fix(miroflow): fallback robusto para resultado do agente + modelo llama3.2:3b

- miroflow_service.py: extrai resposta de summary, final_boxed_answer,
  exceed_max_turn_summary ou último message_history do assistente
- Troca deepseek-r1:14b por llama3.2:3b nos agentes Statistician e Fiscal Auditor
- shared/schema.ts: adiciona tabelas OpenClaw (detectedPatterns, skillSuggestions)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

server/python/miroflow_service.py

@@ -25,8 +25,8 @@ OLLAMA_BASE_URL = os.getenv("OLLAMA_BASE_URL", "http://localhost:11434")
 RESEARCHER_MODEL = os.getenv("MIROFLOW_RESEARCHER_MODEL", "llama3.1:8b")
 
 AGENT_MODELS = {
-    "statistician": "deepseek-r1:14b",
+    "statistician": "llama3.2:3b",
-    "fiscal_auditor": "deepseek-r1:14b",
+    "fiscal_auditor": "llama3.2:3b",
     "researcher": RESEARCHER_MODEL,
 }
 
@@ -75,7 +75,19 @@ async def run_agent(agent_type: str, task: str) -> tuple[str, str]:
         initial_user_message=task,
     )
     result = await agent.run(ctx)
-    text = result.get("summary", str(result)) if isinstance(result, dict) else str(result)
+    if isinstance(result, dict):
+        text = (
+            result.get("summary")
+            or result.get("final_boxed_answer")
+            or result.get("exceed_max_turn_summary")
+            or next(
+                (m["content"] for m in reversed(result.get("message_history") or []) if m.get("role") == "assistant"),
+                None,
+            )
+            or str(result)
+        )
+    else:
+        text = str(result)
     return text, cfg["llm"]["model_name"]
 
 

shared/schema.ts

@@ -7525,3 +7525,67 @@ export type ArcadiaSkill = typeof arcadiaSkills.$inferSelect;
 export type InsertArcadiaSkill = z.infer<typeof insertArcadiaSkillSchema>;
 export type SkillExecution = typeof skillExecutions.$inferSelect;
 export type InsertSkillExecution = z.infer<typeof insertSkillExecutionSchema>;
+
+// ============================================================
+// OpenClaw — Fase 4: Detecção de Padrões e Emergência de Skills
+// ============================================================
+
+export const detectedPatterns = pgTable("detected_patterns", {
+  id: uuid("id").primaryKey().defaultRandom(),
+  tenantId: integer("tenant_id").references(() => tenants.id),
+  userId: varchar("user_id").references(() => users.id),
+
+  // Padrão detectado
+  actionType: varchar("action_type", { length: 100 }).notNull(),
+  description: text("description"),
+  frequency: integer("frequency").notNull().default(0),
+  confidence: numeric("confidence", { precision: 4, scale: 3 }).notNull().default("0"),
+
+  // Janela de análise
+  firstSeenAt: timestamp("first_seen_at").default(sql`CURRENT_TIMESTAMP`).notNull(),
+  lastSeenAt: timestamp("last_seen_at").default(sql`CURRENT_TIMESTAMP`).notNull(),
+
+  // Metadados do padrão (contexto, módulos envolvidos, etc.)
+  metadata: jsonb("metadata").default({}),
+
+  // Estado do ciclo de vida
+  status: varchar("status", { length: 20 }).notNull().default("active"), // active | archived | converted
+
+  createdAt: timestamp("created_at").default(sql`CURRENT_TIMESTAMP`).notNull(),
+  updatedAt: timestamp("updated_at").default(sql`CURRENT_TIMESTAMP`).notNull(),
+});
+
+export const skillSuggestions = pgTable("skill_suggestions", {
+  id: uuid("id").primaryKey().defaultRandom(),
+  patternId: uuid("pattern_id").references(() => detectedPatterns.id, { onDelete: "cascade" }),
+  tenantId: integer("tenant_id").references(() => tenants.id),
+  userId: varchar("user_id").references(() => users.id),
+
+  // Skill sugerida
+  suggestedSkillName: varchar("suggested_skill_name", { length: 200 }).notNull(),
+  suggestedDescription: text("suggested_description"),
+  estimatedAutomation: text("estimated_automation"),
+  confidence: numeric("confidence", { precision: 4, scale: 3 }).notNull().default("0"),
+
+  // Skill gerada (após confirmação)
+  generatedSkillId: uuid("generated_skill_id").references(() => arcadiaSkills.id),
+
+  // Estado da sugestão
+  status: varchar("status", { length: 20 }).notNull().default("pending"), // pending | accepted | rejected | expired
+
+  // Rastreabilidade
+  source: varchar("source", { length: 50 }).notNull().default("openclaw"),
+  reviewedBy: varchar("reviewed_by").references(() => users.id),
+  reviewedAt: timestamp("reviewed_at"),
+
+  createdAt: timestamp("created_at").default(sql`CURRENT_TIMESTAMP`).notNull(),
+  updatedAt: timestamp("updated_at").default(sql`CURRENT_TIMESTAMP`).notNull(),
+});
+
+export const insertDetectedPatternSchema = createInsertSchema(detectedPatterns).omit({ id: true, createdAt: true, updatedAt: true });
+export const insertSkillSuggestionSchema = createInsertSchema(skillSuggestions).omit({ id: true, createdAt: true, updatedAt: true });
+
+export type DetectedPattern = typeof detectedPatterns.$inferSelect;
+export type InsertDetectedPattern = z.infer<typeof insertDetectedPatternSchema>;
+export type SkillSuggestionRecord = typeof skillSuggestions.$inferSelect;
+export type InsertSkillSuggestion = z.infer<typeof insertSkillSuggestionSchema>;