Pdd Eval.prompt

---
agent: agent
description: "Run prompt evaluations, track pass/fail results, and level up eval maturity"
---

# Run Prompt Evaluations

This is the Copilot adapter for the shared `Eval` workflow in `core/workflows/eval.md`. Keep shared workflow behavior aligned there; this file exists to preserve Copilot-specific frontmatter, `#file:` references, and `/pdd-*` command wording.

## Purpose

Track prompt quality over time with explicit criteria instead of relying on memory or one successful run.

An eval turns "this felt good once" into something measurable. It gives the team a repeatable signal for prompt reliability and helps catch drift early.

## Use When

- A prompt matters enough to measure repeatedly.
- The team wants a baseline or regression signal.
- A once-good prompt has started to drift.
- The team wants confidence before reusing or promoting a prompt.

## Inputs

- prompt under test
- expected outputs or acceptance criteria
- baseline artifacts when available

## Step 1: Identify What To Evaluate

If the user names a prompt, evaluate that one.

Otherwise, prefer prompts that are:

- high-impact
- used often
- not yet covered by an eval
- suspected of drifting

## Step 2: Check For An Existing Eval

Look in `pdd/evals/` for a matching eval file, usually named `<prompt-name>-eval.md`.

If none exists, create a Level 1 manual checklist like: