skills.rest
whichguy
whichguyCOMMUNITY
Explore all skills by this author →
GitHub

improve-system-prompt

Benchmark system prompt variants for Google Sheets AI assistants via GAS pipelines.

Updated Jul 27, 2026
Search Tags:#testing#prompt engineering#optimization#benchmarking#llm evaluation#gas
One-click install
npx skills add https://github.com/whichguy/skill-craft --skill improve-system-prompt
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: improve-system-prompt
Source: https://github.com/whichguy/skill-craft/tree/main/plugins/improve-system-prompt/skills/improve-system-prompt
Command: npx skills add https://github.com/whichguy/skill-craft --skill improve-system-prompt

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill solves the ambiguity of prompt engineering by providing a rigorous, data-driven framework to test how different system prompt variants and placement strategies impact AI performance in Google Apps Script (GAS) environments.

Core Features & Use Cases

  • Automated Benchmarking: Runs test scenarios through the real GAS-side ClaudeConversation pipeline to measure effectiveness.
  • Dual Evaluation: Combines heuristic scoring (8-dimension rubric) with LLM-as-judge analysis to provide a comprehensive quality score.
  • Use Case: When you have multiple versions of a system prompt (e.g., V2 vs V2a) and need to determine which one yields better accuracy, safety, and conciseness without manual testing.

Quick Start

Run the improve-system-prompt skill to benchmark your current system prompt variants against the default test scenarios.

Frequently Asked Questions about improve-system-prompt

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I benchmark system prompt variants for a Google Apps Script AI assistant?▼

To benchmark system prompt variants for a Google Apps Script AI assistant, execute test scenarios through a live GAS-side pipeline that measures effectiveness using an 8-dimension heuristic rubric and LLM-as-judge analysis.

What is the best way to compare system prompt versions for LLM accuracy and conciseness?▼

The best way to compare system prompt versions is using a dual evaluation framework that combines an 8-dimension heuristic scoring rubric with LLM-as-judge analysis, providing a comprehensive quality score without manual testing.

Does prompt benchmarking require integration with specific Google Sheets modules?▼

Yes, prompt benchmarking requires integration with sheets-chat and chat-core modules to perform inference and execute validation against test scenarios within the real GAS-side ClaudeConversation pipeline.

How does the heuristic scoring mechanism evaluate prompt optimization performance?▼

The heuristic scoring mechanism evaluates prompt optimization performance using an 8-dimension rubric to measure effectiveness across different prompt content variants and placement strategies within Google Apps Script environments.

Can I test prompt placement strategies alongside content variants in Google Apps Script?▼

Yes, you can test prompt placement strategies alongside content variants in Google Apps Script by running the benchmarking pipeline against default test scenarios to measure how different placements impact AI performance.

Why should I use an LLM-as-judge approach for prompt evaluation instead of manual testing?▼

Using an LLM-as-judge approach for prompt evaluation removes the ambiguity of manual prompt engineering by providing a rigorous, data-driven framework to test accuracy, safety, and conciseness across multiple system prompt variants.

Related Skills

Explore More Software Engineering Skills→
devCharuzudevCharuzu

google-workspace-cli

Automate Google Workspace administration tasks across Gmail, Drive, Calendar, and more.

Community
Advanced
gooseworks-aigooseworks-ai
1.1k

lead-qualification

Generate qualification prompts and score leads with confidence ratings.

Official
Advanced
whichguywhichguy
2

optimize-system-prompt

Optimize and refine Sheets Chat ClaudeConversation system prompts for GAS-based workflows.

Community
Advanced
Dianshu-LiaoDianshu-Liao
2

llm-evaluation

Implement LLM evaluation with automated metrics and LLM-as-judge patterns.

Community
Advanced