nexrall-code 0.5.70 → 0.5.72

package/dist/index.js

@@ -9973,6 +9973,36 @@ var require_client = __commonJS({
       const raw = Number(process.env.NEXRALL_MAX_RETRY_MS);
       return Number.isFinite(raw) && raw > 0 ? raw : 5 * 6e4;
     })();
+    var headerTimeoutMs = () => Number(process.env.NEXRALL_HEADER_TIMEOUT_MS) || 12e4;
+    async function fetchWithHeaderTimeout(url, init, outerSignal, timeoutMs = headerTimeoutMs()) {
+      const headerController = new AbortController();
+      const onOuterAbort = () => headerController.abort();
+      if (outerSignal.aborted)
+        headerController.abort();
+      else
+        outerSignal.addEventListener("abort", onOuterAbort, { once: true });
+      let timedOut = false;
+      const timer = setTimeout(() => {
+        timedOut = true;
+        headerController.abort();
+      }, timeoutMs);
+      try {
+        return await (0, node_fetch_1.default)(url, { ...init, signal: headerController.signal });
+      } catch (err) {
+        if (timedOut && !outerSignal.aborted) {
+          throw Object.assign(
+            new Error(`The server accepted the connection but sent no response within ${Math.round(timeoutMs / 1e3)} s. Reconnecting\u2026`),
+            // Retryable by definition: nothing was received, so a fresh attempt
+            // cannot duplicate any rendered output.
+            { retryable: true }
+          );
+        }
+        throw err;
+      } finally {
+        clearTimeout(timer);
+        outerSignal.removeEventListener("abort", onOuterAbort);
+      }
+    }
     function backoffMs(attempt) {
       const exp = Math.min(RETRY_MAX_MS, RETRY_BASE_MS * Math.pow(2, attempt));
       return Math.round(exp / 2 + Math.random() * (exp / 2));
@@ -9982,7 +10012,7 @@ var require_client = __commonJS({
     }
     var MAX_TOTAL_ATTEMPTS = (MAX_RETRIES + 1) * (MAX_RETRIES + 1);
     async function streamChat(messages, options, onEvent) {
-      const { model, env: env3, editorContext, nexrallMd, mode, effort, clientType, abortSignal, extraTools, agents, skills, allowRestartAfterRender, canWriteSharedMemory, hasOwnAgentStore, canSpawnSubAgents } = options;
+      const { model, env: env3, editorContext, nexrallMd, mode, effort, clientType, abortSignal, extraTools, agents, skills, allowRestartAfterRender, canWriteSharedMemory, hasOwnAgentStore, canSpawnSubAgents, canBrowseWeb } = options;
       let turnId = (0, crypto_1.randomUUID)();
       const controller = new AbortController();
       if (abortSignal?.aborted)
@@ -10008,13 +10038,14 @@ var require_client = __commonJS({
       let emittedAnythingAcrossAttempts = false;
       let emittedCharsAcrossAttempts = 0;
       const buildFetchArgs = () => {
+        const signal = controller.signal;
         if (resuming) {
           return [
             `${exports.API_BASE}/api/code/chat/resume?turnId=${encodeURIComponent(turnId)}`,
             {
               method: "GET",
               headers: { ...authHeaders(), "Last-Event-ID": String(lastEventId) },
-              signal: controller.signal
+              signal
             }
           ];
         }
@@ -10023,8 +10054,8 @@ var require_client = __commonJS({
           {
             method: "POST",
             headers: authHeaders(),
-            body: JSON.stringify({ messages, model, env: env3, editorContext, nexrallMd, mode, effort, clientType, extraTools, agents, skills, turnId, canWriteSharedMemory, hasOwnAgentStore, canSpawnSubAgents }),
-            signal: controller.signal
+            body: JSON.stringify({ messages, model, env: env3, editorContext, nexrallMd, mode, effort, clientType, extraTools, agents, skills, turnId, canWriteSharedMemory, hasOwnAgentStore, canSpawnSubAgents, canBrowseWeb }),
+            signal
           }
         ];
       };
@@ -10085,7 +10116,7 @@ var require_client = __commonJS({
         for (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {
           try {
             errorBodyOverride = null;
-            response = await (0, node_fetch_1.default)(...buildFetchArgs());
+            response = await fetchWithHeaderTimeout(...buildFetchArgs(), controller.signal);
             if (response.status === 429 && attempt < MAX_RETRIES && canRetry()) {
               reportRetry("Rate limited by the API \u2014 retrying");
               const retryAfter = parseInt(response.headers.get("retry-after") ?? "0", 10);
@@ -16248,6 +16279,10 @@ var require_loop = __commonJS({
         case "empty-response":
           return `
 \u26A0\uFE0F The model returned an empty response, so nothing was done. This is usually a transient upstream hiccup \u2014 send "continue" to retry.
+`;
+        case "output-limit":
+          return `
+\u26A0\uFE0F The model used its entire output budget on internal reasoning and was cut off before it could reply, so nothing was done. This will repeat identically if you just retry \u2014 lower the reasoning effort (/effort) or split the task into smaller steps.
 `;
         case "no-balance":
           return `
@@ -17140,6 +17175,12 @@ ${tail}`;
         return "";
       return first.content.filter((b) => b.type === "text" && b.text).map((b) => b.text).join("\n").trim();
     }
+    var CompactionUnavailableError = class extends Error {
+      constructor() {
+        super("Compaction summariser was unavailable");
+        this.name = "CompactionUnavailableError";
+      }
+    };
     async function autoCompactMessages(messages, options, ledger) {
       const cut = findSafeCutIndex(messages, messages.length - COMPACT_KEEP_MIN);
       if (cut < 2)
@@ -17178,7 +17219,7 @@ ${tail}`;
         });
         summary = reply.content.filter((b) => b.type === "text").map((b) => b.text ?? "").join("").trim();
       } catch {
-        return false;
+        throw new CompactionUnavailableError();
       }
       if (!summary)
         return false;
@@ -17253,21 +17294,28 @@ Continue the work from here.` }] });
       let guard = 0;
       while (overCompactThreshold() && messages.length > COMPACT_KEEP_MIN + 2 && guard < 5) {
         guard += 1;
-        const did = await autoCompactMessages(messages, {
-          workDir: opts.workDir,
-          model: opts.model,
-          clientType: opts.clientType,
-          env: opts.env,
-          onText: () => {
-          },
-          onToolUse: () => {
-          },
-          onToolResult: () => {
-          },
-          onUsage: () => {
-          },
-          requestPermission: async () => false
-        });
+        let did;
+        try {
+          did = await autoCompactMessages(messages, {
+            workDir: opts.workDir,
+            model: opts.model,
+            clientType: opts.clientType,
+            env: opts.env,
+            onText: () => {
+            },
+            onToolUse: () => {
+            },
+            onToolResult: () => {
+            },
+            onUsage: () => {
+            },
+            requestPermission: async () => false
+          });
+        } catch (err) {
+          if (err instanceof CompactionUnavailableError)
+            break;
+          throw err;
+        }
         if (!did)
           break;
         compacted = true;
@@ -17324,6 +17372,8 @@ ${options.nexrallMd}` : "") : options.nexrallMd;
       const ledger = createLedger();
       let compactFailures = 0;
       let compactDisabled = false;
+      let compactDisabledAtBytes = 0;
+      const COMPACT_REARM_GROWTH = 1.5;
       try {
         for (; iteration < budget; iteration++) {
           if (options.abortSignal?.aborted) {
@@ -17342,11 +17392,23 @@ ${options.nexrallMd}` : "") : options.nexrallMd;
               (options.onNotice ?? options.onText)(`\u267B\uFE0F Trimmed ~${(reclaimed / (1024 * 1024)).toFixed(1)}MB of already-processed tool output to keep this chat cheap to continue.`);
             }
           }
+          if (compactDisabled && bodyBytes > compactDisabledAtBytes * COMPACT_REARM_GROWTH) {
+            compactDisabled = false;
+            compactFailures = 0;
+          }
           if (autoCompact && !compactDisabled && !compacting && (tokenPressure || bytePressure) && messages.length > COMPACT_KEEP_MIN + 2) {
             compacting = true;
             try {
               const bytesBefore = estimateBodyBytes2(messages);
-              const did = await autoCompactMessages(messages, options, ledger);
+              let did = false;
+              let unavailable = false;
+              try {
+                did = await autoCompactMessages(messages, options, ledger);
+              } catch (err) {
+                if (!(err instanceof CompactionUnavailableError))
+                  throw err;
+                unavailable = true;
+              }
               const bytesAfter = did ? estimateBodyBytes2(messages) : bytesBefore;
               const reclaimed = bytesBefore - bytesAfter;
               if (did) {
@@ -17356,11 +17418,13 @@ ${options.nexrallMd}` : "") : options.nexrallMd;
                 bodyBytes = bytesAfter;
                 bytePressure = bodyBytes > MAX_BODY_BYTES;
               }
-              if (did && reclaimed >= COMPACT_MIN_RECLAIM_BYTES) {
+              if (unavailable) {
+              } else if (did && reclaimed >= COMPACT_MIN_RECLAIM_BYTES) {
                 compactFailures = 0;
               } else if (++compactFailures >= COMPACT_MAX_FAILURES) {
                 compactDisabled = true;
-                (options.onNotice ?? options.onText)(`\u26A0\uFE0F Auto-compaction isn't reducing this conversation any further, so it's been switched off for the rest of this run to avoid repeated summarising. If the context fills up, start a fresh chat or run /compact manually.`);
+                compactDisabledAtBytes = bytesAfter;
+                (options.onNotice ?? options.onText)(`\u26A0\uFE0F Auto-compaction isn't reducing this conversation any further, so it's paused to avoid repeated summarising. It will be retried automatically if the conversation grows substantially. If the context fills up before then, start a fresh chat or run /compact manually.`);
               }
             } finally {
               compacting = false;
@@ -17441,6 +17505,16 @@ ${options.nexrallMd}` : "") : options.nexrallMd;
               canWriteSharedMemory: options._allowedTools ? options._allowedTools.has("memory_write") : true,
               // no allowlist = main agent = may write
               hasOwnAgentStore: !!options._agentMemory,
+              // Same derivation, same reason, for the network. The base prompt's
+              // "Web search" section tells the run to reach the network, but
+              // READ_ONLY_TOOLS (explorer, security-auditor, ...) contains neither
+              // `fetch_url` nor `web_search`. MEASURED on the 195-trial 2026-08-10
+              // deepseek-v4-pro benchmark: 29 of 51 `fetch_url` failures and 3 of 14
+              // `web_search` failures were the permission gate refusing a sub-agent
+              // that its own system prompt had just invited to browse — every one a
+              // wasted round-trip on a paid turn.
+              canBrowseWeb: options._allowedTools ? options._allowedTools.has("fetch_url") : true,
+              // no allowlist = main agent = may browse
               // Withholds the `task` schema and its instructions when this run cannot
               // delegate — see canSpawnSubAgents.
               canSpawnSubAgents: maySpawn,
@@ -17481,7 +17555,7 @@ ${options.nexrallMd}` : "") : options.nexrallMd;
               continue;
             }
             runSimpleHooks(hooks.PostMessageComplete, options.workDir);
-            stopReason = "empty-response";
+            stopReason = assistantMessage.stopReason === "max_tokens" ? "output-limit" : "empty-response";
             break;
           }
           const { stopReason: _stopReason, ...historyMessage } = assistantMessage;
@@ -53543,7 +53617,7 @@ var {
 
 // src/index.ts
 import * as path5 from "path";
-import { createRequire } from "module";
+import { createRequire as createRequire2 } from "module";
 
 // ../../node_modules/.pnpm/chalk@5.6.2/node_modules/chalk/source/vendor/ansi-styles/index.js
 var ANSI_BACKGROUND_OFFSET = 10;
@@ -54171,6 +54245,8 @@ import * as path4 from "path";
 import * as fs8 from "fs";
 import * as os4 from "os";
 import * as crypto from "crypto";
+import { createRequire } from "module";
+import { fileURLToPath as fileURLToPath2 } from "url";
 import { execSync } from "child_process";
 var import_code_core3 = __toESM(require_dist2(), 1);
 
@@ -64732,7 +64808,26 @@ var InkReadlineAdapter = class extends EventEmitter3 {
 };
 
 // src/commands/chat.ts
-var CLI_VERSION = "0.5.63";
+var require2 = createRequire(import.meta.url);
+var CLI_VERSION = (() => {
+  let dir = path4.dirname(fileURLToPath2(import.meta.url));
+  for (let i2 = 0; i2 < 6; i2++) {
+    const candidate = path4.join(dir, "package.json");
+    if (fs8.existsSync(candidate)) {
+      try {
+        const v = require2(candidate).version;
+        if (typeof v === "string" && v)
+          return v;
+      } catch {
+      }
+    }
+    const parent = path4.dirname(dir);
+    if (parent === dir)
+      break;
+    dir = parent;
+  }
+  return "unknown";
+})();
 var MODEL_LABELS = {
   "claude-sonnet-5": "Claude Sonnet 5",
   "claude-opus-5": "Claude Opus 5",
@@ -66481,8 +66576,8 @@ function pluginSourceRemoveCommand(name, opts) {
 }
 
 // src/index.ts
-var require2 = createRequire(import.meta.url);
-var NEX_VERSION = require2("../package.json").version;
+var require3 = createRequire2(import.meta.url);
+var NEX_VERSION = require3("../package.json").version;
 var program2 = new Command();
 program2.name("nex").description("Nexrall Code \u2014 AI coding assistant (powered by Nexrall)").version(NEX_VERSION).enablePositionalOptions();
 program2.command("auth").description("Login to your Nexrall account").action(authCommand);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "nexrall-code",
-  "version": "0.5.70",
+  "version": "0.5.72",
   "description": "Nexrall Code — AI coding assistant for your terminal (headless agent for scripts, CI and automation)",
   "keywords": [
     "ai",
@@ -39,7 +39,7 @@
     "release": "node build.js && node scripts/upload-release.cjs"
   },
   "dependencies": {
-    "@nexrall/code-core": "^1.4.39",
+    "@nexrall/code-core": "^1.4.42",
     "chalk": "^5.3.0",
     "commander": "^12.0.0",
     "diff": "^5.2.0",