AgentSkillsCN

agent-change-walkthrough

根据 Anthropic 宪法原则评估插件命令、智能体与技能。在基于推理与基于规则的指令谱上打分，识别改进机会。适用于用户想要审计、打分或审查命令、智能体或技能的质量时使用——例如“审查这个技能”、“审计这个命令”、“这个智能体写得好吗”、“按宪法原则打分”、“检查是否有基于规则的模式”。

SKILL.md
--- frontmatter
name: agent-change-walkthrough
description: Generates a narrative walkthrough of AI-authored code changes. Use after implementation to explain what changed, why, and how it behaves.
tools: Read, Grep, Glob, Bash
model: sonnet

Agent Change Walkthrough Skill

Purpose

Generate a single-story walkthrough of AI-authored code changes explaining implementation from trigger to final behavior.

Core Method

Follow six steps:

  1. •Capture intent — Restate the change in plain language with scope and non-goals
  2. •Build evidence — Collect git diffs using git status, git diff, git show
  3. •Build story stack — Order steps dependency-first (contracts/types before usage, definitions before invocations)
  4. •Write narrative — Each step: clear title, CHANGED/UNCHANGED marker, filename with line number, code snippet, prose explanation
  5. •Integrate analysis — Add trade-offs, alternatives, performance notes, and risks inline at relevant steps
  6. •Close out — Summarize what changed, why behavior differs, what to monitor

Key Principles

  • •Show dependency order first, then runtime flow
  • •No forward references (define before using)
  • •Use mini-diff snippets for changes
  • •Include sanitized example input/output for data-shape changes
  • •Embed analysis naturally in prose, not rigid templates
  • •Never include conversation process as walkthrough steps
  • •No credentials, keys, or sensitive data in output

Output Structure

code
# Implementation Walkthrough
[Setup paragraph with intent + scope]

## Step 1 — [behavior description] [UNCHANGED CONTEXT | CHANGED]
Filename: `path/to/file.ext:line`
[code snippet]
[prose explanation]

## Step N — ...

## Final Outcome
[summary of changes and next validation steps]

Trace from runtime trigger to observable behavior as one coherent story.