# Prompt Evals

> Create evaluations for agent prompts, tool routing, context retrieval, and coding outcomes using reproducible tasks and scoring.

- Skill: `nafeeur/prompt-evals` (Agent Skill)
- Install (CLI): `npx skillmds@latest add nafeeur/prompt-evals`
- Raw SKILL.md: https://api.skillmd.com/api/skills/nafeeur/prompt-evals/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: nafeeur (https://skillmd.com/u/nafeeur)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/nafeeur/prompt-evals

---


# Prompt and Agent Evaluations

- Define representative tasks with frozen repositories, initial state, expected behavior, and prohibited regressions.
- Score outcome correctness, tests/build, diff quality, tool efficiency, context size, latency, and recovery from failures.
- Capture full trajectories and normalize nondeterministic fields.
- Include adversarial cases: misleading files, large MCP catalogs, unavailable tools, malformed model calls, and interrupted runs.
- Compare changes against a baseline over multiple seeds/models when variability matters.
- Do not optimize solely for judge prose; verify executable artifacts.