@loopstack/delegate-error-example-workflow

Demonstrates how delegate tool-call loops handle different tool failure modes and feed structured error results back to the LLM for recovery.

By using this example you’ll get…

• A multi-turn LLM workflow using LlmGenerateTextTool + LlmDelegateToolCallsTool
• Three failure types in one flow: schema validation error, runtime error, and failing sub-workflow
• Error propagation back into the conversation via LlmUpdateToolResultTool

Installation

npm install @loopstack/delegate-error-example-workflow

This example uses @loopstack/claude-module through @loopstack/llm-provider-module, so configure Claude credentials before running.

Then register the module in your app:

import { StudioApp } from '@loopstack/common';
import { DelegateErrorExampleModule, DelegateErrorWorkflow } from '@loopstack/delegate-error-example-workflow';
 
@StudioApp({
  title: 'Delegate Error Example',
  workflows: [DelegateErrorWorkflow],
})
@Module({
  imports: [DelegateErrorExampleModule],
})
export class MyAppModule {}

How It Works

1. The workflow seeds an instruction asking the model to intentionally trigger tool failures first.
2. The LLM generates tool calls (strict_schema, runtime_error, failing_sub_workflow).
3. Delegate execution runs tools and callback updates merge results into pending delegate state.
4. Completed tool results are posted back as tool_result blocks so the LLM can self-correct.
5. The workflow ends when the model reaches end_turn and stores the final message.

Public API

• DelegateErrorExampleModule
• DelegateErrorWorkflow

Dependencies

• @loopstack/common
• @loopstack/llm-provider-module
• @loopstack/claude-module