omnius 1.0.383 → 1.0.385

package/dist/index.js

@@ -8602,12 +8602,14 @@ __export(vision_exports, {
   MOONDREAM3_PREVIEW_HF_MODEL: () => MOONDREAM3_PREVIEW_HF_MODEL,
   VisionTool: () => VisionTool,
   analyzeImageWithVision: () => analyzeImageWithVision,
+  callOllamaVision: () => callOllamaVision,
   formatVisionPointResult: () => formatVisionPointResult,
   getVisionPointDiagnostics: () => getVisionPointDiagnostics,
   locateImagePoints: () => locateImagePoints,
   normalizeVisionModelName: () => normalizeVisionModelName,
   resetMoondreamClient: () => resetMoondreamClient,
   resolveHuggingFaceVisionModelCandidates: () => resolveHuggingFaceVisionModelCandidates,
+  resolveInstalledOllamaVisionModelAlias: () => resolveInstalledOllamaVisionModelAlias,
   resolveOllamaVisionModelCandidates: () => resolveOllamaVisionModelCandidates
 });
 import { mkdirSync as mkdirSync9, readFileSync as readFileSync12, existsSync as existsSync14, statSync as statSync7, unlinkSync as unlinkSync2, writeFileSync as writeFileSync10 } from "node:fs";
@@ -8940,7 +8942,8 @@ function resolveOllamaVisionModelCandidates(options2 = {}) {
     ollamaVisionModelName(options2.preferredModel || ""),
     process.env["OLLAMA_VISION_MODEL"] || "",
     options2.activeModelHasVision && options2.activeModel ? options2.activeModel : "",
-    DEFAULT_OLLAMA_VISION_MODEL
+    DEFAULT_OLLAMA_VISION_MODEL,
+    `${DEFAULT_OLLAMA_VISION_MODEL}:latest`
   ].map((entry) => entry.trim()).filter(Boolean);
   return [...new Set(candidates)];
 }
@@ -9191,6 +9194,26 @@ async function callOllamaVision(ollamaHost, model, prompt, imageBase64, timeoutM
   if (!res.ok && shouldAutoPullOllamaVisionModel(model)) {
     const errText = await res.text().catch(() => "");
     if (res.status === 404 || /not found|does not exist/i.test(errText)) {
+      const installedAlias = await resolveInstalledOllamaVisionModelAlias(ollamaHost, model, timeoutMs);
+      if (installedAlias && installedAlias !== model) {
+        res = await fetch(`${ollamaHost}/api/generate`, {
+          method: "POST",
+          headers: { "Content-Type": "application/json" },
+          body: JSON.stringify({
+            model: installedAlias,
+            prompt,
+            images: [imageBase64],
+            stream: false,
+            think: false,
+            options: { temperature: 0 }
+          }),
+          signal: AbortSignal.timeout(timeoutMs)
+        });
+        if (res.ok) {
+          const data2 = await res.json();
+          return typeof data2.response === "string" && data2.response.trim() ? data2.response : null;
+        }
+      }
       try {
         ensureDiskSpaceForOllamaVisionModel(model);
         pullOllamaVisionModel(model);
@@ -9217,6 +9240,33 @@ async function callOllamaVision(ollamaHost, model, prompt, imageBase64, timeoutM
   const data = await res.json();
   return typeof data.response === "string" && data.response.trim() ? data.response : null;
 }
+async function resolveInstalledOllamaVisionModelAlias(ollamaHost, model, timeoutMs = 5e3) {
+  const requested = model.trim();
+  if (!requested)
+    return null;
+  try {
+    const res = await fetch(`${ollamaHost}/api/tags`, {
+      signal: AbortSignal.timeout(Math.min(Math.max(timeoutMs, 1e3), 5e3))
+    });
+    if (!res.ok)
+      return null;
+    const data = await res.json();
+    const names = (Array.isArray(data.models) ? data.models : []).map((entry) => typeof entry.name === "string" ? entry.name.trim() : "").filter(Boolean);
+    if (names.includes(requested))
+      return requested;
+    if (!requested.includes(":")) {
+      const latest = `${requested}:latest`;
+      if (names.includes(latest))
+        return latest;
+      const prefixMatch = names.find((name10) => name10.startsWith(`${requested}:`));
+      if (prefixMatch)
+        return prefixMatch;
+    }
+  } catch {
+    return null;
+  }
+  return null;
+}
 function shouldAutoPullOllamaVisionModel(model) {
   if (!envFlag2(process.env["OMNIUS_OLLAMA_VISION_AUTO_PULL"], true))
     return false;
@@ -295276,6 +295326,21 @@ function getTodoSessionId() {
     return envSession;
   return "default";
 }
+function validateLargeTaskDecomposition(todos) {
+  if (todos.length < 20)
+    return null;
+  const ids = new Set(todos.map((todo) => todo.id).filter((id) => typeof id === "string" && id.trim().length > 0));
+  const childTodos = todos.filter((todo) => typeof todo.parentId === "string" && todo.parentId.trim().length > 0);
+  const parentIds = new Set(childTodos.map((todo) => todo.parentId.trim()));
+  const hasValidParent = [...parentIds].some((parentId) => ids.has(parentId));
+  if (childTodos.length > 0 && hasValidParent)
+    return null;
+  return [
+    "Large todo lists (20+ items) must be decomposed into a nested tree with stable ids and parentId links.",
+    "Create parent objectives and child leaf tasks instead of a flat checklist.",
+    'Canonical shape: todo_write({"todos":[{"id":"group-1","content":"Steps 01-08","status":"in_progress"},{"id":"step-01","parentId":"group-1","content":"Complete step 01 and verify evidence","status":"in_progress"}]})'
+  ].join(" ");
+}
 function normalizeIncomingTodos(args) {
   const repairNotes = [];
   const record = args;
@@ -295462,6 +295527,15 @@ Mark tasks complete IMMEDIATELY after finishing — don't batch. Never mark comp
               declaredArtifacts: Array.isArray(entry["declaredArtifacts"]) ? entry["declaredArtifacts"].filter((x) => typeof x === "string") : void 0
             });
           }
+          const decompositionError = validateLargeTaskDecomposition(incoming);
+          if (decompositionError) {
+            return {
+              success: false,
+              output: "",
+              error: decompositionError,
+              durationMs: performance.now() - start2
+            };
+          }
           const sessionId = typeof args["session_id"] === "string" && args["session_id"].trim() ? args["session_id"].trim() : typeof args["sessionId"] === "string" && args["sessionId"].trim() ? args["sessionId"].trim() : getTodoSessionId();
           const oldTodos = readTodos(sessionId);
           const canonicalize2 = (todos) => JSON.stringify(todos.map((t2) => ({
@@ -547133,6 +547207,23 @@ function summarizeProcessFailure(stdout, stderr) {
   }
   return parts.join("\n").slice(0, 2200);
 }
+function formatObjectRecognitionResult(result) {
+  const matches = (Array.isArray(result.matches) ? result.matches : []).filter((m2) => m2.recognized);
+  const matchLines = matches.map((m2) => `  ${m2.label}: ${(m2.blended_score * 100).toFixed(0)}% (image=${(m2.image_similarity * 100).toFixed(0)}%, text=${(m2.text_similarity * 100).toFixed(0)}%)`);
+  const extraLabels = Array.isArray(result.extra_labels) ? result.extra_labels : null;
+  if (extraLabels && extraLabels.length > 0) {
+    const extraLines = extraLabels.map((s2) => `  ${s2.label}: ${(s2.score * 100).toFixed(0)}%`);
+    const sections = [`CLIP candidate label scores:
+${extraLines.join("\n")}`];
+    if (matches.length > 0) {
+      sections.push(`Persistent visual memory matches above threshold:
+${matchLines.join("\n")}`);
+    }
+    return sections.join("\n\n");
+  }
+  return matches.length > 0 ? `Recognized ${result.recognized_count} object(s):
+${matchLines.join("\n")}` : "No taught objects recognized in this image.";
+}
 var VMEM_DIR, VENV_DIR2, VENV_PY, VENV_PIP2, VISUAL_MEMORY_ACTIONS, VisualMemoryTool;
 var init_visual_memory = __esm({
   "packages/execution/dist/tools/visual-memory.js"() {
@@ -547668,18 +547759,7 @@ print(json.dumps({
           const payload = JSON.stringify(result);
           return { success: true, output: payload, llmContent: payload, durationMs: performance.now() - start2 };
         }
-        const matches = (result.matches || []).filter((m2) => m2.recognized);
-        const lines = matches.map((m2) => `  ${m2.label}: ${(m2.blended_score * 100).toFixed(0)}% (image=${(m2.image_similarity * 100).toFixed(0)}%, text=${(m2.text_similarity * 100).toFixed(0)}%)`);
-        let output = matches.length > 0 ? `Recognized ${result.recognized_count} object(s):
-${lines.join("\n")}` : "No taught objects recognized in this image.";
-        if (result.extra_labels) {
-          const extraLines = result.extra_labels.map((s2) => `  ${s2.label}: ${(s2.score * 100).toFixed(0)}%`);
-          output += `
-
-CLIP label scores:
-${extraLines.join("\n")}`;
-        }
-        return { success: true, output, durationMs: performance.now() - start2 };
+        return { success: true, output: formatObjectRecognitionResult(result), durationMs: performance.now() - start2 };
       }
       // =========================================================================
       // Memory Management
@@ -569521,8 +569601,8 @@ var init_focusSupervisor = __esm({
             const directive = this.setDirective({
               turn: input.turn,
               state: ignoredManyTimes ? "verify_or_block" : "single_next_action",
-              reason: ignoredManyTimes ? `model ignored ${this.ignoredDirectiveStreak} focus directives; report incomplete or ask for help instead of trying another variant` : `model ignored prior directive ${prior.id}; ${prior.reason}`,
-              requiredNextAction: ignoredManyTimes ? "report_incomplete" : prior.requiredNextAction,
+              reason: ignoredManyTimes ? `model ignored ${this.ignoredDirectiveStreak} focus directives; take the required recovery action before trying another variant` : `model ignored prior directive ${prior.id}; ${prior.reason}`,
+              requiredNextAction: prior.requiredNextAction,
               forbiddenActionFamilies: unique2([
                 ...prior.forbiddenActionFamilies,
                 family
@@ -569626,7 +569706,7 @@ var init_focusSupervisor = __esm({
             turn: input.turn,
             state: "forced_replan",
             reason: `same ${input.toolName} failure family repeated ${next.count} times: ${next.sample}`,
-            requiredNextAction: input.toolName === "shell" ? "read_authoritative_target" : "update_todos",
+            requiredNextAction: input.toolName === "shell" ? "edit_different_target" : "update_todos",
             forbiddenActionFamilies: [actionFamily(input.toolName, input.args)]
           });
         }
@@ -572951,6 +573031,7 @@ ${parts.join("\n")}
           memoryPrefix: options2?.memoryPrefix ?? "",
           memoryPrefixHash: options2?.memoryPrefixHash ?? "",
           stateDir: options2?.stateDir ?? "",
+          surface: options2?.surface ?? "tui",
           artifactMode: options2?.artifactMode ?? "user-task",
           disablePersistentMemory: options2?.disablePersistentMemory ?? false,
           disableCodebaseMap: options2?.disableCodebaseMap ?? false,
@@ -573302,16 +573383,21 @@ ${parts.join("\n")}
       // -------------------------------------------------------------------------
       /** Infer the surface identifier from runner configuration and dynamic context. */
       _inferSurface() {
-        const ctx3 = this._stickyDynamicContext || this.options.dynamicContext || "";
-        if (/Admin Capability/i.test(ctx3) || /telegram-admin/i.test(ctx3))
-          return "telegram-admin";
-        if (/Telegram|telegram|Voice Soul/i.test(ctx3) || this.options.stateDir)
-          return "telegram-public";
-        if (/api/i.test(ctx3))
-          return "api";
-        if (/background/i.test(ctx3))
-          return "background";
-        return "tui";
+        return this.options.surface;
+      }
+      _isTelegramSurface() {
+        return this.options.surface === "telegram-public" || this.options.surface === "telegram-admin";
+      }
+      stickyDynamicContextForActiveSurface() {
+        const ctx3 = this._stickyDynamicContext.trim();
+        if (!ctx3)
+          return "";
+        if (this._isTelegramSurface())
+          return ctx3;
+        const voiceSoul = this.extractDynamicMarkdownBlock(ctx3, "## Voice Soul Context", 6e3);
+        if (!voiceSoul)
+          return "";
+        return /Telegram|telegram|Public Telegram|Admin Capability/i.test(voiceSoul) ? "" : voiceSoul;
       }
       /**
        * Build structured context via the context engine.
@@ -587072,10 +587158,11 @@ ${postCompactRestore.join("\n")}`);
 [Ephemeral skill-pack restore — current run only, do not persist]
 ${this._ephemeralSkillPackContext}
 Use skill_extract for targeted skill unpacking; do not load full skills into the main context unless necessary.` : "";
-        const stickyDynamicContextReminder = this._stickyDynamicContext ? `
+        const scopedStickyDynamicContext = this.stickyDynamicContextForActiveSurface();
+        const stickyDynamicContextReminder = scopedStickyDynamicContext ? `
 
 [Sticky dynamic context restore — surface/persona anchors]
-${this._stickyDynamicContext}` : "";
+${scopedStickyDynamicContext}` : "";
         const compactionMsg = {
           role: "system",
           // WO-CE-03: XML tags for structural clarity on small/medium models
@@ -587092,7 +587179,7 @@ ${fullSummary}
         this.persistCheckpoint(fullSummary);
         let narrowedHead = [...head];
         const EVIDENCE_RULE_COMPACT = `EVIDENCE RULE (PRIORITY 0): never claim something works or is true unless a tool result you saw this turn proves it. A command succeeding only means it ran — not that the intended effect happened; verify the end-state directly before claiming it. A negative, empty, or error result means failed or absent — report it, never explain it away with an untested theory. Never describe how you got a result (tool, command, or source) unless you actually used it. Do not assert relationships the output does not show. Say "I could not verify X" when it is unproven — that is the correct answer, not a guess.`;
-        const telegramPersonaHead = /Telegram|Voice Soul Context|Public Telegram voice profile/.test(this._stickyDynamicContext) ? `You are Omnius replying through Telegram. Your visible assistant text is sent to Telegram; keep it concise, scoped, and user-facing. Do not emit scratch notes, router decisions, internal status, or no_reply text. Use available tools when needed and call task_complete when the Telegram run is complete.
+        const telegramPersonaHead = this._isTelegramSurface() && /Telegram|Voice Soul Context|Public Telegram voice profile/.test(scopedStickyDynamicContext) ? `You are Omnius replying through Telegram. Your visible assistant text is sent to Telegram; keep it concise, scoped, and user-facing. Do not emit scratch notes, router decisions, internal status, or no_reply text. Use available tools when needed and call task_complete when the Telegram run is complete.
 
 ${EVIDENCE_RULE_COMPACT}
 
@@ -587247,7 +587334,12 @@ ${content.slice(0, 8e3)}
             while (trimmedRecent.length > 1 && trimmedRecent[0]?.role === "tool") {
               trimmedRecent = trimmedRecent.slice(1);
             }
-            result = [...head, compactionMsg, ...stickyToKeep, ...trimmedRecent];
+            result = [
+              ...narrowedHead,
+              compactionMsg,
+              ...stickyToKeep,
+              ...trimmedRecent
+            ];
           }
           if (trimmedRecent.length < filteredRecent.length) {
             this.emit({
@@ -680834,6 +680926,7 @@ ${conversationStream}`
         );
         const requestTimeoutMs = config.timeoutMs ?? 3e5;
         const runner = new AgenticRunner(backend, {
+          surface: isAdminDM || isAdminGroup ? "telegram-admin" : "telegram-public",
           // Admin DMs are operator-directed work sessions. A hard turn cap turns
           // active tool progress into a false "completed" Telegram panel when the
           // model has not reached task_complete yet. Public/group runs stay bounded.
@@ -719721,6 +719814,7 @@ Only tools allowed by this profile are visible and executable.`
     ].filter(Boolean).join("");
   }
   const runner = new AgenticRunner(backend, {
+    surface: "tui",
     maxTurns: realtimeEnabled ? Math.min(effectiveMaxTurns, 8) : effectiveMaxTurns,
     maxTokens: realtimeEnabled ? 512 : 16384,
     temperature: realtimeEnabled ? 0.6 : 0,
@@ -722715,6 +722809,7 @@ Respond to the scoped Telegram target when complete.`
       }
       const modelTier2 = getModelTier(currentConfig.model);
       const runner = new AgenticRunner(backend, {
+        surface: "background",
         // 0 = unlimited; halt only on task_complete or abort. Background
         // prompts may legitimately need many turns; an arbitrary cap stalls
         // them mid-task.
@@ -728675,11 +728770,17 @@ var init_serve2 = __esm({
 // packages/cli/src/commands/eval.ts
 var eval_exports = {};
 __export(eval_exports, {
-  evalCommand: () => evalCommand
+  createTempEvalRepo: () => createTempEvalRepo,
+  evalCommand: () => evalCommand,
+  expectedStatusesForEvalTask: () => expectedStatusesForEvalTask
 });
 import { tmpdir as tmpdir23 } from "node:os";
 import { mkdirSync as mkdirSync106, writeFileSync as writeFileSync90 } from "node:fs";
 import { join as join178 } from "node:path";
+function expectedStatusesForEvalTask(task, live) {
+  if (!live) return task.expectedStatuses;
+  return task.liveExpectedStatuses ?? task.expectedStatuses.filter((status) => status !== "needs_human_decision");
+}
 async function evalCommand(opts, config) {
   const suiteName = opts.suite ?? "basic";
   const suite = SUITES[suiteName];
@@ -728693,6 +728794,10 @@ async function evalCommand(opts, config) {
   printKeyValue("Suite", suiteName, 2);
   printKeyValue("Tasks", String(suite.length), 2);
   printKeyValue("Mode", modeLabel, 2);
+  if (useLive) {
+    printKeyValue("Live pass statuses", "success, partial_success", 2);
+    printInfo("Live eval treats needs_human_decision as a failure for concrete coding tasks.");
+  }
   const evalRepoRoot = opts.repoPath ?? createTempEvalRepo();
   let rawBackend;
   if (useLive) {
@@ -728751,22 +728856,27 @@ async function evalCommand(opts, config) {
     let result;
     try {
       const report2 = await loop.run(task.request, evalRepoRoot);
-      const passed2 = task.expectedStatuses.includes(report2.status);
+      const expectedStatuses = expectedStatusesForEvalTask(task, useLive);
+      const passed2 = expectedStatuses.includes(report2.status);
       result = {
         task,
         status: report2.status,
+        expectedStatuses,
         passed: passed2,
         durationMs: Date.now() - start2
       };
       if (passed2) {
         spinner.succeed(`[${task.id}] PASS (${report2.status})`);
       } else {
-        spinner.fail(`[${task.id}] FAIL (got: ${report2.status})`);
+        spinner.fail(
+          `[${task.id}] FAIL (got: ${report2.status}; expected: ${expectedStatuses.join(", ")})`
+        );
       }
     } catch (err) {
       result = {
         task,
         status: "error",
+        expectedStatuses: expectedStatusesForEvalTask(task, useLive),
         passed: false,
         durationMs: Date.now() - start2,
         error: err instanceof Error ? err.message : String(err)
@@ -728792,7 +728902,7 @@ async function evalCommand(opts, config) {
       const icon = r2.passed ? "PASS" : "FAIL";
       printKeyValue(
         `${r2.task.id} [${icon}]`,
-        `${r2.status} (${formatDuration(r2.durationMs)})`,
+        `${r2.status} (${formatDuration(r2.durationMs)}; expected ${r2.expectedStatuses.join(", ")})`,
         2
       );
       if (r2.error) {
@@ -728810,13 +728920,79 @@ async function evalCommand(opts, config) {
 function createTempEvalRepo() {
   const dir = join178(tmpdir23(), `omnius-eval-${Date.now()}`);
   mkdirSync106(dir, { recursive: true });
-  writeFileSync90(
-    join178(dir, "package.json"),
-    JSON.stringify({ name: "eval-repo", version: "0.0.0" }, null, 2) + "\n",
-    "utf8"
-  );
+  mkdirSync106(join178(dir, "src"), { recursive: true });
+  mkdirSync106(join178(dir, "tests"), { recursive: true });
+  writeEvalFile(dir, "package.json", JSON.stringify({
+    name: "eval-repo",
+    version: "0.0.0",
+    type: "module",
+    scripts: { test: "node tests/auth.test.js && node tests/users.test.js && node tests/db.test.js && node tests/payment.test.js" }
+  }, null, 2));
+  writeEvalFile(dir, "src/auth.js", [
+    "export function authenticateUser(user) {",
+    "  return user.active;",
+    "}"
+  ].join("\n"));
+  writeEvalFile(dir, "src/users.js", [
+    "export function listUsers(users) {",
+    "  return users.slice();",
+    "}"
+  ].join("\n"));
+  writeEvalFile(dir, "src/db.js", [
+    "export function getConnection() {",
+    '  return Promise.resolve({ id: "primary", open: true });',
+    "}"
+  ].join("\n"));
+  writeEvalFile(dir, "src/payment.js", [
+    "export function applyDiscount(amount, percent) {",
+    "  return amount - amount * (percent / 100);",
+    "}",
+    "",
+    "export function addTax(amount, taxRate) {",
+    "  return amount + amount * taxRate;",
+    "}"
+  ].join("\n"));
+  writeEvalFile(dir, "src/api.js", [
+    "export function health() {",
+    "  return { ok: true };",
+    "}",
+    "",
+    "export function version() {",
+    '  return "0.0.0";',
+    "}"
+  ].join("\n"));
+  writeEvalFile(dir, "tests/auth.test.js", [
+    "import assert from 'node:assert/strict';",
+    "import { authenticateUser } from '../src/auth.js';",
+    "assert.equal(authenticateUser({ id: 'u1', active: true }), true);",
+    "assert.equal(authenticateUser({ id: 'u2', active: false }), false);",
+    "assert.equal(authenticateUser(null), false);",
+    "assert.equal(authenticateUser(undefined), false);"
+  ].join("\n"));
+  writeEvalFile(dir, "tests/users.test.js", [
+    "import assert from 'node:assert/strict';",
+    "import { paginateUsers } from '../src/users.js';",
+    "const users = ['a', 'b', 'c', 'd', 'e'];",
+    "assert.deepEqual(paginateUsers(users, 1, 2), { items: ['a', 'b'], page: 1, pageSize: 2, totalPages: 3, totalItems: 5 });",
+    "assert.deepEqual(paginateUsers(users, 3, 2).items, ['e']);"
+  ].join("\n"));
+  writeEvalFile(dir, "tests/db.test.js", [
+    "import assert from 'node:assert/strict';",
+    "import { getConnection } from '../src/db.js';",
+    "const conn = await getConnection();",
+    "assert.deepEqual(conn, { id: 'primary', open: true });"
+  ].join("\n"));
+  writeEvalFile(dir, "tests/payment.test.js", [
+    "import assert from 'node:assert/strict';",
+    "import { applyDiscount, addTax } from '../src/payment.js';",
+    "assert.equal(applyDiscount(100, 15), 85);",
+    "assert.equal(addTax(100, 0.0825), 108.25);"
+  ].join("\n"));
   return dir;
 }
+function writeEvalFile(root, relativePath, content) {
+  writeFileSync90(join178(root, relativePath), content.trimEnd() + "\n", "utf8");
+}
 var BASIC_SUITE, FULL_SUITE, SUITES;
 var init_eval = __esm({
   "packages/cli/src/commands/eval.ts"() {
@@ -728828,21 +729004,24 @@ var init_eval = __esm({
     BASIC_SUITE = [
       {
         id: "eval-001",
-        description: "Simple fix request",
-        request: "Fix the null pointer dereference in the auth module",
-        expectedStatuses: ["success", "partial_success", "needs_human_decision"]
+        description: "Boundary bug fix",
+        request: "In src/auth.js, fix authenticateUser so null or undefined users return false instead of throwing. Use tests/auth.test.js as the acceptance evidence.",
+        expectedStatuses: ["success", "partial_success", "needs_human_decision"],
+        liveExpectedStatuses: ["success", "partial_success"]
       },
       {
         id: "eval-002",
         description: "Feature addition request",
-        request: "Add pagination support to the user list endpoint",
-        expectedStatuses: ["success", "partial_success", "needs_human_decision"]
+        request: "In src/users.js, add paginateUsers(users, page, pageSize) with 1-based page indexing, stable slicing, and totalPages metadata. Use tests/users.test.js as the acceptance evidence.",
+        expectedStatuses: ["success", "partial_success", "needs_human_decision"],
+        liveExpectedStatuses: ["success", "partial_success"]
       },
       {
         id: "eval-003",
         description: "Refactor request",
-        request: "Refactor the database connection pool to use async/await",
-        expectedStatuses: ["success", "partial_success", "needs_human_decision"]
+        request: "In src/db.js, refactor getConnection to async/await while preserving the exported API behavior covered by tests/db.test.js.",
+        expectedStatuses: ["success", "partial_success", "needs_human_decision"],
+        liveExpectedStatuses: ["success", "partial_success"]
       }
     ];
     FULL_SUITE = [
@@ -728850,14 +729029,16 @@ var init_eval = __esm({
       {
         id: "eval-004",
         description: "Test generation request",
-        request: "Write unit tests for the payment processing module",
-        expectedStatuses: ["success", "partial_success", "needs_human_decision"]
+        request: "Add missing unit coverage for src/payment.js discount and tax behavior in tests/payment.test.js without changing production semantics.",
+        expectedStatuses: ["success", "partial_success", "needs_human_decision"],
+        liveExpectedStatuses: ["success", "partial_success"]
       },
       {
         id: "eval-005",
         description: "Documentation request",
-        request: "Add JSDoc comments to all exported functions in the API layer",
-        expectedStatuses: ["success", "partial_success", "needs_human_decision"]
+        request: "Add concise JSDoc comments to the exported functions in src/api.js while preserving behavior.",
+        expectedStatuses: ["success", "partial_success", "needs_human_decision"],
+        liveExpectedStatuses: ["success", "partial_success"]
       }
     ];
     SUITES = {

package/npm-shrinkwrap.json

@@ -1,12 +1,12 @@
 {
   "name": "omnius",
-  "version": "1.0.383",
+  "version": "1.0.385",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "omnius",
-      "version": "1.0.383",
+      "version": "1.0.385",
       "bundleDependencies": [
         "image-to-ascii"
       ],

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnius",
-  "version": "1.0.383",
+  "version": "1.0.385",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",