@thanh01.pmt/curriculum-kit 1.4.38 → 1.4.40

package/dist/ai/index.cjs

@@ -2273,6 +2273,111 @@ Please return a single JSON object matching these exact keys:
 - "actionableRepairPrompts": string[]
 `.trim();
 }
+function buildCurriculumPlanJudgePrompt(input) {
+  const { planJson, briefingContext, expectedLanguage = "Vietnamese", linterFindings = [] } = input;
+  return `
+# System Methodology: Principal Curriculum SME & Macro-Pedagogical Quality Judge (@sme_judge)
+
+You are the **Lead Curriculum Architect & Classroom Practicum SME**. Your role is to rigorously evaluate a candidate \`CURRICULUM_PLAN\` (Schema v3) against empirical pedagogical science, cognitive load capacity, and real classroom delivery constraints.
+
+## COURSE BRIEFING & TARGET CONSTRAINTS
+${briefingContext}
+
+${linterFindings.length > 0 ? `## DETERMINISTIC PRE-LINT FINDINGS (0-TOKEN LINTER)
+The deterministic structural linter detected the following issues in advance:
+${linterFindings.map((f) => `- [${f.severity}] ${f.id}: ${f.title} \u2014 ${f.description}`).join("\n")}
+Consider these deterministic findings when scoring and issuing actionable repair prescriptions.` : ""}
+
+## CANDIDATE CURRICULUM PLAN JSON
+\`\`\`json
+${planJson}
+\`\`\`
+
+## EVALUATION RUBRICS & 5 CRITICAL DIMENSIONS (TOTAL 100 POINTS)
+
+1. **Step Incrementality & Single Leap Principle (20 pts):**
+   - 20 pts: Every session introduces at most ONE new syntax or algorithmic concept. No massive cognitive leaps (e.g. going from basic variables directly to asynchronous networking).
+   - 10 pts: Minor leap detected in 1 session that can be mitigated with starter code.
+   - 0 pts: Severe cognitive gap (e.g. multiple compound paradigms combined in one session).
+
+2. **Vygotsky ZPD Saturation & Cognitive Capacity (20 pts):**
+   - 20 pts: New concept count <= 2 per session (K-12) / 3 (Adults), with explicit prior concept anchor or recap slot.
+   - 10 pts: Session introduces 3 concepts without adequate bridge.
+   - 0 pts: Severe overload (>3 new concepts in a single session) or unanchored concepts.
+
+3. **Classroom Friction-Adjusted Time Budget (20 pts):**
+   - 20 pts: Incorporates 2.5x student friction multiplier. Total content minutes <= 85% of session duration (15% overhead reserved). No step > 76 minutes.
+   - 10 pts: Content is slightly packed, leaving < 10% buffer time.
+   - 0 pts: Overpacked content that will cause students to run out of time in a standard classroom.
+
+4. **Reigeluth Walking Skeleton & Milestone Agency (20 pts):**
+   - 20 pts: Unit 1 culminates in an end-to-end runnable MVP (Epitome). Subsequent units provide modular elaborations.
+   - 10 pts: Walking skeleton present but delayed to Unit 2.
+   - 0 pts: Pure theoretical silo teaching where no working product appears until the end of the course.
+
+5. **Exit Evidence Specificity (20 pts):**
+   - 20 pts: Every session produces an explicit, tangible, user-visible deliverable (e.g. running UI screen, passing test suite, physical LED sequence).
+   - 10 pts: Some sessions have generic deliverables ("completed practice exercise").
+   - 0 pts: Vague or absent deliverables ("student understands topic").
+
+## DECISION RULES
+- **PASS & AUTO-APPROVE:** Total Score >= 85 AND overallVerdict is "PASS" AND no CRITICAL actionable repairs.
+- **NEEDS_REVISION:** Total Score between 70 and 84, or fixable flaws. Prescribe concrete structured \`actionableRepairs\` (\`SPLIT_STEP\`, \`INSERT_RECAP\`, \`ADJUST_TIME_BUDGET\`, \`REVISE_EXIT_EVIDENCE\`).
+- **FAIL:** Total Score < 70, structural collapse, or unbridgeable pedagogical gaps.
+
+## OUTPUT FORMAT
+Return a valid JSON object matching the MacroPedagogyPlanAuditSchema.
+`.trim();
+}
+function buildProjectGraphJudgePrompt(input) {
+  const { bundleJson, briefingContext, linterFindings = [] } = input;
+  return `
+# System Methodology: Lead Systems Engineer & Lab Infrastructure Auditor (@infra_judge)
+
+You are the **Lead Lab Systems Engineer**. Your role is to rigorously evaluate a candidate \`PROJECT_GRAPH\` and its associated \`technology_scope\` against real-world lab environment feasibility, toolchain friction, and hardware constraints.
+
+## COURSE BRIEFING & INFRASTRUCTURE CONSTRAINTS
+${briefingContext}
+
+${linterFindings.length > 0 ? `## DETERMINISTIC PRE-LINT FINDINGS
+${linterFindings.map((f) => `- [${f.severity}] ${f.id}: ${f.title} \u2014 ${f.description}`).join("\n")}` : ""}
+
+## CANDIDATE PROJECT GRAPH BUNDLE JSON
+\`\`\`json
+${bundleJson}
+\`\`\`
+
+## EVALUATION RUBRICS & 4 DIMENSIONS (TOTAL 100 POINTS)
+
+1. **Toolchain & Installation Feasibility (25 pts):**
+   - 25 pts: Toolchain installs cleanly without complex admin rights, OS friction, or fragile build scripts.
+   - 10 pts: Toolchain requires moderate admin setup, suitable for guided lab but risky for BYOD.
+   - 0 pts: High-friction setup that frequently breaks in student environments.
+
+2. **Environment Provisioning Strategy (25 pts):**
+   - 25 pts: Heavy technologies (PostgreSQL, Docker, ROS2, etc.) declare an explicit, realistic provisioning model (\`PRE_INSTALLED_LAB\`, \`CLOUD_MANAGED\`, or \`DEDICATED_SETUP\`). Core course topics are NEVER downgraded; instead, proper infrastructure is specified.
+   - 10 pts: Heavy tech used with ambiguous provisioning plan.
+   - 0 pts: Heavy tech dumped on students with no provisioning strategy, causing setup collapse.
+
+3. **Hardware & Runtime Envelope (25 pts):**
+   - 25 pts: Code, libraries, and algorithms fit comfortably within target device RAM, Flash, and pinout budget (<= 75% envelope).
+   - 10 pts: Memory or resource limits are tight.
+   - 0 pts: Guaranteed hardware crash or pinout conflict.
+
+4. **Time to Hello World (25 pts):**
+   - 25 pts: First working runnable milestone achieved in <= 15 minutes of student effort.
+   - 10 pts: First milestone takes 20-30 minutes.
+   - 0 pts: Student spends the entire first session on environment configuration without seeing running code.
+
+## DECISION RULES
+- **PASS:** Score >= 85 AND overallVerdict is "PASS".
+- **NEEDS_REVISION:** Score between 70 and 84. Prescribe structured actionable repairs.
+- **FAIL:** Score < 70 or critical environment blocker.
+
+## OUTPUT FORMAT
+Return a valid JSON object matching ProjectGraphAuditSchema.
+`.trim();
+}
 
 // src/ai/prompts/projectInstructionPrompt.ts
 function buildProjectInstructionPrompt(input) {
@@ -3084,11 +3189,51 @@ var ExtensionSchema = zod.z.object({
 });
 var QualityAuditVerdictSchema = zod.z.enum(["PASS", "NEEDS_REVISION", "FAIL"]);
 var JudgeCriterionSchema = zod.z.object({
-  name: zod.z.string().describe("Criterion name, e.g. Bloom Alignment, Code Validity, Word Count Gate, Language Adherence"),
+  name: zod.z.string().describe("Criterion name, e.g. Bloom Alignment, Code Validity, Word Count Gate, Language Adherence, Step Incrementality"),
   passed: zod.z.boolean(),
   score: zod.z.number().min(0).max(100).describe("Score out of 100"),
   feedback: zod.z.string().describe("Constructive feedback or citation of issues")
 });
+var ActionableRepairActionSchema = zod.z.enum([
+  "SPLIT_STEP",
+  "REVISE_PROVISIONING",
+  "INSERT_RECAP",
+  "ADJUST_TIME_BUDGET",
+  "SUBSTITUTE_FALLBACK",
+  "REVISE_EXIT_EVIDENCE"
+]);
+var ActionableRepairPromptSchema = zod.z.object({
+  action: ActionableRepairActionSchema,
+  targetId: zod.z.string().describe("ID of the session, feature, or step requiring repair, e.g. U01_M01_L02 or F1-S3"),
+  severity: zod.z.enum(["CRITICAL", "WARNING", "INFO"]).default("WARNING"),
+  rationale: zod.z.string().describe("Pedagogical or technical rationale for why this repair is mandated"),
+  instruction: zod.z.string().describe("Exact, actionable text instruction for the planner/generator in the next turn")
+});
+var ProjectGraphAuditSchema = zod.z.object({
+  schema_version: zod.z.number().default(1),
+  projectId: zod.z.string(),
+  overallVerdict: QualityAuditVerdictSchema,
+  totalScore: zod.z.number().min(0).max(100),
+  toolchainFeasibility: JudgeCriterionSchema.describe("Installation friction, admin rights, IDE/compiler matrix"),
+  provisioningStrategy: JudgeCriterionSchema.describe("Clear model for heavy runtimes: Pre-installed lab, Cloud instance, or Dedicated setup"),
+  hardwareRuntimeEnvelope: JudgeCriterionSchema.describe("Memory/Flash limits, pinout conflicts, hardware platform alignment"),
+  timeToHelloWorld: JudgeCriterionSchema.describe("Student time to achieve first working feedback loop (target <= 15 mins)"),
+  actionableRepairs: zod.z.array(ActionableRepairPromptSchema).default([])
+});
+var MacroPedagogyPlanAuditSchema = zod.z.object({
+  schema_version: zod.z.number().default(1),
+  planHash: zod.z.string(),
+  overallVerdict: QualityAuditVerdictSchema,
+  totalScore: zod.z.number().min(0).max(100),
+  autoApproved: zod.z.boolean().default(false).describe("True if totalScore >= 85 and overallVerdict is PASS"),
+  stepIncrementality: JudgeCriterionSchema.describe("Single leap principle: <= 1 new syntax or 1 new algorithm per step"),
+  zpdSaturation: JudgeCriterionSchema.describe("Vygotsky ZPD: <= 2 new concepts per session (K-12) / 3 (Adults), with prior anchor"),
+  frictionAdjustedBudget: JudgeCriterionSchema.describe("Student time multiplier (2.5x): total content <= 85% session minutes"),
+  epitomeViability: JudgeCriterionSchema.describe("Reigeluth Walking Skeleton: Unit 1 produces an end-to-end runnable MVP"),
+  exitEvidenceSpecificity: JudgeCriterionSchema.describe("Tangible, measurable student deliverable for every session"),
+  actionableRepairs: zod.z.array(ActionableRepairPromptSchema).default([]),
+  summary: zod.z.string().describe("High-level executive evaluation summary")
+});
 var CurriculumQualityReportSchema = zod.z.object({
   targetArtifactType: zod.z.string().describe("Type of artifact audited, e.g. LESSON, ACT, CODE, WKS, SLIDE"),
   lessonId: zod.z.string(),
@@ -3096,7 +3241,8 @@ var CurriculumQualityReportSchema = zod.z.object({
   totalScore: zod.z.number().min(0).max(100),
   languageAdherencePassed: zod.z.boolean().describe("True if content strictly matches the requested target language"),
   criteria: zod.z.array(JudgeCriterionSchema).min(3),
-  actionableRepairPrompts: zod.z.array(zod.z.string()).default([]).describe("Concrete instruction prompts to fix detected issues in the next repair turn")
+  actionableRepairPrompts: zod.z.array(zod.z.string()).default([]).describe("Concrete instruction prompts to fix detected issues in the next repair turn"),
+  structuredRepairs: zod.z.array(ActionableRepairPromptSchema).optional()
 });
 var ProjectProfileSchema = zod.z.object({
   id: zod.z.string().optional(),
@@ -4256,6 +4402,11 @@ async function generateExtensionFlow(input) {
   });
   return object;
 }
+
+// src/constants/pedagogy.ts
+var JUDGE_AUTO_APPROVE_THRESHOLD = 85;
+
+// src/ai/flows/auditCurriculumQualityFlow.ts
 async function auditCurriculumQualityFlow(input) {
   const model = getAIModel(input.modelOptions);
   const prompt = buildJudgePrompt({
@@ -4271,6 +4422,43 @@ async function auditCurriculumQualityFlow(input) {
     schema: CurriculumQualityReportSchema,
     prompt: `${prompt}
 
+Respond with a valid JSON object matching the schema.`
+  });
+  return object;
+}
+async function auditCurriculumPlanFlow(input) {
+  const model = getAIModel(input.modelOptions);
+  const prompt = buildCurriculumPlanJudgePrompt({
+    planJson: input.planJson,
+    briefingContext: input.briefingContext,
+    expectedLanguage: input.expectedLanguage,
+    linterFindings: input.linterFindings
+  });
+  const { object } = await ai.generateObject({
+    model,
+    schema: MacroPedagogyPlanAuditSchema,
+    prompt: `${prompt}
+
+Respond with a valid JSON object matching the schema.`
+  });
+  const autoApproved = object.totalScore >= JUDGE_AUTO_APPROVE_THRESHOLD && object.overallVerdict === "PASS";
+  return {
+    ...object,
+    autoApproved
+  };
+}
+async function auditProjectGraphFlow(input) {
+  const model = getAIModel(input.modelOptions);
+  const prompt = buildProjectGraphJudgePrompt({
+    bundleJson: input.bundleJson,
+    briefingContext: input.briefingContext,
+    linterFindings: input.linterFindings
+  });
+  const { object } = await ai.generateObject({
+    model,
+    schema: ProjectGraphAuditSchema,
+    prompt: `${prompt}
+
 Respond with a valid JSON object matching the schema.`
   });
   return object;
@@ -5385,9 +5573,12 @@ exports.SLIDE_CANONICAL_METHODOLOGY = SLIDE_CANONICAL_METHODOLOGY;
 exports.activityTools = activityTools;
 exports.analystTools = analystTools;
 exports.assessorTools = assessorTools;
+exports.auditCurriculumPlanFlow = auditCurriculumPlanFlow;
 exports.auditCurriculumQualityFlow = auditCurriculumQualityFlow;
+exports.auditProjectGraphFlow = auditProjectGraphFlow;
 exports.buildActivityPrompt = buildActivityPrompt;
 exports.buildCodeLabPrompt = buildCodeLabPrompt;
+exports.buildCurriculumPlanJudgePrompt = buildCurriculumPlanJudgePrompt;
 exports.buildDiagnosticQuizPrompt = buildDiagnosticQuizPrompt;
 exports.buildExtensionPrompt = buildExtensionPrompt;
 exports.buildHandoutPrompt = buildHandoutPrompt;
@@ -5395,6 +5586,7 @@ exports.buildHtmlDeckSlidePrompt = buildHtmlDeckSlidePrompt;
 exports.buildJudgePrompt = buildJudgePrompt;
 exports.buildLessonMasterPrompt = buildLessonMasterPrompt;
 exports.buildMarpMarkdownSlidePrompt = buildMarpMarkdownSlidePrompt;
+exports.buildProjectGraphJudgePrompt = buildProjectGraphJudgePrompt;
 exports.buildProjectInstructionPrompt = buildProjectInstructionPrompt;
 exports.buildSelfLabPrompt = buildSelfLabPrompt;
 exports.buildSlidesPrompt = buildSlidesPrompt;