Find and Fix Errors in Agent Conversation Logs
Review 50 conversations, identify five incorrect responses, group them by cause, and suggest specific system-prompt rules to fix them.
Prompt template
- Format
- Text Parameterized
- Inputs
- Agent_conversations (text, document) · Required
- Outputs
- Conversation_error_analysis (text)
Here are 50 conversations from my agent: ${agent-conversations}. Find 5 instances where it gave the wrong answer and group them by cause: missing information, an unclear rule, or incorrect tone. For each cause, provide a specific rule to add to the system prompt. Give only the fixes, with no praise.
Replace the ${…} placeholders with your values
before using this prompt. See parameters
and defaults below
Parameters
Replace each placeholder with your own value, or use its default when one is provided
${agent-conversations}
agent-conversations
Required
What you'll create
A conversation audit classifies wrong answers by knowledge gaps, unclear rules or tone. The output focuses on corrective rules rather than praise.
Frequently asked questions
Does every error need an unrelated new policy?
The requested rules should address the grouped causes of the actual failures.
Comments
Tried this prompt? Share what you made, compare results, or ask a question