nexrall-code 0.5.114 → 0.5.115

package/dist/index.js

@@ -3078,9 +3078,15 @@ var require_types = __commonJS({
   "../core/dist/types.js"(exports2) {
     "use strict";
     Object.defineProperty(exports2, "__esModule", { value: true });
-    exports2.AgentTurnError = void 0;
+    exports2.AgentTurnError = exports2.RUNTIME_CONTEXT_OPEN_TAG = void 0;
+    exports2.isRuntimeContextBlock = isRuntimeContextBlock;
     exports2.isAgentTurnError = isAgentTurnError;
     exports2.salvageHistory = salvageHistory2;
+    exports2.RUNTIME_CONTEXT_OPEN_TAG = "<runtime_context>";
+    function isRuntimeContextBlock(b) {
+      const x2 = b;
+      return !!x2 && x2.type === "text" && typeof x2.text === "string" && x2.text.startsWith(exports2.RUNTIME_CONTEXT_OPEN_TAG);
+    }
     var AgentTurnError = class extends Error {
       constructor(message, messages, progressCount, cause) {
         super(message);
@@ -10023,6 +10029,7 @@ var require_client = __commonJS({
     var RETRY_MAX_MS = 3e4;
     var HOUSEKEEPING_FRAMES = /* @__PURE__ */ new Set([
       "resumable",
+      "runtime_context",
       "context_warning",
       "tool_progress"
     ]);
@@ -10524,6 +10531,19 @@ var require_client = __commonJS({
                     serverResumable = true;
                     break;
                   }
+                  case "tool_use_ready": {
+                    if (typeof evt.id === "string" && typeof evt.name === "string") {
+                      const input = evt.input && typeof evt.input === "object" ? evt.input : {};
+                      onEvent({ type: "tool_use_ready", id: evt.id, name: evt.name, input });
+                    }
+                    break;
+                  }
+                  case "runtime_context": {
+                    if (typeof evt.text === "string" && evt.text) {
+                      onEvent({ type: "runtime_context", text: evt.text });
+                    }
+                    break;
+                  }
                   case "balance_status": {
                     const balance = typeof evt.balance === "number" ? evt.balance : 0;
                     const zero = !!evt.zero;
@@ -103978,6 +103998,88 @@ var require_readDedupe = __commonJS({
   }
 });
 
+// ../core/dist/agent/toolPrefetch.js
+var require_toolPrefetch = __commonJS({
+  "../core/dist/agent/toolPrefetch.js"(exports2) {
+    "use strict";
+    Object.defineProperty(exports2, "__esModule", { value: true });
+    exports2.ToolPrefetcher = exports2.PREFETCHABLE_TOOLS = void 0;
+    exports2.isPrefetchSafeMessage = isPrefetchSafeMessage;
+    exports2.toolPrefetchEnabled = toolPrefetchEnabled;
+    exports2.PREFETCHABLE_TOOLS = /* @__PURE__ */ new Set([
+      "read_file",
+      "list_directory",
+      "glob",
+      "search_files"
+    ]);
+    var NON_MUTATING_COMPANIONS = /* @__PURE__ */ new Set([
+      ...exports2.PREFETCHABLE_TOOLS,
+      "notebook_read",
+      "todo_write",
+      "todo_read",
+      "memory_read",
+      "use_skill",
+      "fetch_url",
+      "web_search",
+      "get_diagnostics",
+      "go_to_definition",
+      "find_references",
+      "get_symbols",
+      "get_workspace_symbols",
+      "get_hover"
+    ]);
+    function isPrefetchSafeMessage(toolNames) {
+      return toolNames.every((n) => NON_MUTATING_COMPANIONS.has(n));
+    }
+    function toolPrefetchEnabled() {
+      return process.env.NEXRALL_TOOL_PREFETCH !== "0";
+    }
+    function stableKey(value) {
+      if (value === null || typeof value !== "object")
+        return JSON.stringify(value) ?? "undefined";
+      if (Array.isArray(value))
+        return `[${value.map(stableKey).join(",")}]`;
+      const obj2 = value;
+      return `{${Object.keys(obj2).sort().map((k) => `${JSON.stringify(k)}:${stableKey(obj2[k])}`).join(",")}}`;
+    }
+    var ToolPrefetcher = class {
+      constructor(run) {
+        this.run = run;
+        this.entries = /* @__PURE__ */ new Map();
+      }
+      /** Start `name` early if it is a pure read. Idempotent per id. */
+      start(id, name, input) {
+        if (!id || !exports2.PREFETCHABLE_TOOLS.has(name) || this.entries.has(id))
+          return false;
+        const promise = Promise.resolve().then(() => this.run(name, input)).catch((err) => ({ error: `Tool execution failed: ${err?.message ?? String(err)}` }));
+        this.entries.set(id, { name, key: stableKey(input), promise });
+        return true;
+      }
+      /**
+       * The prefetched result for exactly this call, or null. Always removes the entry,
+       * so a result is handed out at most once.
+       */
+      take(id, name, input) {
+        const e2 = this.entries.get(id);
+        if (!e2)
+          return null;
+        this.entries.delete(id);
+        if (e2.name !== name || e2.key !== stableKey(input))
+          return null;
+        return e2.promise;
+      }
+      /** Drop everything (stream restarted — the ids it announced will never be dispatched). */
+      clear() {
+        this.entries.clear();
+      }
+      get size() {
+        return this.entries.size;
+      }
+    };
+    exports2.ToolPrefetcher = ToolPrefetcher;
+  }
+});
+
 // ../core/dist/agent/worktree.js
 var require_worktree = __commonJS({
   "../core/dist/agent/worktree.js"(exports2) {
@@ -106121,6 +106223,7 @@ var require_loop = __commonJS({
     exports2.estimateBodyBytes = estimateBodyBytes2;
     exports2.allowsTestOnlyWrite = allowsTestOnlyWrite;
     exports2.findSafeCutIndex = findSafeCutIndex;
+    exports2.persistRuntimeContext = persistRuntimeContext;
     exports2.transcriptOf = transcriptOf;
     exports2.createLedger = createLedger;
     exports2.ledgerRecord = ledgerRecord;
@@ -106134,6 +106237,7 @@ var require_loop = __commonJS({
     exports2.runAgentLoop = runAgentLoop2;
     exports2.trimToResumableBoundary = trimToResumableBoundary;
     var readDedupe_1 = require_readDedupe();
+    var toolPrefetch_1 = require_toolPrefetch();
     var worktree_1 = require_worktree();
     var backgroundAgents_1 = require_backgroundAgents();
     var types_1 = require_types();
@@ -107448,10 +107552,22 @@ ${partial}` : "",
       return -1;
     }
     var MAX_TRANSCRIPT_CHARS = 6e5;
+    function persistRuntimeContext(messages, idx, text) {
+      const m2 = messages[idx];
+      if (!m2 || m2.role !== "user" || !Array.isArray(m2.content))
+        return false;
+      const lastBlock = m2.content[m2.content.length - 1];
+      if (lastBlock && lastBlock.type === "text" && lastBlock.text === text)
+        return false;
+      messages[idx] = { ...m2, content: [...m2.content, { type: "text", text }] };
+      return true;
+    }
     function transcriptOf(messages) {
       const parts = [];
       for (const m2 of messages) {
         for (const b of m2.content) {
+          if ((0, types_1.isRuntimeContextBlock)(b))
+            continue;
           if (b.type === "text" && b.text) {
             parts.push(`${m2.role.toUpperCase()}: ${(0, safeSlice_1.sliceSafeEnd)(b.text, 2e3)}`);
           } else if (b.type === "tool_use") {
@@ -107604,7 +107720,7 @@ ${tail}`;
       const first = messages.find((m2) => m2.role === "user");
       if (!first || !Array.isArray(first.content))
         return "";
-      return first.content.filter((b) => b.type === "text" && b.text).map((b) => b.text).join("\n").trim();
+      return first.content.filter((b) => b.type === "text" && b.text && !(0, types_1.isRuntimeContextBlock)(b)).map((b) => b.text).join("\n").trim();
     }
     var CompactionUnavailableError = class extends Error {
       constructor() {
@@ -107662,7 +107778,14 @@ Do NOT call any tools and do NOT continue the task. Reply with text only: a conc
             // surprise charge on a different provider. Falls back to the same
             // default as the main loop (see `runAgentLoop`) only when no model was
             // set at all.
-            model: options.model ?? "turbo",
+            //
+            // Cheapest model of the SAME vendor (Claude Code runs this kind of work on
+            // Haiku). Safe to downgrade HERE because this fallback sends a fresh
+            // transcript — it shares no cached prefix with the conversation, unlike
+            // summarizeCached above, which must stay on the conversation's own model.
+            // Never crosses providers; falls back to the session model if the
+            // catalogue is not loaded.
+            model: (0, modelCatalogue_1.cheapestSameVendorModel)(options.model ?? "turbo") ?? options.model ?? "turbo",
             mode: "ask",
             // summariser must not call tools; ask-mode discourages action
             env: options.env,
@@ -107705,7 +107828,9 @@ Continue the work from here.` }] });
             // Same reasoning as autoCompactMessages' summariser: run on the same
             // model as the actual conversation rather than forcing a fixed tier
             // (which silently billed Anthropic regardless of the user's provider).
-            model: options.model ?? "turbo",
+            // Cheapest SAME-vendor model: a fresh prompt with no shared cache prefix,
+            // and merging a few memory bullets does not need the session's top tier.
+            model: (0, modelCatalogue_1.cheapestSameVendorModel)(options.model ?? "turbo") ?? options.model ?? "turbo",
             mode: "ask",
             env: options.env,
             clientType: options.clientType,
@@ -107989,7 +108114,6 @@ Continue the work from here.` }] });
             if (reclaimed > 0) {
               bodyBytes = estimateBodyBytes2(messages);
               bytePressure = bodyBytes > MAX_BODY_BYTES;
-              (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) {
@@ -108014,7 +108138,6 @@ Continue the work from here.` }] });
               if (did) {
                 lastPromptTokens = 0;
                 const reason = bytePressure ? `body ~${(bodyBytes / (1024 * 1024)).toFixed(1)}MB` : "context window";
-                (options.onNotice ?? options.onText)(`\u267B\uFE0F Auto-compacted earlier conversation to stay within the ${reason}.`);
                 bodyBytes = bytesAfter;
                 bytePressure = bodyBytes > MAX_BODY_BYTES;
               }
@@ -108035,6 +108158,10 @@ Continue the work from here.` }] });
 \u26A0\uFE0F Approaching the ${budget}-step limit (step ${iteration + 1}). Please wrap up and summarise what has been done.
 `);
           }
+          let pendingRuntimeContext = null;
+          const sentUserIdx = messages.length - 1;
+          const prefetcher = new toolPrefetch_1.ToolPrefetcher((toolName, toolInput) => (0, executor_1.executeTool)(toolName, toolInput, options.abortSignal, sandboxCfg, options.workDir, agentScope, void 0, options.selfPeer));
+          const prefetchOn = (0, toolPrefetch_1.toolPrefetchEnabled)();
           const onEvent = (event) => {
             switch (event.type) {
               case "text":
@@ -108066,11 +108193,19 @@ Continue the work from here.` }] });
                 options.onRetryResolved?.();
                 break;
               case "stream_restart":
+                prefetcher.clear();
                 options.onStreamRestart?.(event.reason, event.discardedChars);
                 break;
               case "balance_status":
                 options.onBalanceStatus?.(event.balance, event.zero);
                 break;
+              case "runtime_context":
+                pendingRuntimeContext = event.text;
+                break;
+              case "tool_use_ready":
+                if (prefetchOn && !options.abortSignal?.aborted)
+                  prefetcher.start(event.id, event.name, event.input);
+                break;
               case "done":
                 break;
               case "error":
@@ -108090,6 +108225,8 @@ Continue the work from here.` }] });
               allowRestartAfterRender: !!options.onStreamRestart
             }, onEvent);
             options.onApiCallDuration?.(Date.now() - _apiCallStartedAt);
+            if (pendingRuntimeContext)
+              persistRuntimeContext(messages, sentUserIdx, pendingRuntimeContext);
           } catch (err) {
             options.onApiCallDuration?.(Date.now() - _apiCallStartedAt);
             if (options.abortSignal?.aborted || err.name === "AbortError") {
@@ -108244,6 +108381,7 @@ Continue the work from here.` }] });
             stopReason = "clean";
             break;
           }
+          const prefetchSafe = prefetchOn && (0, toolPrefetch_1.isPrefetchSafeMessage)(toolUseBlocks.map((b) => b.name));
           const toolResults = await Promise.all(toolUseBlocks.map(async (block) => {
             const { id, name, input } = block;
             const toolMeta = { id };
@@ -108331,8 +108469,14 @@ Continue the work from here.` }] });
                     const run = () => (0, executor_1.executeTool)(name, input, options.abortSignal, sandboxCfg, options.workDir, agentScope, onStream, options.selfPeer);
                     const locks = name === "bash" ? (0, bashClassify_1.bashNeedsRepoLock)(String(input.command ?? "")) ? [REPO_STATE_LOCK] : [] : lockPathsFor(name, input, options.workDir);
                     const dedupeStub = name === "read_file" ? readDedupe.check(input, messages) : null;
+                    const prefetched = !dedupeStub && prefetchSafe ? prefetcher.take(id, name, input) : null;
                     if (dedupeStub) {
                       result = { output: dedupeStub };
+                    } else if (prefetched) {
+                      result = await prefetched;
+                      if (name === "read_file" && result.error === void 0 && result.output?.startsWith("[File:")) {
+                        readDedupe.record(input, id);
+                      }
                     } else {
                       result = locks.length ? await (0, crossProcessLock_1.withCrossProcessLocks)(locks, () => withFileLocks(locks, run), options.workDir) : await run();
                       if (name === "read_file" && result.error === void 0 && result.output?.startsWith("[File:")) {
@@ -156640,12 +156784,24 @@ function collectEnv(workDir) {
     platform: `${os5.platform()} ${os5.release()}`,
     shell: process.env.SHELL ?? process.env.ComSpec ?? "unknown",
     date: (/* @__PURE__ */ new Date()).toISOString().slice(0, 10),
+    ...collectGitEnv(workDir)
+  };
+}
+function collectGitEnv(workDir) {
+  return {
     gitBranch: tryExec("git branch --show-current", workDir),
     gitStatus: tryExec("git status --short", workDir),
     gitDiff: tryExec("git diff --stat HEAD", workDir),
     recentCommits: tryExec("git log --oneline -5", workDir)
   };
 }
+function refreshEnvForTurn(env4, workDir) {
+  return {
+    ...env4,
+    date: (/* @__PURE__ */ new Date()).toISOString().slice(0, 10),
+    ...collectGitEnv(workDir)
+  };
+}
 var NEXRALL_MD_CANDIDATES = [
   "nexrall.md",
   "NEXRALL.md",
@@ -156878,7 +157034,7 @@ async function runTurn(messages, modelAlias, workDir, abortSignal, env4, nexrall
   const result = await (0, import_code_core5.runAgentLoop)(messages, {
     workDir,
     model: modelAlias,
-    env: env4,
+    env: refreshEnvForTurn(env4, workDir),
     nexrallMd,
     mode,
     // `/mode plan` previously only hinted to the backend — the footer said "plan
@@ -157087,7 +157243,7 @@ async function runTurnHeadless(messages, modelAlias, workDir, _abortSignal, env4
   const result = await (0, import_code_core5.runAgentLoop)(messages, {
     workDir,
     model: modelAlias,
-    env: env4,
+    env: refreshEnvForTurn(env4, workDir),
     nexrallMd,
     mode,
     planMode: mode === "plan",
@@ -157403,8 +157559,8 @@ Resume it with: nex --worktree ${worktreeState.name}
     startRowCount();
   }
   if (options.showBanner !== false && !headless) {
-    const branch = tryExec("git branch --show-current", workDir);
-    const gitStatus = tryExec("git status --short", workDir);
+    const branch = env4.gitBranch;
+    const gitStatus = env4.gitStatus;
     const changedLines = gitStatus ? gitStatus.split("\n").filter(Boolean).length : 0;
     const tips = ["Run /init to create a nexrall.md file with instructions for Nexrall Code."];
     if (branch) {
@@ -157472,11 +157628,8 @@ Resume it with: nex --worktree ${worktreeState.name}
       const compacted = await (0, import_code_core5.compactMessagesForResume)(messages, {
         workDir,
         model: modelAlias,
-        clientType: "cli",
-        onNotice: (text) => {
-          if (!headless)
-            console.log(source_default.dim(text.trim()));
-        }
+        clientType: "cli"
+        // Compaction is silent housekeeping — no "♻️ Auto-compacted …" line.
       });
     } catch {
     }
@@ -157709,8 +157862,8 @@ ${text}` : "");
             await (0, import_code_core5.compactMessagesForResume)(messages, {
               workDir,
               model: modelAlias,
-              clientType: "cli",
-              onNotice: (text) => console.log(source_default.dim(text.trim()))
+              clientType: "cli"
+              // Silent housekeeping — no "♻️ Auto-compacted …" line.
             });
           } catch {
           }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "nexrall-code",
-  "version": "0.5.114",
+  "version": "0.5.115",
   "description": "Nexrall Code — AI coding assistant for your terminal (headless agent for scripts, CI and automation)",
   "keywords": [
     "ai",