@autobe/agent/src/orchestrate/prisma/orchestratePrismaCorrect.ts

AI backend server code generator

import { IAgenticaController, MicroAgentica } from "@agentica/core";
import {
  AutoBePrisma,
  IAutoBeCompiler,
  IAutoBePrismaValidation,
} from "@autobe/interface";
import { ILlmApplication, ILlmSchema } from "@samchon/openapi";
import { IPointer } from "tstl";
import typia from "typia";

import { AutoBeContext } from "../../context/AutoBeContext";
import { assertSchemaModel } from "../../context/assertSchemaModel";
import { enforceToolCall } from "../../utils/enforceToolCall";
import { transformPrismaCorrectHistories } from "./transformPrismaCorrectHistories";

export function orchestratePrismaCorrect<Model extends ILlmSchema.Model>(
  ctx: AutoBeContext<Model>,
  application: AutoBePrisma.IApplication,
  life: number = 4,
): Promise<IAutoBePrismaValidation> {
  const unique: Set<string> = new Set();
  for (const file of application.files)
    file.models = file.models.filter((model) => {
      if (unique.has(model.name)) return false;
      unique.add(model.name);
      return true;
    });
  application.files = application.files.filter((f) => f.models.length !== 0);
  return step(ctx, application, life);
}

async function step<Model extends ILlmSchema.Model>(
  ctx: AutoBeContext<Model>,
  application: AutoBePrisma.IApplication,
  life: number,
): Promise<IAutoBePrismaValidation> {
  const compiler: IAutoBeCompiler = await ctx.compiler();
  const result: IAutoBePrismaValidation =
    await compiler.prisma.validate(application);
  if (result.success)
    return result; // SUCCESS
  else if (life <= 0) return result; // FAILURE

  // VALIDATION FAILED
  const schemas: Record<string, string> = await compiler.prisma.write(
    application,
    "postgres",
  );
  ctx.dispatch({
    type: "prismaValidate",
    result,
    schemas,
    compiled: await compiler.prisma.compile({
      files: schemas,
    }),
    step: ctx.state().analyze?.step ?? 0,
    created_at: new Date().toISOString(),
  });

  const pointer: IPointer<IModifyPrismaSchemaFilesProps | null> = {
    value: null,
  };
  const agentica: MicroAgentica<Model> = new MicroAgentica({
    model: ctx.model,
    vendor: ctx.vendor,
    config: {
      ...(ctx.config ?? {}),
      executor: {
        describe: null,
      },
    },
    histories: transformPrismaCorrectHistories(result),
    controllers: [
      createApplication({
        model: ctx.model,
        build: (next) => {
          pointer.value = next;
        },
      }),
    ],
  });
  enforceToolCall(agentica);

  // REQUEST CORRECTION
  await agentica
    .conversate(
      "Resolve the compilation errors in the provided Prisma schema files.",
    )
    .finally(() => {
      const tokenUsage = agentica.getTokenUsage();
      ctx.usage().record(tokenUsage, ["prisma"]);
    });
  if (pointer.value === null) {
    console.error(
      "Unreachable error: PrismaCompilerAgent.pointer.value is null",
    );
    return result; // unreachable
  }

  const correction: AutoBePrisma.IApplication = {
    files: application.files.map((file) => ({
      filename: file.filename,
      namespace: file.namespace,
      models: file.models.map((model) => {
        const newbie = pointer.value!.models.find((m) => m.name === model.name);
        return newbie ?? model;
      }),
    })),
  };
  ctx.dispatch({
    type: "prismaCorrect",
    failure: result,
    correction,
    planning: pointer.value.planning,
    step: ctx.state().analyze?.step ?? 0,
    created_at: new Date().toISOString(),
  });
  return step(
    ctx,
    {
      files: correction.files,
    },
    life - 1,
  );
}

function createApplication<Model extends ILlmSchema.Model>(props: {
  model: Model;
  build: (next: IModifyPrismaSchemaFilesProps) => void;
}): IAgenticaController.IClass<Model> {
  assertSchemaModel(props.model);
  const application: ILlmApplication<Model> = collection[
    props.model
  ] as unknown as ILlmApplication<Model>;
  return {
    protocol: "class",
    name: "Prisma Compiler",
    application,
    execute: {
      correctPrismaSchemaFiles: (next) => {
        props.build(next);
      },
    } satisfies IApplication,
  };
}

const claude = typia.llm.application<
  IApplication,
  "claude",
  { reference: true }
>();
const collection = {
  chatgpt: typia.llm.application<
    IApplication,
    "chatgpt",
    { reference: true }
  >(),
  claude,
  llama: claude,
  deepseek: claude,
  "3.1": claude,
};

interface IApplication {
  /**
   * Fixes validation errors in specific AutoBePrisma models while preserving
   * ALL existing business logic and model descriptions.
   *
   * ## Core Rules
   *
   * 1. Fix ONLY validation errors in provided models - never remove business
   *    descriptions
   * 2. Apply minimal changes to error models only - preserve original design
   *    intent
   * 3. Return ONLY corrected models - unchanged models remain in original schema
   * 4. Maintain referential integrity with unchanged models
   *
   * ## Targeted Scope
   *
   * - Process ONLY models with validation errors from IError[] array
   * - Exclude models without errors from processing and output
   * - Minimize context usage by returning corrected models only
   * - Preserve unchanged models in their original state
   *
   * ## Fix Strategy
   *
   * - Resolve validation errors within specific models only
   * - Fix field duplications, invalid references, and type mismatches
   * - Update cross-model references without modifying target models
   * - Ensure naming conventions and index rules compliance in corrected models
   */
  correctPrismaSchemaFiles(props: IModifyPrismaSchemaFilesProps): void;
}

interface IModifyPrismaSchemaFilesProps {
  /**
   * Detailed execution plan for fixing `AutoBePrisma` validation errors in
   * specific models.
   *
   * 🎯 Purpose: Enable systematic reasoning and step-by-step error resolution
   * approach for targeted model validation issues
   *
   * 📋 Required Planning Content:
   *
   * 1. **Error Scope Analysis**
   *
   *    - List all validation errors from IAutoBePrismaValidation.IError[] array
   *    - Extract unique table names from errors to identify affected models
   *    - Categorize errors by type (field duplications, references, types, indexes)
   *    - Identify which models need correction vs. which remain unchanged
   * 2. **Targeted Fix Strategy**
   *
   *    - Focus ONLY on models mentioned in validation errors
   *    - Outline minimal changes needed for each affected model
   *    - Plan cross-model reference updates (if any) without modifying non-error
   *         models
   *    - Ensure unchanged models maintain valid references to corrected models
   * 3. **Model-Specific Fix Plan**
   *
   *    - Model-by-model modification plan for ONLY affected models
   *    - Exact field additions, renames, or type corrections required
   *    - Reference updates within corrected models only
   *    - Index corrections limited to affected models
   * 4. **Minimal Scope Validation**
   *
   *    - Confirm which models will be included in output (error models only)
   *    - List models that will remain unchanged in original schema
   *    - Identify cross-model dependencies without including unchanged models
   *    - Preserve all business logic within corrected models
   * 5. **Targeted Impact Assessment**
   *
   *    - Potential effects of fixes on unchanged models (reference validation)
   *    - Verification points for corrected models only
   *    - Ensure no new validation errors in targeted models
   *    - Confirm minimal output scope compliance
   *
   * 💡 Example Planning Structure:
   *
   *     ## Error Scope
   *     - Target Models: shopping_customers, shopping_orders (2 models only)
   *     - Unchanged Models: All others remain in original schema
   *
   *     ## Targeted Fixes
   *     - shopping_customers: Remove duplicate 'email' field
   *     - shopping_orders: Update targetModel reference to 'shopping_customers'
   *
   *     ## Output Scope
   *     - Return: Only shopping_customers and shopping_orders models
   *     - Preserve: All other models unchanged in original schema
   *
   *     ## Cross-Model Impact
   *     - Verify: shopping_orders still references shopping_customers correctly
   *     - No changes needed in other models referencing these
   */
  planning: string;

  /**
   * ONLY the specific models that contain validation errors and need
   * correction.
   *
   * 📥 Input Structure:
   *
   * - Contains ONLY models mentioned in IAutoBePrismaValidation.IError[] array
   * - Each model has specific validation errors that need targeted correction
   * - Models not mentioned in errors are excluded from this input
   * - Represents minimal scope for error correction
   *
   * 🔍 Expected Validation Issues (Model-Specific):
   *
   * - Duplicate field/relation names within these specific models
   * - Invalid foreign key references from these models to other models
   * - Single foreign key fields in index arrays within these models
   * - Invalid naming conventions within these specific models
   * - Type validation errors in fields of these models
   *
   * 📝 Model Content Analysis (Targeted Scope):
   *
   * - Complete field definitions for each error model only
   * - Relationships from these models (may reference unchanged models)
   * - Indexes within these models that need correction
   * - Business descriptions specific to these models
   * - Cross-model references that need validation (read-only for targets)
   *
   * ⚠️ Processing Notes (Focused Approach):
   *
   * - Input contains ONLY models with validation errors
   * - May reference other models not included in this input
   * - Cross-model references must be validated but target models won't be
   *   modified
   * - Output should return corrected versions of ONLY these input models
   * - All business logic and descriptions within these models must be preserved
   * - Corrections must not break references from unchanged models
   *
   * 🎯 Correction Scope:
   *
   * - Fix validation errors within these specific models
   * - Update internal model structure (fields, relations, indexes)
   * - Correct references to external models (without modifying targets)
   * - Maintain compatibility with unchanged models in the full schema
   * - Return corrected versions of ONLY these models
   */
  models: AutoBePrisma.IModel[];
}