Prompt Eval Loop
NewIterates prompts with a small eval set, assertions and a regression table, not one-off gut edits.
Prompt Eval Loop is an Agent skill for Development. Iterates prompts with a small eval set, assertions and a regression table, not one-off gut edits. Tags: 提示词 · 评测 · eval · LLM.
What is Prompt Eval Loop
Prompt Eval Loop packages a Development workflow for agents. Iterates prompts with a small eval set, assertions and a regression table, not one-off gut edits.
Install Prompt Eval Loop in Cursor / Claude / Codex
Claude Code Install with the command on this page, or place the skill in ~/.claude/skills or project .claude/skills so Claude can detect Prompt Eval Loop.
Cursor After the install command writes the skill directory, invoke Prompt Eval Loop by name in the Cursor agent.
Codex Install into the Codex skills directory and call it by name. If only a generic command is listed, run it locally and confirm Codex can see the skill.
This page has no install command. Check the project home for how to install Prompt Eval Loop.
Please visit the Project Home to see installation and usage instructions.
How to Use
Claude Code: Place the skill in ~/.claude/skills (personal) or the project .claude/skills directory, and Claude will auto-detect it.
Cursor / Windsurf: After installing via npx skillsadd, the skill directory is placed in the configured path.
Codex CLI: After installation it loads automatically from the skills directory and is invoked by name.
Tags
Similar recommendations
Web Artifacts Builder
Anthropic's official skill that turns natural-language descriptions into complete interactive web apps — single-file HTML/CSS/JS artifacts for rapid prototyping and demos.
Systematic Debugging
系统化调试方法论技能，引导 Agent 按复现、定位、假设、验证、修复的闭环流程排查 Bug，避免盲目改代码。
Test-Driven Development
测试驱动开发（TDD）工作流技能，遵循红-绿-重构循环，先写失败测试再实现功能，保障代码质量与可维护性。
Code Review Workflow
结构化代码审查工作流，从命名、架构、安全、性能、可读性等维度逐项检查，并输出规范的评审意见与改进建议。
MCP Builder
A Model Context Protocol server-building skill that guides agents to quickly stand up MCP-compliant servers with tools, resources and prompts.
Skill Creator
A skill creator that helps developers author SKILL.md metadata, organize reference scripts and assets, and publish packages agents can load.
FAQ
What is Prompt Eval Loop for?
Prompt Eval Loop is a Development skill. Iterates prompts with a small eval set, assertions and a regression table, not one-off gut edits.
Can Claude, Cursor, and Codex all use Prompt Eval Loop?
Yes if the client supports Agent Skills and has loaded the skill directory. Check the project home for compatibility notes.
What similar skills exist?
Related items come from the same Development group so you can pick a closer workflow.