Promptfoo Evaluation

What it does

This skill integrates Promptfoo to automatically evaluate AI code generation outputs against defined test cases. It validates responses for correctness, security, and adherence to specific constraints before deployment.

When to use it

• Before committing code generated by an agent to ensure it passes functional requirements.
• To detect hallucinations or logic errors in complex algorithms immediately after generation.
• When enforcing strict security policies, such as preventing hardcoded secrets or SQL injection patterns.
• During iterative development cycles to maintain consistent quality standards across multiple agent runs.

Key capabilities

• Executes automated test suites against AI-generated code snippets.
• Provides pass/fail metrics based on custom evaluation criteria.
• Supports various assertion types including unit tests, regex matching, and semantic similarity.
• Generates detailed reports highlighting specific failures in the output.

Example prompts

"Generate a Python function to sort a list of integers and run it through promptfoo-evaluation with standard sorting test cases." "Evaluate this JavaScript API response handler for security vulnerabilities using promptfoo-evaluation before I integrate it into production." "Create a SQL query to fetch user data and validate the output against a set of schema constraints using promptfoo-evaluation."

Tips & gotchas

Ensure your test cases cover edge scenarios, as generic prompts may yield false positives. The skill requires a pre-configured Promptfoo environment with valid test definitions to function correctly.