What causes a non-determinism error in Output SDK?

These errors occur when a workflow produces different results during a replay than it did during the original execution, often due to using Math.random(), Date.now(), or dynamic imports inside the workflow function.

How do I generate a random ID safely inside a workflow?

You should never generate random values directly in the workflow function. Instead, generate the ID inside a 'step' function (which records its result) or pass it as an input to the workflow.

Why must workflows be deterministic?

Output SDK (based on Temporal) replays workflow history to recover state after a crash. If the code path changes during replay because of a random value or external state, the system cannot reliably reconstruct the workflow.

Is it safe to use environment variables in workflows?

No, environment variables can change between different workers or across restarts. It is best practice to pass configuration values as workflow input to ensure consistency across all replays.

Output SDK Determinism Fixer

Name: Output SDK Determinism Fixer
Author: growthxai

bygrowthxai

0•

API 开发

Resolves non-determinism errors and replay failures in Output SDK workflows to ensure reliable execution.

This skill provides specialized guidance for diagnosing and fixing non-determinism errors within Output SDK workflows, which are built on Temporal. It helps developers identify and refactor prohibited operations—such as direct calls to random number generators, system timestamps, and environment variables—that break workflow replay mechanisms. By migrating these side effects into activity steps or workflow inputs, the skill ensures that workflows remain deterministic, stable, and capable of recovering state after system failures or restarts.

主要功能

01Offers grep-based commands to scan repositories for determinism violations

020 GitHub stars

03Explains verification steps to ensure consistent workflow replay

04Provides refactoring patterns for deterministic UUID generation

05Identifies non-deterministic patterns like Math.random() and Date.now()

06Guides migration of side effects from workflow logic to steps

使用场景

01Refactoring workflow code to use Temporal-safe time and ID generation

02Fixing 'non-deterministic' error messages during workflow recovery

03Debugging inconsistent workflow results between initial runs and retries

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add growthxai/output-claude-plugins output-error-nondeterminism

For use in Claude.ai and ChatGPT

Download Skill