flow-convert-activities-to-steps

name	flow-convert-activities-to-steps
description	Convert Flow SDK activities.ts to Output SDK steps.ts. Use when migrating activity functions to step definitions with typed parameters.
allowed-tools	["Bash","Read","Write","Grep","Edit"]

Convert Flow Activities to Output Steps

Overview

This skill guides the conversion of Flow SDK activity functions (activities.ts) to Output SDK step definitions (steps.ts). This is one of the core migration tasks.

When to Use This Skill

During Migration:

• Converting activities.ts file to steps.ts
• Transforming individual activity functions to step definitions
• Setting up typed input/output schemas for steps

Key Differences

Aspect	Flow SDK (activities.ts)	Output SDK (steps.ts)
Definition	Function with direct parameters	step() with inputSchema
Parameters	Individual function arguments	Single typed input object
Return Type	Direct Promise return	outputSchema validation
Imports	Various Flow SDK imports	@outputai/core
LLM Calls	Custom completion functions	generateText() with Output.object()

Conversion Pattern

Flow SDK Activity (Before)

// activities.ts
import { z } from 'zod';
import { completion } from '@flow/sdk';

export async function analyzeDocument(
  documentText: string,
  analysisType: string,
  maxLength?: number
): Promise<AnalysisResult> {
  const prompt = buildPrompt( documentText, analysisType );

  const response = await completion( {
    model: 'gpt-4',
    messages: [ { role: 'user', content: prompt } ],
    maxTokens: maxLength || 2000
  } );

  return parseAnalysisResult( response );
}

Output SDK Step (After)

// steps.ts
import { z, step } from '@outputai/core';
import { generateText, Output } from '@outputai/llm';
import { AnalysisResultSchema, AnalysisResult } from './types.js';

const AnalyzeDocumentInputSchema = z.object( {
  documentText: z.string(),
  analysisType: z.string(),
  maxLength: z.number().optional()
} );

export const analyzeDocument = step( {
  name: 'analyzeDocument',
  inputSchema: AnalyzeDocumentInputSchema,
  outputSchema: AnalysisResultSchema,
  fn: async ( input ) => {
    const { documentText, analysisType, maxLength } = input;

    const { output } = await generateText( {
      prompt: 'analyzeDocument@v1',
      variables: {
        documentText,
        analysisType
      },
      output: Output.object( {
        schema: AnalysisResultSchema
      } )
    } );

    return output;
  }
} );

Step-by-Step Conversion Process

Step 1: Identify All Activities

List all exported functions in activities.ts:

grep -E "^export (async )?function" src/workflows/my-workflow/activities.ts

Step 2: Create Input Schema for Each Activity

For each function, create a Zod schema for its parameters:

// Original function signature
async function processUser( userId: string, options: ProcessOptions ): Promise<Result>

// Convert to input schema
const ProcessUserInputSchema = z.object( {
  userId: z.string(),
  options: ProcessOptionsSchema
} );

Step 3: Create Output Schema (If Needed)

If the function returns structured data, create an output schema:

// types.ts
export const ResultSchema = z.object( {
  success: z.boolean(),
  data: z.any().optional(),
  error: z.string().optional()
} );

export type Result = z.infer<typeof ResultSchema>;

Step 4: Convert Function to Step

Wrap the function body in a step() definition:

export const processUser = step( {
  name: 'processUser',
  inputSchema: ProcessUserInputSchema,
  outputSchema: ResultSchema,
  fn: async ( input ) => {
    const { userId, options } = input;
    // Original function body here
  }
} );

Step 5: Update LLM Calls

Replace Flow SDK completion calls with Output SDK generators:

// Flow SDK
const response = await completion( { model: 'gpt-4', messages: [...] } );

// Output SDK
const { result } = await generateText( {
  prompt: 'myPrompt@v1',
  variables: { ... }
} );

Complete Migration Example

Before: activities.ts (Flow SDK)

import { z } from 'zod';
import { completion } from '@flow/sdk';

const UserSchema = z.object( {
  id: z.string(),
  name: z.string(),
  email: z.string()
} );

type User = z.infer<typeof UserSchema>;

export async function fetchUser( userId: string ): Promise<User> {
  const response = await fetch( `https://api.example.com/users/${userId}` );
  return response.json();
}

export async function generateGreeting(
  user: User,
  style: 'formal' | 'casual'
): Promise<string> {
  const prompt = style === 'formal'
    ? `Write a formal greeting for ${user.name}`
    : `Write a casual greeting for `;

   response =  ( {
    : ,
    : [ { : , : prompt } ]
  } );

   response.;
}

   (): <{ : ; :  }> {
   result =  emailService.( { to, subject, body } );
   { : , : result. };
}

After: steps.ts (Output SDK)

import { z, step } from '@outputai/core';
import { generateText } from '@outputai/llm';
import { UserSchema, User } from './types.js';

// Step 1: Fetch User
const FetchUserInputSchema = z.object( {
  userId: z.string()
} );

export const fetchUser = step( {
  name: 'fetchUser',
  inputSchema: FetchUserInputSchema,
  outputSchema: UserSchema,
  fn: async ( input ) => {
    const { userId } = input;
    const response = await fetch( `https://api.example.com/users/${userId}` );
    return response.json();
  }
} );

// Step 2: Generate Greeting
const GenerateGreetingInputSchema = z.object( {
  user: UserSchema,
  style: z.enum( [ 'formal', 'casual' ] )
} );

export const generateGreeting = step( {
  name: ,
  : ,
  : z.(),
  :  ( input ) => {
     { user, style } = input;

     { result } =  ( {
      : ,
      : {
        : user.,
        style
      }
    } );

     result;
  }
} );


  = z.( {
  : z.(),
  : z.(),
  : z.()
} );

  = z.( {
  : z.(),
  : z.()
} );

  sendEmail = ( {
  : ,
  : ,
  : ,
  :  ( input ) => {
     { to, subject, body } = input;
     result =  emailService.( { to, subject, body } );
     { : , : result. };
  }
} );

After: types.ts (Shared Types)

import { z } from '@outputai/core';

export const UserSchema = z.object( {
  id: z.string(),
  name: z.string(),
  email: z.string()
} );

export type User = z.infer<typeof UserSchema>;

Calling Steps from Workflows

Steps are called with a single input object:

// Flow SDK (direct parameters)
const user = await fetchUser( userId );
const greeting = await generateGreeting( user, 'formal' );

// Output SDK (object parameter)
const user = await fetchUser( { userId } );
const greeting = await generateGreeting( { user, style: 'formal' } );

Common Pitfalls

1. Forgetting to Destructure Input

// WRONG
fn: async ( userId, name ) => { ... }

// CORRECT
fn: async ( input ) => {
  const { userId, name } = input;
  ...
}

2. Missing File Extensions in Imports

// WRONG
import { UserSchema } from './types';

// CORRECT
import { UserSchema } from './types.js';

3. Not Moving Types to types.ts

Keep schemas and types in types.ts for reuse across steps and workflows.

Verification Steps

1. All activities converted to steps
2. Each step has inputSchema defined
3. Imports use @outputai/core for z
4. LLM calls use generateText() with Output.object() for structured output
5. File imports have .js extension

Related Skills

• flow-convert-workflow-definition - Workflow conversion
• flow-convert-prompts-to-files - Prompt file creation
• flow-error-zod-import - Zod import issues
• flow-error-eslint-compliance - Code style compliance