Reprompting

---
title: Reprompting
sidebar_position: 1
---

# Reprompting

What happens when an LLM returns malformed JSON or misses a required field? Without intervention, your agentic workflow fails. But often, the model just needs a second chance with clearer guidance.

Reprompting is Agent Actions' automatic retry system for validation errors. When an action's output fails validation, reprompting retries with the error context included in the prompt—giving the model specific feedback on what went wrong and how to fix it.

## Overview

The reprompting system provides:

- **Automatic retries** - Retry failed validations up to a configurable limit
- **Self-reflection** - Optionally instruct the model to analyze its failure before retrying
- **Configurable exhaustion** - Control what happens when all attempts fail

:::info Retry vs Reprompt
**Retry** handles transient errors (rate limits, network issues) — same request, wait, retry.
**Reprompt** handles validation errors (bad JSON, schema violations) — modify prompt with feedback, retry.

When both are configured, retry runs inside each reprompt attempt. If retry exhausts during a reprompt cycle, the `on_exhausted` policy is respected. API-failed records that reach validation are rejected and reprompted, not silently graduated.

See [Retry & Error Handling](../execution/retry.md) for transient error handling.
:::

## Configuration