skills.rest
ShineBreaker
ShineBreakerCOMMUNITY
Explore all skills by this author →
GitHub

diff-explain-self-review

Walks through every changed file in a git diff to expose silent scope creep.

2|Updated Jan 2, 2026
Search Tags:#code review#scope creep#git diff#self-review#ai coding agent#change verification#diff walkthrough
One-click install
npx skills add https://github.com/ShineBreaker/Guix-configs --skill diff-explain-self-review-shinebreaker
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: diff-explain-self-review
Source: https://github.com/ShineBreaker/Guix-configs/tree/main/dotfiles/mutable/agents/hermes/.local/share/hermes/skills/hermes-agent-ops/diff-explain-self-review
Command: npx skills add https://github.com/ShineBreaker/Guix-configs --skill diff-explain-self-review-shinebreaker

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? After an AI model edits code, it often silently widens scope with unrequested refactors, renames, and cleanups that look reasonable but were never asked for. This Skill forces a post-edit, file-by-file walkthrough of the real git diff so hidden changes surface before the user trusts the result. ## Core Features & Use Cases - Six-Question Diff Walkthrough: Prompts the model to justify every changed file, classify each block as required, taste, or risky, and flag the highest-risk lines for review. - Anchored Reading Rubric: Requires every claim to cite a real file:line reference, cross-checked against the actual diff to catch hallucinated answers. - Drift Detection: Compares claimed files-touched against git diff --name-only output to expose silent scope expansion. - Use Case: After an agent completes a multi-file change touching auth or schema boundaries, run the walkthrough before committing to a shared branch so the user knows exactly which 3-5 lines to review first. ## Quick Start Ask the agent to walk through every file it just changed in the git diff, one by one, explaining why each block exists and which lines are highest-risk to review.

Frequently Asked Questions about diff-explain-self-review

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

FAQPage Schema
How do I review what an AI coding agent actually changed?▼

Capture the real diff with git diff or git diff --cached, then prompt the model to walk through every changed file one by one, justifying each block. Cross-check its claims against git diff --name-only to catch files it failed to mention.

How to detect scope creep in AI-generated code changes?▼

Compare the model's claimed list of touched files against the actual git diff --stat output. Any file present in the diff but missing from the walkthrough is silent scope expansion and should be reverted or discussed before continuing.

When should I run a diff walkthrough after code edits?▼

Run it after non-trivial edits: 3 or more files, more than 50 lines, or changes touching boundary layers like auth, schema, or public APIs. Skip it for single-file typo fixes or one-line config tweaks where review overhead exceeds value.

Does this self-review work without git or version control?▼

No, the walkthrough requires a real, queryable diff from git diff, git diff --cached, or a saved patch file. Fabricating a diff defeats the purpose, so without version control you must perform the equivalent review manually.

Why does the model claim it changed nothing outside the task?▼

That single sentence is the most common self-deception in AI code edits. The Six Questions force enumeration of every file and block, which makes hidden renames, formatting changes, and speculative refactors visible instead of hand-waved.

Related Skills

Explore More Software Engineering Skills→
herren-officialherren-official

b2c-android-create-commit

Generates Git commit messages and commits changes for Android projects using .docs/commit-convention.md.

Official
Intermediate
trevin-jtrevin-j
6

git-basics

Run tests, linters, and static analysis before committing with Conventional Commits.

Community
Intermediate
EarlBearEarlBear

qa-test

Automate functional QA and regression testing for EarlBear wireframes.

Official
Advanced
manuscomanusco

second-opinion

Runs an independent second-model review of a diff reconciled with the primary review.

Community
Basic