-            totalTokens: data.usage.total_tokens ?? 0
+            totalTokens: data.usage.total_tokens ?? 0,
+            cache: parseCacheUsage(data.usage, this.cacheReport)
           }
         });
       }
@@ -6469,6 +6600,7 @@ function openaiCompat(opts) {
     rateLimits,
     maxCompletionTokens: opts.maxCompletionTokens,
     getSessionId: opts.getSessionId,
+    cache: opts.capabilities?.cache,
     logger: opts.logger
   });
 }
@@ -6486,7 +6618,17 @@ var init_presets = __esm(() => {
       reasoningStrategy: "prompt-tag",
       listModels: true,
       requiresKey: false,
-      auth: "bearer"
+      auth: "bearer",
+      cache: {
+        mechanism: "local",
+        report: "openai",
+        sessionHeader: false,
+        requestCachePrompt: false,
+        requestPromptCacheKey: false,
+        requestSessionId: false,
+        requestCacheControl: false,
+        requestStreamUsage: true
+      }
     },
     create: openaiCompat
   };
@@ -6501,7 +6643,17 @@ var init_presets = __esm(() => {
       reasoningStrategy: "openai-effort",
       listModels: true,
       requiresKey: true,
-      auth: "bearer"
+      auth: "bearer",
+      cache: {
+        mechanism: "auto",
+        report: "openai",
+        sessionHeader: false,
+        requestCachePrompt: false,
+        requestPromptCacheKey: false,
+        requestSessionId: true,
+        requestCacheControl: false,
+        requestStreamUsage: true
+      }
     },
     create: openaiCompat
   };
@@ -6516,7 +6668,17 @@ var init_presets = __esm(() => {
       reasoningStrategy: "openai-effort",
       listModels: true,
       requiresKey: true,
-      auth: "bearer"
+      auth: "bearer",
+      cache: {
+        mechanism: "auto",
+        report: "openai",
+        sessionHeader: false,
+        requestCachePrompt: false,
+        requestPromptCacheKey: true,
+        requestSessionId: false,
+        requestCacheControl: false,
+        requestStreamUsage: true
+      }
     },
     create: openaiCompat
   };
@@ -6531,7 +6693,17 @@ var init_presets = __esm(() => {
       reasoningStrategy: "none",
       listModels: false,
       requiresKey: true,
-      auth: "header"
+      auth: "header",
+      cache: {
+        mechanism: "explicit",
+        report: "anthropic",
+        sessionHeader: false,
+        requestCachePrompt: false,
+        requestPromptCacheKey: false,
+        requestSessionId: false,
+        requestCacheControl: true,
+        requestStreamUsage: false
+      }
     },
     create: openaiCompat
   };
@@ -6546,7 +6718,17 @@ var init_presets = __esm(() => {
       reasoningStrategy: "openai-effort",
       listModels: true,
       requiresKey: false,
-      auth: "bearer"
+      auth: "bearer",
+      cache: {
+        mechanism: "auto",
+        report: "openai",
+        sessionHeader: true,
+        requestCachePrompt: false,
+        requestPromptCacheKey: false,
+        requestSessionId: false,
+        requestCacheControl: false,
+        requestStreamUsage: true
+      }
     },
     create: openaiCompat
   };
@@ -6561,7 +6743,17 @@ var init_presets = __esm(() => {
       reasoningStrategy: "openai-effort",
       listModels: true,
       requiresKey: false,
-      auth: "bearer"
+      auth: "bearer",
+      cache: {
+        mechanism: "auto",
+        report: "openai",
+        sessionHeader: true,
+        requestCachePrompt: false,
+        requestPromptCacheKey: false,
+        requestSessionId: false,
+        requestCacheControl: false,
+        requestStreamUsage: true
+      }
     },
     create: openaiCompat
   };
@@ -6624,6 +6816,25 @@ var init_create = __esm(() => {
   init_presets();
 });
 
+// src/modules/providers/cache.ts
+function resolveCacheCapability(spec, override) {
+  const base = spec?.capabilities.cache ?? NO_CACHE_CAPABILITY;
+  return { ...base, ...override ?? {} };
+}
+var NO_CACHE_CAPABILITY;
+var init_cache = __esm(() => {
+  NO_CACHE_CAPABILITY = {
+    mechanism: "none",
+    report: "none",
+    sessionHeader: false,
+    requestCachePrompt: false,
+    requestPromptCacheKey: false,
+    requestSessionId: false,
+    requestCacheControl: false,
+    requestStreamUsage: false
+  };
+});
+
 // src/modules/providers/manager.ts
 class ProviderManager {
   entries;
@@ -6747,17 +6958,25 @@ class ProviderManager {
       rateLimits: entry.rateLimits ?? this.opts.rateLimits,
       maxCompletionTokens: entry.maxCompletionTokens,
       getSessionId: this.opts.getSessionId,
+      capabilities: this.resolveCapabilities(entry),
       logger: this.opts.logger
     }, this.registry);
     this.cache.set(key, provider);
     return provider;
   }
+  resolveCapabilities(entry) {
+    const spec = this.registry.get(entry.type);
+    if (!spec)
+      return { cache: resolveCacheCapability(undefined, entry.cache) };
+    return { ...spec.capabilities, cache: resolveCacheCapability(spec, entry.cache) };
+  }
   resetCache() {
     this.cache.clear();
   }
 }
 var init_manager = __esm(() => {
   init_create();
+  init_cache();
   init_presets();
 });
 
@@ -8160,8 +8379,8 @@ function buildDiff(oldLines, newLines) {
   return result;
 }
 function formatLine(line, maxNumWidth) {
-  const num = line.type === "remove" ? line.oldNum : line.newNum;
-  const numStr = num !== null ? String(num).padStart(maxNumWidth) : " ".repeat(maxNumWidth);
+  const num2 = line.type === "remove" ? line.oldNum : line.newNum;
+  const numStr = num2 !== null ? String(num2).padStart(maxNumWidth) : " ".repeat(maxNumWidth);
   if (line.type === "remove") {
     return `${numStr} ${pc2.red("-")} ${line.content}`;
   } else if (line.type === "add") {
@@ -10296,7 +10515,12 @@ class SessionLogger {
       completionTokens: usage.completionTokens,
       totalTokens: usage.totalTokens,
       source: usage.source,
-      durationMs: usage.durationMs
+      durationMs: usage.durationMs,
+      cacheStable: usage.prefix ? Number(usage.prefix.stableRatio.toFixed(4)) : undefined,
+      cacheCause: usage.prefix?.cause,
+      cachedTokens: usage.cache?.cachedTokens,
+      cacheWriteTokens: usage.cache?.cacheWriteTokens,
+      cacheSource: usage.cache?.source
     });
   }
   logError(message) {
@@ -10946,52 +11170,52 @@ ${output}
   }
   function parseNumber() {
     const start = i;
-    let num = "";
+    let num2 = "";
     let invalid = false;
     if (text[i] === "-") {
-      num += text[i];
+      num2 += text[i];
       i++;
       if (!isDigit(text[i]) && atEndOfNumber()) {
-        num += "0";
+        num2 += "0";
       }
     }
     if (text[i] === "0" && isDigit(text[i + 1])) {
       invalid = true;
     }
     while (isDigit(text[i])) {
-      num += text[i];
+      num2 += text[i];
       i++;
     }
     if (text[i] === ".") {
-      if (num === "" || num === "-") {
-        num += "0";
+      if (num2 === "" || num2 === "-") {
+        num2 += "0";
       }
-      num += text[i];
+      num2 += text[i];
       i++;
       if (!isDigit(text[i])) {
-        num += "0";
+        num2 += "0";
       }
       while (isDigit(text[i])) {
-        num += text[i];
+        num2 += text[i];
         i++;
       }
     }
     if (i > start) {
       if (text[i] === "e" || text[i] === "E") {
-        if (num === "-") {
+        if (num2 === "-") {
           invalid = true;
         }
-        num += text[i];
+        num2 += text[i];
         i++;
         if (text[i] === "-" || text[i] === "+") {
-          num += text[i];
+          num2 += text[i];
           i++;
         }
         if (!isDigit(text[i])) {
-          num += "0";
+          num2 += "0";
         }
         while (isDigit(text[i])) {
-          num += text[i];
+          num2 += text[i];
           i++;
         }
       }
@@ -10999,7 +11223,7 @@ ${output}
         i = start;
         return false;
       }
-      output += invalid ? `"${text.substring(start, i)}"` : num;
+      output += invalid ? `"${text.substring(start, i)}"` : num2;
       return true;
     }
     return false;
@@ -12791,6 +13015,24 @@ var init_moe_executor = __esm(() => {
 // src/modules/lsp/project-root.ts
 import { existsSync as existsSync24 } from "fs";
 import { dirname as dirname9, join as join14, relative as relative3, isAbsolute as isAbsolute3 } from "path";
+import { platform as platform4 } from "os";
+function resolveTscCommand(startDir, maxLevels = 8) {
+  const runtime = process.execPath;
+  let dir = startDir;
+  for (let i = 0;i < maxLevels; i++) {
+    const js = join14(dir, "node_modules", "typescript", "bin", "tsc");
+    if (existsSync24(js))
+      return `"${runtime}" "${js}"`;
+    const stub = join14(dir, "node_modules", ".bin", platform4() === "win32" ? "tsc.cmd" : "tsc");
+    if (existsSync24(stub))
+      return `"${stub}"`;
+    const parent = dirname9(dir);
+    if (parent === dir)
+      break;
+    dir = parent;
+  }
+  return null;
+}
 function findProjectRoot(filePath, baseDir, markers) {
   if (!markers || markers.length === 0)
     return baseDir;
@@ -13163,7 +13405,10 @@ function findTypecheckRoot(baseDir, existingFiles = []) {
   return best?.root ?? null;
 }
 async function runTypecheck(baseDir) {
-  const entry = processRegistry.start("npx --no-install tsc --noEmit --skipLibCheck", baseDir);
+  const tsc = resolveTscCommand(baseDir);
+  if (!tsc)
+    return null;
+  const entry = processRegistry.start(`${tsc} --noEmit --skipLibCheck`, baseDir);
   const exited = await processRegistry.waitForExit(entry.id, 90000);
   const output = entry.log.join(`
 `);
@@ -13176,6 +13421,7 @@ var SKIP_DIRS, TEST_EXT_RE, PY_TEST_RE, TEST_STEP_RE;
 var init_audit_runners = __esm(() => {
   init_bash();
   init_processes();
+  init_project_root();
   SKIP_DIRS = new Set([
     "node_modules",
     ".git",
@@ -13354,7 +13600,7 @@ var init_auditor = __esm(() => {
 
 // src/modules/execution/verifier.ts
 import { existsSync as existsSync27, readFileSync as readFileSync14 } from "fs";
-import { resolve as resolve13, extname as extname3, join as join17 } from "path";
+import { resolve as resolve13, extname as extname3, dirname as dirname11, join as join17 } from "path";
 import { spawn as spawn4 } from "child_process";
 
 class StepVerifier {
@@ -13386,8 +13632,11 @@ class StepVerifier {
     if (!existsSync27(tsconfigPath)) {
       return { passed: true, message: "No tsconfig.json found — skipping type check" };
     }
+    const tsc = resolveTscCommand(this.baseDir);
+    if (!tsc)
+      return { passed: true, message: "tsc not installed — skipping type check" };
     try {
-      await this.runAsync("npx tsc --noEmit", this.baseDir, 60000);
+      await this.runAsync(`${tsc} --noEmit`, this.baseDir, 60000);
       return { passed: true, message: "TypeScript type check passed" };
     } catch (e) {
       const stderr = e.stderr?.toString() || e.stdout?.toString() || e.message;
@@ -13400,8 +13649,11 @@ class StepVerifier {
     if (!existsSync27(tsconfigPath)) {
       return { passed: true, message: "No tsconfig.json found — skipping type check" };
     }
+    const tsc = resolveTscCommand(projectRoot);
+    if (!tsc)
+      return { passed: true, message: "tsc not installed — skipping type check" };
     try {
-      await this.runAsync("npx tsc --noEmit --skipLibCheck", projectRoot, 60000);
+      await this.runAsync(`${tsc} --noEmit --skipLibCheck`, projectRoot, 60000);
       return { passed: true, message: "TypeScript type check passed" };
     } catch (e) {
       const stderr = e.stderr?.toString() || e.stdout?.toString() || e.message;
@@ -13554,8 +13806,11 @@ class StepVerifier {
   async validateSyntax(filePath) {
     const ext = extname3(filePath);
     if (ext === ".ts" || ext === ".tsx" || ext === ".cts" || ext === ".mts") {
+      const tsc = resolveTscCommand(dirname11(filePath));
+      if (!tsc)
+        return true;
       try {
-        await this.runAsync(`npx tsc --noEmit --skipLibCheck "${filePath}"`, this.baseDir, 1e4);
+        await this.runAsync(`${tsc} --noEmit --skipLibCheck "${filePath}"`, this.baseDir, 20000);
         return true;
       } catch (err) {
         if (err.status === 127 || err.message.includes("not found") || err.message.includes("ENOENT")) {
@@ -13885,11 +14140,18 @@ function resolvePrice(model, config) {
     return direct;
   return ZEN_PRICES[bare];
 }
-function calculateCost(model, promptTokens, completionTokens, config) {
+function calculateCostDetailed(model, promptTokens, completionTokens, config, cache) {
   const price = resolvePrice(model, config);
   if (!price)
     return;
-  return promptTokens / 1e6 * price.input + completionTokens / 1e6 * price.output;
+  const cached = Math.max(0, cache?.cachedTokens ?? 0);
+  const write = Math.max(0, cache?.cacheWriteTokens ?? 0);
+  const uncached = Math.max(0, promptTokens - cached);
+  const cachedRate = price.cachedInput ?? price.input;
+  const writeRate = price.cacheWrite ?? 0;
+  const cost = uncached / 1e6 * price.input + cached / 1e6 * cachedRate + write / 1e6 * writeRate + completionTokens / 1e6 * price.output;
+  const saved = cached / 1e6 * (price.input - cachedRate);
+  return { cost, saved };
 }
 function formatCost(cost) {
   if (cost === 0)
@@ -13909,12 +14171,12 @@ var init_prices = __esm(() => {
     "nemotron-3-ultra-free": { input: 0, output: 0 },
     "nemotron-3.5-lightning-free": { input: 0, output: 0 },
     "muse-spark-1.2-contributor-free": { input: 0, output: 0 },
-    "minimax-m3": { input: 0.3, output: 1.2 },
-    "minimax-m2.7": { input: 0.3, output: 1.2 },
-    "minimax-m2.5": { input: 0.3, output: 1.2 },
-    "glm-5.2": { input: 1.4, output: 4.4 },
-    "glm-5.1": { input: 1.4, output: 4.4 },
-    "glm-5": { input: 1, output: 3.2 },
+    "minimax-m3": { input: 0.3, output: 1.2, cachedInput: 0.06 },
+    "minimax-m2.7": { input: 0.3, output: 1.2, cachedInput: 0.06 },
+    "minimax-m2.5": { input: 0.3, output: 1.2, cachedInput: 0.06 },
+    "glm-5.2": { input: 1.4, output: 4.4, cachedInput: 0.26 },
+    "glm-5.1": { input: 1.4, output: 4.4, cachedInput: 0.26 },
+    "glm-5": { input: 1, output: 3.2, cachedInput: 0.2 },
     "glm-5.3": { input: 1.4, output: 4.4 },
     "kimi-k2.7-code": { input: 0.95, output: 4 },
     "kimi-k3": { input: 3, output: 15 },
@@ -13976,6 +14238,7 @@ class CostTracker {
   config;
   model;
   _total = 0;
+  _saved = 0;
   _known = false;
   byProvider = new Map;
   constructor(model, config) {
@@ -13985,24 +14248,28 @@ class CostTracker {
   setModel(model) {
     this.model = model;
   }
-  record(promptTokens, completionTokens, provider) {
+  record(promptTokens, completionTokens, provider, cache) {
     const key = provider || "unknown";
-    const cost = calculateCost(this.model, promptTokens, completionTokens, this.config);
-    if (cost === undefined)
+    const result = calculateCostDetailed(this.model, promptTokens, completionTokens, this.config, cache);
+    if (!result)
       return;
     this._known = true;
-    this._total += cost;
+    this._total += result.cost;
+    this._saved += result.saved;
     let entry = this.byProvider.get(key);
     if (!entry) {
       entry = { total: 0, known: false };
       this.byProvider.set(key, entry);
     }
     entry.known = true;
-    entry.total += cost;
+    entry.total += result.cost;
   }
   get total() {
     return this._known ? this._total : undefined;
   }
+  get saved() {
+    return this._known ? this._saved : undefined;
+  }
   totalFor(provider) {
     const entry = this.byProvider.get(provider);
     return entry && entry.known ? entry.total : undefined;
@@ -14083,6 +14350,10 @@ function createLoopState() {
     apiCompletionChars: 0,
     estimatedPromptTokensTotal: 0,
     totalLlmDuration: 0,
+    cacheCachedTotal: 0,
+    cacheUncachedTotal: 0,
+    cacheWriteTotal: 0,
+    cacheSource: "none",
     auditRetries: 0,
     lastAuditSummary: "",
     emptyResponseRetries: 0,
@@ -14189,6 +14460,7 @@ class TokenTracker {
   }
   beginIteration(state, estimatedPromptTokens) {
     state.estimatedPromptTokensTotal += estimatedPromptTokens;
+    state.cacheUsage = undefined;
     return {
       promptBefore: state.apiPromptTokens,
       completionBefore: state.apiCompletionTokens
@@ -14197,6 +14469,7 @@ class TokenTracker {
   recordApiUsage(state, baseline, usage) {
     state.apiPromptTokens = baseline.promptBefore + usage.promptTokens;
     state.apiCompletionTokens = baseline.completionBefore + usage.completionTokens;
+    state.cacheUsage = usage.cache;
   }
   addCompletionChars(state, chars) {
     state.apiCompletionChars += chars;
@@ -14207,14 +14480,22 @@ class TokenTracker {
     const source = usagePrompt > 0 || usageCompletion > 0 ? "api" : "estimate";
     const prompt = source === "api" ? usagePrompt : contextTokens;
     const completion = source === "api" ? usageCompletion : estimateTokens(textContent);
+    if (state.cacheUsage) {
+      state.cacheCachedTotal += state.cacheUsage.cachedTokens;
+      state.cacheUncachedTotal += state.cacheUsage.uncachedTokens;
+      state.cacheWriteTotal += state.cacheUsage.cacheWriteTokens;
+      state.cacheSource = state.cacheUsage.source;
+    }
     slog.logLlmUsage(iteration, {
       promptTokens: prompt,
       completionTokens: completion,
       totalTokens: prompt + completion,
       source,
-      durationMs
+      durationMs,
+      prefix: state.prefixDelta,
+      cache: state.cacheUsage
     });
-    this.costTracker.record(prompt, completion, this.getProviderName());
+    this.costTracker.record(prompt, completion, this.getProviderName(), source === "api" ? state.cacheUsage : undefined);
   }
   resolveFinal(state) {
     if (state.apiPromptTokens > 0 || state.apiCompletionTokens > 0) {
@@ -14233,6 +14514,75 @@ var init_token_tracker = __esm(() => {
   init_token_counter();
 });
 
+// src/core/agent/prefix-monitor.ts
+function serializeTools(tools) {
+  return tools.map((t2) => `${t2.name}\x01${t2.description}\x01${JSON.stringify(t2.parameters)}`).join("\x02");
+}
+function serializeMessages(messages) {
+  return messages.map((m) => `${m.role}\x01${typeof m.content === "string" ? m.content : JSON.stringify(m.content)}`).join("\x02");
+}
+function buildPromptSnapshot(messages, tools) {
+  const system = [];
+  const rest = [];
+  for (const m of messages) {
+    if (m.role === "system")
+      system.push(m);
+    else
+      rest.push(m);
+  }
+  return {
+    system: serializeMessages(system),
+    tools: serializeTools(tools),
+    history: serializeMessages(rest)
+  };
+}
+function frame(s) {
+  const full = s.system + SEP + s.tools + SEP + s.history;
+  const toolsStart = s.system.length + SEP.length;
+  const historyStart = toolsStart + s.tools.length + SEP.length;
+  return { full, toolsStart, historyStart };
+}
+function diffPrompt(prev, curr, countTokens) {
+  const c = frame(curr);
+  const currTokens = countTokens(c.full);
+  if (!prev) {
+    return {
+      prevTokens: 0,
+      currTokens,
+      commonPrefixTokens: 0,
+      stableRatio: 0,
+      cause: "unknown"
+    };
+  }
+  const p = frame(prev);
+  const prevTokens = countTokens(p.full);
+  const maxCp = Math.min(p.full.length, c.full.length);
+  let cp = 0;
+  while (cp < maxCp && p.full[cp] === c.full[cp])
+    cp++;
+  const commonPrefixTokens = countTokens(c.full.slice(0, cp));
+  const stableRatio = currTokens === 0 ? 0 : Math.min(1, commonPrefixTokens / currTokens);
+  if (cp === maxCp) {
+    return { prevTokens, currTokens, commonPrefixTokens, stableRatio, cause: "none" };
+  }
+  let cause;
+  if (cp >= c.historyStart)
+    cause = "history";
+  else if (cp >= c.toolsStart)
+    cause = "tools";
+  else
+    cause = "system";
+  const window = c.full.slice(Math.max(0, cp - 40), cp + 160) + p.full.slice(Math.max(0, cp - 40), cp + 160);
+  if (VOLATILE_RE.test(window))
+    cause = "volatile";
+  return { prevTokens, currTokens, commonPrefixTokens, stableRatio, cause };
+}
+var SEP = `
+`, VOLATILE_RE;
+var init_prefix_monitor = __esm(() => {
+  VOLATILE_RE = /(\d{4}-\d{2}-\d{2}|\d{2}:\d{2}:\d{2}|[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}|[A-Za-z]:\\|\\AppData\\|\\Temp\\|\/tmp\/)/;
+});
+
 // src/core/agent/reasoning-resolver.ts
 class ReasoningEffortResolver {
   resolve(deps) {
@@ -15261,6 +15611,12 @@ class Agent {
       compactionService.compactIfNeeded(state);
       this.refreshSystemPrompt();
       const history = contextManager.getActiveHistory();
+      const promptSnapshot = buildPromptSnapshot(history, allToolsForBudget);
+      state.prefixDelta = diffPrompt(state.prevPrompt, promptSnapshot, estimateTokens);
+      state.prevPrompt = promptSnapshot;
+      if (state.prefixDelta.cause === "system" || state.prefixDelta.cause === "tools" || state.prefixDelta.cause === "history" || state.prefixDelta.cause === "volatile") {
+        state.lastPrefixBreak = state.prefixDelta;
+      }
       slog.logToolDefs(allToolsForBudget.length, allToolsForBudget.map((t2) => t2.name), state.iteration);
       if (state.iteration === 1) {
         const { blocks } = this.buildSystemPrompt();
@@ -15509,6 +15865,7 @@ class Agent {
     const tokensUsed = contextManager.getEstimatedTokens();
     const budget = contextManager.getBudget();
     const usageTokens = tokenTracker.resolveFinal(state);
+    const cacheStats = this.buildCacheStats(state);
     if (state.iteration >= config.maxToolIterations && !state.finalAnswerAccepted) {
       return {
         success: false,
@@ -15522,6 +15879,7 @@ class Agent {
         totalTokens: usageTokens.total,
         totalCost: this.costTracker.total,
         costBreakdown: this.costTracker.breakdown(),
+        cache: cacheStats,
         compactionCount: contextManager.getCompactionCount(),
         contextQuality: contextManager.getQuality()
       };
@@ -15538,11 +15896,29 @@ class Agent {
       totalTokens: usageTokens.total,
       totalCost: this.costTracker.total,
       costBreakdown: this.costTracker.breakdown(),
+      cache: cacheStats,
       compactionCount: contextManager.getCompactionCount(),
       contextQuality: contextManager.getQuality(),
       llmDurationMs: state.totalLlmDuration
     };
   }
+  buildCacheStats(state) {
+    const hasTokens = state.cacheCachedTotal > 0 || state.cacheUncachedTotal > 0 || state.cacheWriteTotal > 0;
+    const prefix = state.lastPrefixBreak ?? state.prefixDelta;
+    if (!hasTokens && prefix?.stableRatio === undefined)
+      return;
+    const denom = state.cacheCachedTotal + state.cacheUncachedTotal;
+    return {
+      cachedTokens: state.cacheCachedTotal,
+      uncachedTokens: state.cacheUncachedTotal,
+      cacheWriteTokens: state.cacheWriteTotal,
+      hitRate: denom > 0 ? state.cacheCachedTotal / denom : 0,
+      saved: this.costTracker.saved,
+      prefixStable: prefix?.stableRatio,
+      prefixCause: prefix?.cause,
+      source: state.cacheSource
+    };
+  }
   clearContext() {
     this.deps.contextManager.clear();
     this.systemPromptAdded = false;
@@ -15647,6 +16023,7 @@ var init_agent = __esm(() => {
   init_loop_state();
   init_compaction();
   init_token_tracker();
+  init_prefix_monitor();
   init_reasoning_resolver();
   init_context_renderer();
   init_tool_batch();
@@ -16538,8 +16915,8 @@ var init_detector = __esm(() => {
 // src/modules/lsp/command.ts
 import { delimiter, join as join21 } from "path";
 import { existsSync as existsSync30 } from "fs";
-import { platform as platform4 } from "os";
-function resolveSpawnCommand(command, platformName = platform4(), pathEnv = process.env.PATH ?? "") {
+import { platform as platform5 } from "os";
+function resolveSpawnCommand(command, platformName = platform5(), pathEnv = process.env.PATH ?? "") {
   if (platformName !== "win32")
     return command;
   if (command.includes("/") || command.includes("\\") || WIN_EXTS.some((ext) => command.toLowerCase().endsWith(ext))) {
@@ -16571,7 +16948,7 @@ var init_command = __esm(() => {
 });
 
 // src/modules/updater/checker.ts
-import { platform as platform5 } from "os";
+import { platform as platform6 } from "os";
 function semverGt(a, b) {
   const pa = a.replace(/^v/, "").split(/[.-]/).map((p) => Number.parseInt(p, 10) || 0);
   const pb = b.replace(/^v/, "").split(/[.-]/).map((p) => Number.parseInt(p, 10) || 0);
@@ -16642,7 +17019,7 @@ class Updater {
   buildInstallCommand(latest) {
     const npmArgs = ["install", "-g", `${this.packageName}@${latest}`];
     const command = resolveSpawnCommand("npm");
-    if (platform5() === "win32" && /\.(cmd|bat)$/i.test(command)) {
+    if (platform6() === "win32" && /\.(cmd|bat)$/i.test(command)) {
       return { command: "cmd.exe", args: ["/c", command, ...npmArgs] };
     }
     return { command, args: npmArgs };
@@ -20167,11 +20544,11 @@ async function readClipboardImage() {
   return readClipboardFallback();
 }
 async function readClipboardFallback() {
-  const { platform: platform6 } = await import("os");
+  const { platform: platform7 } = await import("os");
   const { execSync } = await import("child_process");
   const { readFileSync: readFileSync21, unlinkSync: unlinkSync5 } = await import("fs");
   const { join: join30 } = await import("path");
-  if (platform6() !== "linux")
+  if (platform7() !== "linux")
     return null;
   const tmpPath = join30(process.env.TEMP || process.env.TMP || "/tmp", `mma-clip-${Date.now()}.png`);
   const commands = [
@@ -20847,7 +21224,7 @@ var init_auto_fixer = __esm(() => {
 import { spawn as spawn7, execSync } from "child_process";
 import { existsSync as existsSync37, readFileSync as readFileSync21, writeFileSync as writeFileSync13 } from "fs";
 import { resolve as resolve17, extname as extname6, join as join31 } from "path";
-import { platform as platform6 } from "os";
+import { platform as platform7 } from "os";
 function lintCacheKey(baseDir, lintScript) {
   return `${baseDir}::${lintScript}`;
 }
@@ -20868,7 +21245,7 @@ function contentHash2(content) {
 function getWinDecoder() {
   if (_winDecoder !== undefined)
     return _winDecoder;
-  if (platform6() !== "win32") {
+  if (platform7() !== "win32") {
     _winDecoder = new TextDecoder("utf-8");
     return _winDecoder;
   }
@@ -21029,6 +21406,10 @@ ${stdout}`;
     if (!existsSync37(tsconfigPath)) {
       return;
     }
+    const tsc = resolveTscCommand(projectRoot);
+    if (!tsc) {
+      return;
+    }
     const now = Date.now();
     if (this._checkPromise && now - this._checkTimestamp < TYPE_CHECK_DEBOUNCE_MS) {
       const error2 = await this._checkPromise;
@@ -21040,10 +21421,10 @@ ${stdout}`;
       return;
     }
     this._checkTimestamp = now;
-    this._checkPromise = this.runTscCheck(projectRoot, signal);
+    this._checkPromise = this.runTscCheck(tsc, projectRoot, signal);
     const error = await this._checkPromise;
     if (error) {
-      const { stdout, stderr } = await runAsync(`npx tsc --noEmit --skipLibCheck 2>&1`, projectRoot, 30000, signal).catch(() => ({ stdout: "", stderr: error }));
+      const { stdout, stderr } = await runAsync(`${tsc} --noEmit --skipLibCheck 2>&1`, projectRoot, 30000, signal).catch(() => ({ stdout: "", stderr: error }));
       const output = stderr || stdout;
       const errors = parseTscOutput(output);
       if (errors.length > 0) {
@@ -21072,9 +21453,9 @@ ${stdout}`;
       }
     }
   }
-  async runTscCheck(baseDir, signal) {
+  async runTscCheck(tsc, baseDir, signal) {
     try {
-      const { stdout, stderr } = await runAsync(`npx tsc --noEmit --skipLibCheck`, baseDir, 30000, signal);
+      const { stdout, stderr } = await runAsync(`${tsc} --noEmit --skipLibCheck`, baseDir, 30000, signal);
       return null;
     } catch (err) {
       if (err.status === 127 || err.message.includes("not found") || err.message.includes("ENOENT")) {
@@ -21560,7 +21941,7 @@ function switchStepToDelete(plan, step, save) {
 }
 
 // src/modules/execution/execution-plugin.ts
-import { platform as platform7 } from "os";
+import { platform as platform8 } from "os";
 function normalizeBrokenPath(p) {
   return toForwardSlash(p).replace(/^\.\//, "");
 }
@@ -21793,7 +22174,7 @@ Last compile error: ${first[1]}`;
         if (call.name === "bash") {
           deps.stuckDetector.recordBashAttempt(false);
         }
-        if (call.name === "bash" && platform7() === "win32") {
+        if (call.name === "bash" && platform8() === "win32") {
           const cmd = String(call.arguments?.command ?? "");
           const forbidden = forbiddenWindowsCommand(cmd);
           if (forbidden) {
@@ -23687,7 +24068,7 @@ class ProfileCompressor {
 // src/modules/user-profile/profile.ts
 import { readFileSync as readFileSync26, writeFileSync as writeFileSync17, existsSync as existsSync42, mkdirSync as mkdirSync18 } from "fs";
 import { join as join35 } from "path";
-import { homedir as homedir10, hostname, platform as platform8, type } from "os";
+import { homedir as homedir10, hostname, platform as platform9, type } from "os";
 import { env } from "process";
 
 class UserProfile {
@@ -23699,7 +24080,7 @@ class UserProfile {
   }
   collect() {
     this.info = {
-      platform: platform8(),
+      platform: platform9(),
       os: `${type()} ${hostname()}`,
       hostname: hostname(),
       shell: env.SHELL || env.ComSpec || "unknown",
@@ -24021,7 +24402,7 @@ var init_browser2 = __esm(() => {
 // src/modules/lsp/client.ts
 import { spawn as spawn8, execSync as execSync2 } from "child_process";
 import { resolve as resolve19 } from "path";
-import { platform as platform9 } from "os";
+import { platform as platform10 } from "os";
 
 class FileOnlyLogger {
   logger;
@@ -24141,7 +24522,7 @@ class LspClient {
     if (config.command === "npx" && effectiveArgs.length > 0) {
       const binaryName = this.extractBinaryFromNpxArgs(effectiveArgs);
       if (binaryName) {
-        const whichCmd = platform9() === "win32" ? `where ${binaryName}` : `which ${binaryName}`;
+        const whichCmd = platform10() === "win32" ? `where ${binaryName}` : `which ${binaryName}`;
         try {
           execSync2(whichCmd, { stdio: "pipe", timeout: 3000 });
           effectiveCommand = binaryName;
@@ -24154,7 +24535,7 @@ class LspClient {
     }
     return new Promise((resolve20, reject) => {
       const args = effectiveArgs;
-      const isWin = platform9() === "win32";
+      const isWin = platform10() === "win32";
       let spawnCommand = resolveSpawnCommand(effectiveCommand);
       let spawnArgs = args;
       const spawnOpts = {
@@ -24760,8 +25141,11 @@ async function runCheck(config, baseDir, deps) {
   return { lines: lines.slice(0, STARTUP_CHECK_ERROR_CAP) };
 }
 async function runTscDefault(projectRoot, timeoutMs) {
+  const tsc = resolveTscCommand(projectRoot);
+  if (!tsc)
+    return [];
   return new Promise((resolve22) => {
-    const child = spawn9("npx", ["tsc", "--noEmit", "--skipLibCheck"], {
+    const child = spawn9(`${tsc} --noEmit --skipLibCheck`, {
       cwd: projectRoot,
       shell: true,
       windowsHide: true,
@@ -25165,7 +25549,7 @@ class IndexCache {
     }
   }
 }
-var init_cache = () => {};
+var init_cache2 = () => {};
 
 // src/modules/indexer/map-select.ts
 function isNoiseFile(file) {
@@ -25637,7 +26021,7 @@ ${stackLine}` : summary;
 var MAP_FILE_LIMIT = 80;
 var init_module6 = __esm(() => {
   init_walker();
-  init_cache();
+  init_cache2();
   init_map_select();
   init_project_profile();
   init_i18n();
@@ -25648,7 +26032,7 @@ var init_module6 = __esm(() => {
 // src/modules/indexer/index.ts
 var init_indexer = __esm(() => {
   init_walker();
-  init_cache();
+  init_cache2();
   init_module6();
 });
 
@@ -25993,7 +26377,7 @@ import { spawnSync as spawnSync3 } from "child_process";
 import { createRequire as createRequire2 } from "module";
 import { join as join43, dirname as dirname16 } from "path";
 import { fileURLToPath as fileURLToPath3 } from "url";
-import { arch, homedir as homedir12, hostname as hostname2, platform as platform10, release } from "os";
+import { arch, homedir as homedir12, hostname as hostname2, platform as platform11, release } from "os";
 import { env as env2 } from "process";
 function readEngineRequirement() {
   const here = dirname16(fileURLToPath3(import.meta.url));
@@ -26139,7 +26523,7 @@ function collectEnvironment(opts) {
     mmaVersion: readMmaVersion(),
     runtime,
     os: {
-      platform: platform10(),
+      platform: platform11(),
       arch: arch(),
       release: release(),
       hostname: hostname2(),
@@ -26496,6 +26880,7 @@ var init_set_thinking = __esm(() => {
 // src/core/bootstrap.ts
 var exports_bootstrap = {};
 __export(exports_bootstrap, {
+  setOneShotMode: () => setOneShotMode,
   buildSystemInfo: () => buildSystemInfo,
   bootstrap: () => bootstrap
 });
@@ -26611,6 +26996,9 @@ function buildSystemInfo(config, baseDir, profileCompressed) {
   return lines.join(`
 `);
 }
+function setOneShotMode(value) {
+  oneShotMode = value;
+}
 async function bootstrap(configDir, projectDir, noAgentsMd, exitOnComplete, reasoningLevel) {
   const dir = configDir || process.env.MMA_CONFIG_DIR || join46(homedir15(), ".mma");
   const projectConfigPath = projectDir ? join46(projectDir, ".mmrc") : join46(process.cwd(), ".mmrc");
@@ -26846,7 +27234,7 @@ async function bootstrap(configDir, projectDir, noAgentsMd, exitOnComplete, reas
   moduleRegistry.register(lspModule);
   registerBuiltinPlugin(pluginManager, lspModule.getPlugin());
   let startupCheckBlock = null;
-  const startupCheckPromise = !exitOnComplete ? runStartupHealthCheck(config.lsp ?? DEFAULT_LSP_CONFIG, baseDir, { logger: lspLogger }) : Promise.resolve(null);
+  const startupCheckPromise = !oneShotMode && !exitOnComplete ? runStartupHealthCheck(config.lsp ?? DEFAULT_LSP_CONFIG, baseDir, { logger: lspLogger }) : Promise.resolve(null);
   const moduleTools = moduleRegistry.collectToolDefinitions();
   for (const tool of moduleTools) {
     toolRegistry.register(tool);
@@ -27001,6 +27389,7 @@ async function bootstrap(configDir, projectDir, noAgentsMd, exitOnComplete, reas
     contextProbe: contextProbePromise
   };
 }
+var oneShotMode = false;
 var init_bootstrap = __esm(() => {
   init_config2();
   init_app_logger();
@@ -27476,9 +27865,9 @@ class Spinner {
   tick() {
     if (!this.timer)
       return;
-    const frame = FRAMES[this.frame % FRAMES.length];
+    const frame2 = FRAMES[this.frame % FRAMES.length];
     this.frame++;
-    this.stream.write("\r" + pc2.cyan(frame) + " " + this.message + "\x1B[K");
+    this.stream.write("\r" + pc2.cyan(frame2) + " " + this.message + "\x1B[K");
   }
 }
 var FRAMES;
@@ -30074,9 +30463,9 @@ var require_stringifyNumber = __commonJS((exports) => {
   function stringifyNumber({ format, minFractionDigits, tag, value }) {
     if (typeof value === "bigint")
       return String(value);
-    const num = typeof value === "number" ? value : Number(value);
-    if (!isFinite(num))
-      return isNaN(num) ? ".nan" : num < 0 ? "-.inf" : ".inf";
+    const num3 = typeof value === "number" ? value : Number(value);
+    if (!isFinite(num3))
+      return isNaN(num3) ? ".nan" : num3 < 0 ? "-.inf" : ".inf";
     let n = Object.is(value, -0) ? "-0" : JSON.stringify(value);
     if (!format && minFractionDigits && (!tag || tag === "tag:yaml.org,2002:float") && /^-?\d/.test(n) && !n.includes("e")) {
       let i = n.indexOf(".");
@@ -30113,8 +30502,8 @@ var require_float = __commonJS((exports) => {
     test: /^[-+]?(?:\.[0-9]+|[0-9]+(?:\.[0-9]*)?)[eE][-+]?[0-9]+$/,
     resolve: (str) => parseFloat(str),
     stringify(node) {
-      const num = Number(node.value);
-      return isFinite(num) ? num.toExponential() : stringifyNumber.stringifyNumber(node);
+      const num3 = Number(node.value);
+      return isFinite(num3) ? num3.toExponential() : stringifyNumber.stringifyNumber(node);
     }
   };
   var float = {
@@ -30518,8 +30907,8 @@ var require_float2 = __commonJS((exports) => {
     test: /^[-+]?(?:[0-9][0-9_]*)?(?:\.[0-9_]*)?[eE][-+]?[0-9]+$/,
     resolve: (str) => parseFloat(str.replace(/_/g, "")),
     stringify(node) {
-      const num = Number(node.value);
-      return isFinite(num) ? num.toExponential() : stringifyNumber.stringifyNumber(node);
+      const num3 = Number(node.value);
+      return isFinite(num3) ? num3.toExponential() : stringifyNumber.stringifyNumber(node);
     }
   };
   var float = {
@@ -30709,23 +31098,23 @@ var require_timestamp = __commonJS((exports) => {
   function parseSexagesimal(str, asBigInt) {
     const sign = str[0];
     const parts = sign === "-" || sign === "+" ? str.substring(1) : str;
-    const num = (n) => asBigInt ? BigInt(n) : Number(n);
-    const res = parts.replace(/_/g, "").split(":").reduce((res2, p) => res2 * num(60) + num(p), num(0));
-    return sign === "-" ? num(-1) * res : res;
+    const num3 = (n) => asBigInt ? BigInt(n) : Number(n);
+    const res = parts.replace(/_/g, "").split(":").reduce((res2, p) => res2 * num3(60) + num3(p), num3(0));
+    return sign === "-" ? num3(-1) * res : res;
   }
   function stringifySexagesimal(node) {
     let { value } = node;
-    let num = (n) => n;
+    let num3 = (n) => n;
     if (typeof value === "bigint")
-      num = (n) => BigInt(n);
+      num3 = (n) => BigInt(n);
     else if (isNaN(value) || !isFinite(value))
       return stringifyNumber.stringifyNumber(node);
     let sign = "";
     if (value < 0) {
       sign = "-";
-      value *= num(-1);
+      value *= num3(-1);
     }
-    const _60 = num(60);
+    const _60 = num3(60);
     const parts = [value % _60];
     if (value < 60) {
       parts.unshift(0);
@@ -37566,6 +37955,63 @@ init_version();
 init_map_command();
 init_budget();
 
+// src/llm/provider-budget.ts
+var DEFAULT_TIMEOUT_MS = 5000;
+function num2(value) {
+  return typeof value === "number" && Number.isFinite(value) ? value : undefined;
+}
+async function getJson(url, apiKey, timeoutMs) {
+  const response = await fetch(url, {
+    headers: { Authorization: `Bearer ${apiKey}` },
+    signal: AbortSignal.timeout(timeoutMs)
+  });
+  if (!response.ok) {
+    throw new Error(`HTTP ${response.status}`);
+  }
+  return response.json();
+}
+async function fetchProviderBudget(provider, baseUrl, apiKey, options) {
+  if (provider !== "openrouter") {
+    options.log("debug", `budget: provider "${provider}" does not expose an API balance`);
+    return { budget: null, reason: "unsupported" };
+  }
+  if (!apiKey) {
+    options.log("debug", "budget: no api key configured");
+    return { budget: null, reason: "no-key" };
+  }
+  const base = baseUrl.replace(/\/$/, "");
+  const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
+  try {
+    const keyBody = await getJson(`${base}/key`, apiKey, timeoutMs);
+    const data = keyBody?.data ?? {};
+    const budget = { provider, source: "openrouter" };
+    budget.keyLabel = typeof data.label === "string" ? data.label : undefined;
+    budget.keyUsageUsd = num2(data.usage);
+    budget.keyLimitUsd = num2(data.limit);
+    budget.keyRemainingUsd = num2(data.limit_remaining);
+    budget.isFreeTier = typeof data.is_free_tier === "boolean" ? data.is_free_tier : undefined;
+    try {
+      const creditsBody = await getJson(`${base}/credits`, apiKey, timeoutMs);
+      const c = creditsBody?.data ?? {};
+      budget.totalCreditsUsd = num2(c.total_credits);
+      budget.totalUsageUsd = num2(c.total_usage);
+      if (budget.totalCreditsUsd !== undefined && budget.totalUsageUsd !== undefined) {
+        budget.balanceUsd = budget.totalCreditsUsd - budget.totalUsageUsd;
+      }
+    } catch (creditsErr) {
+      options.log("debug", `budget: /credits unavailable (${creditsErr instanceof Error ? creditsErr.message : String(creditsErr)})`);
+    }
+    return { budget, reason: "ok" };
+  } catch (err) {
+    const message = err instanceof Error ? err.message : String(err);
+    options.log("warn", `budget: request failed (${message})`);
+    return { budget: null, reason: "error", error: message };
+  }
+}
+
+// src/cli/commands.ts
+init_prices();
+
 // src/modules/updater/changelog-reader.ts
 import { readFileSync as readFileSync35, existsSync as existsSync53 } from "fs";
 import { dirname as dirname20, join as join48 } from "path";
@@ -37854,6 +38300,53 @@ function buildContextCommand(program2) {
     console.log(t("cli.context_set", { size: contextWindow }));
   });
 }
+function buildUsageCommand(program2) {
+  program2.command("usage").description(t("cli.usage")).action(async () => {
+    const { config, logger: logger4 } = await bootstrap();
+    const { type: provider, baseUrl, apiKey } = config.provider;
+    const result = await fetchProviderBudget(provider, baseUrl, apiKey, {
+      log: (level, message) => level === "warn" ? logger4.warn(message) : logger4.debug(message)
+    });
+    if (result.reason === "unsupported") {
+      console.log(t("cli.usage_unsupported", { provider }));
+      return;
+    }
+    if (result.reason === "no-key") {
+      console.log(t("cli.usage_no_key"));
+      return;
+    }
+    if (result.reason === "error" || !result.budget) {
+      console.log(t("cli.usage_error", { error: result.error ?? "" }));
+      return;
+    }
+    const b = result.budget;
+    let printed = false;
+    if (b.keyUsageUsd !== undefined) {
+      console.log(t("cli.usage_key_usage", { usage: formatCost(b.keyUsageUsd) }));
+      printed = true;
+    }
+    if (b.keyLimitUsd !== undefined && b.keyRemainingUsd !== undefined) {
+      console.log(t("cli.usage_key_limit", {
+        limit: formatCost(b.keyLimitUsd),
+        remaining: formatCost(b.keyRemainingUsd)
+      }));
+      printed = true;
+    }
+    if (b.balanceUsd !== undefined) {
+      console.log(t("cli.usage_balance", { balance: formatCost(b.balanceUsd) }));
+      printed = true;
+    }
+    if (b.totalCreditsUsd !== undefined && b.totalUsageUsd !== undefined) {
+      console.log(t("cli.usage_account", {
+        credits: formatCost(b.totalCreditsUsd),
+        used: formatCost(b.totalUsageUsd)
+      }));
+      printed = true;
+    }
+    if (!printed)
+      console.log(t("cli.usage_empty"));
+  });
+}
 function buildMapCommand(program2) {
   program2.command("map").description(t("cli.map.description")).argument("[action]", t("cli.map.action"), "summary").argument("[query]", t("cli.map.query")).action(async (action, query) => {
     const { agent } = await bootstrap();
@@ -38010,6 +38503,31 @@ function buildSessionCommands(program2) {
         }
       }
     }
+    const usageEvents = sessionManager.loadSessionLog(id).filter((e) => e.type === "llm_usage");
+    if (usageEvents.length > 0) {
+      let prompt = 0;
+      let completion = 0;
+      let cached = 0;
+      for (const e of usageEvents) {
+        prompt += e.promptTokens ?? 0;
+        completion += e.completionTokens ?? 0;
+        cached += e.cachedTokens ?? 0;
+      }
+      console.log("");
+      console.log(pc2.cyan(t("cli.session_usage", {
+        prompt,
+        completion,
+        total: prompt + completion
+      })));
+      if (cached > 0 && prompt > 0) {
+        const hit = Math.round(cached / prompt * 100);
+        console.log(pc2.dim(t("cli.session_cache", {
+          hit,
+          cached,
+          uncached: Math.max(0, prompt - cached)
+        })));
+      }
+    }
   });
   session2.command("delete").argument("<id>", "Session id").description(t("cli.delete_session")).action(async (id) => {
     const { sessionManager } = await bootstrap();
@@ -38049,6 +38567,7 @@ function createProgram() {
   buildConfigCommands(program2);
   buildModelCommands(program2);
   buildContextCommand(program2);
+  buildUsageCommand(program2);
   buildMapCommand(program2);
   buildProviderCommands(program2);
   buildSessionCommands(program2);
@@ -38056,11 +38575,15 @@ function createProgram() {
   createPluginCommand(program2);
   buildChangelogCommand(program2);
   program2.argument("[prompt...]", "Prompt to execute").description("Run a single prompt").action((prompt) => {});
-  for (const cmd of program2.commands) {
+  const attachOneShotExit = (cmd) => {
     cmd.hook("postAction", () => {
       process.exit(0);
     });
-  }
+    for (const sub of cmd.commands)
+      attachOneShotExit(sub);
+  };
+  for (const cmd of program2.commands)
+    attachOneShotExit(cmd);
   return program2;
 }
 
@@ -39693,6 +40216,28 @@ function stepContextForTool(plan, tool, args) {
 // src/cli/repl.ts
 init_prices();
 
+// src/cli/cache-line.ts
+init_i18n();
+init_prices();
+function formatCacheLine(cache) {
+  if (!cache)
+    return;
+  const total = cache.cachedTokens + cache.uncachedTokens;
+  const hit = Math.round(cache.hitRate * 100);
+  if (total > 0) {
+    if (cache.saved !== undefined && cache.saved > 0) {
+      return t("repl.cache", { hit, saved: formatCost(cache.saved) });
+    }
+    return t("repl.cache_nosave", { hit });
+  }
+  const broken = cache.prefixCause !== undefined && cache.prefixCause !== "none" && cache.prefixCause !== "unknown";
+  if (broken && cache.prefixStable !== undefined && cache.prefixStable < 0.98) {
+    const stable = Math.round(cache.prefixStable * 100);
+    return t("repl.prefix", { stable, cause: t(`cache.cause.${cache.prefixCause}`) });
+  }
+  return;
+}
+
 // src/ui/output.ts
 function writeWarning(text) {
   process.stdout.write(formatWarning(text) + `
@@ -40199,6 +40744,10 @@ ${t("image.clipboard_empty")}`));
       if (result.totalCost !== undefined && result.totalCost > 0) {
         console.log(pc2.yellow(`  ${t("repl.cost", { cost: formatCost(result.totalCost) })}`));
       }
+      const cacheLine = formatCacheLine(result.cache);
+      if (cacheLine) {
+        console.log(pc2.dim(`  ${cacheLine}`));
+      }
     } else if (ui?.showCompaction && result.compactionCount !== undefined) {
       if (result.compactionCount > this.lastCompactionShown) {
         this.lastCompactionShown = result.compactionCount;
@@ -40383,6 +40932,10 @@ function printRunResult(result, flush) {
     } else if (result.totalTokens !== undefined && result.totalTokens > 0) {
       console.log(pc2.dim(`${t("repl.tokens", { tokens: result.totalTokens })}`));
     }
+    const cacheLine = formatCacheLine(result.cache);
+    if (cacheLine) {
+      console.log(pc2.dim(`  ${cacheLine}`));
+    }
     if (!result.text) {
       console.log(pc2.yellow(t("cli.no_output")));
     }
@@ -40621,6 +41174,22 @@ function startAutoUpdate(config) {
 async function main() {
   installCrashHandlers();
   const program2 = createProgram();
+  const valueOpts = new Set(["-d", "--dir", "--reasoning"]);
+  const argv = process.argv.slice(2);
+  let firstPositional;
+  for (let i = 0;i < argv.length; i++) {
+    const a = argv[i];
+    if (a.startsWith("-")) {
+      if (valueOpts.has(a))
+        i++;
+      continue;
+    }
+    firstPositional = a;
+    break;
+  }
+  if (firstPositional && program2.commands.some((c) => c.name() === firstPositional)) {
+    setOneShotMode(true);
+  }
   program2.parse(process.argv);
   const cmdNames = new Set(program2.commands.map((c) => c.name()));
   const opts = program2.opts();
@@ -40672,7 +41241,8 @@ async function main() {
         completionTokens: result2.completionTokens ?? null,
         totalTokens: result2.totalTokens ?? null,
         totalCost: result2.totalCost ?? null,
-        costBreakdown: result2.costBreakdown ?? []
+        costBreakdown: result2.costBreakdown ?? [],
+        cache: result2.cache ?? null
       }, null, 2));
       process.stdout.write(`
 `);