The problem

Prompts often drift from their intended behavior once they hit production environments. Without a structured audit, identifying why an agent failed or where the prompt logic is leaking remains guesswork.

What it does

• Evaluates real conversation logs against skill definitions and expected behavior specs.
• Generates an evidence-labeled scorecard with coverage, accuracy, and alignment metrics.
• Diagnoses root causes for failures like prompt confusion, parameter errors, or logic gaps.
• Proposes a minimal repair set to fix high-severity issues without bloated scope creep.
• Builds a regression suite to ensure specific failures do not reappear after updates.

Why this beats prompting it yourself

Manual review is subjective and inconsistent. This skill enforces a rigorous, evidence-based framework that quotes your own documentation against real outputs, ensuring your release decisions are based on data rather than vibes.

Use cases

• Perform a pre-release quality check on a new AI agent to determine if it is production-ready.
• Debug recurring hallucinations by identifying specific input patterns that trigger failures.
• Audit third-party prompt templates against your internal success criteria and safety thresholds.
• Generate technical debt reports for existing agents to prioritize prompt engineering tasks.

Known limitations

Requires structured input including the skill definition and at least three real user message samples to provide a meaningful audit.