- Home
- Skills
- Code Quality & Review
- Regression Review — Evidence-Based Git Diff Review
Regression Review — Evidence-Based Git Diff Review
Review Git diffs for breaking contracts and regressions. Trace affected callers and get prioritized, evidence-backed findings without changing your repository.
$15
Works with the AI Agents you already use
Regression Review — Evidence-Based Git Diff Review
Example session with this skill installed
Use Regression Review to review the committed change in this synthetic Python repository. Compare the baseline with the feature commit, trace unchanged callers at the feature commit, and report confirmed regressions. Exclude staged and unstaged local edits. Do not change files or execute application code.
Baseline money.py:
def format_total(cents):
return f"${cents / 100:.2f}"
Feature-commit money.py:
def format_total(cents):
return {"formatted": f"${cents / 100:.2f}"}
Unchanged receipt.py at the feature commit:
from money import format_total
def render(cents):
return format_total(cents).upper()
- Read your context and instructions
- Compiled the regression review
Regression Review — local behavioral demonstration
Evaluator: the preparing Codex task (not an independent model or benchmark).
Input: synthetic Python repository and the packaged workflow.
Requested mode: committed branch comparison. Local edits are excluded.
Base/merge-base: 05ca3cbb09a030d11761492611d41d7edaa5b4ea
Head: bdec2fc9766a5b5dd69aa565a732da3164d411d1
Scope
The committed diff changes money.py. The unchanged caller receipt.py was read at the same head commit. The fixture also contains local edits and untracked files, explicitly excluded by this comparison. No application code was run.
Finding
P1 — Preserve the formatter return contract (money.py:2).
Calling render(1250) reaches receipt.py:4, which invokes .upper() on format_total(cents). Previously the formatter returned a string; the reviewed head returns a dictionary. This produces an AttributeError on the receipt path. Preserve the string return contract or migrate the caller in the same change. Propose a receipt-rendering regression test covering the existing public behavior.
Evidence
Base: return f"$..." (string).
Head: return {"formatted": f"$..."} (dictionary).
Caller at head: return format_total(cents).upper().
The staged local fix does not remove the committed defect: this report reviews head, not the index or working tree.
Validation and limitations
Git diff and snapshot-aware source reads confirmed the source-level path. Nine Git procedure scenarios passed in git-scope-results.json. Application execution/tests were not performed. No GitHub review, independent model evaluation, Agensi delivery, or broader product compatibility test was performed. The same task authored the synthetic case, so this demonstrates procedure behavior, not unbiased detection accuracy.
Connects securely to your tools. The creator never sees your data.
About this skill
Catch compatibility mistakes before a commit or pull request. Regression Review guides your coding agent through staged changes, local edits, branch changes, or explicit commits, then traces changed behavior to the callers and data it can affect.
Get a review that identifies its exact comparison, explains each confirmed defect with a triggering scenario, and separates unresolved risks from demonstrated regressions. It handles staged/unstaged differences, untracked source files, missing base history, and snapshot changes so the report stays tied to the code actually reviewed.
Includes a focused SKILL.md, conditional regression checks, a calibrated example, source attribution, and the MIT license. No bundled executable, GitHub login, second model, API key, or package installation is required. Your agent needs repository read/search tools and local Git, or you can provide a patch and relevant source files for a limited review. This is a static review workflow; it does not execute project code, implement fixes, post reviews, or merge changes.
Created from my public claude-workflows project: https://github.com/lhermoso/claude-workflows . The original MIT-licensed review-changes command is freely available there. This edition adds explicit review-scope selection, snapshot-specific context, evidence-calibrated priorities, and focused review references. You are paying for the adapted standalone workflow and its supporting guidance.
AI review can miss defects. The report states coverage and validation limits and does not replace tests or release checks.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Code Quality & Review.
Static AI-assisted review can miss defects. Findings depend on the selected diff, available callers and repository context. Missing history, unavailable external consumers, generated/binary files and supplied-patch-only reviews reduce coverage and are disclosed. The skill does not execute application tests or prove release readiness. The local fixture validation was performed by the preparing agent, not an independent evaluator. Agensi delivery and cross-client behavior have not been tested.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 26 days ago
- Software
Git
Permissions
File Scopes
Reads relevant source, tests, manifests and Git metadata within the user-selected repository. The broad repository scope supports tracing callers across languages and directories; it is not permission to inspect the rest of the filesystem. Excludes credentials, private keys, .env contents, unrelated personal data and ignored files unless safe relevant material is specifically supplied. Uses local Git read operations; does not edit files, run project code, publish reviews or contact external services.