How to use Replit Agent to debug a game alpha
Replit Agent helped debug a Phaser escape-game alpha, while Zhou supplied art, logs and review; a broken cleanup required Git rollback and manual repair.
Original prompt
Shared by the source author. See source notes for context.
Don’t code; just walk me through the logic
Reach people exploring AI agents
See placements →What happened
A working first desktop alpha level emerged, with corrected rendering and image state but a failed agent cleanup requiring human rollback.
Reported by H.J. Zhou — SubstackHow the agent did it
- Zhou constrained the game stack, supplied artwork and broke the specification into smaller tasks.
- Replit’s tools debugged high-density rendering and a lingering item image using observed behavior and logs.
- After a log-cleanup change broke the game, Zhou restored a working commit and completed that cleanup manually.
Where a person stepped in
Zhou supplied assets, reviewed code differences, reproduced bugs and gathered player feedback; they used Git rollback and manual cleanup when the agent’s change failed.
H.J. Zhou built the first level of The Train West, a point-and-click escape game, with Replit’s agent tools. They supplied hand-drawn assets and chose a limited Phaser, HTML, CSS and JavaScript scope with JSON configuration. The account follows actual debugging, including changes that made the game worse.
A large handoff specification created mismatched assumptions. Zhou moved toward smaller implementation steps and conversations about the intended logic before coding. Runtime observations proved more useful than asking repeatedly for a fix: blurry rendering involved double scaling on a high-density screen, while a lingering item image required clearing state when its event ended.
Not every cleanup succeeded. An agent request to remove console logs broke the game. Zhou restored a working Git revision and removed relevant lines manually. The author also reviewed diffs and gathered player feedback rather than treating generated code as final.
The result was a working desktop-first alpha level, with human artwork, scope choices and substantial testing effort. This is a mixed implementation account: useful agent repairs alongside a failed cleanup and human rollback. It does not show an entire finished commercial game.
Prompt
Original prompt
Source: Substack · H.J. Zhou
Frequently asked questions
Was the game entirely generated?
No. Zhou supplied artwork, constraints, review and player testing.
Did every agent fix work?
No. Removing console logs broke the game and required rollback.
What development scope was completed?
The account describes the first desktop-focused alpha level.
More details
About this account
Zhou constrained the game stack, supplied artwork and broke the specification into smaller tasks. Replit’s tools debugged high-density rendering and a lingering item image using observed behavior and logs.
Opened original dated development diary and checked primary publication metadata. The prompt excerpt is a planning instruction, not the full game brief. Human rollback and manual repair are retained; claimed time and cost are not extrapolated. Original prompt excerpt.
Credit: H.J. Zhou — Substack
Source published 7 Oct 2025
Source checked 14 Oct 2026
Tools and topics
Replit Agent · Phaser · JavaScript · Git
point-and-click · incremental debugging · rollback
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.