empirical-prompt-tuning - Iteratively evaluate and improve agent prompts
Runs unbiased subagents through scenarios, evaluates outputs with self-reports and metrics, and iteratively improves prompts until progress plateaus.
Tags
Updated: 2026-09-30Capabilities
What this skill does
- Check description-body alignment
- Prepare evaluation scenarios
- Define requirement checklists
- Dispatch fresh subagents
- Run prompt scenarios
- Evaluate outputs and metrics
- Apply minimal prompt revisions
- Check convergence and overfitting
Inputs
- Target prompt text
- Evaluation scenarios
- Requirement checklists
- Subagent dispatch access
Outputs
- Iteration evaluation reports
- Prompt revision proposals
- Convergence decisions
- Overfitting findings
Requirements
- Task tool access
- Ability to dispatch fresh subagents
- Agent usage metadata
- A scenario-capable execution environment