Review Changes Skill

This skill is designed to identify dangerous code changes that could violate the application's business logic. It is especially useful after automated test fixes by an AI agent.

When to use

• •After test fixes by an AI agent
• •Before committing changes
• •When code reviewing large diffs

Instructions

Step 1: Get changes

Run the command to get unstaged changes:

git diff

Or for staged changes:

git diff --staged

Step 2: Analyze changes

Switch to the role of Senior Code Reviewer / Safety Gatekeeper.

Analyze the obtained diff according to the following criteria:

Critical Checks (CRITICAL RISK)

1. •

   Relaxing Constraints

   • •Removed checks, validations, or if conditions
   • •Removed throw statements for errors
   • •Expanded allowable input parameter values
2. •

   Logic Alteration

   • •Changed control flow
   • •Modified state transition logic
   • •Changed side effects
3. •

   Contract Breaking

   • •Changed function signatures
   • •Changed return types
   • •Broken API backward compatibility
4. •

   Hardcoding

   • •Added values specific only to test data
   • •Replaced dynamic values with constants
5. •

   Cheating Fixes

   • •Changed return value to match specific test expectations
   • •Removed validation that prevents the test from passing

Exceptions from check

• •Files in test/, spec/, __tests__/, e2e/ directories — these are test files
• •Test configuration changes (jest.config, wdio.conf, etc.)

Allowed changes (OK)

• •Fixing off-by-one errors
• •Adding null/undefined checks
• •Fixing typos
• •Refactoring without changing behavior

Step 3: Report Generation (Structured & Concise)

Your goal is to provide a report that can be read in 10 seconds. Use tables and Emoji Scores.

1. Overall Score (Score Meter)

At the very beginning, display the overall status and safety "traffic light":

• •🟢 SAFETY SCORE: 10/10 (All changes are safe)
• •🟡 SAFETY SCORE: 6/10 (There are suspicious areas)
• •🔴 SAFETY SCORE: 0/10 (Critical vulnerabilities detected)

2. Detailed Changes Table

Group changes into a table. No fluff, just facts.

3. Recommendations (Action Items)

Brief list of actions (if there are issues):

• • 🔴 Restore check in auth.ts (line 45)
• • 🟡 Run tests for utils.js

Step 4: Final Verdict

Finish the report with one of the status commands:

• •> ✅ STATUS: SAFE (Ready to merge)
• •> ⚠️ STATUS: WARN (Requires careful review)
• •> 🛑 STATUS: BLOCK (Unacceptable changes, fix needed)

Usage Example

User writes:

/review-changes

or

Check unstaged changes for dangerous logic changes

The Agent must:

1. •Run git diff
2. •Analyze changes according to the criteria above
3. •Generate a report with found issues
4. •Output the final status

Key Self-Check Question

"Does this change fix a legitimate bug, or does it redefine how the feature works?"

If the answer is "redefines how the feature works" — it is a BLOCK.