@prestyj/core 5.24.0 → 5.25.1

package/dist/index.cjs

@@ -48,6 +48,7 @@ __export(index_exports, {
   clearLocalDiscoveryCache: () => clearLocalDiscoveryCache,
   clearRuntimeModels: () => clearRuntimeModels,
   closeLogger: () => closeLogger,
+  compareClaudeCodeVersions: () => compareClaudeCodeVersions,
   createAutoUpdater: () => createAutoUpdater,
   decodeOggOpus: () => decodeOggOpus,
   discoverLocalModels: () => discoverLocalModels,
@@ -101,6 +102,7 @@ __export(index_exports, {
   probeEndpoint: () => probeEndpoint,
   providerStorageKeys: () => providerStorageKeys,
   readStoredBaseUrlSync: () => readStoredBaseUrlSync,
+  recordRequiredClaudeCodeVersion: () => recordRequiredClaudeCodeVersion,
   refreshAnthropicToken: () => refreshAnthropicToken,
   refreshGeminiToken: () => refreshGeminiToken,
   refreshKimiToken: () => refreshKimiToken,
@@ -365,7 +367,7 @@ async function refreshKimiToken(refreshToken) {
 
 // src/auth-storage.ts
 var import_promises4 = __toESM(require("fs/promises"), 1);
-var import_node_fs3 = require("fs");
+var import_node_fs4 = require("fs");
 var import_node_crypto6 = __toESM(require("crypto"), 1);
 
 // src/oauth/anthropic.ts
@@ -391,6 +393,7 @@ async function generatePKCE() {
 
 // src/claude-code-version.ts
 var import_promises = __toESM(require("fs/promises"), 1);
+var import_node_fs3 = require("fs");
 var import_node_path4 = __toESM(require("path"), 1);
 
 // src/logger.ts
@@ -518,19 +521,41 @@ function closeLogger(opts) {
 
 // src/claude-code-version.ts
 var NPM_LATEST_URL = "https://registry.npmjs.org/@anthropic-ai/claude-code/latest";
-var CACHE_TTL_MS = 24 * 60 * 60 * 1e3;
+var CACHE_FRESH_MS = 60 * 60 * 1e3;
+var CACHE_HARD_MS = 24 * 60 * 60 * 1e3;
 var FETCH_TIMEOUT_MS = 3e3;
-var FALLBACK_VERSION = "2.1.88";
+var FALLBACK_VERSION = "2.1.280";
 var memoryCache = null;
 var inflight = null;
+var backgroundRefresh = null;
+var versionFloor = null;
 function cachePath() {
   return import_node_path4.default.join(getAppPaths().agentDir, "claude-code-version.json");
 }
+function compareClaudeCodeVersions(a, b) {
+  const parse = (v) => v.split(".").map((part) => Number.parseInt(part, 10)).map((n) => Number.isFinite(n) ? n : 0);
+  const left = parse(a);
+  const right = parse(b);
+  for (let i = 0; i < Math.max(left.length, right.length); i += 1) {
+    const diff = (left[i] ?? 0) - (right[i] ?? 0);
+    if (diff !== 0) return diff;
+  }
+  return 0;
+}
+function applyFloor(version) {
+  if (!versionFloor) return version;
+  return compareClaudeCodeVersions(versionFloor, version) > 0 ? versionFloor : version;
+}
 async function readDiskCache() {
   try {
     const raw = await import_promises.default.readFile(cachePath(), "utf-8");
     const parsed = JSON.parse(raw);
     if (typeof parsed.version === "string" && typeof parsed.fetchedAt === "number") {
+      if (typeof parsed.minVersion === "string") {
+        if (!versionFloor || compareClaudeCodeVersions(parsed.minVersion, versionFloor) > 0) {
+          versionFloor = parsed.minVersion;
+        }
+      }
       return parsed;
     }
     return null;
@@ -541,7 +566,11 @@ async function readDiskCache() {
 async function writeDiskCache(data) {
   try {
     await import_promises.default.mkdir(getAppPaths().agentDir, { recursive: true, mode: 448 });
-    await import_promises.default.writeFile(cachePath(), JSON.stringify(data), { mode: 384 });
+    await import_promises.default.writeFile(
+      cachePath(),
+      JSON.stringify({ ...data, ...versionFloor ? { minVersion: versionFloor } : {} }),
+      { mode: 384 }
+    );
   } catch (err) {
     log(
       "WARN",
@@ -567,6 +596,19 @@ async function fetchLatest() {
     clearTimeout(timer);
   }
 }
+function refreshInBackground() {
+  if (backgroundRefresh) return;
+  backgroundRefresh = (async () => {
+    const fetched = await fetchLatest();
+    if (fetched) {
+      await writeDiskCache({ version: fetched, fetchedAt: Date.now() });
+      memoryCache = { version: applyFloor(fetched), expiresAt: Date.now() + CACHE_FRESH_MS };
+    }
+  })().catch(() => {
+  }).finally(() => {
+    backgroundRefresh = null;
+  });
+}
 async function getClaudeCodeVersion() {
   if (memoryCache && Date.now() < memoryCache.expiresAt) {
     return memoryCache.version;
@@ -574,18 +616,26 @@ async function getClaudeCodeVersion() {
   if (inflight) return inflight;
   inflight = (async () => {
     const disk = await readDiskCache();
-    const diskFresh = disk && Date.now() - disk.fetchedAt < CACHE_TTL_MS;
-    if (disk && diskFresh) {
-      memoryCache = { version: disk.version, expiresAt: Date.now() + CACHE_TTL_MS };
-      return disk.version;
+    const age = disk ? Date.now() - disk.fetchedAt : Infinity;
+    if (disk && age < CACHE_FRESH_MS) {
+      const version = applyFloor(disk.version);
+      memoryCache = { version, expiresAt: Date.now() + CACHE_FRESH_MS };
+      return version;
+    }
+    if (disk && age < CACHE_HARD_MS) {
+      const version = applyFloor(disk.version);
+      memoryCache = { version, expiresAt: Date.now() + 5 * 60 * 1e3 };
+      refreshInBackground();
+      return version;
     }
     const fetched = await fetchLatest();
     if (fetched) {
       await writeDiskCache({ version: fetched, fetchedAt: Date.now() });
-      memoryCache = { version: fetched, expiresAt: Date.now() + CACHE_TTL_MS };
-      return fetched;
+      const version = applyFloor(fetched);
+      memoryCache = { version, expiresAt: Date.now() + CACHE_FRESH_MS };
+      return version;
     }
-    const resolved = disk?.version ?? FALLBACK_VERSION;
+    const resolved = applyFloor(disk?.version ?? FALLBACK_VERSION);
     memoryCache = { version: resolved, expiresAt: Date.now() + 5 * 60 * 1e3 };
     log(
       "WARN",
@@ -600,6 +650,44 @@ async function getClaudeCodeVersion() {
     inflight = null;
   }
 }
+var REQUIRED_VERSION_PATTERN = /version\s+(\d+(?:\.\d+)+)\s+or\s+newer\s+is\s+required/i;
+function recordRequiredClaudeCodeVersion(message) {
+  const required = REQUIRED_VERSION_PATTERN.exec(message ?? "")?.[1];
+  if (!required) return null;
+  if (versionFloor && compareClaudeCodeVersions(required, versionFloor) <= 0) return versionFloor;
+  versionFloor = required;
+  memoryCache = null;
+  log(
+    "WARN",
+    "claude-code-version",
+    `Provider requires Claude Code >= ${required}; pinning User-Agent floor`
+  );
+  try {
+    const dir = getAppPaths().agentDir;
+    (0, import_node_fs3.mkdirSync)(dir, { recursive: true, mode: 448 });
+    let existing = {};
+    try {
+      existing = JSON.parse((0, import_node_fs3.readFileSync)(cachePath(), "utf-8"));
+    } catch {
+    }
+    (0, import_node_fs3.writeFileSync)(
+      cachePath(),
+      JSON.stringify({
+        version: existing.version ?? required,
+        fetchedAt: existing.fetchedAt ?? 0,
+        minVersion: required
+      }),
+      { mode: 384 }
+    );
+  } catch (err) {
+    log(
+      "WARN",
+      "claude-code-version",
+      `Failed to persist required version: ${err instanceof Error ? err.message : String(err)}`
+    );
+  }
+  return required;
+}
 async function getClaudeCliUserAgent() {
   const version = await getClaudeCodeVersion();
   return `claude-cli/${version} (external, cli)`;
@@ -1587,7 +1675,7 @@ function activeBaseUrlEntry(data, provider) {
 }
 function readStoredBaseUrlSync(authFile, provider) {
   try {
-    const data = JSON.parse((0, import_node_fs3.readFileSync)(authFile, "utf-8"));
+    const data = JSON.parse((0, import_node_fs4.readFileSync)(authFile, "utf-8"));
     return activeBaseUrlEntry(data, provider)?.baseUrl;
   } catch {
     return void 0;
@@ -2078,11 +2166,37 @@ var MODELS = [
   //   costTier: "high",
   //   maxThinkingLevel: "max",
   // },
+  {
+    // Released 2026-09-22 — "For long-running agentic coding and knowledge
+    // work". Fable-class capability at $4/$20 MTok (cheaper than the Opus 5 it
+    // replaces, $5/$25). Adaptive thinking with the full effort ladder
+    // (low→max, xhigh included), but thinking can no longer be disabled: a
+    // `thinking: {type: "disabled"}` or budget_tokens request 400s. @prestyj/ai
+    // omits the field entirely when thinking is off, so that path is safe.
+    // Forced tool use (`tool_choice` any/tool) also 400s — see
+    // `toAnthropicToolChoice`, which downgrades it to `auto` for this model.
+    // Anthropic declares the server-side default effort as `medium` (Opus 5
+    // ran `high`), and 5.5 thinks more per turn at a given level, so a fresh
+    // session starts at `medium` rather than the ladder ceiling.
+    id: "claude-opus-5-5",
+    name: "Claude Opus 5.5",
+    provider: "anthropic",
+    contextWindow: 1e6,
+    maxOutputTokens: 128e3,
+    supportsThinking: true,
+    defaultThinkingLevel: "medium",
+    supportsImages: true,
+    supportsVideo: false,
+    costTier: "high",
+    maxThinkingLevel: "max"
+  },
   {
     // Released 2026-07-24 — "For complex agentic coding and enterprise work".
     // Near-Fable capability at half the price ($5/$25 vs $10/$50). Adaptive
     // thinking with the full effort ladder (low→max, xhigh included); dateless
-    // ID is the canonical pinned snapshot (post-4.6 naming scheme).
+    // ID is the canonical pinned snapshot (post-4.6 naming scheme). Kept as a
+    // legacy option now that Opus 5.5 leads the line: it's the last Opus that
+    // accepts disabled thinking and forced tool use.
     id: "claude-opus-5",
     name: "Claude Opus 5",
     provider: "anthropic",
@@ -2231,16 +2345,21 @@ var MODELS = [
     maxThinkingLevel: "max"
   },
   // ── xAI (Grok) ─────────────────────────────────────────
-  // Grok 4.6 (released 2026-08-12) is xAI's flagship for coding, agentic tasks,
-  // and knowledge work, with a focus on long-running agents — 500K context,
-  // text+image input, and a `reasoning_effort` ladder that adds a new `xhigh`
-  // top rung (low/medium/high default/xhigh; reasoning still can't be fully
+  // Grok 4.7 (released 2026-09-21) — xAI's flagship for coding, agentic tasks,
+  // and knowledge work: a new, larger base model with a longer RL run weighted
+  // toward hours-long tasks, plus stronger self-verification and long-context
+  // management. 500K context, text+image input, and a `reasoning_effort`
+  // ladder of low/medium/high default/xhigh (reasoning still can't be fully
   // disabled). $2/$6 per MTok under 200K prompt tokens ($4/$12 at or above),
-  // and it's the default model of the Grok Build coding agent. xAI advertises "no text output limit"; we keep the same
-  // 131K practical cap as 4.5 for budget predictability and input headroom.
+  // and it's the default model of the Grok Build coding agent. xAI advertises
+  // "no fixed text output limit"; we keep the 131K practical cap for budget
+  // predictability and input headroom. (A faster "Grok 4.7 Fast" variant
+  // exists but is Cursor/Grok Build-only — not on the public API — so it isn't
+  // registered.) Only the newest Grok ships — 4.6/4.5 are superseded and
+  // retired; saved sessions on them fall back to this default.
   {
-    id: "grok-4.6",
-    name: "Grok 4.6",
+    id: "grok-4.7",
+    name: "Grok 4.7",
     provider: "xai",
     contextWindow: 5e5,
     maxOutputTokens: 131072,
@@ -2250,24 +2369,6 @@ var MODELS = [
     costTier: "medium",
     maxThinkingLevel: "xhigh"
   },
-  // Grok 4.5 (released 2026-07-08) — superseded by 4.6 but retained as an explicit option. 500K context, text+image input,
-  // configurable `reasoning_effort` (low/medium/high, server default high;
-  // reasoning can't be fully disabled). Served over the OpenAI-compatible API
-  // at https://api.x.ai/v1 (API key from console.x.ai). xAI hasn't published an
-  // official max-output cap for 4.5; 131K matches the Grok Responses ceiling
-  // third-party integrations use.
-  {
-    id: "grok-4.5",
-    name: "Grok 4.5",
-    provider: "xai",
-    contextWindow: 5e5,
-    maxOutputTokens: 131072,
-    supportsThinking: true,
-    supportsImages: true,
-    supportsVideo: false,
-    costTier: "medium",
-    maxThinkingLevel: "high"
-  },
   // ── Gemini ─────────────────────────────────────────
   {
     id: "gemini-3.1-flash-lite",
@@ -2456,11 +2557,15 @@ var MODELS = [
     maxThinkingLevel: "high"
   },
   // ── Xiaomi (MiMo) ──────────────────────────────────────
-  // V2.6 series (2026-09) supersedes V2.5 one-for-one: pro → pro, the omni
-  // `mimo-v2.5` → flash, ultraspeed → ultraspeed. The whole series is now
-  // full-modality, so unlike V2.5-Pro the flagship no longer needs a separate
-  // omni sibling for attachments. V2.5 entries are retired here — a session
-  // that still has one saved falls back to the provider default on next start.
+  // V2.6 series (released 2026-09-22, open-weight: Pro 1.02T/42B-A, Flash
+  // 309B/15B-A, plus a 9B Qwen distill not served over the API) supersedes V2.5
+  // one-for-one: pro → pro, the omni `mimo-v2.5` → flash, ultraspeed →
+  // ultraspeed. Every V2.6 text model is natively full-modality, so unlike
+  // V2.5-Pro the flagship no longer needs a separate omni sibling for
+  // attachments — image/video ride the same OpenAI-compatible base64 transport
+  // the old omni model used. API prices are unchanged from V2.5. The V2.5 ids
+  // deprecate on the platform 2026-10-21 and are retired here — a session that
+  // still has one saved falls back to the provider default on next start.
   //
   // Capabilities below are measured against the Token Plan host, not taken
   // from marketing copy: image and video both come back with `image_tokens` /
@@ -2470,6 +2575,8 @@ var MODELS = [
   // binary 1M below (2^20) and not the decimal 1e6 V2.5 was listed with — the
   // few-token gap is the chat envelope the server adds on top of the content.
   {
+    // Coding/agentic flagship — highest open-weight score on Artificial
+    // Analysis at launch (46, tied with Grok 4.7).
     id: "mimo-v2.6-pro",
     name: "MiMo-V2.6-Pro",
     provider: "xiaomi",
@@ -2483,9 +2590,10 @@ var MODELS = [
     maxThinkingLevel: "high",
     authStorageKeys: ["xiaomi", XIAOMI_CREDITS_KEY]
   },
-  // Flash: the cheap, high-frequency sibling at the same modality surface and
-  // window as Pro. It is the provider's `low` tier, so scout sub-agents and
-  // compaction summaries route here instead of paying Pro rates.
+  // Flash: the cheap, high-frequency sibling (~10% of Pro's price class) at the
+  // same modality surface and window as Pro. It is the provider's `low` tier,
+  // so scout sub-agents and compaction summaries route here instead of paying
+  // Pro rates.
   {
     id: "mimo-v2.6-flash",
     name: "MiMo-V2.6-Flash",
@@ -2506,9 +2614,9 @@ var MODELS = [
   // authStorageKeys doc). The Token Plan host rejects it with "Not supported
   // model" — the known-model/wrong-host reply — where an invented id gets
   // "Unsupported model", which is how this id was confirmed without a
-  // Credits key. Attachment support is inferred from the series announcement
-  // ("full modality across the series") rather than measured: images are
-  // enabled, video stays off until it can be verified on the platform host.
+  // Credits key. Attachments can't be probed directly for the same reason, so
+  // this entry tracks the rest of the V2.6 series ("full modality across the
+  // series") and mirrors Pro's verified image+video surface.
   {
     id: "mimo-v2.6-pro-ultraspeed",
     name: "MiMo-V2.6-Pro-UltraSpeed",
@@ -2517,7 +2625,8 @@ var MODELS = [
     maxOutputTokens: 131072,
     supportsThinking: true,
     supportsImages: true,
-    supportsVideo: false,
+    supportsVideo: true,
+    maxVideoBytes: 36 * 1024 * 1024,
     costTier: "high",
     maxThinkingLevel: "high",
     authStorageKeys: [XIAOMI_CREDITS_KEY]
@@ -2667,7 +2776,7 @@ function getDefaultModel(provider) {
     return MODELS.find((m) => m.id === "Qwen/Qwen3-Coder-480B-A35B-Instruct");
   if (provider === "openrouter") return MODELS.find((m) => m.id === "qwen/qwen3.6-plus");
   if (provider === "sakana") return MODELS.find((m) => m.id === "fugu");
-  if (provider === "xai") return MODELS.find((m) => m.id === "grok-4.6");
+  if (provider === "xai") return MODELS.find((m) => m.id === "grok-4.7");
   if (provider === "local") {
     return getModelsForProvider("local")[0] ?? PLACEHOLDER_LOCAL_MODEL;
   }
@@ -2849,7 +2958,7 @@ var FALLBACK_CONTEXT_WINDOW = 4096;
 var LOCAL_API_KEY_PLACEHOLDER = "local";
 var DEFAULT_PROBE_TIMEOUT_MS = 1200;
 var ENRICH_CONCURRENCY = 6;
-var CACHE_TTL_MS2 = 3e4;
+var CACHE_TTL_MS = 3e4;
 var NON_CHAT_ID_PATTERN = /(?:^|[-_/])(?:embed|embedding|rerank|reranker|bge|nomic-embed)/i;
 var LOCAL_ID_PREFIX = "local/";
 function formatLocalModelId(endpointId, rawId) {
@@ -3090,7 +3199,7 @@ function cacheKey(endpoints) {
 }
 async function discoverLocalModels(endpoints = DEFAULT_LOCAL_ENDPOINTS, options = {}) {
   const key = cacheKey(endpoints);
-  if (!options.force && cache && cache.key === key && Date.now() - cache.at < CACHE_TTL_MS2) {
+  if (!options.force && cache && cache.key === key && Date.now() - cache.at < CACHE_TTL_MS) {
     return cache.result;
   }
   const probes = await Promise.all(endpoints.map((endpoint) => probeEndpoint(endpoint, options)));
@@ -3663,7 +3772,7 @@ async function transcribeVoice(fileUrl) {
 
 // src/auto-update.ts
 var import_node_child_process2 = require("child_process");
-var import_node_fs4 = __toESM(require("fs"), 1);
+var import_node_fs5 = __toESM(require("fs"), 1);
 var import_node_path5 = __toESM(require("path"), 1);
 var CHECK_INTERVAL_MS = 60 * 60 * 1e3;
 var FETCH_TIMEOUT_MS2 = 1e4;
@@ -3697,7 +3806,7 @@ function createAutoUpdater(config) {
   }
   function readState() {
     try {
-      const raw = import_node_fs4.default.readFileSync(stateFilePath(), "utf-8");
+      const raw = import_node_fs5.default.readFileSync(stateFilePath(), "utf-8");
       return JSON.parse(raw);
     } catch {
       return null;
@@ -3706,8 +3815,8 @@ function createAutoUpdater(config) {
   function writeState(state) {
     try {
       const filePath = stateFilePath();
-      import_node_fs4.default.mkdirSync(import_node_path5.default.dirname(filePath), { recursive: true, mode: 448 });
-      import_node_fs4.default.writeFileSync(filePath, JSON.stringify(state));
+      import_node_fs5.default.mkdirSync(import_node_path5.default.dirname(filePath), { recursive: true, mode: 448 });
+      import_node_fs5.default.writeFileSync(filePath, JSON.stringify(state));
     } catch {
     }
   }
@@ -3849,6 +3958,7 @@ function createAutoUpdater(config) {
   clearLocalDiscoveryCache,
   clearRuntimeModels,
   closeLogger,
+  compareClaudeCodeVersions,
   createAutoUpdater,
   decodeOggOpus,
   discoverLocalModels,
@@ -3902,6 +4012,7 @@ function createAutoUpdater(config) {
   probeEndpoint,
   providerStorageKeys,
   readStoredBaseUrlSync,
+  recordRequiredClaudeCodeVersion,
   refreshAnthropicToken,
   refreshGeminiToken,
   refreshKimiToken,