skills.rest
elie222
elie222COMMUNITY
Explore all skills by this author →
Website𝕏@elie2222GitHub

review-ai-prompt-changes

Review prompt and tool-description diffs in a branch for redundancy and eval-shaped guidance.

12.1k|1.5k|Updated Jul 12, 2023
Search Tags:#Prompt Engineering#System Prompts#prompt review#Git diff#tool descriptions#LLM evals#Code review
One-click install
npx skills add https://github.com/elie222/inbox-zero --skill review-ai-prompt-changes
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: review-ai-prompt-changes
Source: https://github.com/elie222/inbox-zero/tree/main/.claude/skills/review-ai-prompt-changes
Command: npx skills add https://github.com/elie222/inbox-zero --skill review-ai-prompt-changes

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Prompt and tool-description changes in AI products often accumulate bloated, repetitive, or eval-driven instructions that hurt real product behavior. This Skill audits prompt-related diffs in a branch or PR so reviewers can judge whether each change belongs in the prompt layer where it was added.

Core Features & Use Cases

  • Prompt Diff Audit: Scans changed files against a base branch (default main...HEAD) and prioritizes system prompts, prompt builders, tool descriptions, and schema field descriptions.
  • Guidance Classification: Buckets findings into behavior changes, product-helpful guidance, heavy or eval-shaped guidance, and repeated instructions across prompt layers.
  • Structured Review Output: Produces Findings (ordered by severity with file and line references), a What Changed summary, and a Net Read with trim targets.
  • Use Case: Before merging a PR that edits system prompts and tool descriptions in apps/web/utils/ai/**, run this review to catch duplicated instructions between the main prompt and tool descriptions and flag wording added only to satisfy evals.

Quick Start

Use the review-ai-prompt-changes skill to audit this branch for prompt, tool-description, and eval-shaped instruction changes.

Frequently Asked Questions about review-ai-prompt-changes

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

FAQPage Schema
How do I review AI prompt changes in a pull request?▼

Run this Skill against the branch, and it diffs against main by default, narrows to prompt-relevant files like system prompts and tool descriptions, then outputs findings ordered by severity with file and line references plus a summary of what changed.

What counts as eval-shaped prompt guidance?▼

Eval-shaped guidance is cross-cutting language added mainly to defend a known eval failure mode, such as repeated instructions like treat as evidence or do not invent causes across multiple prompt layers, or wording that reads like judging criteria instead of operating guidance.

Can I review prompt diffs against a branch other than main?▼

Yes, the Skill defaults to main...HEAD but accepts a different base branch when the user names one. It then runs git diff --name-only against that base to find changed files.

Does this Skill edit prompts automatically?▼

No, it is an audit and review tool only. It does not edit prompts by default and will only make changes if the user explicitly asks for them.

When should repeated instructions between system prompt and tool descriptions be flagged?▼

Repetition should be flagged when the same instruction appears in both layers without adding meaning. Tool-local guidance belongs in the tool description, while cross-cutting behavior rules belong in the main prompt.

Related Skills

Explore More Software Engineering Skills→
g2i-aig2i-ai
51

pr-description

Generate PR descriptions from git diffs and commits against a base branch.

Official
Intermediate
can1357can1357
28.8k

tool-prompt-optimization

Measures schema-prompt overlap to prune redundant lines from AI agent tool prompts.

Community
Advanced
samymitysamymity
1

eval-harness

Implement eval-driven development workflows for Claude Code with pass@k tracking.

Community
Intermediate
adamatdevopsadamatdevops
1

write-judge-prompt

Generate binary Pass/Fail LLM-as-judge prompts with JSON critique output.

Community
Advanced