@thanh01.pmt/curriculum-kit 1.0.6 → 1.0.8

package/dist/index.cjs

@@ -14,35 +14,16 @@ var supabaseJs = require('@supabase/supabase-js');
 var url = require('url');
 var jsonrepair = require('jsonrepair');
 var rest = require('@octokit/rest');
-var workflowPkg = require('workflow');
+var workflow = require('workflow');
 
 var _documentCurrentScript = typeof document !== 'undefined' ? document.currentScript : null;
 function _interopDefault (e) { return e && e.__esModule ? e : { default: e }; }
 
-function _interopNamespace(e) {
-  if (e && e.__esModule) return e;
-  var n = Object.create(null);
-  if (e) {
-    Object.keys(e).forEach(function (k) {
-      if (k !== 'default') {
-        var d = Object.getOwnPropertyDescriptor(e, k);
-        Object.defineProperty(n, k, d.get ? d : {
-          enumerable: true,
-          get: function () { return e[k]; }
-        });
-      }
-    });
-  }
-  n.default = e;
-  return Object.freeze(n);
-}
-
 var fs2__default = /*#__PURE__*/_interopDefault(fs2);
 var path3__default = /*#__PURE__*/_interopDefault(path3);
 var pLimit__default = /*#__PURE__*/_interopDefault(pLimit);
 var fsPromises__default = /*#__PURE__*/_interopDefault(fsPromises);
 var crypto__default = /*#__PURE__*/_interopDefault(crypto);
-var workflowPkg__namespace = /*#__PURE__*/_interopNamespace(workflowPkg);
 
 var __defProp = Object.defineProperty;
 var __getOwnPropNames = Object.getOwnPropertyNames;
@@ -9576,14 +9557,14 @@ init_errors();
 
 // src/workflow/gateSettings.ts
 var DEFAULT_GATE_SETTINGS = Object.freeze({
-  LEARNER_PROFILE: "HITL",
-  PROJECT_BRIEF: "HITL",
-  REFERENCE_PACK: "HITL",
-  CURRICULUM_FRAMEWORK: "HITL",
-  KNOWLEDGE_EXPOSITION: "LLM_JUDGE",
-  STYLE_GUIDE: "HITL",
-  ART_DIRECTION: "HITL",
-  LESSON: "LLM_JUDGE",
+  LEARNER_PROFILE: "OFF",
+  PROJECT_BRIEF: "OFF",
+  REFERENCE_PACK: "OFF",
+  CURRICULUM_FRAMEWORK: "OFF",
+  KNOWLEDGE_EXPOSITION: "OFF",
+  STYLE_GUIDE: "OFF",
+  ART_DIRECTION: "OFF",
+  LESSON: "OFF",
   ACT: "OFF",
   QUIZ: "OFF",
   SLIDE: "OFF",
@@ -9591,7 +9572,7 @@ var DEFAULT_GATE_SETTINGS = Object.freeze({
   WKS: "OFF",
   HANDOUT: "OFF",
   EXT: "OFF",
-  CODE: "LLM_JUDGE",
+  CODE: "OFF",
   MEDIA_SCRIPT: "OFF",
   EXIT_TICKET: "OFF"
 });
@@ -11480,9 +11461,10 @@ Fix ALL issues above and output the complete corrected document.` }],
   const taskState = pipelineState.tasks[lessonCode] || {};
   const currentLessonInfo = taskState.LESSON;
   const lessonGate = gateModeFor(gates, "LESSON");
-  let isLessonApproved = lessonGate !== "HITL" || // OFF/LLM_JUDGE: không cần duyệt người
-  currentLessonInfo?.state === "approved" || currentLessonInfo?.review?.decision === "APPROVED" || approvedArtifacts.includes("LESSON") || approvedArtifacts.includes(lessonCode);
-  if (lessonGate === "LLM_JUDGE" && !isLessonApproved && lessonContent) {
+  const humanApproved = currentLessonInfo?.state === "approved" || currentLessonInfo?.review?.decision === "APPROVED" || approvedArtifacts.includes("LESSON") || approvedArtifacts.includes(lessonCode);
+  const needsJudge = lessonGate === "LLM_JUDGE" && !humanApproved;
+  let isLessonApproved = lessonGate === "OFF" || humanApproved;
+  if (needsJudge && lessonContent) {
     onProgress?.("@reviewer", `\u{1F916} Executing LLM-as-Judge academic evaluation on LESSON_${lessonCode}.md...`);
     try {
       let injectYamlReviewMetadata2 = function(content, meta) {
@@ -12521,6 +12503,8 @@ async function generateSingleArtifact(req) {
     extraContext,
     config,
     contextSot = {},
+    mediaPolicy,
+    researchContext,
     options = {},
     onChunk
   } = req;
@@ -12558,7 +12542,10 @@ ${langRule}
 3. PEDAGOGICAL GROUNDING: Strictly align with the ${pedagogy.toUpperCase()} model, Bloom's Revised Taxonomy, and IEEE/ACM CS2023 guidelines.
 4. SCAFFOLDING LEVEL: "${scaffoldingLevel.toUpperCase()}".${scaffoldingRule}
 5. TARGET AUDIENCE: ${gradeLevel ? `Grade ${gradeLevel}, ` : ""}${targetAge} on ${hwStr}.${deviceRule}${classDynamicsRule}${extraContextRule}
-6. OUTPUT INTEGRITY: Output clean Markdown starting directly with YAML frontmatter. Do not wrap entire response in triple backticks.${artifactSpecializedPrompt}`;
+6. OUTPUT INTEGRITY: Output clean Markdown starting directly with YAML frontmatter. Do not wrap entire response in triple backticks.${artifactSpecializedPrompt}${mediaPolicy ? `
+
+IMAGE PLANNING (media ledger) \u2014 QUOTA GEN: t\u1ED1i \u0111a ${mediaPolicy.maxGenImagesPerArtifact} \u1EA3nh AI cho artifact n\xE0y (\u1EA3nh search kho kh\xF4ng gi\u1EDBi h\u1EA1n).
+Khi n\u1ED9i dung c\u1EA7n minh h\u1ECDa (diagram quy tr\xECnh, s\u01A1 \u0111\u1ED3 kh\xE1i ni\u1EC7m, step-by-step visual), ch\xE8n placeholder \u0111\xFAng format [IMAGE: slug] (slug ch\u1EEF th\u01B0\u1EDDng-g\u1EA1ch ngang, unique trong b\xE0i, vd img-for-loop-diagram) T\u1EA0I \u0110\xDANG CH\u1ED6 c\u1EA7n \u1EA3nh, tr\xEAn d\xF2ng ri\xEAng. KH\xD4NG t\u1EF1 sinh \u1EA3nh, KH\xD4NG d\xF9ng markdown image \u2014 placeholder s\u1EBD \u0111\u01B0\u1EE3c thay b\u1EB1ng \u1EA3nh th\u1EADt sau khi Media Curator duy\u1EC7t. Ch\u1EC9 \u0111\u1EB7t cho \u1EA3nh TH\u1EF0C S\u1EF0 c\u1EA7n thi\u1EBFt.` : ""}`;
   const userPrompt = `### CONTEXT GROUND TRUTH:
 - Topic / Concept Domain: "${topic}"
 - Lesson ID: "${lessonId || "ON_DEMAND"}"
@@ -12573,6 +12560,9 @@ ${selectedDevices ? `- Device Mode: "${selectedDevices}"
 ` : ""}${extraContext ? `- User Specific Guidance & Requirements: "${extraContext}"
 ` : ""}${config ? `- Full Task Config: ${JSON.stringify(config)}
 ` : ""}
+${researchContext ? `### RESEARCH GROUND TRUTH (fact-checked context \u2014 \u01B0u ti\xEAn d\xF9ng, KH\xD4NG b\u1ECBa s\u1ED1 li\u1EC7u):
+${researchContext.slice(0, 4e3)}
+` : ""}
 ${contextSot.framework ? `### SOT CURRICULUM FRAMEWORK:
 ${contextSot.framework.slice(0, 1500)}
 ` : ""}
@@ -12613,13 +12603,171 @@ Generate the complete, thorough, publication-grade "${artifactType.toUpperCase()
       }
     );
     const { thought, content } = extractThoughtAndContent(rawResult);
-    return {
-      artifactType,
-      filename: meta.filename,
-      content: content || accumulatedContent,
-      persona: meta.persona,
-      thought: thought || accumulatedThought
+    const finalContent = content || accumulatedContent;
+    const finalThought = thought || accumulatedThought;
+    const gates = resolveGateSettings(req.gateSettings);
+    const kitGateType = {
+      lesson: "LESSON",
+      slide: "SLIDE",
+      act: "ACT",
+      quiz: "QUIZ",
+      guide: "GUIDE",
+      handout: "HANDOUT",
+      worksheet: "WKS",
+      ext: "EXT",
+      codelab: "CODE",
+      exit_ticket: "EXIT_TICKET"
+    }[artifactType];
+    const gateMode = kitGateType ? gateModeFor(gates, kitGateType) : "OFF";
+    if (gateMode === "OFF") {
+      return { artifactType, filename: meta.filename, content: finalContent, persona: meta.persona, thought: finalThought };
+    }
+    const injectReviewMeta = (doc, metaMap) => {
+      const fm = doc.match(/^---\s*\r?\n([\s\S]*?)\r?\n---(?:\r?\n|$)/);
+      if (!fm) {
+        return `---
+${Object.entries(metaMap).map(([k, v]) => `${k}: "${v}"`).join("\n")}
+---
+
+${doc}`;
+      }
+      let yaml = fm[1] ?? "";
+      for (const [k, v] of Object.entries(metaMap)) {
+        const re = new RegExp(`^${k}:.*$`, "m");
+        if (re.test(yaml)) yaml = yaml.replace(re, `${k}: "${v}"`);
+        else yaml += `
+${k}: "${v}"`;
+      }
+      return doc.replace(fm[0], `---
+${yaml.trim()}
+---
+`);
     };
+    if (gateMode === "HITL") {
+      return {
+        artifactType,
+        filename: meta.filename,
+        content: injectReviewMeta(finalContent, { review_status: "PENDING_REVIEW", review_decision: "PENDING" }),
+        persona: meta.persona,
+        thought: finalThought,
+        pausedForReview: true,
+        gateStatus: "awaiting_approval"
+      };
+    }
+    const judgeObjectives = [{ code: "A1", description: `Artifact fulfills its pedagogical contract for topic "${topic}" with complete required sections`, bloomLevel: "Analyze" }];
+    try {
+      const verdict = await LLMJudgeEngine.evaluateArtifactWithLLM({
+        targetArtifactType: artifactType.toUpperCase(),
+        lessonId: lessonId || "ON_DEMAND",
+        expectedLanguage: language.toLowerCase().includes("viet") ? "vietnamese" : "english",
+        targetObjectives: judgeObjectives,
+        generatedContentJson: JSON.stringify({ content: finalContent, artifactType, lessonId: lessonId || "ON_DEMAND", title: displayTitle })
+      });
+      const passed = Boolean(verdict.passed) && verdict.score >= 85;
+      if (passed) {
+        return {
+          artifactType,
+          filename: meta.filename,
+          content: injectReviewMeta(finalContent, {
+            review_status: "APPROVED",
+            reviewed_by: "@reviewer (LLM-as-Judge)",
+            review_decision: "APPROVED"
+          }),
+          persona: meta.persona,
+          thought: finalThought,
+          judgeScore: verdict.score,
+          judgeCritique: verdict.critique
+        };
+      }
+      const repairFeedback = [verdict.autoRepairInstructions, ...verdict.weaknesses || []].filter(Boolean).join("\n");
+      const repairedRaw = await streamCurriculumAIInference(
+        [
+          { role: "user", content: `You previously authored \`${meta.filename}\` and the academic Judge REJECTED it with the critique below.
+Rewrite the COMPLETE corrected document: fix every listed issue, keep the mandatory structure, frontmatter and all approved sections. Output 100% clean Markdown only.
+
+JUDGE CRITIQUE:
+${repairFeedback}
+
+CURRENT DOCUMENT:
+${finalContent}` }
+        ],
+        userPrompt,
+        { ...options, systemPersona: systemPrompt },
+        (chunk, type) => {
+          if (onChunk) onChunk(chunk, type === "thought" ? "thought" : "content");
+        }
+      );
+      const repairedExtract = extractThoughtAndContent(repairedRaw);
+      const repaired = repairedExtract.content || repairedRaw;
+      if (repaired && repaired.trim().length > 100) {
+        const reVerdict = await LLMJudgeEngine.evaluateArtifactWithLLM({
+          targetArtifactType: artifactType.toUpperCase(),
+          lessonId: lessonId || "ON_DEMAND",
+          expectedLanguage: language.toLowerCase().includes("viet") ? "vietnamese" : "english",
+          targetObjectives: judgeObjectives,
+          generatedContentJson: JSON.stringify({ content: repaired, artifactType, lessonId: lessonId || "ON_DEMAND", title: displayTitle })
+        });
+        if (reVerdict.passed && reVerdict.score >= 85) {
+          return {
+            artifactType,
+            filename: meta.filename,
+            content: injectReviewMeta(repaired, {
+              review_status: "APPROVED",
+              reviewed_by: "@reviewer (LLM-as-Judge)",
+              review_decision: "APPROVED"
+            }),
+            persona: meta.persona,
+            thought: repairedExtract.thought || "",
+            judgeScore: reVerdict.score,
+            judgeCritique: reVerdict.critique
+          };
+        }
+        return {
+          artifactType,
+          filename: meta.filename,
+          content: injectReviewMeta(repaired, {
+            review_status: "NEEDS_REVISION",
+            reviewed_by: "@reviewer (LLM-as-Judge)",
+            review_decision: "NEEDS_REVISION"
+          }),
+          persona: meta.persona,
+          thought: repairedExtract.thought || "",
+          pausedForReview: true,
+          gateStatus: "judge_needs_revision",
+          judgeScore: reVerdict.score,
+          judgeCritique: reVerdict.critique
+        };
+      }
+      return {
+        artifactType,
+        filename: meta.filename,
+        content: injectReviewMeta(finalContent, {
+          review_status: "NEEDS_REVISION",
+          reviewed_by: "@reviewer (LLM-as-Judge)",
+          review_decision: "NEEDS_REVISION"
+        }),
+        persona: meta.persona,
+        thought: finalThought,
+        pausedForReview: true,
+        gateStatus: "judge_needs_revision",
+        judgeScore: verdict.score,
+        judgeCritique: verdict.critique
+      };
+    } catch (judgeErr) {
+      return {
+        artifactType,
+        filename: meta.filename,
+        content: injectReviewMeta(finalContent, {
+          review_status: "NEEDS_REVISION",
+          review_decision: "NEEDS_REVISION",
+          critique: `Judge unavailable (fail-closed): ${String(judgeErr?.message || judgeErr).slice(0, 300)}`
+        }),
+        persona: meta.persona,
+        thought: finalThought,
+        pausedForReview: true,
+        gateStatus: "judge_error"
+      };
+    }
   } catch (error) {
     throw createAiInferenceError({
       errorCode: "ERR_AI_ALL_PROVIDERS_FAILED",
@@ -13375,43 +13523,15 @@ function safeParseJson(rawText) {
     return null;
   }
 }
-async function readSseStream(reader, decoder, onData) {
-  let buffer = "";
-  while (true) {
-    const { done, value } = await reader.read();
-    if (done) break;
-    buffer += decoder.decode(value, { stream: true });
-    const lines = buffer.split("\n");
-    buffer = lines.pop() || "";
-    for (const line of lines) {
-      const trimmed = line.trim();
-      if (!trimmed || trimmed.startsWith(":")) continue;
-      if (trimmed === "data: [DONE]") return;
-      if (trimmed.startsWith("data: ")) {
-        const dataStr = trimmed.slice(6).trim();
-        if (dataStr) onData(dataStr);
-      }
-    }
-  }
-}
 async function streamLLMWithFallback(systemPrompt, userPrompt, apiKeys = {}, onChunk) {
-  apiKeys.alibaba || apiKeys.dashscope || process.env.ALIBABA_API_KEY || process.env.DASHSCOPE_API_KEY;
+  const alibabaKey = apiKeys.alibaba || apiKeys.dashscope || process.env.ALIBABA_API_KEY || process.env.DASHSCOPE_API_KEY;
   const nvidiaKey = apiKeys.nvidia || process.env.NVIDIA_API_KEY;
-  apiKeys.deepseek || process.env.DEEPSEEK_API_KEY;
+  const deepseekKey = apiKeys.deepseek || process.env.DEEPSEEK_API_KEY;
   const openrouterKey = apiKeys.openrouter || process.env.OPENROUTER_API_KEY;
-  apiKeys.gemini || process.env.GEMINI_API_KEY;
+  const geminiKey = apiKeys.gemini || process.env.GEMINI_API_KEY;
   const openrouterPreset = process.env.OPENROUTER_FREE_PRESET || "@preset/coding-free";
-  const messages = [
-    {
-      role: "system",
-      content: `${systemPrompt}
-
-THINKING DIRECTIVE: Keep reasoning concise, structured, and focused strictly on the core pedagogical trade-offs in 4-6 concise bullet points (under 120 words). Then output the JSON immediately.`
-    },
-    { role: "user", content: userPrompt }
-  ];
+  const candidates = [];
   if (openrouterKey && isProviderEnabled("openrouter")) {
-    const t0 = Date.now();
     const orModels = [
       openrouterPreset,
       "openrouter/free",
@@ -13421,174 +13541,82 @@ THINKING DIRECTIVE: Keep reasoning concise, structured, and focused strictly on
       "minimax/minimax-m3:free",
       "cohere/north-mini-code:free"
     ].filter((m) => isModelAllowed(m));
-    const uniqueOrModels = Array.from(new Set(orModels));
-    for (const model of uniqueOrModels) {
-      try {
-        const controller = new AbortController();
-        const timeout = setTimeout(() => controller.abort(), 25e3);
-        const res = await fetch("https://openrouter.ai/api/v1/chat/completions", {
-          method: "POST",
-          headers: {
-            "Content-Type": "application/json",
-            "Authorization": `Bearer ${openrouterKey}`
-          },
-          body: JSON.stringify({
-            model,
-            messages,
-            stream: true,
-            temperature: 0.3
-          }),
-          signal: controller.signal
-        });
-        clearTimeout(timeout);
-        if (res.ok && res.body) {
-          const reader = res.body.getReader();
-          const decoder = new TextDecoder();
-          let fullContent = "";
-          await readSseStream(reader, decoder, (dataStr) => {
-            try {
-              const parsed = JSON.parse(dataStr);
-              const delta = parsed.choices?.[0]?.delta;
-              if (delta?.content) {
-                fullContent += delta.content;
-                onChunk?.(delta.content, "content");
-              }
-            } catch {
-            }
-          });
-          if (fullContent.trim()) {
-            console.log(`[curriculum-kit:LayerPrefill] \u2705 Tier 1 OpenRouter (${model}) streamed in ${Date.now() - t0}ms`);
-            return fullContent.trim();
-          }
-        }
-      } catch (e) {
-        console.warn(`[curriculum-kit:LayerPrefill] Tier 1 OpenRouter stream failed for ${model} in ${Date.now() - t0}ms:`, e?.message || e);
-      }
+    for (const m of Array.from(new Set(orModels))) {
+      candidates.push({ provider: "openrouter", model: m, apiKey: openrouterKey });
     }
   }
   if (nvidiaKey && isProviderEnabled("nvidia")) {
-    const t0 = Date.now();
     const nemotronModels = [
       "nvidia/nemotron-3-ultra-550b-a55b",
       "nvidia/nemotron-3.5-lightning-30b-a3b",
       "nvidia/nemotron-3-super-120b-a12b",
       "nvidia/llama-3.1-nemotron-70b-instruct"
     ].filter((m) => isModelAllowed(m));
-    for (const model of nemotronModels) {
-      try {
-        const controller = new AbortController();
-        const timeout = setTimeout(() => controller.abort(), 25e3);
-        const res = await fetch("https://integrate.api.nvidia.com/v1/chat/completions", {
-          method: "POST",
-          headers: {
-            "Content-Type": "application/json",
-            "Authorization": `Bearer ${nvidiaKey}`
-          },
-          body: JSON.stringify({
-            model,
-            messages,
-            stream: true,
-            temperature: 0.2
-          }),
-          signal: controller.signal
-        });
-        clearTimeout(timeout);
-        if (res.ok && res.body) {
-          const reader = res.body.getReader();
-          const decoder = new TextDecoder();
-          let fullContent = "";
-          await readSseStream(reader, decoder, (dataStr) => {
-            try {
-              const parsed = JSON.parse(dataStr);
-              const delta = parsed.choices?.[0]?.delta;
-              if (delta?.content) {
-                fullContent += delta.content;
-                onChunk?.(delta.content, "content");
-              }
-            } catch {
-            }
-          });
-          if (fullContent.trim()) {
-            console.log(`[curriculum-kit:LayerPrefill] \u2705 Tier 2 NVIDIA NIM (${model}) streamed in ${Date.now() - t0}ms`);
-            return fullContent.trim();
-          }
-        }
-      } catch (e) {
-        console.warn(`[curriculum-kit:LayerPrefill] Tier 2 NVIDIA NIM stream failed for ${model} in ${Date.now() - t0}ms:`, e?.message || e);
-      }
+    for (const m of nemotronModels) {
+      candidates.push({ provider: "nvidia", model: m, apiKey: nvidiaKey });
+    }
+  }
+  if (alibabaKey && isProviderEnabled("alibaba")) {
+    const aliModels = ["qwen-plus", "qwen-turbo", "qwen3.7-plus"].filter((m) => isModelAllowed(m));
+    for (const m of aliModels) {
+      candidates.push({ provider: "alibaba", model: m, apiKey: alibabaKey });
+    }
+  }
+  if (deepseekKey && isProviderEnabled("deepseek")) {
+    const dsModels = ["deepseek-chat", "deepseek-coder"].filter((m) => isModelAllowed(m));
+    for (const m of dsModels) {
+      candidates.push({ provider: "deepseek", model: m, apiKey: deepseekKey });
+    }
+  }
+  if (geminiKey && (isProviderEnabled("gemini") || isProviderEnabled("google"))) {
+    const gModels = ["gemini-1.5-flash", "gemini-2.0-flash", "gemini-1.5-pro"].filter((m) => isModelAllowed(m));
+    for (const m of gModels) {
+      candidates.push({ provider: "google", model: m, apiKey: geminiKey });
     }
   }
   const fallbackChain = getDesignatedFallbackChain();
   for (const target of fallbackChain) {
-    const { provider, model } = target;
-    if (provider === "nvidia" && nvidiaKey && isProviderEnabled("nvidia")) {
-      try {
-        const res = await fetch("https://integrate.api.nvidia.com/v1/chat/completions", {
-          method: "POST",
-          headers: {
-            "Content-Type": "application/json",
-            "Authorization": `Bearer ${nvidiaKey}`
-          },
-          body: JSON.stringify({
-            model,
-            messages,
-            stream: true,
-            temperature: 0.2
-          })
-        });
-        if (res.ok && res.body) {
-          const reader = res.body.getReader();
-          const decoder = new TextDecoder();
-          let fullContent = "";
-          await readSseStream(reader, decoder, (dataStr) => {
-            try {
-              const parsed = JSON.parse(dataStr);
-              const delta = parsed.choices?.[0]?.delta;
-              if (delta?.content) {
-                fullContent += delta.content;
-                onChunk?.(delta.content, "content");
-              }
-            } catch {
-            }
-          });
-          if (fullContent.trim()) return fullContent.trim();
+    if (!candidates.some((c) => c.provider === target.provider && c.model === target.model)) {
+      candidates.push({
+        provider: target.provider,
+        model: target.model,
+        apiKey: apiKeys[target.provider]
+      });
+    }
+  }
+  const systemInstructions = `${systemPrompt}
+
+THINKING DIRECTIVE: Keep reasoning concise, structured, and focused strictly on the core pedagogical trade-offs in 4-6 concise bullet points (under 120 words). Then output the JSON immediately.`;
+  for (const candidate of candidates) {
+    const t0 = Date.now();
+    try {
+      const modelInstance = getAIModel({
+        provider: candidate.provider,
+        modelName: candidate.model,
+        apiKey: candidate.apiKey
+      });
+      const streamResult = ai.streamText({
+        model: modelInstance,
+        system: systemInstructions,
+        prompt: userPrompt,
+        temperature: 0.2,
+        abortSignal: AbortSignal.timeout(25e3)
+      });
+      let fullContent = "";
+      for await (const part of streamResult.fullStream) {
+        if (part.type === "reasoning-delta") {
+          onChunk?.(part.text, "thought");
+        } else if (part.type === "text-delta") {
+          fullContent += part.text;
+          onChunk?.(part.text, "content");
         }
-      } catch {
       }
-    } else if (provider === "openrouter" && openrouterKey && isProviderEnabled("openrouter")) {
-      try {
-        const res = await fetch("https://openrouter.ai/api/v1/chat/completions", {
-          method: "POST",
-          headers: {
-            "Content-Type": "application/json",
-            "Authorization": `Bearer ${openrouterKey}`
-          },
-          body: JSON.stringify({
-            model,
-            messages,
-            stream: true,
-            temperature: 0.2
-          })
-        });
-        if (res.ok && res.body) {
-          const reader = res.body.getReader();
-          const decoder = new TextDecoder();
-          let fullContent = "";
-          await readSseStream(reader, decoder, (dataStr) => {
-            try {
-              const parsed = JSON.parse(dataStr);
-              const delta = parsed.choices?.[0]?.delta;
-              if (delta?.content) {
-                fullContent += delta.content;
-                onChunk?.(delta.content, "content");
-              }
-            } catch {
-            }
-          });
-          if (fullContent.trim()) return fullContent.trim();
-        }
-      } catch {
+      if (fullContent.trim()) {
+        console.log(`[curriculum-kit:VercelAI] \u2705 ${candidate.provider} (${candidate.model}) streamed in ${Date.now() - t0}ms`);
+        return fullContent.trim();
       }
+    } catch (e) {
+      console.warn(`[curriculum-kit:VercelAI] Candidate ${candidate.provider} (${candidate.model}) failed in ${Date.now() - t0}ms:`, e?.message || e);
     }
   }
   return null;
@@ -14455,7 +14483,7 @@ template_contract: "artifact-template-v1"
 async function withRateLimitBackoff(options) {
   let attempt = 1;
   try {
-    const meta = workflowPkg.getStepMetadata();
+    const meta = workflow.getStepMetadata();
     if (meta && typeof meta.attempt === "number") {
       attempt = meta.attempt;
     }
@@ -14471,7 +14499,7 @@ async function withRateLimitBackoff(options) {
     if (isRateLimited || isTransientError) {
       const retryAfterSeconds = Math.min(120, Math.pow(2, attempt) * 2);
       const reason = isRateLimited ? "Rate limit exceeded (429)" : `Transient server error (${status || "network"})`;
-      throw new workflowPkg.RetryableError(`[${options.stepName}] ${reason}. Retrying attempt ${attempt + 1}...`, {
+      throw new workflow.RetryableError(`[${options.stepName}] ${reason}. Retrying attempt ${attempt + 1}...`, {
         retryAfter: `${retryAfterSeconds}s`
       });
     }
@@ -14756,10 +14784,10 @@ var approvalPayloadSchema = zod.z.object({
   feedback: zod.z.string().optional().describe("Actionable feedback or required changes if rejected"),
   timestamp: zod.z.string().optional().describe("ISO timestamp of the approval action")
 });
-var hookFactory = workflowPkg__namespace.defineHook || workflowPkg__namespace.createHook || ((opts) => opts);
-var lessonApprovalHook = hookFactory({
-  schema: approvalPayloadSchema
-});
+var lessonApprovalHook = workflow.defineHook();
+function approvalHookToken(projectId, lessonId, artifactType) {
+  return `approval:${projectId}:${lessonId}:${artifactType}`;
+}
 
 // src/standards/standardsCoverageGate.ts
 function resolveStatementRef(ref, packs) {
@@ -15109,9 +15137,9 @@ async function generateMilestoneWorkflow(input) {
   } else if (lessonGate === "HITL") {
     const hookToken = "gate:lesson:" + milestoneId;
     console.log("  \u23F8 [Gate] LESSON awaiting HUMAN review (HITL) \u2014 hook " + hookToken);
-    const workflowPkg2 = await import('workflow');
-    const createHook2 = workflowPkg2.createHook;
-    const hook = createHook2({ token: hookToken });
+    const workflowPkg = await import('workflow');
+    const createHook = workflowPkg.createHook;
+    const hook = createHook({ token: hookToken });
     const review = await hook;
     console.log("  \u2705 [Gate] Human verdict: " + (review.approved ? "APPROVED" : "REJECTED") + (review.feedback ? " \u2014 " + review.feedback : ""));
     if (!review.approved) {
@@ -15131,7 +15159,7 @@ async function generateMilestoneWorkflow(input) {
         standardsContext: standardsContext || void 0,
         modelOptions: input.modelOptions
       });
-      const hook2 = createHook2({ token: hookToken + ":v2" });
+      const hook2 = createHook({ token: hookToken + ":v2" });
       const review2 = await hook2;
       console.log("  \u2705 [Gate] Human verdict (v2): " + (review2.approved ? "APPROVED" : "REJECTED"));
       if (!review2.approved) {
@@ -15319,7 +15347,7 @@ async function generateRoadmapWorkflow(input) {
       }
     }
     if (i + batchSize < milestones.length) {
-      await workflowPkg.sleep("2s");
+      await workflow.sleep("2s");
     }
   }
   const successfulMilestones = milestoneOutcomes.filter((m) => m.status === "FULFILLED").length;
@@ -16741,6 +16769,36 @@ var MediaLedger = class {
   doneCount() {
     return this.listByStatus("done").length;
   }
+  /**
+   * Scan artifact content và seed pending entries cho mọi placeholder [IMAGE: slug].
+   * Được gọi sau khi agent sinh artifact (trước khi Curator duyệt) — đây là cầu nối
+   * duy nhất giữa "agent chèn placeholder" và "sổ cái có entry để Curator/Fulfiller xử".
+   * Idempotent: slug đã có entry active (pending/approved/done) → giữ nguyên;
+   * chỉ re-describe khi entry trước bị rejected/failed.
+   */
+  seedFromArtifact(artifactId, content) {
+    const seeded = [];
+    const re = /\[IMAGE\s*:\s*([a-z0-9][a-z0-9-]*)\s*\]/gi;
+    const seen = /* @__PURE__ */ new Set();
+    let m;
+    while ((m = re.exec(content)) !== null) {
+      const slug = m[1].toLowerCase();
+      if (seen.has(slug)) continue;
+      seen.add(slug);
+      const existing = this.entries.get(slug);
+      const isActive = existing && (existing.status === "pending" || existing.status === "approved" || existing.status === "done");
+      if (isActive) continue;
+      seeded.push(this.upsert({
+        slug,
+        artifactId,
+        name: slug.replace(/^(img-|image-)/, "").replace(/-/g, " "),
+        description: `Placeholder ch\xE8n b\u1EDFi agent trong artifact ${artifactId} \u2014 ch\u1EDD Curator enrich (m\xF4 t\u1EA3 g\u1ED1c ch\u01B0a c\xF3).`,
+        mode: "generate",
+        status: "pending"
+      }));
+    }
+    return seeded;
+  }
 };
 
 // src/media/curator.ts
@@ -17155,6 +17213,7 @@ exports.WorksheetSchema = WorksheetSchema;
 exports.activityTools = activityTools;
 exports.analystTools = analystTools;
 exports.analyzeProjectCreationIntent = analyzeProjectCreationIntent;
+exports.approvalHookToken = approvalHookToken;
 exports.approvalPayloadSchema = approvalPayloadSchema;
 exports.assertAcyclic = assertAcyclic;
 exports.assessorTools = assessorTools;