Build Log · 2026-07-25

The Trials of Automating Game Analysis

A peek into my attempts to automate the analysis of game mechanics for a new mobile puzzle game, blending lessons from Royal Match.

The Idea Behind the Game

I recently decided to design a new mobile game that combines elements from two existing games, one of which is Royal Match. I've enjoyed playing Royal Match casually, but the thought of restarting my progress to analyze every game mechanic felt daunting. I knew I needed a different approach.

Enter HERMES Agent

I thought about using HERMES Agent to automate the tedious task of documenting every little detail in Royal Match. I set it up in a virtual machine and created an Android emulator with Royal Match installed. At first, the plan was to let HERMES play through 100 levels and report back with all the necessary information. However, I quickly realized this was not going to be as simple as I had hoped.

The Frustrating First Steps

The first hurdle was getting HERMES to perform a single swipe in the game. I can’t tell you how many times I tried to make it work, but it was like teaching a cat to fetch. It just wouldn’t cooperate. After many frustrating attempts, I decided to download another puzzle game that only required tap actions instead of swipes. This seemed like a good compromise, and eventually, HERMES managed to execute a tap motion.

The Capture Conundrum

As I continued, I discovered that HERMES was taking screenshots but the image analyzer was reducing the screenshot size. This caused it to return incorrect screen coordinates, making everything even more complicated. I spent a good chunk of time trying to fix this issue, ensuring that the conversion was accurate.

A Slow March to Progress

Every action took an eternity. I was looking at 1-2 minutes per step, which was painfully slow. If I wanted it to complete a single level of Royal Match, I calculated it would probably take about an hour. At that point, I had to face the music.

Putting It All Aside

After wrestling with HERMES and realizing that automating this process was turning into a lesson in frustration, I decided to put the project aside for now. Sometimes, you have to recognize when a tool isn't right for the job. I opted to go back to doing everything manually, taking notes the old-fashioned way. It may be slower, but at least I won’t have to deal with the headaches of automation gone wrong.

Lessons Learned

Through this experience, I learned that while automation can be a powerful ally, it is not always the solution. Sometimes, the best way forward is to roll up your sleeves and dive into the work yourself. Who knows, maybe there will be a better way next time around.