Case study · Wooden Sword Games

Merge, Attack, Win: BugHunter Plays Mystic Duel From the Public Store Build

A red card slides in beside a red hero. They merge. The battle ends on You Win, with +5 trophies.

One recording captures that full PvP battle in Mystic Duel: Card Battler CCG. An AI agent played it on the public store build of a drag-and-drop merge card battler, the kind of game scripted bots struggle with. Wooden Sword Games runs BugHunter by Nodemori on the same build: no SDK, no hooks, no special build.

Studio: Mobile, registered in London
Game: 1v1 merge card battler
Platform: Mobile: iOS and Android
Integration: None

Three Mystic Duel phone screens from one run: the agent dragging a red card next to its red hero, a battle board taking damage, and a You Win victory screen. — BugHunter on Mystic Duel: one drag to merge, then all the way to You Win.

changes to the build: straight from the store

ways to test: explore, play, test cases, replays

screens mapped with no instructions

5 / 5

replays passed, zero AI calls

Why Merge Play Trips Up Scripted Bots

Wooden Sword Games is a mobile game studio registered in London. Its games include Puzzle Wars: Heroes, a match-3 RPG with competitive PvP, and smaller hybrid-casual titles such as Snack Bag and Snail Wars.

In Mystic Duel, three cards sit face up in the centre row and each player has four slots. Same-colour cards of the same rank merge into stronger heroes. Placing a card triggers an automatic attack, and abilities unlock from merge rank 2.

For a player, that is the fun. For a scripted bot, it is a row of traps: drags instead of taps, colour and rank rules on every move, a turn to read, and a lives gate that locks battles for hours after a loss. Beyond combat lie hero upgrades, leagues, Trophy Road, chests and a shop with secure store payment screens. The agent learned merge play from a short action specification, not a script.

Dealt In Early

BugHunter was playing Mystic Duel before there was a contract. It ran its first test cases, won a full battle, and then a scout run cleared the whole tutorial.

Four Mystic Duel screens from one scout run through the tutorial, numbered and labelled Learn to play, First battles, Tutorial won and New hero. — One scout run through the tutorial: the first lesson, the battles, the final win and a new hero unlocked.

The team settled on the public build, and BugHunter installed it straight from the store. The first package followed: two gameplay videos, a generated design document and two specifications.

With that package in hand, the studio described what it wanted next:

›Brief an agent the way it briefs a tester.
›Play chosen situations to check balance.
›Log wins and losses by league.
›See the game on different screens without owning a hundred devices.

Watching thousands of playtest videos was not an option.

Then came a dedicated environment with Mystic Duel installed and its specification preloaded, a live onboarding session with the studio's team, and a test-payment account so purchase flows could be tested without real charges. Today Wooden Sword runs BugHunter on a paid subscription.

“I needed to see it work without integration. BugHunter installed Mystic Duel straight from the store, with no SDK or special build. Before any contract, it was already merging cards and had won a full battle. That made me take it seriously.”
Sergei Orlov, CEO and Founder, Wooden Sword Games

Four Ways the Agent Works the Board

01Let it explore

With no instructions, the agent goes screen by screen. Its findings feed a structured specification and a design document. Point it at a competitor's title and the same mode turns into fast deconstruction.

Four Mystic Duel screens from one scout run, numbered and labelled Heroes, Shop, Reward and League. — Left to explore with no instructions, the agent found heroes, the shop, a chest reward and a league result.

02Let it play

BugHunter reads the cards, drags them into slots, merges same-colour heroes and uses abilities. That is the battle at the top of this page.

BugHunter console, Run Results for a Mystic Duel free-play session: mode, run and game tags, the session video of a PvP battle, and buttons to download the video and console logs. — Every free-play session comes back with its video and console logs, ready to review.

03Write the test like a tester's brief

Each step has an expected result that the agent checks on the live game. When it could not find a button, it stopped and named the missing element instead of guessing.

Start a battle
Open Settings
Exit
Expect You Lose

Run Results panel for a passed test execution: three plain-English steps (start a battle, open settings, exit), each with its expected result, and no bugs. — A plain-English test case, checked step by step on the live game.

04Record once, replay on demand

A passing test, or a playthrough a tester records by hand, becomes a replay that runs with zero AI calls.

Console table titled Runs History with five passed Mystic Duel runs, one per mode: scout, free play, test execution, recorded regression and manual, with duration and findings. — Every run is logged with its mode, status, duration and findings.

What BugHunter Found

Free play is not a demo mode. While the agent fights battles and walks the menus, it watches the screen for anything a player would notice, then a second automatic pass re-checks each flag before it reaches the team. On the live build, here is what held up.

58%

of flags confirmed as bugs on re-check

100%

of confirmed bugs came with video and a screenshot

100%

of confirmed bugs were player-facing polish

57%

text issues

29%

layout issues

14%

render glitches

42%

of raw flags filtered out by the re-check

What the Agent Brought Back

Across BugHunter's runs on Mystic Duel:

Zero integration work.
The public store build, as players get it: no SDK, no hooks, no special build.
Playing before the contract was signed.
Test cases, a won battle and the whole tutorial.
A map of the game.
23 screen states and 29 recorded transitions, from the tutorial to the shop. The specification lists seven game modes and six resources.
Five of five regression replays passed.
Each one or two steps long, each under 45 seconds, with zero AI calls. Three ran in Wooden Sword's own workspace.

Building a card battler or a merge game?

Start with a free pilot on your own game.